(IJCNS) International Journal of Computer and Network Security, 1

Vol. 2, No. 5, May 2010

A Model-driven Approach for Runtime Assurance

of Software Architecture Model
Yujian Fu1, Xudong He2, Sha Li3, Zhijiang Dong4 and Phil Bording5
1
Alabama A&M University, School of Engineering & Technology,
4900 Meridian Street, Normal AL 35762, USA
Yujian.fu@aamu.edu
2
Middle Tennessee State University, College of Basic and Applied Sciences,
1301 East Main Street, Murfreesboro, TN 37132, USA
zdong@mtsu.edu
3
Florida International University, School of Computing and Information Science,
11200 SW 8th Str, Miami, FL 33199, USA
Hex@cs.fiu.edu
4
Alabama A&M University, School of Engineering & Technology,
4900 Meridian Street, Normal AL 35762, USA
Phil.bording@aamu.edu
5
Alabama A&M University, School of Education,
4900 Meridian Street, Normal AL 35762, USA
Sha.li@aamu.edu

notations of UML. As a result, UML has been adapted for

Abstract: Unified Modeling Language (UML) has been widely
accepted for object-oriented system modeling and design, and
has also been adapted for software architecture descriptions in
recent years. Although, the use of UML for software
architecture representation has the obvious benefits of
facilitating learning and comprehension, UML lacks precise
semantics for defining key features of software architecture
level entities. In this paper, we present an approach to map a
UML architecture description into a formal specification model
called SAM, which serves as a semantic domain for defining the
precise semantics of the UML description and supports formal
analysis. Furthermore, when the UML architecture description
contains sufficient details, an implementation from the resulting
SAM architecture description to Java code can be automatically
generated from our existing SAM translator tool. The generated
Java code contains monitoring checkers for run-time
verification. Therefore, our approach provides not only a
method of formal analysis to detect design errors, but also a
technique to automatically generate executable code with run-
time verification capability, which effectively eliminates the
tedious, labor intensive, and error-prone manual coding process.
Keywords: Software architecture, UML, runtime verification,
temporal logic, Petri nets.
1. Introduction
Software architecture is a high level representation of a
software design, which can significantly impact the overall
system development as well as future system maintenance
and evolution. The importance of software architecture
design and description has been widely recognized in the
past decade. As a result, many architecture description
languages (ADLs) have been proposed. Most of these ADLs
are based on some formal methods. Although the formal
basis of an ADL enables some early design level analysis,
the formality and peculiar syntax of an ADL often hinder its
wide acceptance.
In contrast, UML has been widely accepted for object-
oriented system modeling and design. Software architects
and designers are familiar with the main features and
software architecture descriptions in recent years [21, 16,
30]. Defining software architecture using UML has the
following advantages: graphical, extensible, and capable of
representing both structural and behavioral aspects. In
addition, the UML architecture level representation provides
a simple transition or relation to a detailed object-oriented
design also described in UML.
As a general-purpose object-oriented modeling language,
UML does not directly provide constructs needed for
software architecture modeling, such as architectural
configurations, connectors, and styles. Although a UML
component diagram can be used to describe the organization
of a software system in terms of its components and
interconnections at specification level, it is unclear that how
to instantiate component interfaces and dependency
relationships and how to associate interaction protocols with
component dependencies.
Several research works explored the integration of UML
with some existing ADLs [31]. Among these works, three
approaches can be identified. In the first approach, rules are
provided to translate architectural descriptions between a
particular ADL and UML. In the second approach, key
constructs are added to standard UML to represent software
architecture, which often results in a large and complex
language that is hard to understand and use. In the third
approach, UML’s built-in extension mechanisms such as
stereo-types and tagged values are tailored for architecture
description. None of the above approaches address the issues
of code generation from an architecture description and the
verification of generated code.
In this paper, we present an integrated approach that
combines the ideas of the 1st and the 3rd approaches
mentioned above, as well as code generation from an
architecture description. This approach maps a UML
architecture description into a formal software architecture
model called SAM, which serves as a semantic domain for
defining the precise semantics of the UML description and
supports formal analysis. Furthermore, when the UML
2 (IJCNS) International Journal of Computer and Network Security,
architecture description contains sufficient details, an
implementation from the resulting SAM architecture
description to Java code can be automatically generated
from our existing SAM translator tool.
The paper proceeds as follows: Section 2 gives some
background information. Section 3 describes our approach
to mapping UML architecture description to SAM model;
Section 4 gives an example of the approach applied an
embedded system example and evaluates the approach;
Section 5 describes related work and Section 6 concludes.
2. Preliminaries
In this section, we provide a brief introduction to software
architecture documentation, UML, and SAM.
2.1 Software Architecture Viewpoints
Although, there is not a universally accepted way in
documenting software architecture design. The component
and connector (C&C) view, showing the dynamic behavioral
aspect of a software system, proposed in [35] is no doubt an
essential one, which has been the target used in the
development of many software architecture description
languages. The C&C view is also included in the work of
the Software Engineering Institute (SEI) [30], where the
authors presented three view types - module, component and
connector, and allocation view in documenting software
architecture design; and provided guidelines of how to use
UML to document these architecture view types. With
regard to the representation of the C&C view, three
strategies are demonstrated: using component types as
classes, subsystems, or real-time profiles. Since the behavior
of each class and object is described by the state charts
diagram, the behavior of the system represented by the class
diagram can be a group of state charts diagram with
interactions.
2.1 SAM – Software Architecture Model
SAM is an architectural description model based on Petri
nets [29], which are well-suited for modeling distributed
systems. SAM [15] has dual formalisms underlying – Petri
nets and Temporal logic. Petri nets are used to describe
behavioral models of components and connectors while
temporal logic is used to specify system properties of
components and connectors.
SAM (Software Architecture Model) is hierarchically
defined as follows. A set of compositions C = {C1, C2, …,
Ck} represents different design levels or subsystems. A set of
component Cmi and connectors Cni are specified within each
Figure 1. Overall Structure of Our Approach
Vol. 2, No. 5, May 2010
composition Ci as well as a set of composition constraints Csi,
e.g. Ci = {Cmi, Cni, Csi}. In addition, each component or
connector is composed of two elements, a behavioral model
and a property specification, e.g. Cij = (Bij, P ij). Each
behavioral model is described by a Petri net, while a property
specification by a temporal logical formula. The atomic
proposition used in the first order temporal logic formula is
the ports of each component or connector. Thus each
behavioral model can be connected with its property
specification. A component Cmi or a connector Cni can be
refined to a low level composition Cl by a mapping relation h,
e.g. h(Cmi ) or h(Cmi ) = Cl. SAM is suitable to describe large
scale systems’ description.
SAM gives the flexibility to choose any variant of Petri nets
and temporal logics to specify behavior and constraints
according to system characteristics. In our case, Predicate
Transition (PrT) net [11] and linear temporal logic (LTL)
are chosen.
3. Our Approach
Our approach takes a software architecture description based
on the C&C view and documented using the UML, and
produces a formal software architecture description in SAM.
More specifically, in the UML software architecture
description:
a) A class diagram is used to model the overall
structure of a software architecture,
b) State chart diagrams are used to define the behavior
of individual components and connectors, and
c) OCL is used to specify architecture level
constraints.
The above UML notations are mapped to SAM entities as
follows:
a) The class diagram is mapped to an overall
hierarchical SAM structure,
b) State chart diagrams are mapped to PrT nets, and
c) OCL expressions are mapped to temporal logic
formulae.
The resulting SAM model can be analyzed with various
existing formal analysis techniques including model
checking.
Furthermore, to improve the productivity and quality, we
provide an automated realization of the resulting SAM
model. We have designed and implemented a tool, named
SAM parser, with the PrT-XML and temporal logic-XML
transformation. Figure 1 shows the overall structure of the
approach.
 (IJCNS) International Journal of Computer and Network Security, 3
b)
3.1 Mapping from State-charts to SAM
Composition
It is straightforward to map each class into a composition
in a SAM model. In this section, we mainly focus on the
mapping from state charts to a SAM composition.
A statechart diagram specified the states that an object
can inhabit in, describes the events triggered flow of
control and actions responded from the system, as the
result of the objects reactions to events from the
environment. The semantics of statecharts is based on the
run-to-completion assumption [1]. A State is an abstract
metaclass that models a situation during which some
(usually implicit) invariant conditions hold. The structure
of a state vertex, defined by the following function
children, incarnates the hierarchy of state machines.
Definition 1 (Children). Let S be a set of state vertices.
The function children: S defines for each state the set of
its direct substates.
c)
We define childreni(s) = {s’|s’ ∈ childreni-1 (s)} for i > 1.
Thus the function children+(s) = ∪i>0childreni(s) defines
for each state a set of its transitively nested substates.
Incoming events are processed one at a time and an event
can only be processed if the processing of the previous
event has been completed – when the statechart
component is in a stable state configuration. In our study,
we considered the events and actions defined by the
methods which are specified in the corresponding classes.
Thus we can obtain more detailed implementation
information. To map the state chart diagram to SAM
composition, we present a translation algorithm (fT) in the
following steps.
a) Each super state S, where children(S) ≠ Φ, is
mapped into a composition C in SAM model, i.e.,
fT(S) = Ci , where children(S) ≠ Φ.
Figure 2. Translated SAM Model of States Fork and Join
Vol. 2, No. 5, May 2010
Each substate s ∈ S, where children(S) = Φ, is
mapped into a component Cm ∈ C. We have fT(S)
= Cm, where children(S) = Φ, Cm ∈ C, and Cm =
< Bm, P m> . Let S be < En,E x,Do,A,G> , where En,
E x, Do, and A represent entry, exit, do and other
regular actions taken inside the states
respectively, G denotes the guard of any actions.
b.1 The behavior model of the component (Bm) is
described by following mapping:
fT(S) = <<pt_in, ten>, <pt_out, tex>, ti, gt >, where
pt_in, and pt_out denote input and output port
respectively, ten and tex denotes transitions that
are corresponding to entry and exit action
respectively, ti denotes transitions that
corresponding to the do action and other
internal actions, gt denotes the internal
transition’s guard function.
b.2 The property specified (P m) by OCL can be
transformed into first order (or propositional)
logic and temporal logic.
Pseudo states include initial, History, join, fork,
junction and choice. We use some special tokens
to denote and symbolize the pseudo states.
• Initial state is mapped to a token < “init” > in
an input port of the component as initial
marking.
• History state is mapped to some token in a
place. The sort of the token would be
<History, OtherTokens >.
• Fork and junction states are special states that
represent the splitting of a single flow of
control and the synchronization of concurrent
flows of control. Generally there is no label
for the input and output transitions. In the
component we use a special token < “Fork” >
or < “Join” > to represent them. Fig. 2
illustrates the translation of these two special
states.
4 (IJCNS) International Journal of Computer and Network Security,
d) A transition tS inside the super state S is mapped
into a connector Cn ∈ C. Let S be < E,A,Valr ,G> .
In this case, we have:
fT (< E,A,Valr ,G> ) = <<pt_in, t>, <pt_out, t>, tk,
gt_internal>, where E denotes trigger event or
action, A denotes action, pt_in, and pt_out denote
input and output port respectively, t denotes
transition, tk denotes tokens in the ports, Valr
denotes returned value of the action A, and G
denotes guard of the label, gt_internal denotes the
internal transition’s guard function.
e) Generally, we can merge the place holding the
return value for entry action with the place
holding the parameters for the do action, the
holding the return value for do action with the
place holding the parameters for the exit action.
From above translation algorithm (fT), we can realize a
mapping relation between each item in the state-chart
diagram and the SAM architecture model. This
translation algorithm (fT) is meaningful for the formal
verification of UML architecture document.
3.2 Mapping from OCL to First Order Temporal
Logic
In the translation from OCL to first order and temporal
logic, it is worth to note that following features of OCL
restrict the translation to temporal logic.
1. There is not timing concept in the OCL
specification.
2. There is not object concept in the temporal logic
specification.
3. OCL is an implementation-based and oriented
specification language, while temporal logic is a
high level specification language. Thus there is not
type system, data structure, condition and flow
control sentence in temporal logic.
4. Finally, OCL is 3-value function logic (true, false,
and undefined,) while temporal logic is a 2-value
logic.
Considering the above features and difference between
OCL and temporal logic and first order logic, we decide
to map OCL expression to the first order logic with the
following concerns:
• Context: It indicates the scope of the entity. It can be
translated to the quantifiers ∀ or ∃. The difference is
that the scope of the quantifiers in the OCL context
generally is finite.
• Navigation: Navigation through association either
results in a Set(Object) or in a Seq(Object) if
association is marked as ordered. We can use a
quantifier ∃ on Set(Object), Seq(Object) or Collection,
since the later two also have the set concept. In
addition, select and reject used in the Collection can
be mapped to the quantifiers ∀ and ∃, since they
restrict the scope of instances.
• forall and exists: are obviously translated into
quantifiers ∀ and ∃.
• allinstance: It semantically matches ∃.
• operations, attributes: All predefined boolean
operations can be mapped into first order logic
operations. All mathematic operations can be used
Vol. 2, No. 5, May 2010
directly. Some other predefined operations, such as
toUpper(), Concat() have no images in the logic
domain. Attributes can be (atomic) predicates in the
first order or propositional logic.
• invariants: This means that the constraints are always
holding in the context, thus it is a context-specific
safety property:  invariants
• pre- and post-condition: can be mapped to the
following temporal formula precondition → ⋄
postcondition. In the case that it is an invariants, we
have  (precondition → ⋄ postcondition)
• let-in clause: let clause defines some variables used in
the in clause. In the in clause, you can use any
defined syntax in the OCL.
• if-then-else clause: Let use format if cond then r1 else
r2, we have the logic formula as follows
cond → r1 ∨ r2.
• previous value @pre: The postfix @pre refers to a
value of a property at the start of an operation in a
postcondition. In this case we have to record the
previous value and then invoke the value through
some method since there is not a corresponding
mapping in the logic.
3.3 Tool Support – SAM Translator
Runtime verification, composed of a software module and
an observer, monitors the execution of a program, and
checks its conformity with a requirement specification,
often written in a temporal logic. Runtime verification
can be applied to evaluate automatically test runs, either
on-line or o_-line, analyzing stored execution traces; or it
can be used on-line during operation, potentially steering
the application back to a safety region if a property is
violated. It is highly scalable. Several runtime verification
systems have been developed, such as JPaX [14], Java-
MAC [17], etc..
We have developed a runtime monitoring system, SAM
Parser, on Software Architecture Model (SAM) [15]
whose behavior is modeled by Petri Nets and properties
by temporal logic. To validate the translated state charts
model, we adapted it to SAM parser to reuse that tool. We
introduce the SAM Parser simply introduced in the
following.
Input: The runtime checker has to have two input data:
transformed SAM model and temporal logic. These data
are specified in the PNML and XML format.
Code Generation: The code of SAM model and temporal
logic and first order logic formulae are automatically
generated from the input files. In our work, we construct
a class as a child of templates for each net, place,
transition, arc, inscription, initial marking, and guard.
The reason for this is to make it easier to understand and
maintain. For example, the user can provide a more
efficient way to check the enableness of a transition and
the way to fire it by replacing methods of corresponding
classes without any side effects on other transitions. The
execution of generated code is non-deterministic, i.e. we
choose an enabled transition and a valid assignment
randomly to fire. It is hard to generate code automatically
given a Petri net due to the complexity of sorts, guard
conditions of transition and arc labels [25]. If some
restrictions[10] are satisfied, we can achieve it.
 (IJCNS) International Journal of Computer and Network Security, 5
Temporal logic and first order logic formulae are
transformed into automata by a logic engine Maude ([5]).
These translated automata will feed into our runtime
checker generator to produce monitors for different
formulae.
Runtime Checker Generation: Runtime checkers are
generated by breaking the temporal logic formula into
subformulae and creating a matrix for the formula [33].
In order to generate monitoring codes for properties
(linear temporal formulae), a logic server, Maude [5] in
our case, is necessary. Maude, acting as the main
algorithm generator in the framework, constructs an
e_cient dynamic programming algorithm (i.e. monitoring
code) from any LTL formula [33]. The generated
algorithm can check if the corresponding LTL formula is
satisfied over an event trace.
4. An Application of the MDA Approach
We present an instance of our approach in a statechart
diagram, which is one of the state machine in the UML
models, through a case study. The case study deals with a
simplified cruise control system adapted from [13]. In
this section we will first introduce the cruise control
system and a state chart diagram for the cruise controller.
Then we present some properties of this example. Finally,
the runtime verification results are discussed.
4.1 Cruise Control System and UML Documentation
The purpose of a cruise control system is to accurately
maintain the driver’s desired set speed, without
4.2 Experiment Results and Discussion
The time of generated code with monitors is 4.3s. We
checked 5 properties covering 5 components and 4
connectors.
Since the events and actions are defined by the methods
which are specified in the corresponding classes, each
fired transition represents a method is operated under
some guard condition. The properties specified in the
OCL expressions for the state chart diagram is mapped to
temporal formulae and further used to generate monitors.
5. Related Work
Our MDA approach and the presented framework integrate
two aspects: software architecture with UML design
notation and its extension. Moreover, our approach also
provided a runtime validation and verification technique by
using SAM Parser. The related works are discussed in the
following.
Formal Modeling and Analysis of Architecture
Descriptions in UML: Our work has been influenced by a
large body of research and practical experience. In the
interest of brevity, we only compare it to the most relevant
approaches. The work reported in this paper relates to works
that focus on specifying structural and possibly behavioral
aspects of a software system using UML. The representative
UML architecture description examples are provided by
Kruchten [21], Hofmeister et al. [16] and Clements et al.
[31]. Although several researchers have explored the formal
Vol. 2, No. 5, May 2010
intervention from the driver, by actuating the throttle-
accelerator pedal linkage. A modern automotive cruise
control is a control loop that takes over control of the
throttle, which is normally controlled by the driver with
the gas pedal, and holds the vehicle speed at a set value.
We assume an automatic transmission vehicle. When
turned on by the driver, a cruise-control system (CCS)
automatically maintains the speed of a car over varying
terrain. The CCS can be turned on by pressing Start
button, and enabled by pressing SetSpeed button. Resume
button will enable the CCS at the last maintained speed
when the brake is released. The cruise control function is
disabled when the brake or accelerator pedal is pressed.
Pressed once Resume button can increase the speed with
1mph and the SetSpeed button can decrease the speed
with 1mph when the cruise control function is enabled.
The cruise control system should be automatically
disabled when the speed is below 25mph and above
90mph. For the space limit, we cannot show the UML
diagram and generated code. Each place or port must
carry its state information which is also ignored in the
tables. Finally, we have to point out that the guard
function for a transition has to be added some more
restrictions for the evaluations. For instance, if there is a
token “void” in the input place of a transition, it means
the previous action does not have return value, we have to
justify that field to evaluate the firing condition of the
transition. This means that automatically mapping a
guard condition in a state chart diagram to a guard
function is not sufficient in some cases.
All properties are true if the conditions and guard are
satisfied. We also check some conditions that is not
suitable for the method, such as different parameters
feeding for the method that makes the guard is not
satisfied, in that case the formula is evaluated as false.
The Results are consistent with what we expected from
the state chart diagram. Finally, we also find a mapping
mistake in the component Accelerating and component
Decelerating when we check properties relative to them.
We modified the mapped SAM composition according to
the checking results.
analysis of generic UML object-oriented designs [7, 34, 12,
24, 4, 26, 35], their results may not be ready applicable to
UML architecture level designs.
Code Generation from the UML Description and
Verification in the Implementation:
There is a significant amount of research that considers
mappings from UML to other (mostly formal) modeling
techniques to validate UML models (e.g. using B [22], CSP
[8], SPIN [23], PVS [2], Petri Nets [27, 38], Z/Eves [3] and
the work with Object-Z [19, 20, 18, 39]). These works focus
on the mapping from UML diagrams to formal methods for
model checking or some specification language for the
model checkers. Moreover, less of them address the code
generation and verification in the implementation level.
Property Specification – OCL Expression and Temporal
Logic: Various methodologies proposed to deal with the
property specification of object-oriented systems. There are
two main streams to cooperate OCL with temporal logics,
one is extending OCL with temporal notations ([31, 9, 37]
6 (IJCNS) International Journal of Computer and Network Security,
etc.), another is add object concepts into temporal logics
([6]). The work in both streams increases the complexity of
the extended language and obstacles of the usage.
6. Conclusion and Future Works
In this paper, we have presented an integrated approach for
transferring an architecture description represented in UML
to a formal architecture model represented in SAM, which
not only supports design level analysis but also automated
code generation with run-time verification capability. The
specific details of the approach, outlined in the algorithms
in Section 3, are likely to evolve as our research on the
relationship between UML and software architectures
deepens; however, we believe that the approach is flexible
and general enough to accommodate needed new changes.
As pointed out by Medvidovic in the work [27] ensuring
system properties at the level of architecture is of little value
unless it can also be ensured that those properties will be
preserved in the resulting implementation. This reflects the
importance of the code generation and runtime verification
of system properties in the implementation. Our automated
code generation and run-time verification approach nicely
addresses the above research issue.
Acknowledgements We appreciate for all reviewers to read
this paper. This work is supported by Title III under grant
PO31B085057-08.
References
[1] Uml 2.0 specification. http://www.omg.org/
technology/documents/formal/uml.htm.
[2] Enhancing Structured Review with Model-Based
Verification. IEEE Transaction on Software
Engineering, 30(11):736–753, 2004. Member-Issa
Traore and Member-Demissie B. Aredo.
[3] N. Am´alio, S. Stepney, and F. Polack. Formal proof
from uml models. In ICFEM’04, volume 3308 of
Lecture Notes in Computer Science, pages 418–433,
2004.
[4] D. B. Aredo. Semantics of UML statecharts in PVS. In
Proceeding of 12th Nordic Workshop on Programming
Theory, Bergen, Norway, 2000.
[5] M. Clavel, F. J. Dur´an, S. Eker, P. Lincoln, N. Mart´ı-
Oliet, J. Meseguer, and J. F. Quesada. Maude:
Specification and Programming in Rewriting Logic.
http://maude.csl.sri.com/papers, March 1999.
[6] D. Distefano, J.-P. Katoen, and A. Rensink. On a
Temporal Logic for Object-Based Systems. In S. F.
Smith and C. L. Talcott, editors, Formal Methods for
Open Object-Based Distributed Systems IV - Proc.
FMOODS’2000, Stanford, California, USA, September
2000. Kluwer Academic Publishers.
[7] Z. Dong and X. He. Integrating UML State-chart and
Collaboration Diagrams Using Hierarchical Predicate
Transition Nets. In GI Lecture Notes in Informatics,
2001.
[8] G. Engels, R. Heckel, and J. M. K¨uster. Rule-based
specification of behavioral consistency based on the
UML meta-model. volume 2185, pages 272–284, 2001.
Vol. 2, No. 5, May 2010
[9] S. Flake and W. Mueller. An OCL extension for real-
time constraints. In Object Modeling with the OCL,
pages 150–171, 2002.
[10] Y. Fu, Z. Dong, and X. He. A Methodology of
Automated Realization of a Software Architecture
Design. In Proceedings of The Seventeenth
International Conference on Software Engineering and
Knowledge Engineering (SEKE2005), 2005.
[11] H. J. Genrich. Predicate/Transition Nets. Lecture Notes
in Computer Science, 254, 1987.
[12] M. Gogolla and F. P. Presicce. State Diagrams in UML:
A Formal Semantics using Graph Transformations. In
Proceedings of International Conference of Software
Engineering, Workshop on Precise Semantics of
Modeling Techniques, pages 55–72, 1998.
[13] H. Gomaa. Designing Concurrent, Distributed, and
Real-Time Applications with UML. Addison-Wesley
Professional, 2000.
[14] K. Havelund and G. Rosu. An overview of the runtime
verification tool java pathexplorer. Journal of Formal
Methods in System Design, 2004.
[15] X. He and Y. Deng. A Framework for Specifying and
Verifying Software Architecture Specifications in SAM.
volume 45 of The Computer Journal, pages 111–128,
2002.
[16] C. Hofmeister, R. L. Nord, and D. Soni. Describing
Software Architecture with UML. In Proceedings of the
TC2 1st Working IFIP Conference on Software
Architecture (WICSA1), pages 145 – 160, 1999.
[17] M. Kim, S. Kannan, I. Lee, and O. Sokolsky. Java-
MaC: a Run-time Assurance Tool for Java. In
Proceedings of RV’01: First International Workshop
on Runtime Verification, Paris, France, Electronic
Notes in Theoretical Computer Science. Elsevier
Science, 2001.
[18] S.-K. Kim, D. Burger, and D. Carrington. An mda
approach towards integrating formal and informal
modeling languages. In FM 2005: Formal Methods,
International Symposium of Formal Methods Europe,,
volume 3582 of Lecture Notes in Computer Science,
pages 448–464, 2005.
[19] S.-K. Kim and D. Carrington. Formalizing the UML
Class Diagrams Using Object-Z. In UML’99: The
Unified Modeling Language - Beyond the Standard,
Second International Conference, volume 1723 of
Lecture Notes in Computer Science, 1999.
[20] S.-K. Kim and D. Carrington. A Formal Mapping
between UML Models and Object-Z Specifications.
Lecture Notes in Computer Science, volume 1878,
pages 2–21, 2000.
[21] P. Kruchten. The 4+1 view model of architecture. IEEE
Software, 12(6):42–50, 1995.
[22] K. Lano, D. Clark, and K. Androutsopoulos. UML to B:
Formal Verification of Object-Oriented Models. volume
2999 of Lecture Notes in Computer Science, pages 187–
206,2004.
[23] D. Latella, I. Majzik, and M. Massink. Automatic
Verification of a Behavioural Subset of UML Statechart
Diagrams Using the SPIN Model-checker. Formal
Aspects of Computing, 11(6):637 – 664, 1999.
 (IJCNS) International Journal of Computer and Network Security, 7
[24] D. Latella, I. Majzik, and M. Massink. Towards a
Formal Operational Semantics of UML Statechart
Diagrams. In Proceedings of the 3rd IFIP International
Conference on Formal Methods for Open Object-based
Distributed Systems, pages 331–347, February 1999.
[25] S. W. Lewandowski and X. He. Generating Code for
Hierarchical Predicate Transition Net Based Designs. In
Proceedings of the 12th International Conference on
Software Engineering & Knowledge Engineering, pages
15–22, Chicago, U.S.A., July 2000.
[26] J. Lilius and I. P. Paltor. The Semantics of UML State
Machines. Technical Report 273, Turku Centre for
Computer Science, 1999.
[27] N. Medvidovic, D. S. Rosenblum, and D. F. Redmiles.
Modeling Software Architectures in the Unified
Modeling Language. ACM Transactions on Software
Engineering and Methodology, 11(1):2–57, January
2002.
[28] H. Motameni. Mapping to Convert Activity Diagram in
Fuzzy UML to Fuzzy Petri Net. World Applied Sciences
Journal, 3(3): 514 – 521, 2008. ISBN: 1818-4952.
[29] T. Murata. Petri Nets: Properties, Analysis and
Applications. Proceedings of the IEEE, 77(4):541–580,
1989.
[30] J. S. e. a. Paul Clements, Len Bass. Documenting
Software Architectures: Views and Beyond. Addison-
Wesley, January 2003.
[31] S. Ramakrishnan and J. McGregor. Extending OCL to
Support Temporal Operators. In 21st International
Conference on Software Engineering (ICSE 99),
Workshop on Testing Distributed Component-Based
Systems, May 1999.
[32] J. E. Robbins, N. Medvidovic, D. F. Redmiles, and D.
S. Rosenblum. Integrating architecture description
languages with a standard design method. In ICSE ’98:
Proceedings of the 20th international conference on
Software engineering, pages 209–218, Washington,
DC, USA, 1998. IEEE Computer Society.
[33] G. Rosu and K. Havelund. Rewriting-Based Techniques
for Runtime Verification. Journal of Automated
Software Engineering, 2004.
[34] J. Saldhana and S. M. Shatz. UML Diagrams to Object
Petri Net Models: An Approach for Modeling and
Analysis. In Proceedings of the International
Conference on Software Engineering and Knowledge
Engineering, pages 103–110, 2000.
[35] T. Sch¨afer, A. Knapp, and S. Merz. Model Checking
UML State Machines and Collaborations. Electronic
Notes in Theoretical Computer Science, 55(3):1–13,
2001.
[36] M. Shaw and D. Garlan. Software Architecture:
Perspectives on an Emerging Discipline. Prentice Hall,
1996.
[37] P. Ziemann and M. Gogolla. An Extension of OCL with
Temporal Logic. In Critical Systems Development with
UML – Proceedings of the UML’02 workshop, pages
53–62, TUM, Institut fur Informatik, TUM-I0208,
September 2002.
Vol. 2, No. 5, May 2010
[38] Jiexin Lian, Zhaoxia Hu, Sol M. Shatz. Simulation-
based analysis of UML statechart diagrams: methods
and case studies. Software Quality Journal. 16(1),
March, 2008. ISBN: 0963-9314. Springer Netherlands.
[39] Rafael M. Borges and Alexandre C. Mota. Integrating
UML and Formal Methods. Electronic Notes in
Theoretical Computer Science (ENTCS). Volume 184.
Page 97-112. July 2007. Elsevier Science Publishers.
Authors Profile
Yujian Fu received the B.S. and M.S. degrees in Electrical
Engineering from Tianjin Normal University and Nankai University
in 1992 and 1997, respectively. In 2007, she received her Ph.D.
degree in computer science from Florida International University.
She joined the faculty of Department of Computer Science at the
Alabama A&M University in the same year. Dr. Yujian Fu
conducts research in the software verification, software quality
assurance, runtime verification, and formal methods. Dr. Yujian Fu
also actively serves as reviewers of several top journals and
prestigious conferences. She continuously committed as a member
of IEEE, ACM and ASEE.
Zhijiang Dong received the B.S. and M.S. degrees in Huazhong
Tech University, Ph.D. degree in computer science from Florida
International University. He currently is assistant professor at
Middle Tennessee University. Dr. Dong’s research is mainly in the
software engineering. Dr. Dong also actively serves as reviewers of
several top journals and conferences. He continuously committed as
a member of IEEE, ACM.
Xudong He is a professor of school of computing and
information science and director of center for advanced and
distributed system engineering at Florida International University.
Dr. He’s research are software engineering, formal verification and
specification. Dr. He currently has over one hundred publications in
prestigious journals and conferences.
Sha Li is an associate professor at department of curriculum,
teaching and educational leadership, school of education of
Alabama A&M University. Dr. Sha Li received his doctorial
degree of educational technology from Oklahoma State University,
2001. Sha Li' research interests include distance education,
instructional technology, instructional design and multimedia for
learning.
Phil Bording is an associate professor and chair of department of
computer science at Alabama A&M University. Dr. Phil Bording
received his Ph.D. degree in computer science from University of
Tulsa in 1995 and M.S. degree from University of Alabama at
Huntsville in 1984. Dr. Bording’s research area is parallel
computing.
8 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

Principle Component Analysis

From Multiple Data Representation
Rakesh Kumar yadav1, Abhishek K Mishra 2, Navin Prakash3 and Himanshu Sharma4
1
College of Engineering and Technology, IFTM Campus,
Lodhipur Rajput, Moradabad, UP, INDIA
rakeshoso@yahoo.co.in
2
College of Engineering and Technology, IFTM Campus,
Lodhipur Rajput, Moradabad, UP, INDIA
abhimishra2@gmail.com
3
College of Engineering and Technology, IFTM Campus,
Lodhipur Rajput, Moradabad, UP, INDIA
naveenshran@gmail.com

4
College of Engineering and Technology, IFTM Campus,
Lodhipur Rajput, Moradabad, UP, INDIA
cs_himanshu@rediffmail.com

Abstract: For improving accuracy and increasing efficiency of
classifier, there are available many effective techniques. One of
them combining multiple classifier technique is used for this
purpose. In this paper I present a novel approach to a
combining algorithm designed to improve the accuracy of
principle component classifier. This novel approach combines
multiple PCA classifiers, each of using a subset of feature. In
contrast other combining algorithms usually manipulate the
training pattern.
Keywords: PCA classifier, Eigenface, Eigenvector, Voting,
Covariance matrix, new projected data, face recognition.
1. Introduction
The PCA classifier is one of the oldest and simplest methods
for classification. PCA involves a mathematical course of
action that transform a number of possible correlated
variable into a smaller no of uncorrelated variable called
principle component. PCA is mathematically defined as an
orthogonal linear transform that transform the data to a new
coordinate system such that greatest variance by any
projection of the statistics comes to lie on the first
coordinate. The second greatest variance on the second
coordinate and so on [10].
1.1. Method of Pronouncement PCA.
Step 1: Acquire some data
Step 2: Subtract the mean
Step 3: Compute the covariance matrix
Step4: Compute the eigenvectors and eigenvalues of the
covariance Matrix
Step 5: Choosing components and forming a feature vector
Step 6: Deriving the new data set.
1.2. Related work
Image classification is a thorny task because images are
multidimensional. There are many classifiers although there
are a number of face recognition algorithms which works
well in constrained situation; face recognition is still an
open and very challenging dilemma in real application. The
area of face recognition has focused on detecting individual
feature such as the eyes, nose, mouth, lips and head outline
and defining face model by the position, size, perimeter and
relationship among features. Bledsoe’s [3] and Kanade’s [4]
recognition based on these parameter. A Mathew A. Turk
and Alex P. Pentland [1] track a subject’s and then
recognizes the person by comparing characteristics of the
face to those of known individuals. Face images are
projected onto a feature space that best encodes the variation
among known face images. The face space is defined by the
“eigenface” which are the eigenvectors of the set of faces;
they do not necessarily correspond to isolated feature such as
eyes, ear and nose. The framework provides the ability to
learn to recognize new faces in an unsupervised manner.
In mathematical terms, we want to find the principle
component of the distribution of faces or the eigenvectors of
the covariance matrix of set of face images. These
eigenvectors can be thought of as a set of feature which
together characterizes the variation between face images.
Each image location contributes more or less to each
eigenvector so that we can display the eigenvector as a sort
of ghostly face which we call an eigenfaces.
There are many approaches available to combining the
classifier such as Stephen O.Bay, “Nearest Neighbor
Classification from Multiple Feature Subset” [7] and Bing-
Yu-Sun- Xiao-Ming Zhang and Rujing Wans, “Training
SVMs for Multiple Feature Classifier Problem” [5] but there
in no any approach available to combining the base
classifiers. Actually the combination of classifier can be
implemented at two levels, feature level and decision level.
Xiaoguang Lu, Yunhong Wang, Anil K. Jain, “combining
classifier for face recognition”[4] provide frameworks to
combine different Classifier on feature based .We are using
feature level combination and want to present combination
of classifier which works on low level feature in this
approach.
 (IJCNS) International Journal of Computer and Network Security, 9
Vol. 2, No. 5, May 2010

2. Classification amalgamation 8. Store training feature vector and projected data in a

matrix.
2.1. Training Phase 9. Read test set
In the training phase we extract feature vector for each
subset of images in training data set. Let T1 be a training
subset in which every image has pixel resolution of M by N
(M row, N column). In order to extract PCA features of T1
we will first convert image into a pixel vector by
concatenating each of M rows into single vector V1. The
length of pixel vector V1 will be M*N. we will use the PCA
algorithm as a dimensionality reduction technique which
transform the vector V1 to a vector W1. Which has
dimensionality d where d<=M*N. For each training image
subset Ti, we will calculate and store these feature vector
Wi.

2.2 Recognition Phase

In the recognition phase, we will be given a test set of

images Tj of known person. As in the training phase we will
compute the feature vector of this test set using PCA and
match the identity name of persons. In order to identity we
will compute the similarities between test set and training
subset. For this we will use the combining algorithm
method.

2.3. Combining algorithm. Figure 1

The algorithm for PCA classification from multiple feature
subsets is simple and can be treated as:
Using voting, combining the output from multiple PCA
classifiers each having access each subset of feature.
We select the subset of features by sampling from original
set of features we use two different sampling functions:
sampling with replacement and sampling without
replacement. In sampling with replacement a feature can be
selected more than once which we treat as increasing its
weight.
Each of the PCA classifiers uses the same number of
features. This is parameter of algorithm which we set by
cross confirmation performance estimates on the training
set.
The similarity among new projected data can be calculated
using euclidean distance the identity of most similar Wi will
be output of our classifier. If i=j it means that we have
correctly identified the subset otherwise we have
misclassified the subset. The schematic diagram of this
presented in figure 1.
2.4. Pseudo code of combining algorithm
1. Prepare training set which contains subset of
images
2. Set PCA dimensionality parameter
3. Read training subset
4. Form training data matrix.
5. Form training class label matrix
6. Calculate PCA transform Matrix
7. Calculate feature vector , projected data of all
subset of training set by using PCA transformation
matrix
10. Calculate the feature vector of test set and also
calculate new projected data buy using PCA
transformation matrix.
11. Compute the similarity between subset and test set
by using voting method.
12. store the similarity
13. Now we go the subset test set which have lowest
similarity factors determine the person id and
match the id between training subset and test set.
14. Initialize error count to zero
15. For each test face If the id is match then
recognition done and accuracy will be hundred
percent.
16. If the id of any test face image is not equal to id of
training subset of each image then increment in
error count.
17. Compute the accuracy by using error count value.
3. Experiment Discussion
To access the possibility of this approach to face
recognition, we have performed experiments .we used the
ORL database [9] which available in public domain. We
also used image processing tool of MATLAB 7.6. Using
ORL database we have conducted 3 experiments to access
the accuracy of recognition by formula:
(1-(error count/total test image count))*100
In the experiments we take 5 training subset of face image.
Each subset contains 20 face images of 4 different person
and one test set which contains 5 images of a person. In the
first experiment we set up 60 percent accuracy. In the
10 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

second experiment we set up 80 percent accuracy and at last Abhishek K Mishra received the M.Tech Degree in computer
in third experiment we found 60 % accuracy. Technology and Application from school of Information
As can seen my result are no where near perfect yet with Technology, UTD, RGPV, Bhopal, India.. Currently Working as
my best outcomes in with almost 80 percent matches. Sr. lecturer in college of Engineering and Technology, IFTM
Moradabad. INDIA.
If we study this approach in the view of complexity
then definitely I assured both time and space complexity will Navin prakash received the B Tech Degree in Computer Science
be reduced because we are using multiple feature subset from BIET Jhansi in 2001, India. Currently Working as Sr. lecturer
concepts as like in nearest neighbor classifier from multiple in college of Engineering and Technology, IFTM Moradabad.
feature subsets [7]. The nearest neighbor classifier from INDIA.
multiple feature subsets supports lower complexity.
Himanshu Sharma received the B.Tech Degree in computer
Science and Information technology from MIT, Moradabad, India.
Currently Working as Sr. lecturer in college of Engineering and
4. Conclusion and future work Technology, IFTM Moradabad, INDIA.
As can be seen my results are near to perfect while these
results may be improve. Conceptually this approach reduced
the complexity also but this approach has two limitations
one voting can only improve accuracy if the classifier select
the correct class more often than only other class. Another
algorithm has two parameter value need to be set first is size
of features subset and second is number of classifier. So in
the near future we can also improve the accuracy and
complexity of this approach.

5. References
[1] M. Turk and A. Pentland,”Face recognition using
eigenface”, In Computer Vision and Pattern
Recognition, 1991.
[2] W.W.Bledsoe, “the model method in facial recognition”,
panoramic research inc. Palo alto.ca roe. Pr: 15, Aug
1966
[3] T. Kanade, “picture processing system by computer
complex and recognition of human faces”, dept pf
information science, Kyoto university, Nov 1973.
[4] Xiaoguang Lu, Yunhong Wang, Anil K. Jain,
“combining classifier for face recognition”.
[5] Bing-Yu-Sun- Xiao-Ming Zhang and Ru-jing Wang,
“Training SVMs for Multiple feature classifier
problem”.
[6] Julein Meynet Vlad Popovici, Matteo, Sorci and jean
Philippe Thiran, “combining SVMs for face class
modeling”.
[7] Stephen D.Bay,” Nearest neighbor classification from
multiple feature subset”, November 15, 1998.
[8] Linday I smith, “A tutorial on principle component
analysis”, February 26, 2002
[9] http://homepages.cae.wiseedu/~ece533/
[10]http://en.wikipedia.org/wiki/Principal_component_anal
ysis

Authors Profile
Rakesh Kumar Yadav received the B.Tech Degree in
Information Technology from A K G Engineering College
Ghaziabad, INDIA in 2004. Currently pursuing M.Tech and
Working as Sr. Lecturer in college of Engineering and Technology,
IFTM Moradabad. INDIA.
 (IJCNS) International Journal of Computer and Network Security, 11
Vol. 2, No. 5, May 2010

Performance Evaluation of MPEG-4 Video

Transmission over IEEE802.11e
Abdirisaq Mohammed Jama1, Sinzobakwira Issa2 and Othman O. Khalifa3
1
International Islamic University Malaysia,
Department of Electrical & Computer Engineering,
P.O.Box 10, 50728, Kuala Lumpur, Malaysia
wacays91@hotmail.com
2
International Islamic University Malaysia,
Department of Electrical & Computer Engineering,
P.O.Box 10, 50728, Kuala Lumpur, Malaysia
issa10issa@gmail.com
3
International Islamic University Malaysia,
Department of Electrical & Computer Engineering,
P.O.Box 10, 50728, Kuala Lumpur, Malaysia
khalifa@iiu.edu.my

the adaptation of MPEG-4 Video transmission over wireless

Abstract: Transmitting MPEG-4 video over Wireless Local
Area Networks is expected to be an important component of
many emerging multimedia applications. One of the critical
issues for multimedia applications is to ensure that the Quality
of Service (QoS) requirement to be maintained at an acceptable
level. The IEEE802.11 working group developed a standard
called IEEE802.11e to support Quality of Service (QoS) in
WLANs. This standard aims to support QoS by providing
differentiated classes of service at the Medium Access Control
(MAC) layer to enhance the ability of physical layers to deliver
time-critical traffic in the presence of traditional data packets.
In this Paper, Network Simulator 2 (NS-2) & Evalvid are used
as simulation tool to evaluate the Performance of MPEG-4
Video Transmissions over IEEE802.11e. A modification of the
parameters of NS-2 and Evalvid were done to present the
evaluated method. This has allowed us to control the different
design metrics values such as Frame/Packet loss, PSNR &
Decodable Frame Rate (Q). In this paper, we evaluate a
framework which consists of a MAC-centric cross-layer
architecture to allow MAC layer to retrieve video streaming
packet information, and a single-video multi-level queue to
prioritize I/P/B slice (packet) delivery, the evaluated systematic
scheme shows better results for lower packet loss, higher PSNR
& Decodable Frame Rate (Q) for MPEG-4 video transmission
over IEEE 802.11e..
Keywords: IEEE802.11e, MPEG-4, NS-2, Evalvid.
1. Introduction
Efficient video transmission over wireless remains one of
the challenging goals for multimedia communications
because of the scarce wireless resources, high bandwidth
and quality of service (QoS) requirement for video
transmission. However, studies on the MPEG-4 video
transmission in the wireless networks are not well explored.
This paper proposes a study of the transfer of video
sequences coded in MPEG-4 over wireless networks using
the IEEE802.11e technology. Quantifying the effects of
these simulations will help us in the design of a scheme for
channel. In the following section 2, we briefly preview the
related work involved in the transmission of MPEG-4 video
packets over IEEE802.11e.
Section 3 presents the simulation set-up, & Network
topology of our simulation study. In section 4, simulation
results are presented and discussed. The conclusion is given
in Section 5 followed by the References.
2. Related Work
IEEE802.11 is one of most popular medium access control
protocols in the world. However, with IEEE802.11 it is
difficult to guarantee QoS because all stations use the equal
access method and network parameters.
With the development of services such as multimedia
streaming, the demands for QoS guarantees are increasing.
The IEEE standard [13] proposed IEEE802.11e medium
access control protocol for QoS guarantee extends the
existent and enhancements of IEEE802.11 Medium Access
Control protocols and newly defines the HCF (Hybrid
Coordination Function). HCF proposed two methods: EDCA
(Enhanced Distributed Channel Access) and HCCA (HCF
Controlled Channel Access).
The EDCA model is the competition base medium access
method and the HCCA model is a polling base medium
access method to act under the control of the HC (Hybrid
Coordinator). The EDCA model is designed to provide a
differentiation service based on priority, and is similar with
the DCF (Distributed Coordination Function) model of
existing 802.11 MAC. The EDCA model is the defining
concept of AC (Access Category) for QoS support. Each AC
has competition parameters, AIFS [AC] (Arbitration Inter-
frame Space) that alternate DIFS (Distributed Inter-frame
Space), CWmin [AC] and CWmax [AC] [1], [2].
12
Fig. 1 shows the structure of IEEE802.11e EDCA model,
Wireless Station (WSTA) that have QoS parameters to
decide priority and four transmission queues that are
recognized by the virtual station. If more than two values of
the back-off counter in a station reach to 0 at the same time,
the scheduler of WSTA prevents a virtual collision.
Figure 1. Structure of IEEE802.11e model
Fig. 2 shows the medium access method of the EDCA
model. Each frame has an AIFS [AC] of different size
according to the priority. Frames of highest priority have
IFS (Inter-frame Space) such as DIFS, and low priority
frames have longer IFS than others. As a result, the
probability of access to the medium of frame and the priority
of frame are in proportion. The sizes of the competition
windows are different CWmin [AC] and CWmax [AC]
according to each AC.
Figure 2. Medium Access Control method of IEEE802.11e
EDCA
If the priority of the frame is high, it has a small CWmin
[AC] and CWmax [AC]. Therefore, the latency times to
medium access of the high priority frames can be reduced
even if a collision arises, and the data in transmission queue
can have a transmission opportunity easily.
There is much research taking place on transmitting
multimedia video efficiently using the medium access
control method of the IEEE802.11e EDCA model
[3],[4],[5]. Techniques, among the ones being researched to
improve quality of service, use the discriminating medium
access control method of IEEE802.11e EDCA model
according to the importance of the video data type [6].
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
However, these techniques can create attempts at
unnecessary data transmission that exceed available
bandwidth of networks because it does not have an adaptive
transmission rate adjustments scheme. As a result,
transmission queue overflow or loss of data causes
degradation in the quality of the video streaming service.
In [7] the authors classify cross-layer architectures for video
transport over wireless networks into five categories:-
1. Top-down approach: the higher-layer protocols optimize
their parameters and the strategies at the next lower layer.
2. Bottom-up approach: the lower layers try to insulate the
higher layers from losses and bandwidth variations.
3. Application-centric approach: the APP layer optimizes
the lower layer parameters one at a time in a bottom-up
(starting from the PHY) or top-down manner, based on its
requirements.
4. MAC-centric approach: the APP layer passes its traffic
information and requirements to the MAC, which decides
which APP layer packets/flows should be transmitted and at
what QoS level.
5. Integrated approach: strategies are determined jointly by
all the open system interconnection (OSI) layers.
Figure 3. MAC Centric Cross-layer Architecture
As shown in Fig. 3, the MAC layer treats video streams
differently. Thus, the application layer passes its traffic
information (the priority of the streams) with their QoS
requirements to the MAC layer, which maps these partitions
to different traffic categories to improve the perceived video
quality.
 (IJCNS) International Journal of Computer and Network Security, 13
Vol. 2, No. 5, May 2010

Figure 4. Prioritized Packetization 4. Simulation Results

IEEE802.11e EDCA has priority multi-level queue, AC
Prioritized Packetization gives priority to each frame (Access Category). The priorities of the four ACs are AC_I
according to the importance of the MPEG-4 video frames. > AC_P > AV_B > AV_Non-video If most of the
The frame degree of importance becomes different transmissions are MPEG-4 video packets of the same QoS
according to the relativity of the frames and the compression level then the advantage of priority multi-level queue will no
rate [8]. Fig. 4 shows the structure of the prioritized more exist. Hence we use a single-video multilevel queue by
packetization module that gives priority to each frame creating four access categories, AC_0 (highest priority),
according to the degree of importance of the frame; the AC_1 (second priority), and AC_2 (third priority). I-slice,
parser program reads each compressed video stream from P-slice and B-slice packets are assigned to AC_0, AC_1,
the video encoder output and generates a traffic trace file. and AC_2, respectively, and the remaining AC is dedicated
to non-video traffic.
3. Simulation Set-up
To evaluate the performance of IEEE802.11e we have
conducted simulations using a widely adopted network
Simulator (NS-2) [9] integrated with EvalVid [10].
The original NS2 software supports the IEEE802.11e only,
and it was necessary to augment it with the new
IEEE802.11e. The EDCA setup is added using the TKN
implementation of IEEE802.11e [11]. All simulations were
done under Windows XP Service Pack 2 using Cygwin [11]
in order to run NS-2 under Windows Operating Systems.

3.1 Network Topology

We consider a wireless topology having an infrastructure
WLAN with a varying number of wireless stations, and one
access point (AP) as shown in Fig. 5.

Figure 6. Single Video multi-level Queue

As shown in Fig. 6 our proposed scheme is a single-video

multi-level queue to prioritize I/P/B slice (packet) delivery,
this IEEE802.11e priority multi-level queue is able to
enhance bandwidth utilization.

Figure 5. Simulation Network Topology

All stations communicate with each other through the AP.

All stations are located in the same domain with the AP.
There is no mobility in the system to avoid the wireless
problems such as the hidden node problem. All stations send
CBR (Constant Bit Rate) traffic at a fixed sending rate and
interval. UDP is implemented as the transport layer
protocol. Figure 7. Foreman QCIF (176 x 144)

The video source can be either in the YUV QCIF (176 x

144) or in the YUV CIF (352 x 288) formats.
The video source used in the simulation is YUV QCIF (176
14 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

x 144), Foreman. Each video frame was fragmented into 237 149 273 0 1 9
packets before transmission, and the maximum transmission
packet size over the simulated network is 1000 bytes, Fig. 7
shows the video used in our simulations.
Packet Sent Vs. Packet Lost

300
273
The video sequence has temporal resolution 30 frames per 250 237

second, and GoP (Group of Pictures) pattern

200
IBBPBBPBBPBBI. In Table 1 & Table 2 shows the Total

Amount
149 Packet Sent
amount of each frame/packet in the video sequence 150
Packet Lost

respectively. 100

50
9
Table 1: Amount of video frames in the video source 0
0 1

I P B
Pa ck e t type
Video Format Frame Number Total

I P B
Foreman QCIF 45 89 266 400 Figure 8. Sent packets vs. lost packets

Table 2: Amount of video packets in the video source As shown in Table 5 & Fig. 9, the total number of lost video
frame is 7, including 0 I frames, 1 P frames, and 6 B frames.
Video Format Packet Number Total The Decodable Frame Rate (Q) is 0.9625, and the average
PSNR of the result video is 33.67, as shown in Fig. 10 is a
I P B
sample of an output video frame.
Foreman QCIF 237 149 273 659

Table 5: Sent frames vs. lost frames

In table 3 shows the simulation parameters of the Sent Frames Lost Frames
simulation; we simulated two different traffics which is
I P B I P B
Video & Voice.
45 89 266 0 1 6

Table 3: Simulation parameters

Simulation Video Voice Frame Sent Vs. Frame Lost
Parameter
I -frame P-frame B- frame 300
266

Transport 250
Protocol UDP UDP UDP UDP
200
CWmin 3 7 15 7
Amount

CWmax 7 15 1023 15 Frame Sent

150
Frame Lost
AIFSN 2 3 3 4
100 89
Packet Size
(bytes) 1028 1028 1028 160 50
45
Packet Interval 6
0 1
10 ms 10 ms 10 ms 20 ms 0
Data rate I P B
(kbps) 1024 1024 1024 64 Frame type

Figure 9. Sent frames vs. lost frames

As shown in Table 4 & Fig. 8, the total number of lost video

packets is 10, including 0 I frames, 1 P frames, and 9 B
frames.

Table 4: Sent packets vs. lost packets

Sent Packets Lost Packets Figure 10. Sample of Received video frame
I P B I P B
 (IJCNS) International Journal of Computer and Network Security, 15
Vol. 2, No. 5, May 2010
Contention-Based Channel Access,” IEEE JOURNAL
ON SELECTED AREAS IN COMMUNICATIONS,
VOL. 22, No. 10, December 2004.

5. Conclusions
In this paper, we evaluated a framework to improve the
quality of video streaming. This framework consists of
MAC-centric cross-layer architecture to allow MAC-layer to
retrieve video streaming packet information (slice type), to
save unnecessary packet waiting time, and a single-video
multi-level queue to prioritize I/P/B slice (packet) delivery.
We evaluated our simulations based on the decodable frame
rate (Q) and PSNR. Through simulations, we also revealed
that PSNR & decodable frame rate can evaluate the
perceived quality well by an end user. Therefore, we found
that the larger the Q & PSNR value, the better the video
quality perceived by the end user. Simulations show that the
evaluated methodology outperforms IEEE802.11e in packet
loss rate, and PSNR.

References
[1] IEEE 802.11e, “Wireless LAN Medium Access Control
(MAC) Enhancements for Quality of Service (QoS),”
802.11e Draft 8.0, 2004.
[2] S. Choi, “Overview of Emerging IEEE 802.11 Protocols
for MAC and Above,” Telecommunications Review,
Special Edition, November 2003.
[3] D. Gu, and J. Zhang, “QoS Enhancement in IEEE
802.11 Wireless Local Area Networks,” IEEE
Communications Magazine, June 2003.
[4] D. Chen, D. Gu, J. Zhang, “Supporting Real-Time
Traffic with QoS in IEEE802.11e Based Home
Networks,” IEEE Consumer Communications and
Networking Conference, January 2004.
[5] D. Gao, J. Cai, P. Bao, Z. He, “MPEG-4 Video
Streaming Quality Evaluation in IEEE802.11e
WLANs,” IEEE International Conference on Image
Processing, September 2005.
[6] A. Ksentini, A. Gueroui, M. Naimi, “Toward an
improvement of H.264 Video Transmission over
IEEE802.11e through a Cross Layer Architecture,”
IEEE Communications Magazine, Special Issue on
Cross-Layer Protocol Engineering, January 2006.
[7] Sai Shankar N, Mihaela van der Schaar, “Performance
Analysis of Video Transmission over IEEE 802.11a/e
WLANs,” IEEE Transactions on Vehicular, VOL. 56,
Issue 4, 2007.
[8] Pilgyu Shin, Kwangsue Chung, “A Cross-Layer Based
Rate Control Scheme for MPEG-4 Video Transmission
by Using Efficient Bandwidth Estimation in IEEE
802.11e,” School of Electronics Engineering,
Kwangwoon University, Seoul, Korea, 2007.
[9] NS-2, http://www.isi.edu/nsnam/ns.
[10] Chih-Heng, Ke, “How to evaluate MPEG video
transmission using the NS2 simulator,” EE
Department, NCKU, Taiwan, 2007.
[11] S. Wiethölter and C. Hoene, “Design and Verification
of An IEEE802.11e EDCF Simulation Model in ns-
2.26,” Tech. Rep. TKN-03-019, Technische Universität
Berlin, 2003.
[12] Cygwin, http://www.cygwin.com
[13] IEEE Std 802.11e-2005, “Wireless LAN Medium
Access Control (MAC) and Physical Layer (PHY)
specifications Amendment 8: Medium Access Control
(MAC) Quality of Service Enhancements,” November
2005.
[14] Zhen-ning Kong, Danny H. K. Tsang, Brahim Bensaou,
Deyun Gao, “Performance Analysis of IEEE802.11e
16 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

Face Recognition Using Subspace LDA

Sheifali Gupta 1, O.P.Sahoo 2, Ajay Goel3 and Rupesh Gupta4
1
Department of Electronics & Communication Engineering,
Singhania University, Rajasthan,India
sheifali@yahoo.com
2
Department of Electronics & Communication Engineering,
N.I.T. Kurukshetra, India,
opsahu_reck@yahoo.co.in
3
Department of Computer Science & Engineering
Singhania University, Rajasthan, India
goelajay1@gmail.com

4
Department of of Mechanical Engineering
Singhania University, Rajasthan, India
rup_esh100@yahoo.co.in

2. Background and Related Work

Abstract: In this paper we have developed subspace LDA
approach for face recognition. This approach consists of two
steps; the face image is projected into the eigenface space which
is constructed by PCA, and then the eigenface space projected
vectors are projected into the LDA classification space to
construct a linear classifier. We have focused on the effects of
taking the number of significant eigen vectors in LDA approach.
Experimental results using MATLAB are presented in this paper
to demonstrate the viability of the proposed face recognition
method. It shows that a high recognition rate is observed when
the eigenface space’s dimension is small (40-60) and it is less
when eigenface space’s dimension is large (175-200). It also
shows that if the minimum Euclidian distance of the test image
from other images is zero, then the test image completely
matches the existing image in the database. If minimum
Euclidian distance is non-zero but less than threshold value,
then it is a known face but having different face expression else
it is an unknown face.
Keywords: Face Recognition, Eigenvalues, Principle
component analysis (PCA), LDA (Linear Discriminant Analysis)
1. Introduction
In recent years considerable progress has been made on
the problem of face detection and recognition. The best
results have been attained with frontal facial images. The
problem of recognizing a human face from a general view
remains largely unsolved. Biological changes such as aging
cause similar problems.
An information theory approach to face recognition
decomposes face images into a small set of characteristic
feature images called ‘eigenfaces’, which are actually the
principal components of the initial training set of face
images. Then recognition is performed by projecting a new
image into the subspace spanned by the eigenfaces (‘face
space’) and then classifying the face by comparing its
position in the face space with the positions of the known
individuals [1].
Much of the work in computer recognition of faces has
focused on detecting individual features such as eyes,
nose, mouth and head outline and defining a face model by
position, size and relationships among features. Eigenfaces
and Fisherfaces are reference face recognition methods
which build reduced subspaces in order to generate optimal
features of facial images.
2.1 Principal Component Analysis
PCA is a general method used for the identification of linear
directions in which a set of vectors is best represented. It is
commonly used as a pre-processing technique for
dimensional reduction. Kirby and Sirovich used PCA in
1987 in order to obtain a reduced image space [2]. After that
in 1991 Turk, Pentland, Moghaddam and Starner extended
the idea to eigenspace projections in the solution of face
recognition [3].
2.2 Fisher Discriminant Analysis
R. A. Fisher developed Linear/Fisher Discriminant Analysis
(LDA) in 1936 [4]. Fisher Discriminant Analysis (FDA) is a
powerful method for face recognition yielding an effective
representation that linearly transforms the original data
space into a low-dimensional feature space where the data is
as well separated as possible under the assumption that the
data classes are Gaussian with equal covariance structure.
While PCA tries to generalize the input data to extract the
features, LDA tries to discriminate the input data by
dimension reduction. The Linear Discriminant Analysis
searches for the best projection to project the input data on a
lower dimensional space, in which the patterns are
discriminated as much as possible. For this purpose, LDA
tries to maximize the scatter between different classes and
minimize the scatter between the input data in the same
class.
 (IJCNS) International Journal of Computer and Network Security, 17
Vol. 2, No. 5, May 2010

3. Subspace LDA Method each image in the database is calculated. The test image will
match the image in the database having the minimum
It is an alternative method which combines PCA and Euclidean distance with it.
LDA [5]. This method consists of two steps; the face image
is projected into the eigenface space which is constructed by In figure 2, Euclidean distance of one test image from all the
PCA, and then the eigenface space projected vectors are 400 images is shown. The Euclidean distance is zero with
projected into the LDA classification space to construct a the image no. 7 in the database as clear from the figure 2.
linear classifier. In this method, the choice of the number of As the Euclidean distance is zero, so test image completely
eigenfaces used for the first step is critical; the choice match the image from our database as shown in figure 3.
enables the system to generate class separable features via
LDA from the eigenface space representation. Projecting the
data to the eigenface space generalizes the data, whereas
implementing LDA by projecting the data to the
classification space discriminates the data. Thus, Subspace
LDA approach seems to be a complementary approach to
the Eigenface method.

4. Experimental Results
To analyze the performance of human face recognition
using subspace LDA, we performed experiments on ORL
databases using MATLAB. There are 10 different images of
40 distinct subjects in ORL database. For some of the
subjects, the images were taken at different times, varying
lighting slightly, facial expressions (open/closed eyes,
smiling/non-smiling) and facial details (glasses/no-glasses).
All the images are taken against a dark homogeneous Figure 2. Euclidean Distance of a test image from other
background and the subjects are in up-right, frontal position images in database
(with tolerance for some side movement). The files are in
.TIF file format (Tagged Image File Format). The size of
each image is 92 x 112 (width x height), 8-bit grey levels.

In this part of the experiments, the effect of using different

number of eigenvectors while projecting the images is
studied. From Fig 1, it can be observed that when the
eigenface space’s dimension is small (40-60), it performs
better. When the eigenface space’s dimension is large (175-
200) there is a decrease in recognition rate.

Figure 3.

For another test image no. 3, Euclidean distance from other

images is shown in figure 4. The minimum Euclidean
distance is with the Image no.5 in the database. This
distance is less than threshold value; hence it is a known
face.

Figure 1. Recognition Rate Vs Number of EigenFace

Now we have taken the Euclidean distance as a metric for

face recognition. The Euclidean distance of test image from
18 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

References
[1] S.A. Rizvi, P.J. Phillips, and H. Moon, “A verification
protocol and statistical performance analysis for face
recognition algorithms”, pp. 833-838, IEEE Proc.
Conf. Computer Vision and Pattern Recognition
(CVPR), Santa Barbara, June 1998.
[2] M.Kirby and L. Sirovich “Application of the
Karhunen-Loeve procedure for the characterization of
human faces". IEEE Transactions on Pattern analysis
and Machine Intelligence 12 (1): 103–108. doi:
10.1109 /34.41 39
[3] A.Pentland, B.Moghaddam, T.Starner, O. Oliyide, and
M. Turk (1993), “View - based and modular
Eigenspaces for face recognition". Technical Report
245, M.I.T Media Lab.
[4] R.A.Fisher, “The Use of Multiple Measurements in
Taxonomic Problems”, Annals of Eugenics, Vol. 7,
No. 2. (1936), pp. 179-188.\
[5] P. N. Belhumeur, J. P. Hespanha and D. J. Kriegman,
“Eigenfaces vs. Fisherfaces: Recognition Using Class
Spesific Linear Projection”, IEEE Transactions on
Pattern Analysis and Machine Intelligence, Vol. 19,
Figure 4. No. 7, July 1997.

The test image matches the image in the database having

different face expression as clear from the figure 5.

Figure 6.

5. Conclusions
The subspaceLDA method is a combination of eigenface
method and LDA approach. Eigenface method is used for
dimension reduction to decrease the processing time and
LDA approach is used for classification due to its
discrimination power to enhance the recognition rate. From
the observations, it is clear that a high recognition rate is
observed when the eigenface space’s dimension is small (40-
60) and it is less when eigenface space’s dimension is large
(175-200).
It is also clear that if the minimum Euclidian distance of
the test image from other images is zero, then the test image
completely matches the existing image in the database. If
minimum Euclidian distance is non-zero but less than
threshold value, then it is a known face but having different
face expression else it is an unknown face.
 (IJCNS) International Journal of Computer and Network Security, 19
Travelling Salesman’s Problem: A Petri net
Approach
Sunita Kumawat 1 , G.N.Purohit 2
1, 2
Department of Mathematics and statistics, Centre of Mathematical Sciences (CMS),
Banasthali University,Banasthali (Rajasthan)-304022, India.
email: 1sunita_1586@yahoo.co.in, 2gn_purohitjaipur@yahoo.co.in
Abstract: In this paper, we consider the Travelling salesman’s
problem (TSP). Where TSP is modelled as a TSP-graph, in
which each edge is treated as a parallel combination of
oppositely directed edges. We model TSP-graph as a Petri Net-
graph, where Petri Net- graph is an underlying graph of TSP-
graph. Then solve TSP by defining suitable binary operation on
elements of columns in sign incidence matrix representation of
Petri Net-graph. In Petri Net-graph, we find a set of places
which is both Siphon and Trap with minimum sum of
capacities, whose set of input transitions equals to the set of
output transitions, and both of them are equal to the set of all
transitions in Petri Net. Then edges in TSP-graph corresponding
to these places in Petri Net-graph will form a shortest route for
the salesman to return the point of origin, after traversing all
the cities exactly ones. For the solution of TSP, we describe a
new algorithm, based on siphon-trap and bounded-ness property
of the Petri Nets.
2000 Mathematics Subject Classification: 68R10, 90C35,
94C15.
Keywords: Travelling Salesman’s Problem, Weighted
Directed Graph, Spanning Cycle, Petri Net, Siphon and Trap.
1. Introduction
The Travelling Salesman’s Problem (TSP) is one of the
most intensely studied problems in computational
mathematics [5]. Mathematical problems related to the
TSP were treated in the early nineteenth century by W.R
Hamilton and British mathematician T. P. Kirkman.
Although there are many algorithms given for the solution
of TSP [6, 7, 8, 13, 14], yet no effective solution is known
for the general case for the TSP. In this paper, we address
the same problem with a different approach, using Petri
Net model. Here we present a new algorithm to solving a
TSP using the siphon-trap and bounded-ness property of
the places in the One-one Petri Net model of given TSP-
graph. For the TSP we find a set of places in Petri Net,
which is both Siphon and Trap [1, 3], with minimum sum
of capacities, having the property that set of input
transitions equals to the set of output transitions, and both
of them are equal to the set of all transitions given in the
Net. Then edges in TSP-graph corresponding to these
places form a shortest route for the salesman.
In Petri Net theory, Petri Net is a formal tool which is
particularly well suited for discrete event systems. Its
application has emerged from the initial seminal PhD
thesis of C. A. Petri, so C.A. Petri is considered as the
originator of Petri Net applications [10, 11]. The
computational algorithmic aspects of graph theory are
emphasized in the study of the Petri Nets. The most
Vol. 2, No. 5, May 2010
interesting connections between graph theory and Petri
Nets have been brought out by T. Murata [9].
This paper is organized as follows: Section 2 provides
the necessary preliminaries, Section 3 formulates the
problem, Section 4 describes the algorithm for TSP with
illustrative example and Section 5 briefly concludes this
paper.
2 . An Overview Of Petri Net Approach
This section of the paper provides the necessary
preliminaries for the readers who are not familiar with
Petri nets.
As Petri Nets are also called place-transition net (PT-
Net), it is a particular kind of directed graphs together
with an initial state called the initial marking. In general a
Petri Net is an underlying graph of any directed graph,
which is in essence a directed bipartite graph with two
types of nodes called places and transitions. The arcs are
either from places to transitions (output of places) or from
transitions to places (input of places). In the Petri Net
graph a place is denoted by a circle, a transition by a box or
a bar and an arc by a directed line. A Petri Net is a PT-Net
with tokens assigned to its places denoted by black dots,
and the token distribution over its places is done initially
by a marking function denoted by M0. A token is
interpreted as a command given to a condition (place) for
the firing of an event (transition). An event can happen,
when its all input conditions are fulfilled, See Fig.1.
Figure 1.
There are many subclasses of Petri Nets such as One-one
Petri Nets, free choice Petri Nets, colored and stochastic
Petri Nets etc. Here we introduce only One-one Petri Net, as
it is an ordinary Petri Net such that each place P has exactly
one input transition and one output transition having weight
one on each edge, but we ignore these weights generally in
the model representation of the Net PN. One-one Petri Net
is also called as Marked Graph [4]. In our paper standard
notation PN is treated as One-one Petri Net. More detailed
and formal description of Petri Nets is given in [2, 9, 10,
20 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

15]. We include here some basic definitions, which are goes through every city exactly once is a feasible tour with
relevant to this paper. a given cost, if it is smaller than the other minimum cost
Definition 2.1: A place-transition net (PT-Net) is a tour.
quadruplet PN = 〈P, T, F, W〉, where P is the set of places, T 3.1 Modelling TSP as a Graph (TSP-graph):
is the set of transitions, such that P ∪ T≠∅ and P ∩ T=∅, A pair G = {V, E}, where V= {v1,v2,v3,…,vn} is the set of
F⊆ (P x T) ∪ (T x P) is the set of arcs and W: F à{1, 2 ...} vertices and E = {e1,e2,…,em} is the set of edges such that
is the weight function. PN is said to be an ordinary PT-Net
each edge ei having some weights wi∈W, where W:Eà R+
if and only if W: Fà {1}.
is the weight function and wi =w(ei) is the weight
A marking is a function M0: Pà {0, 1, 2 ...}, which
distributes the tokens to the places initially. Here M0 (p) is associated with edge ei . Further when the edges vi vj and
the number of tokens in the place p at initial marking M0, it vj vi are considered different then G={V,E} is called a
is a non- negative integer less then or equal to the capacity weighted directed graph. In weighted directed graph, a
of the place. Capacity of the place is defined as the directed cycle is a closed sequence of directed edges
capability of holding the maximum no. of tokens at any without repetition of vertices except terminals and it said
reachable marking M from M0. A marking M is said to to be Spanning if it contains all the vertices of the graph. If
reachable to M0 if there exist a firing sequence σ = {t1, t2… any edge from the sequence is deleted then cycle becomes
tn} such that M can be obtained from M0 as firing of open. Thus solving a TSP amounts to finding a minimum
transitions t1, t2,…, tn. A Petri Net structure PN = 〈P, T, F, weight spanning cycle.
W〉 without any specific initial marking is denoted by PN As an illustration we consider a weighted graph on
and Petri Net with the given initial marking is denoted by 〈 four vertices denoting the four cities A, B, C, D, where
PN, M0〉. *x and x* are the set of input transitions (or each city has a direct link with the other three. The
places) and the set of output transitions (or places)
numbers (weights) associated with the edges denotes the
respectively, as x∈ P (or T). Here | *x | and | x* | stands for
physical distance between the cities, See Fig. 2. We
number of the input transitions (or places) and the output
transitions (or places) respectively. Thus a One-one Petri construct a directed weighted graph (TSP-graph) of the
Net is an ordinary PT-Net such that ∀ p ∈ P: |*p|=|p*| =1, same graph in Fig. 3.
i.e., the number of the input transitions for p ∈ P equals to A
the number of the output transitions and both of them are 298 128
equal to one.
163
For a PT-Net, a path is a sequence of nodes ρ = 〈 x1, x2…
xn 〉 where (xi, xi+1) ∈F for i = 1, 2… n-1. ρ is said to be D 206
B
elementary if and only if it does not contain the same node 137
more than once and a cycle is a sequence of places 〈 p1 , p2 , 98
…pn 〉 such that there exist t1, t2,…, tn ∈ T : 〈 p1, t1, p2 ,
t2,…., pn , tn 〉 forms an elementary path and (t n, p1) ∈F. C
Definition 2.2: For a PT-Net 〈PN, M0〉, a place p is said to Figure 2. Network of cities
be k-bounded (or bounded by k) where k∈ R+, if and only if
V1
M0 (p) < k., denotes the capacity of the place in the Net. e1 e2 e8
(PN, M0〉 is said to be k-bounded if and only if every place is
k-bounded. e11 e9 e7
Definition 2.3: A non-empty subset of places S is called a V4 e12 e10
Siphon if *p⊆ p* ∀ p∈S denoted also *S⊆ S*; i.e., every e3 e4 V2
transition having an output place in S has an input place in e6
S. Likewise a non-empty subset of places Q is called a Trap
if p*⊆ *p ∀ p∈Q denoted also Q* ⊆ *Q; i.e., every V3 e5
transition having an input place in Q has an output place in
Q. Figure 3. TSP- Graph

3. Problem Formulation: Travelling Salesman’s Distance between two cities in Fig.3 can be represented in
Problem (TSP) terms of adjacency matrix as depicted below:

For a given network of cities and the cost of travel V1 V2 V3 V4

between each pair of them, the Travelling Salesman’s
Problem or TSP for short, is to find the shortest route for V1 --- 128 163 298
the salesman, visiting to all of the cities exactly ones and V2 128 --- 98 206
returning to the starting point. In the standard version of V3 163 98 --- 137
TSP, the travel costs or distances are symmetric in the V4 298 206 137 ---
sense that travelling from city X to city Y costs just as
much as travelling from Y to X. Any round-trip tour that
 (IJCNS) International Journal of Computer and Network Security, 21
Vol. 2, No. 5, May 2010

3.2 Modelling TSP-graph as a Petri Net I of PN, which is both siphon and trap and also *X= X* = T,
A Petri Net PN is modelled from TSP-graph given in Fig. 3 i.e., set of the input transitions of X is equals to the set of
as follows: edges es are transformed into places ps and output transitions of X, and both of them equal to the set
vertices vk are transformed into transitions tk so that the of all transition in T. This equality holds only if the
place ps has an input from a transition ti, and an output to a addition under the operation ⊕, of the k column vectors
transition tj, if es is an directed edge vi vj in graph, then say C1, C2, C3,…, Ck i.e., C1 ⊕ C2 ⊕ C3⊕…..⊕ Ck contains
weights (distance between two cities) of es’s are replaced only ± entries everywhere, where Cj , j = 1, 2... k, denotes
by capacities ks of corresponding places ps and number of the column vector corresponding to the place pj in I.
tokens for places ps ‘s is the value of ks  . Now let C1 ⊕ C2⊕ C3⊕…..⊕ Ck will be a column vector
denoted by γ = [γi] where γi denotes the
Fig. 4 in Appendix (1), shows the One-one Petri Net
i
th
element of the column vector γ, have as elements from
model say PN, of the TSP-graph, having the set of
set U. Then from the definition of I under the operation⊕,
transitions T= {t1 t2, t3, t4} and the set of places is P = {p1,
p2,…,p12 } corresponding to the set of vertices{v1,v2 ,v3,v4} we interprets about γi as. γi = 0 means no place in X is an
input or output place of transition i.
and the set of edges {e1,e2 ,…,e12} respectively in the given
γi = - means some place in X is an input place of
TSP-graph. For the sake of clarity, we observe that t1 is the
transition i.γi = + means some place in X is an output place
input transitions for the places p2, p7, p10 and the output
transition for the places p1, p8, and p9. Similarly p1 is the of transition i.
γi = ± means some place in X is an input place as well as
input place for the transition t1 and output place for the
output place for transition i.
transition t4. Using this same procedure, we find a set of the
From the above it can be seen that every transition having
places in Net which is both siphon and trap with minimum
sum of capacities, whose set of the input transitions equals an output place in X has an input place in X only if γi ≠ +,
and likewise every transition having an input place in X
to the set of output transitions and both of them are equal
has an output place in X only if γi ≠-. So X is both siphon
to the set of all transitions in the Net PN.
and trap if and only if γ has either 0 or ± entries. And if γ
4. Description of algorithm for Travelling
has only ± entries everywhere, then the places
Salesman’s Problem corresponding to columns C1, C2, C3,…, Ck in C1 ⊕ C2 ⊕
As in above we model the TSP-graph as a Petri Net-graph.
C3⊕…..⊕ Ck forms a set of places, which is both siphon
Now here we present an algorithm for solving the
and trap, whose input transitions equals to the output
Travelling Salesman’s Problem. For the description of the
transitions and both of them are equal to the set of all
algorithm for TSP; we introduce some new notations as
transitions T, with having some capacities. We select only
follows:
those set of places X such that *X = X* = T, having
In a Petri Net PN with n-transitions and m-places, the sign
minimum sum of capacities among all. Algorithm
incidence matrix I = [aij] is the
discussed below, gives us a siphon–trap set of places X with
n x m matrix, whose entries are defined as,
minimum sum of capacities, having the property *X = X* =
aij = + if place j is an output place of transition i.
T, whose corresponding edges in TSP-graph will form a
aij = - if place j is an input place of transition i.
shortest route for the salesman.
aij = ± if place j is both input and output places of
transition i, (i.e., transition i and place j form a self loop) ALGORITHM:
Input Sign incidence matrix I of order n x m.
aij = 0 otherwise.
Step 1 Select Cj, the first column in the sign incidence
matrix I having ‘+’ entry whose corresponding place and
As an illustration, the Sign incidence matrix I of PN is:
capacity is denoted as PLACEj and CAPACITYj
p1 p2 p3 p4 p5 p6 p7 p8 p9 p10 p11 p12
Set recursion level r to 1
t1 - + 0 0 0 0 + - - + 0 0 Set Vjr = Cj
Set PLACE jr = PLACEj;
t2 0 0 0 0 + - - + 0 0 + - Set CAPACITYjr =CAPACITYj;
t3 0 0 + - - + 0 0 + - 0 0 Set X =∅, W=∅, Sum =0 and K =∅.
t4 + - - + 0 0 0 0 0 0 - +
Step 2 If Vjr has a ‘±’ entry at the ith row then PLACEjr is a
Hare we introduce a commutative binary operation, self loop with transition i, Go to Step 5.
denoted by ⊕ on the set U= (0, +, - ,±).
Step 3 If Vjr has a ‘+’ entry in the kth row find a column Cs
⊕ defined as following: + ⊕ - = ± (i.e., ‘+’ entry is said
which contains a ‘–‘entry at kth row.
to be neutralized by adding a ‘–‘entry to get a ‘±’ entry).
a⊕a=a ∀ a∈U (i) If no such column Cs exists then go to Step 5.
±⊕a=± ∀ a∈
(ii) If such Cs exists, add it to Vjr to obtain Vj(r+1) = Vjr ⊕
0⊕ a=a ∀ a∈U
Cs, containing a ‘±’ entry at kth row. Then
For the Travelling salesman’s problem, we choose a
subset of k places X = {p1, p2 …pk} in sign incidence matrix PLACEj(r+1) = PLACEjr ∪ PLACEs
22 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

CAPACITYj(r+1) = CAPACITYjr ∪ CAPACITYs V12( 2) has ‘+’ entry at 3rd row. The neutralizing columns
(iii) Repeat this step for all neutralizing columns Cs. This are C4, C5 and C10.
gives a new set of Vj(r+1)’s, PLACEj(r+1)‘s and
−  0  − 
CAPACITYj(r+1) 0  0  0 
V13(1) = V12( 2) ⊕ C4 =   ⊕   =  ;
Step 4 Increase r by 1, Repeat Step 3 until there are no ‘+’entries +  −  ± 
in each Vjr = C1 ⊕ C2⊕ C3⊕…..⊕ Cjr.      
±  +  ± 
     
Step 5 Any Vjr with all entries as ‘±’ represents both siphon
(1) (1)
and trap such that their input transitions equal to the output PLACE 13 = {pl,p3, p4}; CAPACITY 13 =
transitions and both of them equal to the set of all {137,206,206}
transitions T.
X = X ∪ PLACEjr −  0  − 
0  +  + 
K= K ∪ CAPACITYjr V13( 2) = V12( 2) ⊕ C5 =   ⊕   =  ;
W=Sum + Sumw K, where Sumw K is the sum of the +  −  ± 
     
capacities in the set K, Store it any other set and compare it ±  0  ± 
to minimum weight set at each iteration.      
Step 6 Delete Cj (2) ( 2)
PLACE 13 ={pl, p3, p5}; CAPACITY 13 = {137,
j = j + 1 Go to Step 1. 206,128}
Output: Set X has places both siphon and trap such that −  +  ± 
their input transitions equal to the output transitions and 0     
V13(3 ) = V12( 2 ) ⊕ C10 =   ⊕ 0  = 0  ;
both of them equal to the set of all transitions T with +  −  ± 
minimum sum of capacities. Whose corresponding edges      
±  0  ± 
set in TSP-graph, forms a shortest route for the salesman.      
As an illustration consider the graph in Fig 2. ( 3) ( 3)
PLACE 13 = {pl, p3, p10}; CAPACITY 13 = {137, 206, 98}
Step 1 Select first column having ‘+’ entry. Here is C1,
then V12( 3) has ‘+’ entry at 2nd row. The neutralizing column is
− 
  C6, C7 and C12.
V11 =  0  ; PLACE11 = {pl};
0  −  0  − 
  +  −  ± 
⊕ C6 = ⊕
( 4) (3 )
+ 
V 13 =V 12     =  ;
  0  +  + 
     
CAPACITY11 = {137} ±  0  ± 
     
Steps 2, 3 and 4, V11 has a ‘+’ entry at 4th row. The (4) (4)
neutralizing columns are C2, C3 and C11. PLACE 13 = {pl, p11, p6}; CAPACITY 13 = {137,298,128}
−  +  ±  −  +  ± 
(1) 0  0    +  −  ± 
V = V11 ⊕ C2 =   ⊕   = 0  ; V13(5 ) = V12(3 ) ⊕ C7 = ⊕
12     =  ;
0  0  0  0  0  0 
           
+  −  ±  ±  0  ± 
           
(1) (1) ( 5) ( 5)
PLACE 12 = {pl , p2}; CAPACITY 12 ={137,137 } PLACE 13 = {pl , p11, p7}; CAPACITY 13 = {137,298,163}
−  0  − 
0      −  0  − 
V ( 2)
= V11 ⊕ C3 =   ⊕  0  =  0  ; +  −  ± 
⊕ C12 = ⊕
( 6) ( 3)
12 V =V     =  ;
0  +  +  13 12
      0  0  0 
+
  −  ±       
      ±  +  ± 
     
(2) (2) ( 6) ( 6)
PLACE 12 = {pl, p3}; CAPACITY 12 = {137,206} PLACE 13 = {pl, p11, p12}; CAPACITY 13 = {137,298,298}
−  0  − 
0  +  +  V13(1) and V13( 6 ) has no ‘+’ entry , also all the entries are not
V12(3 ) = V11 ⊕ C11 =   ⊕   =  ;
0  0  0  ‘±’ only, but those sets which have all entries ‘±’ or ‘0’ are
      both siphon and trap ( as V12(1) , V13( 3) and V13( 5) )
+
  −  ± 
     
V13( 2) has ‘+’ entry at 2rd row. The neutralizing columns are
( 3) ( 3)
PLACE 12 = {pl, p11}; CAPACITY 12 = {137,298} C6, C7, and C12.
 (IJCNS) International Journal of Computer and Network Security, 23
Vol. 2, No. 5, May 2010

−  0  −  Step 6 now delete C1 from sign incidence matrix, choose

+  −  ±  next column and repeat all the steps again in similar way,
V14(1) = V13( 2) ⊕ C6 =   ⊕   =  ;
±  +  ±  we get another different sets of places {p2, p4, p6, p8},{p2, p9,
     
±
  0  ±  p5, p12 }, {p3, p8, p10, p12} and{p4, p11, p7, p9 } as choosing
      columns C2 , C3 and C4 respectively having sum of the
capacities 634, 661, 765 and 765. The set of places which
(1) (1)
PLACE 14 = {pl, p3, p5, p6}; CAPACITY 14 = {137,206,128,128} have minimum sum of capacities 634 is either {pl, p3, p5, p7 }

−  +  ± 
or {p2, p4, p6, p8}. The edges set {el, e3, e5, e7} or {e2, e4, e6, e8 }
      are corresponding to places set {pl, p3, p5, p7 } and {p2, p4, p6,
V14( 2 ) = V13( 2 ) ⊕ C7 =  +  ⊕  −  =  ±  ;
±  0  ±  p8} respectively, will form a shortest spanning cycle in
      underlying graph of Petri Net-graph, which give an
±  0  ± 
      optimal route for the salesman in TSP-graph.
The finding of this can be applied to a company’s logistic
(2) (2) problem of delivering petrol or similar to different petrol
PLACE 14 = {pl, p3, p5, p7}; CAPACITY 14 = {137,206,128,163}
sub-stations. This delivery system is formulated as the Petri
−  0  −  Net model, which involves finding an optimal route for
+     
V14 = V13 ⊕ C12 =   ⊕  −  =  ±  ;
( 3) ( 2)
visiting stations and returning to point of origin, where the
±  0  ±  inter-station distance is symmetric and known. As a
     
±
  +
    ± standard problem, we defined it simply as the time spent or
     
distance traveled by salesman visiting n cities (or nodes)
cyclically, where vehicle visits each station just once and
( 3) ( 3)
PLACE 14 ={pl, p3, p5, p12}; CAPACITY 13 ={137,206,128,298} returns the starting station. This real world application is a
deceptive simple combinatorial problem and our approach is
V13( 4 ) has ‘+’ entry at 3rd row. The neutralizing columns are to develop solutions of such type of distribution problems,
C , C , and C . based on concept of the Petri nets.
4 5 10

−  0  −  5. Conclusion
±     
C4 =   ⊕  0  =  ±  ;
( 4) ( 4)
V 14 =V 13 ⊕ In this paper, while solving the travelling salesman
+  −  ±  problem, we have exploited the potentials of siphons and
     
±  +  ±  traps. Our analysis is based on the notion of sign incidence
      matrix; this helps us to relate Petri Net theory to graph
(4) (4) theory. The complexity of the TSP is part of a deep
PLACE 14 ={pl, p6, p11, p4}; CAPACITY 14 ={137,298,128,206}
question in mathematics, as TSP is a NP- complete
−  0  −  problem. Here we have developed an algorithm using Petri
     
V14(5) = V13( 4) ⊕ C5 =  ±  ⊕  +  =  ±  ; net model, which can be executed by computer for any finite
+  −  ±  number of nodes.
     
±  0  ± 
      Acknowledgement
I would like to thank my supervisor Prof.G.N. Purohit for
( 5) (5 )
PLACE 14 ={pl, p6, p11, p5}; CAPACITY 14 = {137,298,128,128} his continuous support and valuable suggestions for this
research work. I would also like to thank my friends,
−  +  ± 
      colleagues and Centre for mathematical sciences for their
V14( 6 ) = V13( 4 ) ⊕ C10 =  ±  ⊕  0  =  ±  ; guidance and feedback to drive me in the proper direction of
+  −  ±  the research is highly appreciable. Last but not the least I
     
±  0  ±  would like to thank my parents for their love and support.
     

( 6) ( 6)
PLACE 14 = {pl, p6, p11, p10}; CAPACITY 14 = {137, 298, 128, 98}
References
V14(1) , V14( 3) , V14( 4) and V14(5) has no + entry. V14( 2) , V14( 6)
have all the entries as ‘±’, Hence PLACE 14 and PLACE 14
(2) ( 6) [1] R. Boer and T. Murata, “Generating basis siphons and
traps of Petri nets using sign incidence matrix”, IEEE
form both siphon and trap whose input transitions equal to
Transactions on circuits and systems-1, Vol.41, No. 4,
the output transitions and both of equal to the set of all
pp. 266-271,(1994).
transitions.
Step 5 the subsets of places, which are both siphon and [2] B. Baugarten,”Petri Nets basics and application” , 2nd
trap, whose input transitions equal the output transitions ed., Berlin: spectrum akademischer Verlag, (1996).
and both of them equal to the set of all transitions are{pl, [3] K.S. Cheung and K.O. Chow, “Cycle-Inclusion Property
p3, p5, p7 }and {pl,, p6, p11, p10}with capacities sum 634 and 661 of Augmented Marked Graph” , Information Processing
as choosing column first C1. Letters, Vol. 94, No. 6, pp. 271-276, (2005).
24 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

[4] F. Commoner, A. W. Holt, S. Even and A. Pnueli, her P.hd in guidance of Prof. G. N. Purohit in research area of
“Marked directed graph” , Journal of Computer and Petri net theory, As the Junior Research Scholar under the centre
System Sciences- 5, pp. 511-523, (1971). for mathematical sciences(CMS), governed by DST-New Delhi, in
[5] Narasingh Deo, “Graph theory with Applications to Banasthali University(Rajasthan).
Engineering and Computer Science”, Prentice-Hall,
India, (1984).
[6] S. Hiller Frederick and J Lieberman Gerald,
Appendix
“Operations Research “, Eighth edition, Tata Mcgraw
Hill, (1990),.
[7] M. Jünger, G. Reinelt and G. Rinaldi, "The Traveling
Salesman Problem," in Ball, Magnanti,Monma and
Nemhauser (eds.) , Handbook on Operations Research
and the Management Sciences North Holland Press, pp.
225-330, (1994).
[8] D. S. Johnson, "Local Optimization and the Traveling
Salesman Problem," Proc. 17th Colloquium on
Automata, Languages and Programming, Springer
Verlag, pp. 446-461, (1990).
[9] T. Murata, “Petri nets, properties, analysis and
application” , Proceedings of IEEE, Vol.77 (4), pp. 541-
580, (1989).
[10] J. L., Peterson, “Petri net Theory and the Modeling of
Systems” , Prentice Hall, INC, Englewood, Cliffs, and
New Jercy, (1981).
[11] C.A. Petri, "Kommunikation mit Automaten." Bonn:
lnsti-tut fur Instrumentelle Mathematik, Schriften des
MM Nr. 3, 1962. Also, English translation,
"Communication with Auto-mata." New York: Griffiss
Air Force Base.Tech. Rep.RADC-TR-65-377, vol.1,
(1966).
[12] C.A. Petri, "Fundamentals of a theory of asynchronous
information flow," in Proc. of IP Congress 62,pp. 386-
390, (1963).
[13] J.V. Potvin, "The Travelling Salesman Problem: A
Neural Network Perspective", INFORMS Journal on
Computing 5, pp. 328-348, (1993).
[14] J.V. Potvin,"Genetic Algorithms for the Traveling
Salesman Problem", Annals of Operations Research 63,
pp. 339-370, (1996).
[15] W. Reisig, “Petri Nets and introduction “, Heidelberg:
springer-verlag, (1985).

Authors Profile
Prof. G. N. Purohit, a renowned
Mathematician of the state, joined Banasthali
University in April, 2001 as Professor in the
Department of Mathematics & Statistics and
also Coordinator of Centre for mathematical
sciences in Banasthali University. He received
the B.Sc, M.Sc and P.hd degree in
mathematics from Rajasthan University during
1957-1968. His research interests are in Fluid
Dynamics, Partial Differential Equations, Petri
net theory and wireless networks.

Sunita kumawat received the B.Sc and M.Sc

degree in pure mathematics. And pursuing
 (IJCNS) International Journal of Computer and Network Security, 25
PET/CT and MR Image Registration using
Normalized Cross Correlation Algorithm and
Spatial Transformation Techniques
1
B.Balasubramanian 2 Dr. K. Porkumaran
1
Associate Professor, Sri Ramakrishna Engg. College, Affiliated to Anna University Coimbatore, India
2
Vice Principal, Dr.NGP Institute of Technology, Affiliated to Anna University Coimbatore, India
balasubramanian@ieee.org, balasrec2000@gmail.com
Abstract: This paper presents the current status of PET/CT
brain image registration. Registration of brain images is a more
complex problem, Automatic registration, based on computer
programs, might, however, offer better accuracy and
repeatability and save time. Computed Tomography (CT) is used
for the attenuation correction of Positron Emission Tomography
(PET) to enhance the efficiency of data acquisition process and
to improve the quality of the reconstructed PET data in the
brain. Due to the use of two different modalities, chances of
misalignment between PET and CT images are quite significant.
Template matching is used for applications in image processing.
Cross Correlation is the basic statistical approach to image
registration. It is used for template matching or pattern
recognition. Template can be considered a sub-image from the
reference image, and the image can be considered as a sensed
image. The objective is to establish the correspondence between
the reference image and sensed image. It gives the measure of
the degree of similarity between an image and template. This
paper describes medical image registration by template
matching based on Normalized Cross-Correlation (NCC) using
Cauchy-Schwartz inequality.
Keywords: Positron Emission Tomography (PET),
Computed Tomography (CT), Magnetic Resonance (MR),
Normalized Cross-Correlation (NCC)
1. Introduction
Image registration is the process of overlaying two
or more images of the same scene taken at different times,
from different viewpoints, and/or by different sensors. It
geometrically aligns two images—the reference and sensed
images. The present differences between images are
introduced due to different imaging conditions. Image
registration is a crucial step in all image analysis tasks in
which the final information is gained from the combination
of various data sources like in image fusion, change
detection, and multichannel image restoration. Typically,
registration is required in remote sensing (multispectral
classification, environmental monitoring, change detection,
weather forecasting, creating super-resolution images,
integrating information into geographic information
systems, in medicine (combining computer tomography
(CT) and NMR data to obtain more complete information
about the patient, monitoring tumor growth, treatment
verification, comparison of the patient’s data with
anatomical atlases), in cartography (map updating), and in
Vol. 2, No. 5, May 2010
computer vision (target localization, automatic quality
control), to name a few. During the last decades, image
acquisition devices have undergone rapid development and
growing amount and diversity of obtained images invoked
the research on automatic image registration. A
comprehensive survey of image registration methods was
published in 1992 by Brown . According to the database of
the Institute of Scientific Information, in the last 10 years
more than 1000 papers were published on the topic of image
registration. Methods published before 1992 that became
classic or introduced key ideas, which are still in use, are
included as well to retain the continuity and to give
complete view of image registration research.
Biomedical imaging based on different physical
principles has played a more and more important role in
medical diagnosis. Multiform methods of image diagnosis
provide doctors and clinicians with various anatomical and
functional information to carry out exact diagnosis and
effective treatment. Image registration is the process of
combination the different sets of data of the same object,
which come from different modalities (CT, MRI, SPECT,
PET etc.). Pre-processing algorithms improve image quality
and image registration algorithms transform the object of
images into one coordinate system.
Nowadays, medical image diagnosis is considered
as one of the fields taking advantage of high-technology and
modern instrumentation. DSA, Ultrasound, CT, MRI,
SPECT, PET etc. have played an important role in
diagnosis, research, and treatment. In general, there are two
basic types of medical images [1]: functional images (such
as SPECT or PET scans), which provide physiological
information, and anatomical images (such as X-ray,
Ultrasound, CT or MRI), which provide anatomic structure
of the body.The combination process consists of alignment
(registration) and data fusion steps. After the combination
process, the physician could use his knowledge to determine
which anatomical areas are normal or abnormal in their
functional characteristics.
2. Image Registration Methodology
26
Figure 1. Concept of Image registration
Image registration, as it was mentioned above, is
widely used in remote sensing, medical imaging, computer
vision etc. In general, its applications can be divided into
four main groups according to the manner of the image
acquisition: Different viewpoints (multiview analysis).
Images of the same scene are acquired from different
viewpoints. The aim is to gain larger a 2D view or a 3D
representation of the
scanned scene. Examples of applications: Remote sensing—
mosaicing of images of the surveyed area. Computer
vision—shape recovery (shape from stereo). Different times
(multitemporal analysis). Images of the same scene are
acquired at different times, often on regular abasis, and
possibly under different conditions. The aim is to find and
evaluate changes in the scene which appeared between the
consecutive image acquisitions. Examples of applications:
Remote sensing—monitoring of global land usage,
landscape planning. Computer vision—automatic change
detection for security monitoring, motion tracking. Medical
imaging - monitoring of the healing therapy, monitoring of
the tumor evolution. Different sensors (multimodal analysis)
- Images of the same scene are acquired by different sensors.
The aim is to integrate the information obtained from
different source streams to gain more complex and detailed
scene representation.
Examples of applications: Remote sensing—fusion
of information from sensors with different characteristics
like panchromatic images, offering better spatial resolution,
color/multispectral images with better spectral resolution, or
radar images independent of cloud cover and solar
illumination. Medical imaging—combination of sensors
recording the anatomical body structure like magnetic
resonance image (MRI), ultrasound or CT with sensors
monitoring functional and metabolic body activities like
positron emission tomography (PET), single photon
emission computed tomography (SPECT) or magnetic
resonance spectroscopy (MRS). Results can be applied, for
instance, in radiotherapy and nuclear medicine. Scene to
model registration. Images of a scene and a model of the
scene are registered. The model can be a computer
representation of the scene, for instance maps or digital
elevation models (DEM) in GIS, another scene with similar
content (another patient), ‘average’ specimen, etc. The aim
is to localize the acquired image in the scene/model and/or
to compare them. Examples of applications: Remote
sensing—registration of aerial or satellite data into maps or
other GIS layers. Computer vision—target template
matching with real-time images, automatic quality
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
inspection. Medical imaging— comparison of the patient’s
image with digital anatomical atlases, specimen
classification. Due to the diversity of images to be registered
and due to various types of degradations it is impossible to
design a universal method applicable to all registration
tasks. Every method should take into account not only the
assumed type of geometric deformation between the images
but also radiometric deformations and noise corruption,
required registration accuracy and application-dependent
data characteristics.
The detection methods should have good
localization accuracy and should not be sensitive to the
assumed image degradation. In an ideal case, the algorithm
should be able to detect the same features in all projections
of the scene regardless of the particular image deformation.
Physically corresponding features can be dissimilar due to
the different imaging conditions and/or due to the different
spectral sensitivity of the sensors. The choice of the feature
description and similarity measure has to consider these
factors. The feature descriptors should be invariant to the
assumed degradations. Simultaneously, they have to be
discriminable enough to be able to distinguish among
different features as well as sufficiently stable so as not to be
influenced by slight unexpected feature variations and noise.
The matching algorithm in the space of invariants should be
robust and efficient. Single features without corresponding
counterparts in the other image should not affect its
performance. The type of the mapping functions should be
chosen according to the a priori known information about
the acquisition process and expected image degradations. If
no a priori information is available, the model should be
flexible and general enough to handle all possible
degradations which might appear. The accuracy of the
feature detection method, the reliability of feature
correspondence estimation, and the acceptable
approximation error need to be considered too. Moreover,
the decision about which differences between images have to
be removed by registration has to be done. It is desirable not
to remove the differences we are searching for if the aim is a
change detection. This issue is very important and extremely
difficult. Finally, the choice of the appropriate type of
resampling technique depends on the trade-off between the
demanded accuracy of the interpolation and the
computational complexity.
3. PET/CT Image Processing Steps
The PET and CT data are acquired using a GE
discovery STE scanner and exported as DICOM images for
the post processing. The specifications and the parameters
of the acquisition device and the PET and CT data are
provided in [5]. The PET and CT images are aligned as
described below.
3.1 Image Segmentation
In the first step of the segmentation process, the
image data is low pass filtered to smooth out the boundary
of the brain in the PET as well as the CT images. To
segment the brain in the PET images, adaptive thresholding
 (IJCNS) International Journal of Computer and Network Security, 27
Vol. 2, No. 5, May 2010

is applied based on the minimum and the maximum the brain from two different modalities is calculated and the
intensity of each image, where different intensity levels CT image is shifted based on the average distance of the two
correspond to different anatomical structures in the PET edges.
brain images. The range of intensities representing the brain
is selected after testing the range on several datasets. By
applying an adaptive threshold based on the intensity range
of the brain, unwanted structures are removed from the
image. Figure 2 (a)-(b)show the result of the segmentation
process on the PET data. To segment the brain in the CT
images, the lower and upper limit of the intensity range is
defined for the brain. Starting from the center point of pixels
belonging to this intensity range, a region growing
technique is used and the entire brain region is filled
iteratively from pixel to pixel until the threshold boundary is
reached. Once all of the pixels inside the brain are
iteratively marked, the rest of the image is set to zero.

Figure 3. Horizontal slices from an MRI data scan

4. Results and Discussion

4.1 Normalized Cross Correlation

(a)
We present the results for medical image
registration by template matching. The experiments are
conducted on gray scale brain image of size 512 x 512. The
different size of template is extracted from the original
image.. Fig. 4(a) shows the original image and the template
of size 80 x 80 are extracted from original image is shown
in Fig.4(b) and 4(c). Using normalized cross-correlation the
generated image for registration in 4(d) shows the NCC plot
(b)
. So we can conclude that simple cross correlation is not
used for image registration.. For perfect template matching
Figure 2. The result of the segmentation process applied to
the value of maximum cross correlation coefficient is 0.99.
the PET and CT images, (a) PET image and segmentation
With Noise the maximum cross correlation coefficient is not
(b) PET image and segmentation
0.99 but this value can be considered for match.
3.2 PET/CT Image Registration

Once the preprocessing is complete, edge detection

[6] is applied to the filtered images to obtain the boundaries
of the PET and CT brain. The registration process is
performed in the axial plane first to obtain the shift values
along the x and y axes followed by the registration in the
sagittal plane for the shift values along the z axis. Using the
axial image, the average distance of the left and right
boundary between the PET and CT brain is calculated. The
magnitude of the average provides the amount of shift
required for the CT, and the sign of the average value
provides the direction in which the CT needs to be shifted. (a)
Then, the average distance for the top and bottom boundary
between the PET and CT brain is calculated which gives the
amount of shift required for the CT along the y axis. To
register the PET and CT images along the z axis, sagittal
plane images are used. In this case, the top boundary of the
PET brain is aligned with the corresponding boundary of the
CT brain. The average distance between the boundaries of
28 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

The image in Fig.5(c) is obtained by applying projective

transformation to the original image

(b) (c)

(a) (b)
0.6

0.4

0.2

-0.2

-0.4
300

200
200
250 (c)
150
100 100
50
0 0

Figure 5. a) original image b) affine transform c) projective

(d)
transform

The MR image in various views using rigid registration is

Figure 4. (a) Original image (b), (c)Templates (d) NCC
shown in Figure 6. Fig. 6(a) shows the coronal view.
Plot
Fig.6(b) and (c) shows the sagittal and axial views of an MR
image
The misregistration between PET and CT is the
major drawback for the CT based attenuation correction of
the PET. In PET/CT studies of the human brain, the
misregistration occurs mainly due to the physical motion of
the patient, as most patients are unable to keep their heads
still during the entire exam. The misregistration between the
PET and CT brain geometries results in erroneous
attenuation correction parameters for the PET
reconstruction. The reconstructed PET images obtained
using the misaligned CT attenuation map show areas of
abnormal metabolic activity in the brain, where the PET
brain is projected outside the CT brain. One approach to (a) (b)
address the PET/CT misalignment problem is to repeat the
CT scan of the brain in case of a misalignment between the
PET and the original CT. In this approach, the PET and CT
is downloaded to the processing unit and overlaid. If the
brains geometries of the PET are CT align properly, then
this CT is used for the reconstruction of the PET. However,
if the PET and CT have a significant misregistration, then
the patient has to go through another CT scan of the brain.
In this case, the patient is exposed to an unnecessary (c)
radiation dose that can be avoided if software-aligned PET
and CT is used. Also, this process can be very time- Figure 6. a) coronal view b) sagittal view c) axial view
consuming and inefficient in terms of clinical resources. In
contrast, software based PET/CT brain registration has the
advantage of being fast with no need for operator
intervention, which increases the potential for routine
5. Conclusion
Image registration is an important task that enables
clinical application. Only one CT scan is needed in this
comparative analysis of images acquired at different
case, which is aligned with the PET and then used as the
occasions. We presented a simple and fast registration
attenuation correction map. This approach also reduces the
method that can be applied to 3DMR images of the brain. It
radiation dose.
is based on the normalized cross correlation and spatial
The original PET image and its transformed
transformation techniques. We have described a algorithm
versions are shown in Figure 5. The original image is shown
based on Normalized Cross-Correlation (NCC) is the ideal
in Fig,5(a). Fig. 5(b) shows the affine transformed image.
approach to image registration by template matching. It
 (IJCNS) International Journal of Computer and Network Security, 29
Vol. 2, No. 5, May 2010

gives perfect template matching in the given reference a Student member of IEEE Society. He has contributed in
image. The Maximum Cross-correlation coefficient values publishing several research papers in national Journals and
from the NCC plot indicate the perfect template matching Conferences.
with noise and without noise condition. This approach gives
registered image, if the sensed images do not have any Dr. K. Porkumaran is presently Vice-
rotation or scaling. We have also presented a simple and Principal cum Professor and Head, EEE
fast rigid registration method that can be applied to 3DMR Department of Dr. N G P Institute of
images of the brain. It is based on automatic brain Technology, Kalapatti Road, Coimbatore
segmentation, automatic mid-sagittal plane location, image 641035. He has a bachelor’s degree in
alignment. Finally the various transformation techniques are Instrumentation & control Engineering, a
discussed master’s degree in Control systems and Ph.D in Control
Systems. He has 10 years of teaching experience and has
References guided many UG and PG projects. His research and teaching
interests include Modeling and Simulation, Instrumentation
[1] M. V. Green, J. Seidel, J. J. Vaquero, E. Jagoda, I. Lee, System Design, Linear and Non-Linear Control Systems,
and W. C. Eckelman, “High resolution PET, SPECT Process Dynamics and Control, Neural Networks. He is a
and projection imaging in small animals,” Comput. Life member of Indian Society for Technical Education
Med. Imaging Graph., vol. 25, pp. 79–86, 2001. (ISTE) and Associate member of Institution of Engineers
[2] C. A. Pelizzari, G. T. Y. Chen, D. R. Spelbring, R. R. (AMIE).He has published several research papers in
Weichselbaum, and C. T. Chen, “Accurate 3- International and national Journals. He has organized many
dimensional registration of CT, PET, and or MR conferences and chaired several technical session. He is the
images of the brain,” J. Comput. Assist. Tomog., vol. reviewer for international conferences and journals.
13, pp. 20–26, 1989.
[3] R. P. Woods, J. C. Mazziotta, and S. R. Cherry, “MRI-
PET registration with automated algorithm,” J.
Comput. Assist. Tomog., vol. 17, pp. 536–546, 1993.
[4] B. A. Ardekani, M. Braun, B. F. Hutton, I. Kanno, and
H. Iida, “A fully-automatic multimodality image
registration algorithm,” J. Comput.Assist. Tomog., vol.
19, pp. 615–623, 1995.
[5] W. M. Wells, P. Viola, H. Atsumi, S. Nakajima, and S.
Kikinis, “Multi-modal volume registration by
maximization of mutual information,” Med. Image
Anal., vol. 1, pp. 35–51, 1996.
[6] F. Maes, A. Collignon, D. Vandermeulen, G. Marchal,
and P. Suetens, “Multimodality image
registration by maximization of mutual information,” IEEE
Trans. Med. Imaging, vol. 16, pp. 187–198, 1997.
[7] C. Studholme, D. L. G. Hill, and D. J. Hawkes, “An
overlap invariant entropy measure of 3D medical
image alignment,” Pattern Recognit., vol. 32, pp. 71–
86, 1999.

Authors Profile

Prof. B.Balasubramanian is a research

scholar. He is working as an Associate
professor in the Dept. of Biomedical Engg
at Sri Ramakrishna Engg. College,
Coimbatore-22. . He has a bachelor’s
degree in Instrumentation & control
Engineering, a master’s degree in Medical
Electronics and currently pursing PhD in Biomedical
Engineering . He has 11 years of industry experience and
has 5 years of teaching experience & has guided many UG
and PG projects. His research and teaching interests include
Human Anatomy & Physiology, Medical Physics,
Biomedical Instrumentation, Bio-signal Processing,
Biomedical Image Processing, Artificial Neural Networks,
Human Assist Devices and Radiological Equipments .He is
30 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

Effect of Community Based Mobility Model and

SMS Mobility Model on the Performance of
AOMDV, DYMO, FSR, LAR, OLSR and TORA
Routing Protocols in MANET Under Diversified
Scenarios
Gowrishankar.S1, T.G.Basavaraju2 and SubirKumarSarkar3
1
Department of Electronics and Telecommunication Engineering,
Jadavpur Universisty, Kolkata 700032, West Bengal, India.
gowrishankarsnath@acm.org, sksarkar@etce.jdvu.ac.in
2
Department of Computer Science and Engineering,
Acharya Institute of Technology, Visvesvaraya Technological University,
Belgaum 590014, Karnataka, India.
basavaraju@acharya.edu.in

make use of the resources provided in their current setup.

Abstract: In this paper we have compared six different routing
protocols namely AOMDV, DYMO, FSR, LAR, OLSR and
TORA using Community Based Mobility Model and SMS
mobility Model. Of these LAR, AOMDV and TORA routing
protocols are analyzed with Community Mobility Model and
OLSR, DYMO and FSR protocols are analyzed using SMS
mobility Model. The reason for bifurcation of routing protocols
was to see whether these routing protocols were able to take
finical advantage of the mobility model on which they were
being applied. This allows us to analyze the application of
routing protocols in a network scenario for a particular
application like identifying the location of the moving nodes.
The Community Based Mobility Model and SMS Mobility Model
are modern day mobility models with close resemblance to real
world vehicular or human group mobility. It is still not clear
how these routing protocols behave under these mobility models.
Our research paper tries to address various issues like effect of
mobility, traffic, energy consumption and quality of services of
these routing protocols on the mentioned mobility models. The
results obtained in this paper also colligates with the results
obtained in [16]. Various metrics like packet delivery ratio,
average network delay, network throughput, routing overhead
and average energy consumption have been considered by
varying the mobility speed, number of nodes and the traffic load
in the network. We also claim that our paper is the first to
compare these six routing protocols with Community mobility
model and SMS mobility model as the underlying mobility
models. We were also able to unravel some of the amplifications
inherently associated with these routing protocols.
Keywords: Routing Protocols, Performance, Mobility Models,
Ad Hoc Networks, Simulation.
1. Introduction
Ad hoc networks are a specialized class of networks
comprising of hundreds of nodes which are able to
communicate among each other in a dynamic arbitrary
manner. Ubiquitous computing is intertwined with the ad
hoc networking. In this type of environment the devices
Also the mobile nodes are crippled with the amount of
power used in mobility and for sending and receiving the
messages. Infrastructure less setup of ad hoc networks can
be used in military deployment and emergency operations.
The main contribution of this paper is that we have made an
inviolable effort to study the performance of six routing
protocols namely AOMDV, DYMO, FSR, LAR, OLSR and
TORA over Community Based Mobility Model and SMS
Mobility Model. To the best of our knowledge, no work has
been reported that compares and studies the performance of
all these six Routing Protocols with Community Mobility
Model and SMS mobility Model in a single research paper.
The rest of the paper is organized as follows. Previous
related work is discussed in section two. In the Third section
a brief description of the routing protocols, Mobility models
and the energy model is given. Various simulation
parameters used in this paper is mentioned in section four.
Analysis of the result is done in section five and finally we
conclude the paper.
2. Related Work
In [1] the routing protocols are compared in MANET and
Sensor network scenario. In MANET scenario protocols like
AODV, DSDV and TORA are compared. In sensor scenario
AODV, DSDV, TORA and LEACH protocols are
considered. The authors come to conclusion that AODV is
the best routing protocol in MANET with less overhead and
higher packet delivery. In sensor network environment the
LEACH protocol has improved performance over AODV. It
is also shown that that TORA is the worst protocol in both
the scenarios.
Five different mobility models like Random Waypoint,
Random Direction, Gauss Markov Model, City Section and
 (IJCNS) International Journal of Computer and Network Security, 31
Manhattan Mobility Model are compared in [2]. These
mobility models are compared using link disjoint and node
disjoint multipath routing algorithms. With lifetime as the
parameter the Gauss Markov mobility model and Random
Waypoint mobility model has highest lifetime at link
disjoint and node disjoint multipaths at a low speed. The
lifetime of these mobility models decreases as the speed
increases. For a multi path set size parameter the maximum
number of paths are generated for city section and random
waypoint at both low speed and high speed.
A performance analysis of AODV and OLSR is done traffic
load and node density in [3]. The parameters considered are
packet delivery and the number of collisions per packet. The
authors show that with less mobility and more traffic load
the delivery of packet is highest. But with increase in
mobility and the traffic load the packet delivery decreases.
This trend is observed for AODV and OLSR both. There is
an increase in the number of collisions that occurs in the
network when the number of neighboring node is increased.
This is found to be severe in OLSR routing protocol.
Popular mobility models like Random Waypoint, Random
Walk and Random Direction Mobility Models are compared
in [4]. Packet delivery ratio, routing overhead and
throughput are the metrics considered. These mobility
models are compared by varying the number of nodes and
the mobility speed. For a varying mobile density the
Random waypoint has highest packet delivery and
throughput and Random direction has the highest overhead.
Increase in mobility speed results in Random Walk having
highest routing overhead. Same as in varying node density
even with varying mobility speed the highest packet delivery
and throughput is associated with Random Waypoint
mobility model.
A new group mobility model based on “virtual track” is
proposed in [5]. This mobility model considers the military
scenarios that can be applied in ad hoc networks. This
mobility model is then compared with random waypoint
mobility model. Through simulation it is shown that this
group mobility model based on real world scenario provides
more realistic results than random way point model.
Routing protocols like AODV, FSR, DSR and TORA are
compared in a city section environment [6]. In a city section
model the AODV routing protocol scores over other routing
protocols and FSR comes a close second.
3. Description of Routing Protocols, Mobility
Models and Energy Model
In this section we give a description of routing protocols like
AOMDV, DYMO, FSR, LAR, OLSR and TORA.
Community Based Mobility Model and SMS Mobility
Model are also discussed along with the Energy Model.
3.1 Adhoc On Demand Multipath Distance Vector
Routing Algorithm (AOMDV)
Adhoc On Demand Multipath Distance Vector Routing
Algorithm (AOMDV) is proposed in [7]. AOMDV employs
the “Multiple Loop-Free and Link-Disjoint path” technique.
In AOMDV only disjoint nodes are considered in all the
Vol. 2, No. 5, May 2010
paths, thereby achieving path disjointness. For route
discovery RouteRequest packets are propagated through out
the network thereby establishing multiple paths at
destination node and at the intermediate nodes. Multiples
Loop-Free paths are achieved using the advertised hop count
method at each node. This advertised hop count is required
to be maintained at each node in the route table entry. The
route entry table at each node also contains a list of next hop
along with the corresponding hop counts. Every node
maintains an advertised hop count for the destination.
Advertised hop count can be defined as the “maximum hop
count for all the paths”. Route advertisements of the
destination are sent using this hop count. An alternate path
to the destination is accepted by a node if the hop count is
less than the advertised hop count for the destination. We
have used the AOMDV implementation for NS-2 provided
by [8].
3.2 DYMO Routing Protocol
DYMO Routing protocol is an improved version of AODV
routing protocol. Routes are established on demand. When a
source node needs to establish a route to the destination the
RouteRequest (RREQ) messages are flooded through out the
network. During broadcasting only those nodes that have
not broadcasted previously will forward the messages. The
RREQ message includes its own address (source address),
sequential number, a hop count and the destination node
address. A hop count of one is added to a RREQ packet
originating at the source node. Each node that forwards the
RREQ packet adds its own address and the sequence
number. After the source node has sent the RREQ packet it
waits for RREQ_WAIT_TIME duration. This is a constant
value and is fixed at 1000 milliseconds.
Processing of RREQ by other nodes is done as follows.
When a node receives the RREQ message it is compared
with its own routing table. If the routing table does not
contain any information regarding the source node, then an
entry is created for the source node. The next hop is to the
node from which it had received the packets. This helps in
establishing a route back to the originator.
Figure 1. Route Breakage Scenario in Dymo
Since the network is flooded with RREQ messages almost
all the nodes do contain information regarding the source
node. If an entry is found to the source node, the node which
had received the RREQ packet compares its sequence
number and hop count with the information found in its own
routing table. If an entry exists and if it is not stale then the
node updates its routing table with the latest information or
otherwise the RREQ is dropped. Every node that evaluates a
valid RREQ is capable of creating a reverse route to all the
32
nodes whose addresses are found in the RREQ. After
updating the route table, the node increments the hop count
by a value of one to imply the number of hops the RREQ has
traveled. Once the packet reaches the destination a RREP
message is sent back to the source node through the reverse
path that had been created. This RREP packet contains the
information regarding its sequence number, number of hop
counts and addresses. The same information is added by all
the nodes that processes the RREP along the reverse path.
Consider fig 1. When ever there is a break in the path it has
to be modified by generating a RERR message. Consider
that the path between node 4 and node 6 is broken. Then
node 4 has to generate RERR message containing the
address and sequence number of the node which cannot be
reached. This RERR is broadcasted through out the network.
When node 2 receives the RERR message it compares the
information found in the RERR message with its own
routing table entries. If it has an entry then the route
information has to be removed if the next hop node (i.e.
node 4) is same as the node from which it had received the
RERR message. When node 3 and node 5 receive the RERR
message, the information provided in RERR is checked with
their corresponding entries. Since they do not use node 4 to
reach 6 the RERR message is discarded. In this way by
broadcasting the RERR message, the RERR messages the
concerned nodes are informed about any breakage in the
path to the destination node [9, 10, 11].
3.3 Fisheye State Routing Protocol (FSR)
The FSR protocol employs a concept called “FISH EYE”
technique. The idea behind fish eye technique is that a fish’s
eye is able to observe maximum number of pixels near its
focal point. The amount of pixels observed decreases as the
distance increases from the focal point. If we think of the
same concept from a network point of view it means that a
node maintains a high quality of information about its
immediate neighboring nodes while the information
decreases with the increase in the distance. A node
maintains up to date information of the link table by
obtaining the latest values from its neighboring nodes. This
information is exchanged with the neighboring nodes
periodically. These periodical exchanges are not event
driven thus reducing the overhead. When the nodes
exchange the information the entries with the highest
sequence number replaces the entries of a node with the
lowest sequence number. FSR routing protocol uses
something called “SCOPE”. The Scope of a network can be
defined as the group of nodes among whom the
communication takes place frequently as they are with in the
assumed number of hops (or less number of hops).
Figure 2. FSR Demonstration
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
In fig 2 for source node 1 all the neighboring nodes have
one hop except from node 2. For source node 2, the number
of hops is 2 for nodes 3, 4 and 5 while it is 1 for node 1 and
6. For source node 3, the number of hops from node 2, 4 and
5 is 2 and 3 hops from node 6 while the number of hops
from remaining nodes is less than 1. For source node 4, the
number of hops from 2, 3 and 5 is 2 and for 6 it is 3 while
from remaining nodes it is less than 2. For source node 5
the number of hops is less than 2 only for node 1. For source
node 6 the number of hops is less than 2 only for node 1.
These results are obtained by assuming that the scope is less
than 2. The nodes with less than 2 hops are the nodes with
whom the packets are exchanged at a high rate of frequency
than the nodes having a hop count of 2 or more. This may
result in keeping improper information regarding the nodes
that are far away. But as the packet tends to get closer and
closer to the destination node the routes are established [12,
13, 14].
3.4 Location Aided Routing (LAR) Protocol
Location Aided Routing (LAR) makes use of the Global
Positioning System (GPS). Using GPS the LAR routing
protocols is able to obtain the information regarding the
location of a node. In the LAR scheme routes are established
by using the flooding in an intelligent way. When a path is
broken or when a node needs to establish a route to the
destination node then it initiates the route discovery by
using RouteRequest (RREQ) messages.
When an intermediate node receives the same RREQ from
two different nodes then it discards one of the messages and
forwards the other message. When a node receives a RREQ
message it compares the address with its own. On finding
that it is the same intended destination node, it triggers a
RouteReply (RREP) in the reverse direction. A timeout is
specified within which the source node should receive the
RREP packet. Otherwise the source node will trigger the
route discovery process once again. In the above process the
route RREQ is sent to every node. In LAR the number of
packets sent to other nodes is reduced using various
schemes.
(a) (b)
Figure 3. (a) LAR 1 Scheme (b) LAR 2 Scheme [15]
LAR makes use of a concept called expected zone and
request zone. An expected zone can be defined as a circular
region with in which the destination node is expected to be
present. The source node determines the expected zone of
the destination node based on the average speed and the past
location of the destination node. If the actual speed of the
destination node is more than the average speed then the
 (IJCNS) International Journal of Computer and Network Security, 33
destination node may be farther from the expected zone.
Thus expected zone is just an estimation region in which the
destination node may be present. If the source node is not
able to obtain any information regarding the previous
position of the destination node, then the entire network is
considered as the expected zone. Thus the rule is; more the
information is collected regarding the destination node the
more the accurate the expected zone is.
Request zone is defined as the rectangular region with in
which a node forwards a packet. If a node is outside the
request zone then the node discards any packet it receives.
Usually the request zone includes the expected zone. If the
source node is not able to find a path to the destination node
then this request zone size is increased. The size of the
request zone is dependent on the “average speed of the node
and the time elapsed since the last position of the
destination node was found out”. In the LAR 1 scheme as
shown in fig 3 (a) the nodes 1, 3, 4 and 6 is present with in
the request zone and node 5 is the destination node present
with in the expected zone. When source node 1 sends a
RREQ packet it is broadcasted to the neighboring nodes 2, 3
and 4. Since node 2 is not present in the request zone it
discards all the received packets. Node 3 and 4 broadcasts
the packets to node 5 since they are found with in the
request zone. Once the destination node receives the RREQ
it triggers the RREP message containing information
regarding its location. In LAR scheme 2, it is assumed that
the location of the destination node is known to the source
node. The source node sends a RREQ message that contains
the location information of the destination node and the
distance from the source node to the destination node. When
an intermediate node receives this RREQ it checks the
distance mentioned in the RREQ message to its own value.
If the distance is less than the mentioned distance value then
the RREQ message is forwarded or else the RREQ message
is discarded. In fig 3 (b) node 2 on receiving the RREQ
compares its distance with its own routing table
information. Since the distance is less the RREQ packet is
forwarded. When node 6 receives The RREQ packet it finds
that it is the destination node and it triggers a RREP in the
reverse direction and a route is established to the source
node [16, 17, 18, 19]
3.5 Optimized Link State Routing Protocol (OLSR)
Optimized Link State Routing Protocol (OLSR) is a table
driven, proactive based routing protocol. Multipoint Relay
(MPR) nodes are used to optimize the OLSR routing
protocol. By MPRs the number of packets broadcasted in the
network is minimized. A node selects a set of one hop
neighboring nodes to retransmit its packets. This subset of
selected neighboring nodes is called the Multipoint Relays
of that node. The MPR nodes are the only nodes those
forward the packets during broadcasting. All the links
between the nodes are assumed to be bidirectional.
Vol. 2, No. 5, May 2010
Figure 4. Transmission of Packets Using MPR
The MPR node is chosen in such a way that the chosen node
is one hop and this one hop node also covers those
neighboring nodes which are two hops away from the
originating node. The MPR nodes are affiliated to this
original node. This reduces the number of messages that
needs to be retransmitted.
In fig 4 node 4 is two hops away from node 1. So node 3 is
chosen as an MPR. Any node that is not present in the MPR
list does not forward the packets. Every node in the network
maintains information regarding the subset neighboring
nodes that have been selected as MPR nodes. This subset
information is called as MPR Selector List.
Optimization in OLSR is achieved in two ways. First the
amount packets broadcasted in the network is reduced as
only a selected few nodes called MPR broadcast the packets.
Secondly the size of the control packets is reduced as the
information regarding its multipoint relay selector set is
provided instead of providing an entire list of neighboring
nodes [20, 21, 22, 23, 24].
3.6 Temporally Ordered Routing Algorithm (TORA)
TORA comes under a category of algorithms called “Link
Reversal Algorithms”. TORA is an on demand routing
protocol. Unlike other algorithms the TORA routing
protocol does not uses the concept of shortest path for
creating paths from source to destination as it may itself
take huge amount of bandwidth in the network. Instead of
using the shortest path for computing the routes the TORA
algorithm maintains the “direction of the next destination”
to forward the packets. Thus a source node maintains one or
more “downstream paths” to the destination node through
multiple intermediate neighboring nodes. TORA reduces the
control messages in the network by having the nodes to
query for a path only when it needs to send a packet to a
destination. In TORA three steps are involved in
establishing a network. A) Creating routes from source to
destination, B) Maintaining the routes and C) Erasing
invalid routes. TORA uses the concept of “directed acyclic
graph (DAG) to establish downstream paths to the
destination”. This DAG is called as “Destination Oriented
DAG”. A node marked as destination oriented DAG is the
last node or the destination node and no link originates from
this node. It has the lowest height. Three different messages
are used by TORA for establishing a path: the Query (QRY)
message for creating a route, Update (UPD) message for
34
creating and maintaining routes and Clear (CLR) message
for erasing a route. Each of the nodes is associated with a
height in the network
(a) (b)
Figure 5. Directed Path in TORA
. A link is established between the nodes based on the
height. The establishment of the route from source to
destination is based on the DAG mechanism thus ensuring
that all the routes are loop free. Packets move from the
source node having the highest height to the destination
node with the lowest height. It’s the same top to down
approach.
When there is no directed link from source to destination
the source node trigger the QRY packet. The source node
(node 1) broadcasts the QRY packet across all the nodes in
the network. This QRY packet is forwarded by all the
intermediate nodes which may contain a path to the
destination. Consider fig 5(a). When the QRY packet
reaches the destination node (node 9) then the destination
node replies with a UPD message. Each node receiving this
UPD message will set the value of the height to a value
greater than the height of the node from which it had
received. This results in the creation of the directed link
from the source to the destination. This is the concept
involved in the link reversal algorithm. This enables to
establish a number of multiple routes from the source to
destination. Assume that the path between node 5 and node
6 is broken (fig 5(b)). Then node 6 generates an UPD
message with a new height value with in a given “defined
time”. Node 3 reverses its link on receiving the UPD
message. This reverse link indicates that the path to
destination through that directed link is not available. If
there is a break between node 1 and node 3 then it results in
partition of the network where the resulting invalid routes
are erased using the CLR message [25, 26, 27].
3.7 Community Based Mobility Model
A new social network based model called Community
Mobility Model was proposed by [29]. This mobility model
can be used to model humans moving in groups or groups of
humans which are clustered together. The authors of the
community model have evaluated their model by using the
real time synthetic mobility traces provided by the Inter
Research Laboratory, Cambridge. The Community Model
can be conceptualized as below:
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
Figure 6. Steps Involved in the Establishment of the
Community Mobility Model [29]
The first step in establishing a community model is using
the “Social network as input” to the community mobility
model. It involves two ways i.e. “modeling of social
relationships” and “detection of community structures”.
Modeling of social relationship can be represented as a
weighted graph matrix. If any of the elements in the matrix
is greater than the specified threshold value then that
element in the graph is set to 1 and if it is less than the
threshold value then it is set to 0. A value of 1 represents
strong social interaction between the groups and a value of 0
represents no interaction. The concept of 1 and 0 is used to
emphasize the relation between any two members of a group
or any individuals of the group. The next step is to
conceptualize the interaction between groups of a social
community network. The authors of the community model
have implemented this aspect by considering an algorithm
provided by [28]. Groups communicate with each other
through “inter community edges” and this concept is called
as “betweeness of edges”. Once the connection between the
individuals in the communities and the interaction between
the communities itself is established then the next step is the
placement of the communities in a square location on a grid.
This can be represented by Spq i.e. “a square in position of p,
q”. The next step is the dynamics of the mobile host. For
mobility, a host from each group or community is selected.
For each of the host the first goal is randomly chosen inside
the square Spq. Here goal represents the mobility position.
The next goal is selected by the “social attractivity”. Each
host will have a certain attraction for another host
representing another square location. When a host is
attracted to another host then the community moves from
the present square location to the square location of another
host to which the present host is attracted. Finally, the
mobile host needs to be associated with the mobility
dynamics.
3.8 Semi-Markov Smooth (SMS) Mobility Model
Semi Markov Smooth (SMS) mobility model obeys the
physical law of smooth mobility. It has three phases a) speed
up phase b) middle smooth phase and c) slow down phase.
In the speed up phase or α phase the object is accelerated
until it reaches an acceptable speed. The speed up phase
time interval is given by [t0 , tα ] = [t0 + t0 + α∆t ] where t 0
 (IJCNS) International Journal of Computer and Network Security, 35
Vol. 2, No. 5, May 2010

is the initial time and tα is the alpha phase time. mobility models:
Once the object is moving at an acceptable normal speed Packet Delivery Ratio: It is defined as
then it has to maintain this speed for some time. The middle
smooth phase or β phase is represented by
∑ Number of Received Data Packets
[t
α , t β = tα , tα + β∆t ] where β is uniformly distributed ∑ Number of Sent Data Packets
Average Network Delay: It is defined as
over [βmin, βmax]. As we know that every moving object at
∑ (Time packet arrive @dest − Time packet sent @source )
some point of time has to slow down and eventually stop.
Total Number of Connection Pairs
The node reaches γ phase or slow down phase. The node
Throughput of the network: Throughput is defined as
remains in this phase for a random period of time before the
speed reaches to Vγ = 0. Distance between any two points is ∑ Node Throughputs of Data Transmissi on
treated as the Euclidian distance. Trace length is the actual Total Number of Nodes
length traveled by a node during one movement without any Routing Overhead: It is defined as
change in that direction. Distance evolution is the way the
traveling of a node is observed. If the speed of a node
∑[MAC (Control )
pkt sentsize + MAC (Controlpkt ) fwdSize]
ΔT sim
increases continuously then the α phase and β phase of the
trace length is equivalent to the distance evolution. This Table 1: Simulation Parameters
results in a smooth trace. When the speed of the node Simulator NS2
decreases in the γ phase it results in a sharp curve. SMS
mobility model can also be considered as a group mobility Routing Protocols AOMDV, DYMO, FSR, LAR,
model. In this scenario a node is selected as a leader. This OLSR and TORA
group leader selects a speed and direction which the other Mobility Model Community Mobility Model and
nodes do follow. The velocity of a group member m at its SMS Mobility Model
nth step is given by Simulation Time 900
Vnm = VnLeader + (1 − ρ ) ⋅U ⋅ ∆Vmax (sec)

φnm = φnLeader + (1 − ρ ⋅U ⋅ ∆φmax )

Pause Time (sec) 10
Simulation Area 1000 x 1000
Where U is a random variable, ∆Vmax is the maximum speed (m)
and φmax is the difference in the direction of the common
Number of Nodes 10, 20, 30, 40, 50
node and the leader node in one step [30, 31].
Transmission 250 m
3.9 Energy Model Range
The Energy Consumption model considered in this paper is Maximum Speed 5, 10, 15, 20, 25
based on a study done in [32, 33]. The amount of energy (m/s)
spent in transmitting and receiving the packets is calculated
Traffic Rate 5, 10, 15, 20, 25
by using the following equations:
(pkts/sec)
Energytx = (330*PacketSize)/2*106 Data Payload 512
(Bytes)
Energyrx = (230*PacketSize)/2*106 where packet size is
specified in bits. 5. Result Analysis
All the simulations scenarios are averaged for 5 different
Average Energy Consumed is defined as seeds while running independently. We have considered the
∑ PercentageEnergyConsumedbyallNodes effect of mobility speed, node density and traffic load on the
NumberofNodes performance of these routing protocols using Community
based mobility model and SMS mobility model.
Effect of Mobility Speed
4. Simulation Environment
5.1 Effect of Mobility Speed
Simulations are performed using Network Simulator NS-2
[34]. The simulated values of the radio network interface Both the Mobility speed and the type of Mobility model play
card are based on the 914MHz Lucent WaveLan direct a vital role in determining the performance of routing
sequence spread spectrum radio model. This model has a bit protocols. The mobility speed is varied from 5 m/s to 25 m/s
rate of 2 Mbps and a radio transmission range of 250m. The in steps of 5. The routing protocols LAR, AOMDV and
IEEE 802.11 distributed coordinated function with TORA are compared using the Community Based Mobility
CSMA/CA is used as the underlying MAC protocol. Model and the routing protocols OLSR, DYMO and FSR
Interface Queue (IFQ) value of 70 is used to queue the are compared using SMS mobility model.
routing and data packets. In this paper scheme one implementation of route
Following metrics have been selected for evaluating the establishment is employed in LAR routing protocol. The
36 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

LAR routing protocol has the highest packet delivery at the increase in mobility speed. The delay of FSR is in
varying mobile speed (fig 7). This shows that a location between 0.01 to 0.2 seconds (fig 13). The throughput of
based algorithm like the LAR routing protocol is able to DYMO routing protocols remains steady but it decreases in
take advantage of community model. The LAR routing OLSR and FSR with increase in speed (fig 14).
protocol maintains a delivery ratio of 98% to 70%. Only The throughput results of the routing protocols are in
AOMDV can be comparable to the packet delivery consistent with the packet delivery. But DYMO routing
capability of LAR. But there is an increases and decrease in protocol is having higher overhead over OLSR and FSR
the packet delivery of AOMDV thereby indicating that it is which was not expected considering that DYMO routing
not consistent with the mobility speed of the community protocol performs better than AODV routing protocol (fig
model. 15).
The delay in the network increases with the increase in The routing overhead of DYMO ranges from 45 % to 75 %.
speed. This is in sharp contrast to what can be seen from the This overhead in DYMO is mainly due to the flooding of
increase in node density. This shows that both node density RouteRequest to discover the destination node. This
and mobile speed play a vital role in determining the final RouteRequest message is flooded through out the network
outcome of the performance of a routing protocol (fig 8). thereby causing huge overhead. The overhead in OLSR is
The high throughput of LAR routing protocol results in less mainly due to the large selection of MPR resulting in huge
delay as less number of hops are required to reach the amount of information being sent in the network.
destination. The delay of TORA routing protocol is more In SMS mobility model, the energy consumption of all the
than 0.1 seconds while the delay of LAR routing protocol is routing protocols remains fairly the same with increase in
between 0.01 to 0.02 seconds. The delay in TORA routing mobility speed (fig 16).
protocol may be mainly attributed to the creation of directed
links in the network. The links are created across all the
nodes in the network until the destination node is reached
resulting in higher delay in the network.
The network throughput of LAR routing protocol increases
with increase in mobility speed (fig 9). The results obtained
for LAR routing protocol is ratified with [16] confirming
our stand that the LAR routing protocol goes hand in hand
with the Community model. The throughput of AOMDV
and TORA routing protocols remain between 40 to 50 kbps
while the throughput of LAR routing protocol increases
from 100 to 140 kbps. Figure 7. PDR v/s Average Mobility Speed
The AOMDV and TORA routing protocols score over LAR
routing protocol by having less overhead and less energy
consumption (fig 10 and 11). The routing overhead in LAR
is due to the fact that the expected zone is an estimated one
and sometimes the entire network is considered as an
expected zone. This results in flooding the whole network
with the RouteRequest messages. All the mechanisms
employed in finding the destination zone optimally tends to
fail in this case. This leads to clogging of the whole network Figure 8. Average Network Delay v/s Mobility Speed
thereby increasing the routing packet overhead in the
network. Through simulation it has been showed in
literature that TORA routing protocol consumes more
energy than OLSR. But in the present situation with
Community Mobility model the amount of energy consumed
is almost same between OLSR and TORA routing protocols.
The energy consumption of AOMDV and TORA routing
protocols range between 50 to 100 joules where as the LAR
routing protocol consumes energy between 300 to 350 Figure 9. Network Throughput v/s Mobility Speed
joules.
The DYMO routing protocol has highest packet delivery.
OLSR has an increase and decrease pattern but still it is
comparable with DYMO routing protocol. The packet
delivery is around 90% (fig 12).
FSR routing protocol has least packet delivery. The delay of
OLSR and DYMO routing protocols is between 0.01 to 0.02
seconds. But the delay in FSR is very high. It increases with
 (IJCNS) International Journal of Computer and Network Security, 37
Figure 10. Routing Overhead v/s Mobility Speed
Figure 11. Average Energy Consumed v/s Mobility
Speed
Figure 12. PDR v/s Mobility Speed
Figure 13. Average Network Delay v/s Mobility Speed
Figure 14. Network Throughput v/s Mobility Speed
Figure 15. Routing Overhead v/s Mobility Speed
Vol. 2, No. 5, May 2010
Figure 16. Average Energy Consumed v/s Mobility
Speed
5.2 Effect of Node Density
The number of nodes is varied from 10 to 50 in steps of 10.
Our goal is not to find the optimum number of nodes for
these routing protocols but to see how these routing
protocols fare against different node densities. Higher node
density enables us to analyze whether these routing
protocols are able to take the inherent advantages of a
community model especially the LAR routing protocol. The
routing protocols LAR, AOMDV and TORA are mapped
against number of nodes using Community Based Mobility
Model.
At higher node density the connectivity between nodes is
significantly better resulting in better packet delivery. When
the number of nodes is less the LAR routing protocol has
better packet delivery ratio. But when the number of nodes
is increased to more than 30 then all the routing protocols
converge to the same level. At higher node density all the
routing protocols are able to provide nearly 100% packet
delivery (fig 17).
The TORA routing protocol has the highest delay at lower
node density (fig 18). But as the node increases delay
decreases among all the routing protocols. AOMDV has the
least delay when compared to LAR and TORA routing
protocols. The delay in AOMDV is less due to the
availability of alternate routes thus avoiding the necessity of
sending the RouteRequest packets again and again.
It can be noticed that the throughput of all the protocols
increases with the increase in node density (fig 19). This is
due to increase in network connectivity.
The routing overhead of LAR routing protocol is
astronomically high when compared to AOMDV and TORA
routing protocols. This can be attributed to the amount of
RouteRequest packet that is flooded through out the
network. Even though the LAR routing protocol has various
mechanisms to make the flooding of network an intelligent
one still the amount of packet needed to discover the
destination node in the estimated zone itself results in too
much overhead in the network (fig 20). The routing
overhead of AOMDV is due to the additional RouteRequest
that are needed to maintain multiple path in the network.
The energy consumption of all the routing protocols
increases with the increase in the number of nodes. More
energy is consumed by LAR routing protocol (fig 21). The
increase in the energy consumption in LAR routing protocol
is due to the underlying flooding nature of the routing
protocol.
Our simulation result sprang a few quite surprises. We were
38
expecting TORA to take advantage of its location
capabilities when employed under the Community Mobility
Model. But here the best all round performer was AOMDV
routing protocol. This shows that if were to employ a
routing protocol for analyzing realistic mobility models like
community models then AOMDV is the well versed routing
protocol.
Figure 17. PDR v/s Number of Nodes
Figure 18. Average Network Delay v/s
Number of Nodes
Figure 19. Network Throughput v/s Number
of Nodes
Figure 20. Routing Overhead v/s Number of
Nodes
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
Figure 21. Average Energy Consumed v/s
Number of Nodes
5.3 Effect of Traffic Load
The traffic load is varied from 5 pkts/sec to 25 pkts/sec in
steps of 5. The OLSR, DYMO and FSR routing protocols
are analyzed using SMS mobility model with varying traffic
load. The packet delivery of OLSR and DYMO remains
close to 100% up to 10 pkts/sec (fig 22). With the increase
in traffic load the packet delivery of all the routing protocols
tend to decrease. The DYMO Routing protocol is able to
cope with the congestion in the network better than OLSR
and FSR routing protocols.
The delay in FSR is the worst when compared to OLSR and
DYMO. There is an astronomical increase in delay for FSR
routing protocol starting from 5 pkts/sec and for OLSR the
delay increases after 10 pkts/sec. The DYMO routing
protocol has less delay when compared to other two routing
protocols (fig 23). The delay in DYMO routing protocol
increases after 15 pkts/sec. The delay varies from 0.1
seconds to 0.4 seconds.
Even though DYMO has less delay and more packet
delivery to fails to maintain the momentum for network
throughput (fig 24). Surprisingly both OLSR and FSR
routing protocols score over DYMO in network throughput.
The throughput of all the routing protocols increases with
the increase in the traffic load.
The routing overhead of DYMO increases with increase in
traffic load while the routing overhead of OLSR and FSR
decreases with increase in traffic load. FSR is having less
routing overhead. This factor can be attributed to the unique
feature in FSR where it maintains the information of its
immediate neighboring nodes thereby it is able to find the
destination node with minimum number of hops (fig 25).
DYMO protocol’s routing overhead increases up to 15
pkts/sec and maintain a steady state after that indicating that
the network is saturated. The routing overhead of OLSR and
FSR decreases with the increase in traffic load. The routing
overhead of FSR decreases from 10% to 4% with the
increase in traffic load.
Both OLSR and FSR have high energy consumption. Above
10 pkts/sec the energy consumption of OLSR increases at a
very high rate. But the energy consumption of FSR routing
is more than 300 joules starting from 5 pkts/sec (fig 26).
DYMO routing protocol maintains steady energy
consumption between 150 to 200 joules.
For SMS mobility model if high packet delivery, less delay
and less energy consumption are the criterion to employ a
routing protocol then the routing protocol of choice should
 (IJCNS) International Journal of Computer and Network Security, 39
be the DYMO routing protocol.
Figure 22. PDR v/s Traffic Load
Figure 23. Average Network Delay v/s Traffic Load
Figure 24. Network Throughput v/s Traffic Load
Figure 25. Routing Overhead v/s Traffic Load
Figure 26. Average Energy Consumed v/s Traffic
Load
6 Conclusion
In this paper we have evaluated six routing protocols over
two realistic mobility models. We have conducted these
experiments in best possible scenarios conductive for an ad
hoc network to understand the performance of these routing
protocols over different mobility model. The LAR routing
protocol is the protocol of choice for a community model
moving at high speed. We came across few surprises like
Vol. 2, No. 5, May 2010
AOMDV performing better than the LAR routing protocol
at higher node density for a Community Based Mobility
Model. The DYMO routing protocol outperforms both
OLSR and FSR with varying mobility speed and traffic load.
At the end of the day the type of routing protocols that needs
to be deployed depends on the situation in hand in addition
to the various parameters given in our paper.
Acknowledgments
The first author of this paper would like to acknowledge the
help of Tracy Camp, Wenye Wang, Mirco Musolesi in this
work.
References
[1] A.K.Dwivedi et al.,”Performance of Routing Protocols
for Mobile Adhoc and Wireless Sensor Networks: A
Comparative Study”, International Journal of Recent
Trends in Engineering, Vol 2, No. 4, November 2009.
[2] Nicholas Cooper and Natarajan Meganathan,”Impact
of Mobility Models on Multi Path Routing in Mobile
Ad Hoc Networks”, International Journal of
Computer Networks and Communications (IJCNC),
Vol 2, No 1, January 2010.
[3] Anders Nilsson,”Performance Analysis of Traffic Load
and Node Density in Ad hoc Networks”, In
Proceedings of the 5th European Wireless Conference
(EW2004) Mobile and Wireless Systems, Barcelona,
Spain, Feb 24-27, 2004.
[4] M.I.M Saad et al,“Performance Analysis of Random
Based Mobility Models in MANET Routing
Protocol”, European Journal of Scientific Research,
Vol 32, No. 4, pp 444-454.
[5] Biao Zhou, Kaixin Xu and Mario Gerla,”Group and
Swarm Mobility Models for Ad hoc Network
Scenarios Using Virtual Tracks”, IEEE Military
Communications Conference (MILCOM’04),
Moneterey, California, USA, Oct 31-Nov 3, 2004.
[6] Sven Jaap, Marc Bechler and Lars Worf,”Evaluation of
Routing Protocols for Vehicular Ad hoc Networks in
City Traffic Scenarios”, In Proceedings of the 5th
International Conference o Intelligent Transportation
Systems Telecommunications (ITST), Brest, France,
June 2005.
[7] Mahesh K. Marina and Samir R.Das, “ On-Demand
Multipath Distance Vector Routing in Ad-hoc
Networks”, In proceedings of the IEEE International
Conference on Network Protocols, 2001 pp 14-23.
[8] M.Caleffi, G.Ferraiuolo and L.Pauro, “On Reliability
of Dynamic Addressing Protocols in Mobile Ad hoc
Networks”, Proceedings of WRECOM’07, Wireless
Rural and Emergency Communications Conference,
Roma, Italy, October 2007.
[9] Ian D Chakeres and Charles E Perkins,”Dynamic
Manet On Demand Routing Protocol”, IETF Internet
Draft, draft-ietf-manet-dymo-12.txt, February 2008.
[10] Rolf Ehrenreich Thorup,”Implementing and
Evaluating the DYMO Routing Protocol”, Master’s
Thesis, Department of Comptuer Science, University
of Aarhus, Denmark.
40
[11] Siti Rahayu Abdul Aziz et al.”Performance
Evaluation of AODV, DSR and DYMO Routing
Protocols in MANET”, 2009 International
Conference on Scientific and Social Science
Research, Kaula Lampur, Malaysia, 14-15 March.
[12] Guangyu Pei, Mario Gerla and Tsu Wei
Chen,”Fisheye State Routing: A Routing Scheme for
Ad Hoc Wireless Networks”, In Proceedings of IEEE
International Conference on Communications
(ICC2000), New Orleans, Louisiana, USA, June 18-
22, 2000.
[13] Elizabeth M Royer, Sung-Ju Lee and Charles E
Perkins,”The Effects of MAC Protocols on Ad hoc
Network Communication”, In Proceedings of IEEE
WCNC 2000, Chicago, IL, September 2000.
[14] Allen C Sun,”Design and Implementation of Fisheye
Routing Protocol for Mobile Ad hoc Networks”,
http://www.scanner-group.mit.edu/PDFS/SunA.pdf
[15] C.Siva Rama Murthy and B.S. Manoj, “Adhoc
Wireless Networks: Architectures and Protocols”,
Second Edition, Prentice Hall.
[16] Tracy Camp, Jeff Boleng et al.,”Performance
Comparison of Two Location Based Routing
Protocols for Ad Hoc Networks”, In Proceedings of
the IEEE 21st Annual Joint Conference of the IEEE
Computer and Communications Societies (INFOCOM
2002), pp 167801687, 2002.
[17] Young Bae Ko and Nitin H Vaidya,”Location Aided
Routing (LAR) in Mobile Ad Hoc Networks”, In
Proceedings of the 4th Annual ACM/IEEE
International Conference on Mobile Computing and
Networking, Dallas, Texas, United States, 1998.
[18] Ioannis Broustis, Gentian Jakllari et al.,”A
Comprehensive Comparison of Routing Protocols for
Large Scale Wireless MANETs”, In Proceedings of
the 3rd Inernational Workshop on Wireless Ad Hoc
and Sensor Networks, Newyork, June 28-30 2006.
[19] S.K.Sarkar, T.G.Basavaraju et al,”Ad hoc Mobile
Wireless Networks, Principles, Protocols and
Applications”, Auerbach Publications -2008.
[20] T.Clausen and P.Jacquet,”Optimized Link State
Routing Protocol for Ad hoc Networks”,
http://hipercom.inria.fr/olsr/rfc3626.txt
[21] Mounir Frikha and Manel Maamer,” Implementation
and Simulation of OLSR protocol with QOS in ad hoc
Networks”, Proceedings of the Second International
Symposium on Communications, Control and Signal
Processing (ISCCSP’06), 13-15 March, Marrakech,
Morocco.
[22] John Novatnack and Harpreet Arora,”Evaluating ad
hoc Routing Protocols with Restpect to Quality of
Service”, IEEE International Conference on Wireless
and Mobile Computing, Networking and
Communications (WiMob’05), 03 October 2005.
[23] Kenneth Holter,”Comparing AODV and OLSR”,
folk.uio.no/kenneho/studies/essay/essay.html
[24] Ying Ge et al,”Quality of Service Routing in ad Hoc
Networks Using OLSR”, Proceedings of the 36th
IEEE International conference on System Sciences
(HICSS’03), 2002.
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
[25] Vincent D Park and M. Scott Corson,”A Performance
Comparison of the Temporally Ordered Routing
Algorithm and Ideal Link State Routing”, In
Proceedings of the 3rd IEEE Symposium on
Computers and Communications, Washington D.C,
USA, 1998.
[26] Vincent D Park, J P Macket and M Scott
Corson,”Applicability of the Temporally Ordered
Routing Algorithm for use in Mobile Tactical
Networks”, IEEE Military Communications
Conference (MILCOM 98), 1998.
[27] Eryk Dutkiewicz et al,”A Review of Routing
Protocols for Mobile Ad Hoc Networks”, Ad hoc
Networks, Volume 2, Issue 1, January 2004, Pages 1-
22.
[28] Newman and Girvan,”Finding and evaluating
community structure in networks”, Physical Review
E, 69, Februrary 2004.
[29] Mirco Musolesi and Cecilia Mascolo, “A Community
Based Mobility Model for Ad hoc Network Research”,
International Symposium on Mobile Ad hoc
Networking and Computing, Proceedings of the 2nd
International Workshop on Multi hop ad hoc
networks, 2006.
[30] Ming Zhao and Wenye Wang,”A Novel Semi Markov
Smooth Mobility Model for Mobile Ad Hoc
Networks”, In Proceedings of the IEEE
GLOBECOM’06, San Francisco, CA, November
2006.
[31] Maig Zhao and Wenye Wang,”Design and
Applications of a Smooth Mobility Model for Mobile
Ad hoc Networks”, In Proceedings of the IEEE
Milcom’06, Washington D.C., October 2006.
[32] Juan Carlos Cano and Pietro Manzoni,”A
Performance Comparison of Energy Consumption for
Mobile Ad Hoc Network Routing Protocols”, In
Proceedings of the 8th International Symposium on
Modelling, Analysis and Simulation of Computer and
Telecommunication Systems, San Francisco, CA,
2000.
[33] Laura Marie Feeney,”Investigating the Energy
Consumption of a Wireless Network Interface in an
Adhoc Networking Environment”, In Proceedings of
the INFOCOM01 Twentieth Annual Joint Conference
of the IEEE Computer and Communication Societies,
Vol 3, August 2002, pp 1548-1557.
[34] Information Sciences Institute, “The Network
Simulator Ns-2”, Http://www.isi.edu/nanam/ns/,
University of Southern California.
 (IJCNS) International Journal of Computer and Network Security, 41
Vol. 2, No. 5, May 2010

Security Issues for Voice over IP Systems

Ali Tahir1, Asim Shahzad2
1
Faculty of Telecommunication and Information Engineering, UET TAXILA,
Pakistan
ali.tahir.nsn@hotmail.com

2
Faculty of Telecommunication and Information Engineering, UET TAXILA,
Pakistan
asim.shahzad@uettaxila.edu.pk

Abstract: Firewalls, Network Address Translation (NAT), and

encryption produce Quality of Service (QoS) issues which are
the basic building block to the operations of a VOIP system.
There are two major non proprietary standards used for VoIP
communications. They are H.323 and Session Initiation
Protocol (SIP). Firewalls, NAT, and Intrusion Detection
Systems (IDS) are used to filter unwanted packets in VoIP
environment. To provide defense against an internal hacker,
another layer of defense is necessary at the protocol level to
protect the data itself. In VOIP, as in data networks, this can be
accomplished by encrypting the packets at the IP level using
IPSec. As VoIP is an IP based technology that utilizes the
Internet it also inherits all associated IP vulnerabilities. This
research paper focus on different security issues associated with Figure 1. How VOIP works [1]
voice over IP system.
The VoIP networks replace the traditional public-switched
Keywords: Voice-Over Internet Protocol (VoIP), Firewalls. telephone networks (PSTNs), as these can perform the same
functions as the PSTN networks. The functions performed
include signaling, data basing, call connect and disconnect,
1. Overview of VOIP and coding-decoding.

The Voice-Over Internet Protocol (VoIP) technology allows

the voice information to pass over IP data networks. This
technology results in huge savings on the amount of
physical resources required to communicate by voice over
long distance. It does so by exchanging the information in
packets over a data network. The basic functions performed
by a VoIP include - signaling, data basing, call connect and
disconnect, and coding/decoding. The steps involved in
originating and internet telephone call are the conversion of
the analogue voice signal to digital format and compression
/ translation of the signal into internet protocol (IP) packets
for transmission over the internet; the process is reversed at
the receiving end. VoIP software’s like Vocal TEC or Net 2
Phone are available for the user [1]. With the exception of
phone to phone, the user must posses an array of equipment
which should at minimum include VoIP software, an
internet connection, and a multimedia computer with a
sound card, speakers, a microphone and a modem [1].The
VoIP network acts as a gateway to the existing PSTN
network. This gateway forms the interface for transportation
of the voice content over the IP networks. Gateways are
responsible for all call origination, call detection, analogue
to digital conversion of voice, and creation of voice packets.
2. Quality of Service Issues
Quality of Service (QoS) is fundamental to the operation of
a VOIP network. Despite all the money VOIP can save users
and the network elegance it provides, if it cannot deliver at
least the same quality of call setup and voice relay
functionality and voice quality as a traditional telephone
network, then it will provide little added value. There are
various security measures that can degrade QoS. For
example blocking of call setups by firewalls, encryption-
produced latency and delay variation (jitter). QoS issues are
central to VOIP security. If QoS was assured, then most of
the same security measures currently implemented in
today’s data networks could be used in VOIP networks. But
because of the time-critical nature of VOIP, and its low
tolerance for disruption and packet loss, many security
measures implemented in traditional data networks just
aren’t applicable to VOIP in their current form. The main
QoS issues associated with VOIP that security affects are
presented here [4].
2.1 Latency
Latency in VOIP refers to the time it takes for a voice
transmission to go from its source to its destination. We
would like to keep latency as low as possible for ideal
42
systems but there are practical lower bounds on the delay of
VOIP. The ITU-T Recommendation G.114 establishes a
number of time constraints on one-way latency. The upper
bound is 150 ms in [2] for one-way traffic experienced in
domestic calls across PSTN lines in the continental United
States. For international calls, a delay of up to 400 ms was
deemed tolerable, but since most of the added time is spent
routing and moving the data over long distances, we
consider here only the domestic case and assume our
solutions are upwards compatible in the international realm.
VOIP calls must achieve the 150 ms bound to successfully
emulate the QoS that today’s phones provide. It places a
genuine constraint on the amount of security that can be
added to a VOIP network. The encoding of voice data can
take between 1 and 30 ms and voice data traveling across
the North American continent can take upwards of 100 ms
although actual travel time is often much faster . Assuming
the worst case (100 ms transfer time), 20 –50 ms remain for
queuing and security implementations [4].
Figure 2. Sample Latency Budget [4]
Delay is not confined to the endpoints of the system. Each
hop along the network introduces a new queuing delay and
possibly a processing delay if it is
a security checkpoint (i.e. firewall or encryption/decryption
point). Also, larger packets tend to cause bandwidth
congestion and increased latency. In light of these issues,
VOIP tends to work best with small packets on a logically
abstracted network to keep latency at a minimum.
2.2 Jitter
Jitter refers to non-uniform packet delays. It is often caused
by low bandwidth situations in VOIP and can be
exceptionally detrimental to the overall QoS. Variations in
delays can be more detrimental to QoS than the actual
delays themselves. Jitter can cause packets to arrive and be
processed out of sequence. RTP, the protocol used to
transport voice media, is based on UDP so packets out of
order are not reassembled at the protocol level. However,
RTP allows applications to do the reordering using the
sequence number and timestamp fields. The overhead in
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
reassembling these packets is non-trivial, especially when
dealing with the tight time constraints of VOIP.
When jitter is high, packets arrive at their destination in
spurts. The general prescription to control jitter at VOIP
endpoints is the use of a buffer, but such a buffer has to
release its voice packets at least every 150 ms (usually a lot
sooner given the transport delay) so the variations in delay
must be bounded. The buffer implementation issue is
compounded by the uncertainty of whether a missing packet
is simply delayed an anomalously long amount of time, or is
actually lost.
Jitter can also be controlled throughout the VOIP network
by using routers, firewalls, and other network elements that
support QoS. These elements process and pass along time
urgent traffic like VOIP packets sooner than less urgent data
packets. Another method for reducing delay variation is to
pattern network traffic to diminish jitter by making as
efficient use of the bandwidth as possible. This constraint is
at odds with some security measures in VOIP. Chief among
these is IPSec, whose processing requirements may increase
latency, thus limiting effective bandwidth and contributing
to jitter.
The window of delivery for a VOIP packet is very small, so
it follows that the acceptable variation in packet delay is
even smaller. Thus, although we are concerned with
security, the utmost care must be given to assuring that
delays in packet deliveries caused by security devices are
kept uniform throughout the traffic stream [4].
2.3 Packet Loss
VOIP is exceptionally intolerant of packet loss. Packet loss
can result from excess latency, where a group of packets
arrives late and must be discarded in favor of newer ones. It
can also be the result of jitter, that is, when a packet arrives
after its surrounding packets have been flushed from the
buffer, making the received packet useless. Compounding
the packet loss problem is VOIP’s reliance on RTP, which
uses the unreliable UDP for transport, and thus does not
guarantee packet delivery. However, the time constraints do
not allow for a reliable protocol such as TCP to be used to
deliver media. The good news is that VOIP packets are very
small, containing a payload of only 10-50 bytes, which is
approximately 12.5-62.5 ms in [4], with most
implementations tending toward the shorter range. The loss
of such a minuscule amount of speech is not discernable or
at least not worthy of complaint for a human VOIP user.
The bad news is these packets are usually not lost in
isolation. Bandwidth congestion and other such causes of
packet loss tend to affect all the packets being delivered
around the same time. So although the loss of one packet is
fairly inconsequential, probabilistically the loss of one
packet means the loss of several packets, which severely
degrades the quality of service in a VOIP network.
Despite the infeasibility of using a guaranteed delivery
protocol such as TCP, there are some remedies for the
packet loss problem. One cannot guarantee all packets are
delivered, but if bandwidth is available, sending redundant
information can probabilistically annul the chance of loss.
 (IJCNS) International Journal of Computer and Network Security, 43
Such bandwidth is not always accessible and the redundant
information will have to be processed, introducing even
more latency to the system and ironically, possibly
producing even greater packet loss.
2.4 Bandwidth & Effective Bandwidth
As in data networks, bandwidth congestion can cause packet
loss and a host of other QoS problems. Thus, proper
bandwidth reservation and allocation is essential to VOIP
quality. One of the great attractions of VOIP, data and voice
sharing the same wires, is also a potential headache for
implementers who must allocate the necessary bandwidth
for both networks in a system normally designed for one.
Congestion of the network causes packets to be queued,
which in turn contributes to the latency of the VOIP system.
Low bandwidth can also contribute to non-uniform delays
(jitter) [1], since packets will be delivered in spurts when a
window of opportunity opens up in the traffic.
Because of these issues, VOIP network infrastructures must
provide the highest amount of bandwidth possible. On a
LAN, this means having modern switches running at 100M
bit/sec and other architectural upgrades that will alleviate
bottlenecks within the LAN. Percy and Hommer suggest
that if network latencies are kept below 100 milliseconds,
maximum jitter never more than 40 milliseconds, then
packet loss should not occur. With these properties assured,
one can calculate the necessary bandwidth for a VOIP
system on the LAN in a worst case scenario using statistics
associated with the worst-case bandwidth congesting codec.
This is fine when dealing simply with calls across the LAN,
but the use of a WAN complicates matters. Bandwidth usage
varies significantly across a WAN, so a much more complex
methodology is needed to estimate required bandwidth
usage.
Methods for reducing the bandwidth usage of VOIP include
RTP header compression and Voice Activity Detection
(VAD). RTP compression condenses the media stream
traffic so less bandwidth is used. However, an inefficient
compression scheme can cause latency or voice degradation,
causing an overall downturn in QoS. VAD prevents the
transmission of empty voice packets (i.e. when a user is not
speaking, their device does not simply send out white noise).
However, by definition VAD will contribute to jitter in the
system by causing irregular packet generation [4].
2.5 The Need for Speed
The key to conquering QoS issues like latency and
bandwidth congestion is speed. By definition, faster
throughput means reduced latency and probabilistically
reduces the chances of severe bandwidth congestion. Thus
every facet of network traversal must be completed quickly
in VOIP. The latency often associated with tasks in data
networks will not be tolerated. Chief among these latency
producers that must improve performance are firewall/NAT
traversal and traffic encryption/decryption [4]. Traditionally,
these are two of the most effective ways for administrators to
secure their networks. However, they are also two of the
greatest contributors to network congestion and throughput
delay. Inserting traditional firewall and encryption products
Vol. 2, No. 5, May 2010
into a VOIP network is not feasible, particularly when VOIP
is integrated into existing data networks. Instead, these data-
network solutions must be adapted to support security in the
new fast paced world of VOIP.
2.6 Power Failure and Backup Systems
Conventional telephones operate on 48 volts supplied by the
telephone line itself. This is why home telephones continue
to work even during a power failure. Most offices use PBX
systems with their conventional telephones, and PBXs
require backup power systems so that they continue to
operate during a power failure [4]. These backup systems
will continue to be required with VOIP, and in many cases
will need to be expanded. An organization that provides
uninterruptible power systems for its data network and
desktop computers may have much of the power
infrastructure needed to continue communication functions
during power outages, but a careful assessment must be
conducted to ensure that sufficient backup power is available
for the office VOIP switch, as well as each desktop
instrument. Costs may include electrical power to maintain
UPS battery charge, periodic maintenance costs for backup
power generation systems, and cost of UPS battery
replacement. If emergency/backup power is required for
more than a few hours, electrical generators will be
required. Costs for these include fuel, fuel storage facilities,
and cost of fuel disposal at end of storage life.
2.7 QoS Implications for Security
The strict performance requirements of VOIP have
significant implications for security, particularly denial of
service (DoS) issues. VOIP-specific attacks (i.e., floods of
specially crafted SIP messages) may result in DoS for many
VOIP-aware devices. For example, SIP phone endpoints
may freeze and crash when attempting to process a high rate
of packet traffic SIP proxy servers also may experience
failure and intermittent log discrepancies with a VOIP-
specific signaling attack of under 1Mb/sec. In general, the
packet rate of the attack may have more impact than the
bandwidth; i.e., a high packet rate may result in a denial of
service even if the bandwidth consumed is low [4].
3. Standards and Protocols
Over the next few years, the industry will address the
bandwidth limitations by upgrading the Internet backbone to
asynchronous transfer mode (ATM), the switching fabric
designed to handle voice, data, and video traffic. Such
network optimization will go a long way toward eliminating
network congestion and the associated packet loss. The
Internet industry also is tackling the problems of network
reliability and sound quality on the Internet through the
gradual adoption of standards. Standards-setting efforts are
focusing on the three central elements of Internet telephony:
the audio codec format; transport protocols; and directory
services.
In May 1996, the International Telecommunications Union
(ITU) ratified the H.323 specification, which defines how
44
voice, data, and video traffic will be transported over IP–
based local area networks; it also incorporates the T.120
data conferencing standard. The recommendation is based
on the real-time protocol/real-time control protocol
(RTP/RTCP) for managing audio and video signals.
As such, H.323 addresses the core Internet-telephony
applications by defining how delay-sensitive traffic, (i.e.,
voice and video), gets priority transport to ensure real-time
communications service over the Internet [1].
Figure 3: H.323 Architecture [4]
4. Firewalls, NAT AND IDS
Firewalls and NAT present a formidable challenge to VOIP
implementers. However, there are solutions to these
problems, if one is willing to pay the price. It is important to
note that all three major VOIP protocols, SIP, H.323, and
H.248 all have similar problems with firewalls and NATs.
Although the use of NATs may be reduced as IPv6 is
adopted in [4], they will remain a common component in
networks for years to come, and IPv6 will not alleviate the
need for firewalls, so VOIP systems must deal with the
complexities of firewalls and NATs. Some VOIP issues with
firewalls and NATs are unrelated to the call setup protocol
used. Both network devices make it difficult for incoming
calls to be received by a terminal behind the firewall / NAT.
Also, both devices affect QoS and can impose strong inflict
with the RTP stream.
4.1 Firewalls
Firewalls are a staple of security in today’s IP networks.
Whether protecting a LAN, WAN, encapsulating a DMZ, or
just protecting a single computer, a firewall is usually would
be the first line of defense against attackers. Traffic not
meeting the requirements of the firewall is dropped.
Processing of traffic is determined by a set of rules
programmed into the firewall by the network administrator.
These may include such commands as “Block all FTP traffic
(port 21)” or “Allow all HTTP traffic (port 80)”. Much more
complex rule sets are available in almost all firewalls. A
useful property of a firewall, in this context, is that it
provides a central location for deploying security policies. It
is the ultimate bottleneck for network traffic because when
properly designed, no traffic can enter or exit the LAN
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
without passing through the firewall. The introduction of
firewalls to the VOIP network complicates several aspects of
VOIP, most notably dynamic port trafficking and call setup
procedures.
4.2 Stateful Firewalls
Most VOIP traffic travels across UDP ports. Firewalls
typically process such traffic using a technique called packet
filtering. Packet filtering investigates the headers of each
packet attempting to cross the firewall and uses the IP
addresses, port numbers, and protocol type contained therein
to determine the packets’ legitimacy [4]. In VOIP and other
media streaming protocols, this information can also be used
to distinguish between the start of a connection and
established connection.There are two types of packet
filtering firewalls, stateless and stateful. Stateless firewalls
retain no memory of traffic that has occurred earlier in the
session. Stateful firewalls do remember previous traffic and
can also investigate the application data in a packet. Thus,
stateful firewalls can handle application traffic that may not
be destined for a static port.
4.1.1 VOIP specific Firewall Needs
In addition to the standard firewall practices, firewalls are
often deployed in VOIP networks with the added
responsibility of brokering the data flow between the voice
and data segments of the network. This is a crucial
functionality for a network containing PC-Based IP phones
that are on the data network, but need to send voice
messages. All voice traffic emanating from or traveling to
such devices would have to be explicitly allowed in if no
firewall was present because RTP (Real Time Protocol)
makes use of dynamic UDP ports (of which there are
thousands). Leaving this many UDP ports open is an
egregious breach of security. Thus, it is recommended that
all PC-based phones be placed behind a stateful firewall to
broker VOIP media traffic. Without such a mechanism, a
UDP DoS attack could compromise the network by
exploiting the excessive amount of open ports [10].
Firewalls could also be used to broker traffic between
physically segmented traffic (one network for VOIP, one
network for data) but such an implementation is fiscally and
physically unacceptable for most organizations, since one of
the benefits of VOIP is voice and data sharing the same
physical network.
4.2 Network Address Translation (NAT)
NAT is commonly performed by firewalls to conserve IP
addresses and hide internal IP addresses/ports from direct,
external access. This causes issues for VoIP. When VoIP
endpoints negotiate ports for media exchange, they
communicate these ports to one another in packet payloads.
So a conventional NAT remains unaware of ports selected,
cannot translate them meaningfully and blocks them at the
firewall. A Real Time Mixed Media (RTMM) firewall uses
the Back-to-Back User Agent (B2BUA) to work with the
NAT to rewrite the addresses in the signaling stream,
providing NAT pathways for the media streams. To do this,
 (IJCNS) International Journal of Computer and Network Security, 45
the RTMM must be able to perform NAT at media speeds,
to prevent latency/jitter or loss of packets [11].
All of the benefits of NAT come at a price. NATs “violate
the fundamental semantic of the IP address, that it is a
globally reachable point for communications”. This design
has significant implications for VOIP. For one thing, an
attempt to make a call into the network becomes very
complex when a NAT is introduced. The situation is
analogous to a phone network where several phones have
the same phone number, such as in a house with multiple
phones on one line (see Figure 4). There are also several
issues associated with the transmission of the media itself
across the NAT, including an incompatibility with IPSec
[4].
Figure 4. IP Telephones behind NAT and Firewall [4]
Problem:
• Simple NAT devices, which are not VoIP aware, perform
NAT on the IP headers only.
• VoIP packets contain private IP address in the payload.
• Therefore, VoIP sessions cannot be established.
Solution:
• Translates both the IP header and the packet’s payload
with a routable IP address (Far End NAT)
• Media relay until full session establishment.
Session Border Controller (SBC)
• SBC’s started as a solution to connectivity problems
caused by NAT done by non-VoIP aware devices.
• SBC’s are usually used by carriers and located at the
border of their core networks [7].
Figure 5. Far End NAT [7]
4.3 Intrusion Detection System (IDS)
Vol. 2, No. 5, May 2010
Intrusion detection is a second line of defense behind other
security mechanisms (Firewalls, Encryption). Supplying the
detector engine with application specific knowledge makes
it more effective and powerful.
An Intrusion Detection System (IDS) helps administrators to
monitor and defend against security breaches. Intrusion
detection techniques are generally divided into two
paradigms, anomaly detection and misuse detection. In
anomaly detection techniques, the deviation from normal
system behaviors is detected, whereas misuse detection is
based on the matching of attack signatures. Unlike
signature-based intrusion detection, anomaly detection has
the advantage of detecting previously-unknown attacks but
at the cost of relatively high false alarm rate.
Sekar etal introduced a third category of specification-based
intrusion detection. Specification-based approach takes the
manual development of a specification that captures
legitimate system behavior and detects any deviation
thereof. This approach can detect unseen attacks with low
false alarm rate. However, these previous approaches fall
short of defending VoIP applications, because of the cross-
protocol interaction and distributed nature of VoIP.
Figure 6. IDS Engine sits on or close to the end-point [12]
VoIP systems use multiple protocols for call control and
data delivery. For example, in SIP-based IP telephony,
Session Initiation Protocol (SIP) is used to control call setup
and teardown, while Real-time Transport Protocol (RTP) is
for media delivery. A VoIP system is distributed in nature,
consisting of IP phones, SIP proxies, and many other
servers. Defending against malicious attacks on such a
heterogeneous and distributed environment is far from
trivial [2].
4.3.1 Cross-protocol Methodology for Detection
Is the IDS which use cross-protocol detection accesses
packets from multiple protocols in a system to perform its
detection. This methodology is suitable to systems that use
multiple protocols and where attacks spanning these
multiple protocols are possible. There is the important
design consideration that such access to information across
protocols must be made efficiently. A VoIP system
incorporates multiple protocols. A typical example is the use
of SIP to establish a connection, followed by use of RTP to
transfer voice data. Also, RTCP and ICMP are used to
monitor the health of the connection. VoIP systems typically
have application level software for billing purposes and
therefore may have accounting software and a database.
Using the cross-protocol methodology for detection, one can
create a cross-protocol rule to look at the SIP messages, the
46
transaction messages between the accounting software and
the database, and the RTP flows later on. Specifically, each
of the following three conditions must hold [12].
1. The SIP message should follow the correct format.
2. When the accounting software sends out a transaction
to denote a call from user A to user B, check if user A has
sent a SIP Call Initialization message to user B. If user A
has not set up the call with a legitimate SIP Call
Initialization message, then this condition will be violated.
3. Check the source/destination IP addresses of the
subsequent RTP flows. Together with information from
DNS and SIP Location Servers, we can reconfirm that each
RTP flow has a corresponding legitimate call setup.
4.3.2 Stateful Methodology for Detection
A second abstraction useful for VoIP systems in particular is
stateful detection. Stateful detection implies building up
relevant state within a session and across sessions and using
the state in matching for possible attacks. It is important
that the state aggregation be done efficiently so that the
technique is applicable in high throughput systems, such as
VoIP systems.
A VoIP system maintains considerable amount of system
state. The client side maintains state about all the active
connections – when the connection was initiated, when it
can be torn down, and what the properties of the connection
are. The server side also maintains state relevant to billing,
such as the duration of the call.
5. Encryption and IPsec
The only focus on security of the network, protecting
endpoints, other components, from malicious attacks is not
enough for VoIP systems. Firewalls, IDS, NAT, and other
such devices can help keep intruders from compromising a
network, but firewalls are no defense against an internal
hacker in [4]. Another layer of defense is necessary at the
protocol level to protect the data itself. In VOIP, as in data
networks, this can be accomplished by encrypting the
packets at the IP level using IPsec. This way if anyone on
the network, authorized or not, intercepts VOIP traffic not
intended for them, these packets will be impossible to
understand. The IPsec suite of security protocols and
encryption algorithms is the standard method for securing
packets against unauthorized viewers over data networks
and will be supported by the protocol stack in IPv6. Hence,
it is both logical and practical to extend IPsec to VOIP,
encrypting the signal and voice packets on one end and
decrypting them only when needed by their intended
recipient.
Also, several factors, including the expansion of packet size,
ciphering latency, and a lack of QoS urgency in the
cryptographic engine itself can cause an excessive amount of
latency in the VOIP packet delivery. This leads to degraded
voice quality, so once again there is a tradeoff between
security and voice quality, and a need for speed. Fortunately,
the difficulties are not insurmountable. IPsec can be
incorporated into a SIP network with roughly a three-second
additional delay in call setup times, an acceptable delay for
many applications. This section explains the issues involved
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
in successfully incorporating IPsec encryption into VOIP
services.
5.1 IPSec
IPsec is the preferred form of VPN tunneling across the
Internet. There are two basic protocols defined in IPsec:
Encapsulating Security Payload (ESP) and Authentication
Header (AH) (see Figure 7). Both schemes provide
connectionless integrity, source authentication, and an anti-
replay service. The tradeoff between ESP and AH is the
increased latency in the encryption and decryption of data in
ESP and a “narrower” authentication in ESP, which
normally does not protect the IP header “outside” the ESP
header, although Internet Key Exchange (IKE: Responsible
for key agreement using public key cryptography) can be
used to negotiate the security association (SA), which
includes the secret symmetric keys. In this case, the
addresses in the header (transport mode) or new/outer
header (tunnel mode) are indirectly protected, since only the
entity that negotiated the SA can encrypt/decrypt or
authenticate the packets. Both schemes insert an IPsec
header (and optionally other data) into the packet for
purposes, such as authentication [4].
IPsec also supports two modes of delivery: Transport and
Tunnel. Transport mode encrypts the payload (data) and
upper layer headers in the IP packet. The IP header and the
new IPsec header are left in plain sight. So if an attacker
were to intercept an IPsec packet in transport mode, they
could not determine what it contained; but they could tell
where it was headed, allowing rudimentary traffic analysis.
On a network entirely devoted to VOIP, this would equate to
logging which parties were calling each other, when, and
for how long. Tunnel mode encrypts the entire IP datagram
and places it in a new IP Packet [9]. Both the payload and
the IP header are encrypted. The IPsec header and the new
IP Header for this encapsulating packet are the only
information left in the clear.
Figure 7. IPsec Tunnel and Transport Modes [9]
IPSec supports the Triple DES encryption algorithm (168-
bit) in addition to 56-bit encryption. Triple DES (3DES) is a
strong form of encryption that allows sensitive information
to be transmitted over untrusted networks. It enables
 (IJCNS) International Journal of Computer and Network Security, 47
customers, particularly in the finance industry, to utilize
network layer encryption.
Table 1: Encryption Algorithms for VoIP [9]
Encryption Algorithms for IPSec
Type Key Strengt
h computations [4].
DES Symmetric 56-bit Weak
3DES Symmetric 168-bit Mediu
m
AES Symmetric 128, 192, or 256- Strong
bit
RSA Asymmetric 1024-bit minimum Strong
5.2 The Role of IPSec in VOIP
The prevalence and ease of packet sniffing and other
techniques for capturing packets on an IP based network
makes encryption a necessity for VOIP. Security in VOIP is
concerned both with protecting what a person says as well as
to whom the person is speaking. IPSec can be used to
achieve both of these goals as long as it is applied with ESP
using the tunnel method. This secures the identities of both
the endpoints and protects the voice data from prohibited
users once packets leave the corporate intranet. The
incorporation of IPSec into Ipv6 will increase the
availability of encryption, although there are other ways to
secure this data at the application level. VOIPsec (VOIP
using IPSec) helps reduce the threat of man in the middle
attacks, packet sniffers, and many types of voice traffic
analysis. Combined with the firewall implementations, Ipsec
makes VOIP more secure than a standard phone line. It is
important to note, however, that Ipsec is not always a good
fit for some applications, so some protocols will continue to
rely on their own security features.
5.3 Local VPN Tunnels
Virtual Private Networks (VPNs) are “tunnels” between two
endpoints that allow for data to be securely transmitted
between the nodes. The IPSec ESP tunnel is a specific kind
of VPN used to traverse a public domain (the Internet) in a
private manner. The use and benefits of VPNs in IPSec have
been great enough for some to claim “VOIP is the killer app
for VPNs”. VPN tunnels within a corporate LAN or WAN
are much more secure and generally faster than the IPSec
VPNs across the Internet because data never traverses the
public domain, but they are not scaleable [4]. Also, no
matter how the VPN is set up, the same types of attacks and
issues associated with IPSec VPNs are applicable, so we
consider here only the case of IPSec tunneling and assume
the security solutions can be scaled down to an internal
network if needed.
5.4 Memory and CPU Considerations
Packets that are processed by IPSec are slower than packets
that are processed through classic crypto. There are several
reasons for this and they might cause significant
performance problems:
1. IPSec introduces packet expansion, which is more likely
Vol. 2, No. 5, May 2010
to require fragmentation and the corresponding reassembly
of IPSec datagrams.
2. Encrypted packets are probably authenticated, which
means that there are two cryptographic operations that are
performed for every packet.
3. The authentication algorithms are slow, although work
has been done to speed up things as the Diffie−Hellman
5.5 Difficulties Arising from VOIPsec
IPSec has been included in IPv6. It is a reliable, robust, and
widely implemented method of protecting data and
authenticating the sender. However, there are several issues
associated with VOIP that are not applicable to normal data
traffic. Of particular interest are the Quality of Service
(QoS) issues and some others like latency, jitter, and packet
loss. These issues are introduced into the VOIP environment
because it is a real time media transfer, with only 150 ms to
deliver each packet. In standard data transfer over TCP, if a
packet is lost, it can be resent by request. In VOIP, there is
no time to do this. Packets must arrive at their destination
and they must arrive fast. Of course the packets must also be
secure during their travels. However, the price of this
security is a decisive drop in QoS caused by a number of
factors.
A study by researchers focused on the effect of VOIPsec on
various QoS issues and on the use of header compression as
a solution to these problems. They studied several codecs,
encryption algorithms, and traffic patterns to garner a broad
description of these effects. Some empirical results
developed by Cisco are available as well in [4].
Delay
• Processing—PCM to G.729 to packet
• Encryption — ESP encapsulation + 3DES
• Serialization — time it takes to get a packet out of the
router, each “hop” generally has fixed delay.
• IPsec overhead: about 40 bytes (depending configuration)
• IP header: 20 bytes
• UDP + RTP headers: 20 bytes
• RTP header compression: 3 bytes for IP+UDP+RTP
Effects on 8 kbps CODEC (voice data: 20 bytes)
• clear text voice has an overhead of 3 bytes, which
suggests required bandwidth of approximately 9 kbps
• IPsec encrypted voice: overhead 80 bytes and required
bandwidth 40 kbps
5.6 Encryption / Decryption Latency
The cryptographic engine is bottleneck for voice traffic
transmitted over IPsec. The driving factor in the degraded
performance produced by the cryptography was the
scheduling algorithms in the crypto-engine itself However,
there still was significant latency due to the actual
encryption and decryption.
Encryption/decryption latency is a problem for any
cryptographic protocol, because much of it results from the
computation time required by the underlying encryption.
With VOIP’s use of small packets at a fast rate and
48
intolerance for packet loss, maximizing throughput is
critical. However, this comes with a price, because although
DES is the fastest of these encryption algorithms, it is also
the easiest to crack [4]. Thus, designers are once again
forced to toe the line between security and voice quality.
5.7 Expanded Packet Size
IPsec also increases the size of packets in VOIP, which leads
to more QoS issues. The increase is actually just an increase
in the header size due to the encryption and encapsulation of
the old IP header and the introduction of the new IP header
and encryption information. This leads to several
complications when IPsec is applied to VOIP. First, the
effective bandwidth is decreased as much as 63% [4]. Thus
connections to single users in low bandwidth areas (i.e. via
modem) may become infeasible. The size discrepancy can
also cause latency and jitter issues as packets are delayed by
decreased network throughput or bottlenecked at hub nodes
on the network (such as routers or firewalls).
5.8 IPsec and NAT Incompatibility
IPsec and NAT compatibility is far from ideal. NAT
traversal completely invalidates the purpose of AH because
the source address of the machine behind the NAT is
masked from the outside world. Thus, there is no way to
authenticate the true sender of the data. The same reasoning
demonstrates the inoperability of source authentication in
ESP. We have defined this as an essential feature of
VOIPsec, so this is a serious problem. There are several
other issues that arise when ESP traffic attempts to cross a
NAT. If only one of the endpoints is behind a NAT, the
situation is easier. If both are behind NATs, IKE negotiation
can be used for NAT traversal, with UDP encapsulation of
the IPsec packets.
6. Solutions to the VOIPsec Issues
We have raised a number of significant concerns with
IPsec’s role in VOIP. However, many of these technical
problems are solvable. Despite the difficulty associated with
these solutions it is very important for the establishment of a
secure implementation of VOIPsec.
6.1 Encryption at the End Points
One proposed solution to the bottlenecking at the routers
due to the encryption issues is to handle
encryption/decryption solely at the endpoints in the VOIP
network. One consideration with this method is that the
endpoints must be computationally powerful enough to
handle the encryption mechanism. Though ideally
encryption should be maintained at every hop in a VOIP
packet’s lifetime, this may not be feasible with simple IP
phones with little in the way of software or computational
power. In such cases, it may be preferable for the data be
encrypted between the endpoint and the router (or vice
versa) but unencrypted traffic on the LAN is slightly less
damaging than unencrypted traffic across the Internet [4].
Fortunately, the increased processing power of newer
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
phones is making endpoint encryption less of an issue.
6.2 Secure Real Time Protocol (SRTP)
The Secure Real-time Protocol is a profile of the Real-time
Transport Protocol (RTP) offering not only confidentiality,
but also message authentication, and replay protection for
the RTP traffic as well as RTCP (Real-time Transport
Control Protocol).
SRTP provides a framework for encryption and message
authentication of RTP and RTCP streams. SRTP can
achieve high throughput and low packet expansion.
SRTP is independent of a specific RTP stack
implementation and of a specific key management standard,
but Multimedia Internet Keying (MIKEY) has been
designed to work with SRTP.
In comparison to the security options for RTP there are
some advantages to using SRTP. The advantages over the
RTP standard security and also over the H.235 security for
media stream data are listed below [4, 12].
SRTP provides increased security, achieved by
• Confidentiality for RTP as well as for RTCP by
encryption of the respective payloads;
• Integrity for the entire RTP and RTCP packets, together
with replay protection;
• The possibility to refresh the session keys periodically,
which limits the amount of cipher text produced by a fixed
key, available for an adversary to cryptanalyze;
• An extensible framework that permits upgrading with
new cryptographic algorithms;
• A secure session key derivation with a pseudo-random
function at both ends;
• The usage of salting keys to protect against pre-
computation attacks;
• Security for unicast and multicast RTP applications.
SRTP has improved performance attained by
• Low computational cost asserted by pre-defined
algorithms;
• Low bandwidth cost and a high throughput by limited
packet expansion and by a framework preserving RTP
header compression efficiency;
• Small footprint that is a small code size and data memory
for keying information and replay lists.
6.3 Key Management for SRTP – MIKEY
SRTP uses a set of negotiated parameters from which
session keys for encryption, authentication and integrity
protection are derived. MIKEY describes a key management
scheme that addresses real-time multimedia scenarios (e.g.
SIP calls and RTSP sessions, streaming, unicast, groups,
multicast). The focus lies on the setup of a security
association for secure multimedia sessions including key
management and update, security policy data, etc., such that
requirements in a heterogeneous environment are fulfilled.
MIKEY also supports the negotiation of single and multiple
crypto sessions.
 (IJCNS) International Journal of Computer and Network Security, 49
MIKEY has some important properties in [4]:
• MIKEY can be implemented as an independent software
library to be easily integrated in a multimedia
communication protocol. It offers independency of a specific
communication protocol (SIP, H.323, etc.)
• Establishment of key material within a 2-way handshake,
therefore best suited for real-time multimedia scenarios
There are four options for Key Distribution:
• Pre-shared key
• Public-key encryption
• Diffie-Hellman key exchange protected by public-key
encryption
• Diffie-Hellman key exchange protected with pre-shared-
key and keyed hash functions (using an MIKEY extension
(DHHMAC))
6.4 Better Scheduling Schemes
Without a way for the crypto-engine to prioritize packets,
the engine will still be susceptible to DoS attacks and
starvation from data traffic impeding the time-urgent VOIP
traffic. A few large packets can clog the queue long enough
to make the VOIP packets over 150 ms late (sometimes
called head-of-line blocking), effectively destroying the call.
Ideally, the crypto-engine would implement QoS scheduling
to favor the voice packets, but this is not a realistic scenario
due to speed and compactness constraints on the crypto-
engine. One solution implemented in the latest routers is to
schedule the packets with QoS in mind prior to the
encryption phase.
It is not surprising that for voice traffic the crypto-engine
can be a serious bottleneck. Rather than the expected
constraints on the crypto-engine throughput, the critical
factor turned out to be the impossibility to control and
schedule access to the crypto-engine so as to favor real-time
traffic over regular one. This applies regardless of whether
the scheduler is implemented as a software module or a
hardware component. Therefore, if voice traffic is
interleaved with other types of traffic, e.g., ftp or http traffic,
during a secure session, it may happen that the latter
(usually characterized by big packets) is scheduled in the
crypto-engine before voice traffic. In this case voice traffic
might be delayed to the point that packets are discarded
most of the times.
6.5 Compression of Packet Size
Compression of Packet Size turn results in considerably less
jitter, latency, and better crypto-engine performance. There
is, of course, a price for these speedups. The compression
scheme puts more strain on the CPU and memory
capabilities of the endpoints in order to achieve the
compression, and, of course, both ends of a connection must
use the same compression algorithm.
Efficient solution for packet header compression is known
as cIPsec, for VoIPsec traffic. Simulation results from
different sources show that the proposed compression
scheme significantly reduces the overhead of packet headers,
thus increasing the effective bandwidth used by the
Vol. 2, No. 5, May 2010
transmission. In particular, when cIPsec is adopted, the
average packet size is only 2% bigger, rather than 50%
longer plain VoIPsec packets, which makes VoIPsec and
VoIP equivalent from the bandwidth usage point of view.
However, packet loss does have an exacerbated detrimental
effect on packets compressed under the cIPsec scheme.
When packets are lost, they cannot be re-sent and the
endpoints need to resynchronize. However, the time saved in
the crypto-engine and the security provided may be well
worth this price of this approach [5].
6.6 Resolving NAT/IPSec Incompatibilities
The most likely widespread solution to the problem of NAT
traversal is UDP encapsulation of IPSec. This
implementation is supported by the IETF and effectively
allows all ESP traffic to traverse the NAT. In tunnel mode,
this model wraps the encrypted IPSec packet in a UDP
packet with a new IP header and a new UDP header, usually
using port 500. This port was chosen because it is currently
used by IKE peers to communicate so overloading the port
does not require any new holes to be punched in the
firewall. The SPI field within the UDP-encapsulated packet
is set to zero to differentiate it from an actual IKE
communication. This solution allows IPsec packets to
traverse standard NATs in both directions. The adoption of
this standard method should allow VOIPsec traffic to
traverse NATs cleanly, although some extra overhead is
added in the encapsulation/decapsulation process. IKE
negotiation will also be required to allow for NAT traversal.
The problem still remains that IP-based authentication of
the packets cannot be assured across the NAT, (although
fully qualified domain names could be used) but the use of a
shared secret (symmetric key) negotiated through IKE could
provide authentication. It is important to note that IP-based
authentication is weak compared with methods using
cryptographic protocols.
There are several other solutions to the IPsec/NAT
incompatibility problem, including Realm-Specific IP
RSIP), IPv6 Tunnel Broker, IP Next Layer (IPNL), and UDP
encapsulation. RSIP is designed as a replacement for NAT
and provides a clear tunnel between hosts and the RSIP
Gateway. RSIP supports both AH and ESP, but
implementing RSIP would require a significant overhaul of
the current LAN architecture so while it is quite an elegant
solution, it is currently infeasible. Perhaps as a result of
these problems, RSIP is not widely used. The IPv6 tunnel
broker method uses an IPv6 tunnel as an IPSec tunnel, and
encapsulates an IPv6 packet in an IPv4 packet. But this
solution also requires LAN upgrades and doesn’t work in
situations where multiple NATs are used. IPNL introduces a
new layer into the network protocols between IP and
TCP/UDP to solve the problem, but IPNL is in competition
with IPv6 and IPv6 is a much more widely used standard
[4].
50 (IJCNS) International Journal of Computer and Network Security,
7. Security Threats in VoIP
In the early days of VoIP, there was no big concern about
security issues related to its use. People were mostly
concerned with its cost, functionality and reliability. Now
that VoIP is gaining wide acceptance and becoming one of
the mainstream communication technologies, security has
become a major issue.
The security threats cause even more concern when we think
that VoIP is in fact replacing the oldest and most secure
communication system the world ever known – POTS (Plain
Old Telephone System). Let us have a look at the threats
VoIP users face.
7.1 Identity and Service Theft
Service theft can be exemplified by phreaking, which is a
type of hacking that steals service from a service provider,
or use service while passing the cost to another person.
Encryption is not very common in SIP, which controls
authentication over VoIP calls, so user credentials are
vulnerable to theft.
Eavesdropping is how most hackers steal credentials and
other information. Through eavesdropping, a third party can
obtain names, password and phone numbers, allowing them
to gain control over voicemail, calling plan, call forwarding
and billing information. This subsequently leads to service
theft.
Stealing credentials to make calls without paying is not the
only reason behind identity theft. Many people do it to get
important information like business data.
A phreaker can change calling plans and packages and add
more credit or make calls using the victim’s account. He can
of course as well access confidential elements like voice
mail, do personal things like change a call forwarding
number [10].
7.2 Vishing
Vishing is another word for VoIP Phishing, which involves
a party calling you faking a trustworthy organization (e.g.
your bank) and requesting confidential and often critical
information.
7.3 Viruses and Malware
VoIP utilization involving soft-phones and software are
vulnerable to worms, viruses and malware, just like any
Internet application. Since these soft-phone applications run
on user systems like PCs and PDAs, they are exposed and
vulnerable to malicious code attacks in voice applications.
7.4 DoS (Denial of Service)
A DoS attack is an attack on a network or device denying it
of a service or connectivity. It can be done by consuming its
bandwidth or overloading the network or the device’s
internal resources.
In VoIP, DoS attacks can be carried out by flooding a target
with unnecessary SIP call-signaling messages, thereby
Vol. 2, No. 5, May 2010
degrading the service. This causes calls to drop prematurely
and halts call processing.
Why would someone launch a DoS attack? Once the target
is denied of the service and ceases operating, the attacker
can get remote control of the administrative facilities of the
system [10, 12].
7.5 Spamming over Internet Telephony (SPIT)
If you use email regularly, then you must know what
spamming is. Put simply, spamming is actually sending
emails to people against their will. These emails consist
mainly of online sales calls. Spamming in VoIP is not very
common yet, but is starting to be, especially with the
emergence of VoIP as an industrial tool.
Every VoIP account has an associated IP address. It is easy
for spammers to send their messages (voicemails) to
thousands of IP addresses. Voice mailing as a result will
suffer. With spamming, voicemails will be clogged and
more space as well as better voicemail management tools
will be required. Moreover, spam messages can carry viruses
and spyware along with them. [10]
7.6 Call Tampering
Call tampering is an attack which involves tampering a
phone call in progress. For example, the attacker can simply
spoil the quality of the call by injecting noise packets in the
communication stream. He can also withhold the delivery of
packets so that the communication becomes spotty and the
participants encounter long periods of silence during the
call.
7.7 Man-in-the-Middle Attacks
VoIP is particularly vulnerable to man-in-the-middle
attacks, in which the attacker intercepts call-signaling SIP
message traffic and masquerades as the calling party to the
called party, or vice versa. Once the attacker has gained this
position, he can hijack calls via a redirection server [10].
8. Conclusion
This paper provides an overview of the VoIP and of the
issues related to VoIP security. There are numerous
challenges to the secure implementation, deployment and
use of the VoIP on a same network which is used to carry
other network data. These challenges are related to the
service availability, Quality of Service, Firewalls, Intrusion
Detection, IPSec, VoIP threats, and privacy. The next step
for the ongoing research project is to identify specific areas
that will be addressed in the near term research efforts.
References
[1] Rinzam A. VoIP: Voice Over Internet Protocol.
Department of Computer Science and Engineering, M
G College of Engineering.
[2] H. Sengar, D. Wijesekera. Center for Secure Information
Systems, George Mason University, Fairfax, USA. H.
Wang, S. Jajodia. Department of CS, College of
William and Mary, Williamsburg, USA. “IDS: VoIP
 (IJCNS) International Journal of Computer and Network Security, 51
Vol. 2, No. 5, May 2010

Intrusion Detection through Interacting Protocol State Authors Profile

Machines”.
[3] M. Marjalaakso. Security Requirements and Constraints Ali Tahir is an MS Scholar at University of
of VoIP. Helsinki University of Technology, Engineering and Technology, Taxila, Pakistan.
Department of Electrical Engineering and He did his B.Sc Software Engineering from
Telecommunications. UET TAXILA in 2005. He has worked in
[4] D. Richard Kuhn, Thomas J. Walsh, and Steffen Fries, Nokia Siemens Network for two years as an
implementation engineer. Currently he is
“Security Considerations for Voice Over IP Systems:
working in cryptography and network security
Recommendations of the National Institute of Standards center at a Government organization. His areas
and Technology” Special Publication 800-58, Sections 8 of interest are digital image processing, computer vision, network
and 9, January 2005. security and wireless communication.
[5] R. Barbieri, D. Bruschi, E. Rosti. Voice over IPsec:
Analysis and Solutions. Department of Science,
University of Milano, Itlay. Asim Shahzad is pursuing his PhD from
[6] G. Egeland, “Introduction to IPsec in IPv6”. University of Engineering and Technology
http://www.eurescom.de/~publicwebdeliverables/P1100 Taxila, Pakistan. He has also completed his
series/P1113/D1/pdfs/pir1/41_IPsec_intro.pdf MS in Computer Engineering from UET
[7] Check Point Software Technologies Ltd. “Check Point TAXILA. He has done MS in
Solution for Secure VoIP”. 2003 - 2007. Telecommunication Engineering from
[8] NAT: Network Address Translation and Voice Over institute of communication technologies,
Internet Protocol. http://www.voip- Islamabad, Pakistan. He is working as
Assistant Professor at UET TAXILA in Telecommunication
info.org/wiki/view/NAT+and+VOIP Engineering Department. His areas of interest are optical
[9] Jeremy Stretch. IPSEC: Internet Protocol Security. communication, Data and network Security.
http://www. packetlife.net
[10] M. Hurley. “VoIP VULNERABILITIES”. Centre for
Critical Infrastructure Protection (CCIP). Information
Note, Issue 06, Wellington, New Zealand, January
2007.
[11] Mark D. Collier. Firewall Requirements for Securing
VoIP, SecureLogix Corporation.www.securelogix.com
[12] Yu-Sung Wu, S. Bagchi. School of Electrical &
Computer Eng, Purdue University, S. Garg, N. Singh,
Tim Tsai, Avaya Labs. “A Stateful and Cross Protocol
Intrusion Detection Architecture for VoIP
Environments”.
52 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

A Novel Run-Time Memory Leak Detection and

Recovery (MLDR) Using Aging in Physical
Memory Space
Ahmed Otoom1, Mohammad Malkawi2 and Mohammad Alweh3
1
Royal Jordanian Air Force, IT Directorate,
Amman 11134, Jordan
aotoom@rjaf.mil.jo
2
Middle East University
Amman, Jordan
mmalkawi@aimws.com
3
Southern Illinois State University
Carbondale, Illinois, USA
Alweh@siu.edu

Abstract: This paper provides a novel run-time approach for
memory leak detection and recovery (MLDR) using aging in
physical memory space. MLDR algorithm reflects both the
physical and virtual behavior of memory allocation and benefits
from the hardware support available for tracking physical pages
in real memory in order to detect leak in virtual address space.
MLDR detects stale and unreachable memory objects. A
conservative implementation of MLDR recovers only
unreachable leaky objects in order to avoid the potential false
positives stale objects. MLDR has a coarse granularity in the
sense that it deals with pages rather than memory chunks, thus
enabling it to avoid costly scans of heap allocation. Simulation
results show the effectiveness of MLDR in detecting and
recovering memory leaks at run time.
Keywords: memory leak detection, memory leak recovery,
memory aging, and dynamic memory management.
1. Introduction
Several approaches for solving memory leak have been
developed including static analysis tools, dynamic tools, and
automatic dynamic tools. Static analysis tools [1], [7] are
used to identify leaks in development environment before
executing the program; they do not cause any run time
overhead. These tools lack dynamic information so they
produce false positives and can not find all leaks. Some
dynamic tools [10] analyze memory consumption
interactively by inspecting and comparing snapshots of the
heap. They allow for the search for causes of leaked objects
at runtime environments and facilitate the use of garbage
collection. Automatic dynamic tools report the leaked
objects and leak sites at the end of the program run. Some
tools [4] instrument allocation functions to report immortal
objects. Insure++ [8] uses a reference counting garbage
collection approach to detect where the last reference to an
object disappears. Others [11] instrument allocation function
and use mark and sweep garbage collectors to capture
unreachable objects. These tools either miss many leaks or
incur high overhead cost. Besides operating outside the run
time environment, the current approaches have relatively
high overhead cost. The concept of detecting leak based on
object access is used in SWAT [15]. SWAT is an automatic
dynamic memory leak detection tool that traces memory
allocations/frees to build a heap model and uses an adaptive
statistical profiling approach to monitor loads/stores to these
objects. Objects that are not accessed for a long time are
reported as leaks. This tool detects leaks with a less than
10% false positive rate. Leaks are reported at the end of an
application run; SWAT can not decide on its own to
deallocate them.
Different businesses can deal with software failures in
different ways. In some cases, system administrators simply
restart the system whenever the memory leaks to a point
where a crash is imminent or performance degrades beyond
an acceptable level. Systems with critical applications can
not tolerate the cost of frequent shutdowns or performance
degradation. For example, rebooting and restarting the
software systems such as the Patriot Software [3] every 8
hours was responsible for losing lives during the first Gulf
War. The consequences of unresolved memory leaks in real-
time systems can have a direct impact on human safety,
security and business sustainability.
This paper introduces a new memory leak detection and
recovery (MLDR) algorithm. MLDR provides a run-time
solution; it helps recover leaky objects which results in
enhancing the system performance and making mission
critical programs live longer. The main concept underlying
the design of MLDR is the concept of memory objects aging
in the physical space rather than in the virtual space.
Software aging, in general, refers to source contention
issues that can cause performance degradation or can cause
systems to hang, panic, or crash. Software aging
mechanisms can include memory leaks, unreleased file
locks, accumulation of unterminated threads, data
corruption/round-off accrual, file space fragmentation,
shared memory pool latching, and others [6], [13], [14].
Parnas [9] documents software aging as a phenomenon
wherein legacy software becomes difficult to maintain as it
ages.
Aging, in this paper, refers to the time a piece of memory
object remains untouched. A leaky object by definition will
 (IJCNS) International Journal of Computer and Network Security, 53
begin to age since it will no longer be accessed by any
application program. Aging, in this context, can then be
used to detect memory leakage. Memory leaks and aging
refer to objects in the heap virtual space. Detecting leaky
objects by estimating their age in the virtual space [15] is
time consuming. However, detecting the age of a page in the
physical space is much less time consuming and can be done
using the already available support for virtual memory
systems. MLDR makes use of page aging in the physical
space to detect actual leaks in the virtual space. Hence,
MLDR can be used to detect and recover leaks at run time
reducing the risk of performance degradation or crashes.
This paper is organized as follows. Section 2 presents the
new memory leak detection and recovery algorithm
(MLDR). Section 3 presents a simulation model to measure
the performance and effectiveness of MLDR. Section 4
presents a summary of the conclusions.
2. Memory Leak Detection and Recovery
(MLDR) Using Aging in Physical Memory
2.1 Background
A memory object is said to have leaked if it is not freed by
the program or the system after it becomes no longer used or
accessible. The main job for a memory leak detector is to
determine whether an object has leaked at a given point of
time. In most general terms, memory leak occurs because of
1) unreachable objects: a program either unintentionally or
maliciously neglects to free heap-allocated objects and
therefore these objects are lost. In this case, there is no
reference chain for these objects from the root set starting
from global and stack variables and 2) useless objects: a
program maintains references to objects which are never
used again.
The memory leak detection and recovery (MLDR)
algorithm is able to detect both types (unreachable and
useless/stale objects). In this paper, we present two versions
of MLDR; a conservative one which recovers only
unreachable objects, and a liberal version which recovers
both unreachable and stale objects. The conservative
approach does not produce false positives, i.e, falsely
identified leaks. The liberal approach produces false
positives, mostly related to stale objects. Recovering stale
objects without incurring false positives is more complicated
and requires significant architectural changes, and will not
be discussed in this paper.
MLDR has a coarse granularity in the sense that it deals
with pages instead of memory chunks. Hence, we introduce
the term “leaky page” where a leaky page is a page which
contains potentially leaky chunks. A leaky page may contain
any combination of the following types of objects: 1)
unreachable objects and 2) stale objects. In the context of
this paper, stale objects are objects that have not been
accessed for a given period of time. They are either a)
useless objects or b) active but have not been referenced for
that period of time. In our approach, the MLDR leak
detection mechanism identifies the potential leaky pages.
Depending on the page size and the sizes of allocated
memory chunks, a page may contain several memory
Vol. 2, No. 5, May 2010
chunks or the page may constitute a portion of one memory
chunk.
The recovery mechanism of the conservative MLDR
identifies unreachable objects from the set of leaky pages.
The remaining objects are stale objects. The unreachable
objects are relinquished (freed) and returned to the operating
system.
MLDR is particularly useful in long lived applications.
As time passes, age accumulates and MLDR will be able to
identify a set of unreachable objects. These objects will be
freed which makes enough room for new allocations and
enables the program to continue execution. Another
consequence of memory leak is thrashing. Memory leak can
cause extreme low locality of reference as live objects are
dispersed over a large virtual address space. MLDR will
relinquish unreachable locations causing the operating
system to reuse these locations and enhance the locality of
reference as a result.
2.2 Description of MLDR
Once a memory object has leaked, it will no longer be
accessed by the application program. Hence, a leaky
memory object will begin to age. Aging, in this context, is
related to the time a piece of memory remains untouched. In
virtual memory computer systems, memory allocation is
done both at the virtual and physical levels. A memory
leakage in the virtual space will render the corresponding
mapped physical memory in a “page-out status”. A page-out
status makes a page in the physical memory a target for the
replacement policy that swaps the target page out to disk.
Going backward, a swapped out page can correspond, but
not necessarily, to a leaked chunk in the virtual space. We
define “the age of a page” as the time elapsed since the page
is swapped out from the physical to the virtual space. More
precisely, the age of a page begins to accumulate from the
time a page is marked by the replacement policy as a target
for replacement. Note that a page may remain in the
physical space for a long time after it has been marked as a
replaceable page. A memory chunk will be considered a
candidate leak if the age of the page to which it belongs
exceeds a certain limit (threshold). This threshold can be
either user-defined or tuned by a telemetry tool.
Memory leak detection policies can rely on memory
aging either in the virtual or physical space. Aging in the
virtual space is more accurate than aging in the physical
space; however, using the concept of physical space aging
has several advantages. The physical space is much smaller
than the virtual address space, which makes the time
required to scan for leaky pages much less than the time
needed to search for leaky chunks in the virtual space. Also,
tracing the age of a page in the physical space can greatly
benefit from the available hardware support for virtual
memory organization such as page tables and address
translation buffers. Of course, the age of a page in the
physical space may not be entirely due to the behavior of the
owner of the page; it might well depend on the behavior of
other programs as well as on the page replacement policy
(LRU, FIFO, OPT, LFU) [12] deployed by the operating
system.
It should be noted that several chunks, whose sizes are
smaller than a page, may be mapped to one physical page. If
54
only one of these chunks remains active, while the other
chunks have leaked, then the corresponding page will be
considered active and the leaks in that page will not be
detected; this case is known as “false negative”. We show in
this paper how the problem of false negatives can be
reduced.
2.3 MLDR Algorithm Main Functions
MLDR algorithm utilizes the following functions: Main(),
Initialize(), Bookkeeping(), StalenessDetector(),
StalenessRecovery(), and new versions of malloc() and free()
functions.
2.3.1 Main() and Initialize() Functions
The “main()” function of the algorithm keeps iterating over
monitored processes. Main() calls two functions: Initialize()
and Bookkeeping(). For each monitored process, the
Initialize() function sets some user-configurable parameters
that affect the behavior of the MLDR algorithm. These
parameters can be user-configurable or they can be
automatically tuned using a telemetry tool. These
parameters are: upper heap size threshold(UHST), lower
heap size threshold(LHST), staleness detector sleeps
time(SDST), and a normalized parameter(K). The UHST is
used to initiate the leak detection process once the size of
the heap for the monitored process exceeds the UHST limit.
The LHST is used to suspend the detection and recovery
process once the current heap size drops below LHST. The
SDST is the time (in milliseconds) a StalenessDetector()
will wait between any two successive scans to the page table
of a given process. The page age threshold(PAT) is used to
identify aged pages. When the age of a page exceeds PAT, it
will be marked as a candidate leaky page. PAT is calculated
dynamically as an accumulated average and is fine-tuned
using the K parameter.
2.3.2 Bookkeeping() Function
The page table is augmented with a new entry called
timestamp(TS). For every victim page selected from the
mapped physical space, the bookkeeping() function
timestamps the corresponding page in the virtual address
space by setting the TS to the current time. The TS is used
to calculate the age of a page. Whenever a page is paged
back into physical memory, the TS entry is reset to zero.
Pages whose TS value is larger than zero begin to age as
time progresses.
2.3.3 Memory Allocation and Deallocation Functions
For implementation purposes, we need to modify memory
allocation and deallocation functions, e.g., malloc() and
free(). Other functions are treated in a similar manner, and
hence will not be provided in this paper.
(a) malloc () Function
The malloc() function incorporates two major changes
before it returns a pointer to the newly allocated memory
chunk. The first, Malloc() checks the heap size and if the
UHST is reached, it invokes the StalenessDetector(). The
StalenessDetector() is suspended when the heap size reaches
the lower limit LHST. This modification to malloc() helps
reducing the overhead associated with leak detection and
recovery.
The latter, the execution of malloc() updates the entries
of a table called “MallocTable” which includes the address
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
of the allocated chunk, its size and a flag. The flag is
initially set to one, indicating that the object is alive and
reachable. The flag is reset to zero by the
StalenessRecovery() function as a result of scanning the
stack and registers. If a memory chunk in the MallocTable
is not matched by any address value in the stack and address
registers, then the corresponding flag is set to zero, and the
chunk is deemed unreachable.
(b) free() Function
Before deallocating a chunk of memory and exiting, the
free() function removes the deallocated chunk from the
MallocTable; these are objects with flag value reset to zero.
2.3.4 StalenessDetector() Function
The StalenessDetector() traverses the page table of a given
process in order to identify leaky pages. It identifies all
pages whose age has exceeded the page age threshold
(PAT). Note that any identified page at this phase is
considered to be only a candidate leaky page, meaning that
it may or may not contain leaky chunks. A memory chunk
contained in a leaky page could in fact be active, but has not
been referenced for a relatively long time, such that the
aging algorithm marked their corresponding page as a leaky
page; this is a case of “false positive”.
2.3.5 StalenessRecovery() Function.
StalenessRecovery() works on the set of identified leaky
pages(aged pages). Its main task is to remove unreachable
objects, given that a leaky page is found to include a leak. In
order to identify the set of chunks in a leaky page, the
StalenessRecovery() intersects the MallocTable with the
range of addresses in that page [page_number *
PAGE_SIZE, (page_number+1)*PAGE_SIZE-1]. The result
of intersection (working set), will contain the potential leaky
chunks. Unreachable chunks in the working set will be
identified by scanning the roots (static, stack, and registers)
searching for reachable objects in the given set of leaky
pages. Once reachable objects are identified, the remaining
objects in this set are definitely unreachable. Identified
unreachable chunks are passed to the free function to be
deallocated from the virtual address space and from the
MallocTable, respectively. The remaining chunks in the
working set are stale objects. The conservative version of
MLDR ignores all stale objects in order to avoid running the
risk of false positives.
2.4 False Positives and False Negatives
2.4.1 False Positives
One of the main problems in memory leak detection tools is
the potential error known as “false positives”. In other
words, a detected leak is not a real leak. Since MLDR is a
run-time solution it is not recommended to deallocate
chunks which may turn out to be false positives. This is
especially true for systems running critical applications.
Referencing an object after it has been removed or
deallocated leads to incorrect results or program crash.
MLDR identifies potential leaky pages which might contain
false positive objects. In order to avoid the consequences of
false positives, the set of leaky pages are passed to the
StalenessRecovery() which identifies and recovers
unreachable objects. Of course, the leaky but reachable
(stale) objects will persist in the system and will be counted
as false negatives. The liberal approach would consider all
 (IJCNS) International Journal of Computer and Network Security, 55
Vol. 2, No. 5, May 2010

objects in a leaky page as real leaks. This approach will replacement and the current time is written into its
reduce the number of false negatives at the expense of corresponding page table entry (TS).
generating false positives. One way of eliminating the Memory allocation and deallocation are made using the
negative impact of false positives is to keep a copy of modified malloc() and free() functions as explained earlier.
removed objects on disk, such that the copy can be recovered For memory access event, we developed a function that
and reused in case the removed object is dereferenced. This simulates demand paging with global replacement strategy.
approach will not be further investigated in this paper. The bookkeeping part of the algorithm is implemented
within the scope of the page replacement policy. The
2.4.2 False Negatives simulator implements the least recently used (LRU) strategy.
False negatives are leaky chunks that go undetected. Note
that a page allocated to physical memory may consist of one Run the
start Make necessa ry cha nges
benchmark
or more memory objects in the heap virtual space. If at least
one of these objects remains active, then the corresponding Create trace
page may never age and the MLDR will not be able to files
identify it as a leaky page. As a result, the leaky objects,
which happen to be allocated to the same page, will go Read one NO
undetected. This phenomenon will result in false negatives, trace file a t a
time
i.e., undetected leaky objects.
MLDR may reduce the number of false negatives by Validate trace file using
decreasing the page age threshold allowing more pages to be universally available if results
tools mtrace() and are valid
identified as leaky at the expense of increasing the overhead dma lloc
cost. The number of false negatives can be further reduced Yes
by choosing a smaller page size such that the number of
independent objects allocated to the same page will be Leak injection
small. Also, in the conservative implementation of the
algorithm, where only unreachable objects are detected, the
stale objects will go undetected and they will count towards No more traces
the false negatives. The liberal approach, which treats all While there Print output
are more parameters
objects in a leaky page as real leaks, eliminates this type of traces
false negatives at the risk of running into false positives.
end

3. Performance Evaluation Determine next event

A trace-driven simulation is used to evaluate the MLDR

algorithm. Figure 1 shows an abstract block diagram for the
top-level of the simulation program. The trace-driven
simulation program consists of: trace data collection, leak
injection, and MLDR modeling. Data collection is carried
out using synthetic benchmarks which simulate the behavior
of real application programs in terms of memory allocation,
deallocation, and referencing. The benchmarks are validated
against well known tools such as mtrace() [5] and dmalloc()
[2]. Leak injection is simulated by disabling some free()
functions in the trace files.
Unreachable objects will age because they are no longer
accessed by the application and they will be detected by the
MLDR aging algorithm. The injected leak is known in
advance, so the MLDR effectiveness will be determined
based on how much of the injected leak is recovered.
The simulator iterates over a given trace file. At each
iteration, the simulator processes one of three events:
memory allocation, memory free, or memory access
(load/store). Allocating memory results in increasing the
heap size and consuming virtual address space. Freeing
memory saves virtual address space. A memory reference to
a page that has been allocated to physical memory requires
no further action. A reference to a page not allocated to
physical memory results in a page fault. The required page
is paged-in and its time stamp (TS) is reset to zero. If the
main physical memory is full, a victim page is selected for
Memory alloca tion event
Memory Dea llocation event
Load/store event
Figure 1. Trace-driven simulation model
The next event in the simulator is determined by the next
row in the trace file. Each row contains a sign, return
address, and a size. A “+” means allocation, and “-”means
deallocation. The remaining values represent the return
address in hexadecimal format and the size of each allocated
or deallocated chunk. The distribution of malloc(), free()
and memory access are controlled as random processes with
different probability distribution functions. This gives the
simulator flexibility to model real life applications; where
memory allocation and deallocation are process dependent.
3.1 Results and Discussion
Figure 2 shows the growth of the heap size for process (P0)
and process (P1), where MLDR is turned off and on
respectively. The heap size for P0 continues to grow beyond
the maximum allowed upper limit; in real life, the program
will not be able to run beyond (t1), where the allocated heap
size equals the maximum heap size. With MLDR turned on,
P1 continued to run until (t2) before the allocated heap size
exceeds the upper limit threshold. The figure shows how P1
56
heap allocation oscillates between the upper and lower heap
size thresholds, due to the activation and deactivation of the
StalenessDetector() and StalenessRecovery() functions. This
experiment shows how MLDR extends the life of a process
due to keeping the allocated heap size below an upper limit
threshold.
Time Vs Heap size
MaxHeapSize
Upper_Heap_Size_Threshold
(80%)
Lower_Heap_Size_Threshold
(50%)
Heap size
S t1 ∆t t2
Time
Figure 2. Time vs. heap size
Figure 3 shows the impact of the page age threshold on
the number of false negatives. As expected, the number of
false negatives increases with the increase in the page age
threshold. With relatively small page age, many pages will
be identified as potential leaks. All unreachable objects will
be identified. Stale but reachable objects will persist in the
system. In the conservative implementation of the
algorithm, only unreachable objects will be removed from
the system. For relatively large age threshold, the number of
pages which are deemed leaky is relatively small. As a
result, smaller number of leaks will be detected.
Page Age Threshold Vs False Negatives
False Negatives
Number of
Page_Age_Threshold
Figure 3. Page age threshold vs. false negatives
The liberal implementation of the algorithm allows the
system to treat all objects contained in pages whose age has
surpassed the age threshold as real leaks. This
implementation is less costly because it does not require a
scan of the stack and registry entries. However, it runs the
risk of generating false positives, especially for smaller
values of age threshold.
Figure 4 shows the number of false positives versus the
age threshold. The advantage of this implementation is that
it detects all type of leaks without having to incur the cost of
scanning the stack and registry. The disadvantage though is
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
running the risk of false positives which if exist can cause
the system to fail. The conservative implementation is
preferred when the system runs critical applications. The
liberal implementation may be used in systems running less
critical applications or systems used for testing applications
before deployment.
Page Age Threshold Vs
False Positives
Number of False Positives
Page_Age_Threshold
Figure 4. Page age threshold vs. false positives
Figure 5 shows the overhead cost of running the MLDR
algorithm. The overhead measures the time spent on
executing the StalenessDetector, StalenessRecovery,
bookkeeping, memory allocation, and memory deallocation.
The overhead cost is dominated by the cost of the detection
and recovery functions. As expected, the overhead cost is
smaller for larger values of the page age threshold.
Page Age Threshold Vs
Overhead
Overhead Cost
Page_Age_Threshold
Figure 5. Page age threshold vs. overhead cost
4. Conclusions
We have described a new approach for memory leak
detection and recovery (MLDR) using aging in physical
memory. This algorithm reflects both the physical and
virtual behavior of memory allocation and benefits from the
hardware support available for tracking physical pages in
real memory. MLDR is shown to follow a conservative
approach in removing unreachable objects. A liberal MLDR
detects stale objects at the expense of running the risk of
false positives. The conservative MLDR produces no false
positives and the number of false negatives can be
minimized, albeit at more overhead cost. The false negative
rates and overhead cost were shown to be highly dependent
on the user-configurable input parameters. MLDR is
recommended for mission critical, server, or long live
 (IJCNS) International Journal of Computer and Network Security, 57
applications that need to run 24x7. The trace-driven
simulation results have shown that the MLDR is capable of
removing unreachable objects and providing more room for
new allocations which makes the applications avoid or delay
imminent crashing.
References
[1] William R. Bush, Jonathan D. Pincus, and David J.
Sielaff, “A Static analyzer for finding dynamic
programming errors,” In Software Practice and
Experience, 2000.
[2] “Debug Malloc Library,” [Online]. Available:
http://dmalloc.com/. [Accessed: Jan 20, 2010].
[3] E. Marshall, “Fatal Error: How Patriot Overlooked a
Scud,” Science, page 1347, March 13, 1992.
[4] Erwin Andreasen and Henner Zeller. “LeakTracer,”
[Online]. Available:Http://www.andreasen.org/
leakTracer/ . [Accessed: Mar 20, 2003].
[5] “GNU C Library,” [Online]. Available:
http://www.gnu.org/software/libc/manual/html_node/
Allocation-Debugging.html, [Accessed: Feb 14, 2010].
[6] K. C. Gross, V. Bhardwaj, and R. Bickford, “Proactive
Detection of Software Aging Mechanisms in
Performance Critical Computers,” Software
Engineering Workshop, 2002. Proceedings. 27th
Annual NASA Goddard/IEEE, 2002.
[7] David L. Heine, and Monica S. Lam, “A Practical
Flow-Sensitive and Context-Sensitive C and C++
Memory Leak Detector,” In ACM SIGPLAN
conference on Programming Language Design and
Implementation (PLDI), 2003.
[8] “Parasoft Insure++,” [Online]. Available:
http://www.parasoft.com/jsp/products/home.jsp?produc
t=Insure&. [Accessed: Mar 15, 2010].
[9] D. L. Parnas, "Software Aging", Proc. 16th Int. Conf.
on Software Eng., Sorento, Italy, IEEE Press, pp. 279-
287, May 16-21/94.
[10] “Quest JProbe Memory Debugger,” [Online].
Available: http://www.quest.com/jprobe/. [Accessed:
Apr 10, 2010].
[11] R. Hasting and B. Joyce, “Purify: Fast Detection of
memory leaks and access errors,” In Proceedings of
the Winter USENIX Conference, pages 125-136, 1992.
[12] A. Silberschatz, P. Galvin, and G. Gagne, Operating
System Concepts, 7th edition, John Wiley and Sons,
2005.
[13] K. S. Trivedi, K. Validyanathan, and K. Goseva-
Postojanova, "Modeling and Analysis of Software
Aging and Rejuvenation,” Proc. 33rd Annual
Simulation Symp., pp. 270-279, IEEE Computer
Society Press, 2000.
[14] K. Validyanathan, R. E. Harper, S. W. Hunter, and
K. S. Trivedi, “Analysis and Implementation of
Software Rejuvenation in Cluster Systems,” ACM
Sigmetrics 2001/Performce 2001, June 2001.
[15] T. M. Chilimbi, and M. Hauswirth, “Low-Overhead
Memory Leak Detection Using Adaptive Statistical
Profiling,” ACM, 2004.
Vol. 2, No. 5, May 2010
Authors Profile
Dr. Ahmed Otoom received his Ph.D.
degree in Computer Science from
Amman Arab University in 2007, dual
M.S. degrees in Computer Science and
Information Technology Management
from Naval Postgraduate School, CA in
2000 and a B.S. degree in Computer
Science from Mutah University in 1992.
During 1992-2010, he worked as a system analyst,
developer, and researcher for the IT Directorate in Royal
Jordanian Air Force (RJAF). During the same period, he
also taught many computer science classes in the Air Force
Technical College and Alzaytoonah University of Jordan,
respectively. His major areas of interest are in operating
systems and interoperability in large heterogeneous
information systems.
Dr. Mohammad Malkawi received
his Ph.D. degree in computer
engineering from the University of
Illinois at Urbana-Champaign in
1986. He received his BS and MS in
computer engineering in 1980 and
1983. He was a member of the High Productivity Computing
Systems team at SUN Microsystems, and the high
availability platform team at Motorola. Currently, he is
associate professor at Middle East University in Jordan.
Dr. Mohammad Alweh received his
Ph.D. degree in electrical and
computer engineering from Southern
Illinois University in 2010. He
received his BS and MS in computer
engineering from SIU. Dr. Alweh
worked at Motorola between 1999 and
2003 in the network infrastructure
group. His interests include computer architecture, software
engineering, and system reliability.
58 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

Adaptive Modulation Techniquesfor WIMAX

B.Chaitanya1, T.Sai Ram Prasad2, K.Sruthi3 and T.Tejaswi4
1
Lecturer, E.C.E DEPT V.R.Siddhartha Engineering College, Vijayawada
E-mail:bchaitanya55@gmail.com, bchaitanya_55@yahoo.co.in

2
M.Tech student V.R.Siddhartha Engineering College, Vijayawada
E-mail:srp.tummala@gmail.com, srp.tummala@rocketmail.com

3
B.Tech student V.R.Siddhartha Engineering College, Vijayawada
E-mail:sruthi_rockz@yahoo.com,

3
B.Tech student V.R.Siddhartha Engineering College, Vijayawada
E-mail:tejaswi.tummala@gmail.com,

Abstract: The advancements in broadband and mobile
communication has given many privileges to the subscribers for
instance high speed data connectivity, voice and video
applications in Economical rates with good quality of services.
WiMAX is an eminent technology that provides broadband and
IP connectivity on “last mile” scenario. It offers both line of
sight and non-line of sight wireless communication. Orthogonal
frequency division multiple access is used by WiMAX on its
physical layer. Orthogonal frequency division multiple accesses
use adaptive modulation technique on the physical layer of
WiMAX and it uses the concept of cyclic prefix that adds
additional bits at the transmitter end. The signal is transmitted
through the channel and it is received at the receiver end. Then
the receiver removes these additional bits in order to minimize
the inter symbol interference, to improve the bit error rate and
to reduce the power spectrum.
In this paper, we investigated the physical layer
performance on the basis of bit error rate, signal to noise ratio,
and error probability. These parameters are discussed in two
different Models. The first model is a simple OFDM
communication model without the cyclic prefix, while the
second model includes cyclic prefix.
Keywords: WiMAX, wireless communication, physical layer.
1. Introduction
The demand for broadband mobile services continues to
grow. Conventional high-speed broadband solutions are
based on wired-access technologies such as digital
subscriber line (DSL). This type of solution is difficult to
deploy in remote rural areas, and furthermore it lacks
support for terminal mobility. Mobile Broadband Wireless
Access (BWA) offers a flexible and cost-effective solution to
these problems [1].The IEEE WiMax/802.16 is a promising
technology for broadband wireless metropolitan areas
networks (WMANs) as it can provide high throughput over
long distances and can support different qualities of
services. WiMax/802.16 technology ensures broadband
access for the last mile. It provides a wireless backhaul
network that enables high speed Internet access to
residential, small and medium business customers, as well
as Internet access for WiFi hot spots and cellular base
stations [2]. It supports both point-to-multipoint (P2MP) and
multipoint-to-multipoint (mesh) modes. WiMAX will
substitute other broadband technologies competing in the
same segment and will become an excellent solution for the
deployment of the well-known last mile infrastructures in
places where it is very difficult to get with other
technologies, such as cable or DSL, and where the costs of
deployment and maintenance of such technologies would
not be profitable. In this way, WiMAX will connect rural
areas in developing countries as well as underserved
metropolitan areas. It can even be used to deliver backhaul
for carrier structures, enterprise campus, and Wi-Fi hot-
spots. WiMAX offers a good solution for these challenges
because it provides a cost-effective, rapidly deployable
solution [3]. Additionally, WiMAX will represent a serious
competitor to 3G (Third Generation) cellular systems as
high speed mobile data applications will be achieved with
the 802.16e specification. The original WiMAX standard
only catered for fixed and Nomadic services. It was reviewed
to address full mobility applications, hence the mobile
WiMAX standard, defined under the IEEE 802.16e
specification. Mobile WiMAX supports full mobility,
nomadic and fixed systems [4]. It addresses the following
needs which may answer the question of closing the digital
divide:
• It is cost effective.
• It offers high data rates.
• It supports fixed, nomadic and mobile applications
thereby converging the Fixed and mobile networks.
• It is easy to deploy and has flexible network architectures.
• It supports interoperability with other networks.
• It is aimed at being the first truly a global wireless
broadband network.
1.1 Adaptive Modulation Techniques in WiMAX
The basic idea of Adaptive Modulation is to adapt
different modulation techniques when a wireless
communication system experiences fading and variations on
the link. WiMAX takes full advantage of link adaptation
technique along with coding. This scheme is quite simple, if
the link condition is not good, WiMAX system changes the
modulation automatically. Hence, real time application such
as video and voice can run continuously by varying the
modulation, the amount of data transferred per signal also
varies, i.e., deviation in throughputs and spectral
 (IJCNS) International Journal of Computer and Network Security, 59
Vol. 2, No. 5, May 2010

efficiencies. For instance, 64 QAM is capable of delivering

much higher throughput as compared to QPSK. For using
higher modulation, SNR should be optimum to overcome
noise and interference in the channel. Lower data rates are
accomplish by means of BPSK & QPSK constellations along
with ½ rate convolution error correcting codes. Here it
means that the system generates two codes for transmission
of one bit. Whereas higher data rates are achieved by using
16 QAM & 64 QAM constellations together with. ¾ rate
convolution or LDPC error correcting codes. So QPSK
provide the lowest throughput and 64 QAM ensures highest
throughput.

1.2 Simulation

In our simulation work we investigated the behavior of Figure 1. Model – 1 OFDM transmitter simple model
adaptive modulation technique of WiMAX. The adaptive
modulation used following modulation techniques for
modulating and demodulating the signal:

• Binary Phase Shift Keying (BPSK)

• Quadrature Phase Shift Keying (QPSK)
• 16 - Quadrature Amplitude Modulation (16-
QAM)
• 64 - Quadrature Amplitude Modulation (64-
QAM)

Based on these modulation techniques the following

parameters were investigated.

•B
 it Error Rate (BER) Figure 2. Model – 2 (Model with Cyclic Prefix)
•S ignal to Noise Ratio (SNR) Mersenne Twister - Random Number Generator (RNG)
•P  robability of Error (Pe) Algorithm

The key points, in the simulations are:
• Mat lab 7.4.0 (R2007b).
• Mersenne Twister - Random Number
Generator (RNG) Algorithm
• Noise is characterized as Gaussian
• Fading is characterized as Rayleigh
Probability distribution function.
• Cyclic prefix is used
• All the plotting is done to evaluate the
Performance on the basis of BERVs
SNR.
Mersenne twister is a random number generator that
generates the random number by using the pseudorandom
algorithm. The generator is composed of a large linear
feedback shift register and provides some excellent output
statistical properties. Mersenne twister - RNG comprises a
seed value which is 19,937 bits long and the value is stored
in 624 element array. Mersenne twister has a period of
2^19937 – 1. Mersenne Twister is basically implemented in
C language and utilizes the memory in an efficient way. In
Matlab this algorithm is used in rand function which is used
to generate the input of random number for scientific
applications
3. Simulation Results

OFDM BER Simulation and Adaptive Modulation

2. Simulation Model
There are two types of model that were used in
the simulation. The first model is a simple model while the
other comprises on cyclic prefix.
Technique the simulation result based on the adaptive
modulation technique for BER calculation was observed in
this section. The adaptive modulation techniques used in the
WiMAX are BPSK, QPSK, 16-QAM, 64-QAM and 256-
QAM respectively. We use all modulation techniques in
order to get the results on different models.
OFDM with Adaptive Modulation Techniques in
PURE AWGN. The initial results observed in the pure
AWGN channel condition using adaptive modulation
60
techniques and compared the performance of these
techniques while using the 256 multicarrier OFDM waves.
Figure 3. OFDM with Adaptive Modulation Techniques in
PURE AWGN
3.1 Theoretical Values of BER using Adaptive
Modulation Techniques in OFDM
The theoretical value of BER with respect to adaptive
modulation techniques in the presence of pure AWGN is
used to estimate the theoretical value of SNR with 256 sub
carriers
Figure 4. Theoretical Values of BER using Adaptive
Modulation Techniques in OFDM (CP)
3.2 OFDM with Adaptive Modulation Techniques in
AWGN + Rayleigh Fading Channel with Cyclic
Prefix
There is another model which consists on AWGN and
Rayleigh Fading Channel with the addition of Cyclic Prefix
(CP) at the transmitter as well as receiver end. We
investigate the effects of CP while using adaptive
modulation techniques and compared the performance of
OFDM symbols in terms of BER and SNR
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
Figure 5. OFDM with Adaptive Modulation Techniques in
AWGN + Rayleigh Fading Channel with Cyclic Prefix (CP)
4. Probability of Error (Pe) for Adaptive
Modulation
The Probability of Error (Pe) is the assumption of the
rate of the error that introduce in the system because of
noise and fading effects in the channel and also due to the
cable losses at transmitter and the receiver ends. The
Probability of Error for M-ary PSK has been calculated
using the following formula:
Also for M-ary QAM the error probability given as:
Figure 6. Probability of Error (Pe) for Adaptive
Modulation
5. Conclusion
We concluded that BPSK is more power efficient and
need less bandwidth amongst all other modulation
techniques used in an OFDM adaptive modulation. In case
of bandwidth utilization the 64 and 256QAM modulation
 (IJCNS) International Journal of Computer and Network Security, 61
Vol. 2, No. 5, May 2010

requires higher bandwidth and gives an excellent data rates

as compared to others. While the QPSK and the 16QAM
techniques are in the middle of these two and need higher
bandwidth and less power efficient than BPSK. But they
required lesser bandwidth and lower data rates than
64QAM. Also, BPSK has the lowest BER while the 64-
QAM has highest BER than others.

There are other aspects as well that we conclude:

1 The inclusion of the Cyclic Prefix reduces the Inter

Symbol Interference (ISI) that causes the lower BER in the
OFDM system but increases the complexity of the system.

2. Model with CP requires high power as compared to the

non-CP model.
In future, adaptive modulation technique and
OFDMA as physical layer will be adapted by Long Term
Evaluation (LTE) and High Altitude Platform (HAP)

References
[1] Mai Tran, George Zaggoulos, Andrew Nix and
Angela Doufexi,” Mobile WiMAX Performance
An a l ysi s an d Compar i son with Experimental
Results”
[2] J. El-Najjar, B. Jaumard, C.Assi, “Minimizing
Interference in WiMax/802.16 based Mesh
Networks with Centralized Scheduling”,
Global Telecommunications Conference, 2008.
pp.1-6. http://www.intel.com/netcomms
/technologies/WiMAX/304471.pdf
[3] Intel white paper, Wi-Fi and Wimax Solution:
“Understanding Wi-Fi and Wimax as
Metro-Access Solution, “Intel Corporation
2004.
[4] A. Yarali, B. Mbula, A. Tumula, “WiMAX:
A Key to Bridging the Digital Divide ”IEEE
Volume, 2007, pp. 159 – 164.
[5] WiMAX Forum, “Fixed, Nomadic, Portable and
Mobile Applications for 802.16-2004 and
802.16eWiMAXNetworks”, November,
2005.
[6] T.TAN BENNY BING, "The World Wide Wi-
Fi: Technological Trends and B u s i n e s s
Strategies", J O H N W I L E Y & S O N S , INC.,
2003
[7] M. Nadeem Khan, S. Ghauri, “The WiMAX
802.16e Physical Layer Model”, IET
International Conference on Volume, 2008, pp.117
– 120.
[8] Kavesh Pahlavan and prashant Krishna Murthy
“Principles of Wireless Network “Prentice-Hall, Inc.,
2006
[9] Yango Xiao, “Wimax/Mobile”MobileFi Advanced
Research and Technology Auerbach publications,
2008.
62 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

Performance of DSR Protocol over Sensor

Networks
Khushboo Tripathi1, Tulika Agarwal2 and S. D. Dixit3
1
Department of Electronics and Communications
University of Allahabad, Allahabad-211002, India
khushboo83@live.com
2
Department of Electronics and Communications
University of Allahabad, Allahabad-211002, India
tulikabbs09@gmail.com

Abstract: Routing protocols play crucial role in determining

performance parameters such as throughput parameter, packet 2. Related Work
delivery fraction, end to end (end 2 end) delay, packet loss etc. of Dynamic Source Routing (DSR) Protocol: DSR is
any ad hoc communication network. In this paper the
characterized by source routing i.e. sender knows the
performance of DSR protocol in sensor network of randomly
complete hop-by-hop route to the destination. It emphasizes
distributed static nodes with mobile source node is investigated
with source velocities 10, 20, 40 and 60 m/sec. and node the aggressive use of route caches that store full paths to the
densities 40-100 nodes/km2 . On varying velocity it is observed destination as given in figure1 (a) & (b) [2,7 ]. This protocol
that throughput of received packets is higher which reduces the has the advantages of loop-free routing and avoidance of the
overheads in the network. The delivery ratio is also maximum need for up-to-date routing information in the intermediate
except at node density 60/km2.When the number of dropped nodes.
packet is considered the average end to end delay is decreases
except at 80 nodes/km2 and so throughput is high. Hence DSR
performs better in this scenario and reduces minimum
overhead.

Keywords: Sensor Networks, DSR, Node Density.

1. Introduction
The recent development in small embedded sensing devices
and the wireless sensor network technology has provided
Figure1 (a). RREQ Broadcast
opportunities for deploying sensor networks to a range of
applications such as environmental monitoring, disaster
management, tactical applications etc [1].Main requirement
for such application is that motes carrying onboard sensors
should be physically small, low power consuming and
include wireless radio.
For data collection a straight forward solution is that each
mote transmits its data to a centralized base station.
However in such cases the energy requirement of each node
would be large which reduces mote life and also there would
Figure1 (b). RREP Propagation
be interference problem. Alternative approach for harvesting
data from sensor fields uses mobile data collector such as
It is composed of two main mechanisms: Route Discovery
robots which move in the sensor field to collect data and
and Route Maintenance. For route discovery the source node
transmit the same to base station in real /non-real time [6].
floods the route request (RREQ) packets in the network. The
In the present paper, the performance of dynamic source
nodes receiving RREQ rebroadcast it and the process repeats
routing (DSR) protocol has been analyzed keeping in mind a
until the destination node or an intermediate node having a
sensor network scenario wherein all the nodes are static and
route to the destination is found. Such a node replies back to
source node is moving ( one of which is a data harvester
the source with a RREP packet. When a link-break in an
from static nodes and other one acting as a sink).
established route occurs, the route maintenance phase is
Section 2 describes the related work and Section 3 is about
initiated wherein upstream node at the link-break site
the network scenario and all about the definition of
creates a route error (RERR) message and sends it to the
simulation parameters and details of simulation experiment.
source node. On receiving RERR the source node utilizes
Section 4 gives the results and discussion of the simulation.
alternate routes from the route cache, if they are available, to
Section 5 concludes the work.
prevent another route discovery. The drawback with DSR is
that it needs to place entire route in both the route replies
and the data packets and thus requires greater control
overhead.
 (IJCNS) International Journal of Computer and Network Security, 63
Vol. 2, No. 5, May 2010

3. Network Scenario and Simulation Details

The simulation scenario consists of tcl script that runs over
TCP connections for various number of nodes (40, 60, 80
and of 100 nodes) in an area of size of 1000 1000m2 . Table1: Parameter values for simulation
Figure 2 shows the random node topology of the network.
The simulation time is set to 150 seconds. TCP connection Maximum simulation 150 seconds
time
has been used to set communication between the source
Area size (Flat area) 1000×1000 m2
node and the sink node.
Routing protocol DSR
(proactive)
Propagation Model Two Ray Ground
Propagation
MAC layers protocol IEEE802.11
Node placement Static Random
Distribution
Number of nodes 40,60,80,100
Velocities of source 10,20,40,60 m/sec
and sink nodes

4. Results and Discussion

The throughput vs. node density graph is given in figure3.
According to our topology considerations, in sparse
network, for the velocities 10, 20, 40 and 60 m/sec different
throughput graphs are obtained. For all the velocities at the
Figure 2. NAM Windows showing network topology for
node density 60 per square km minimum throughput is
100 nodes
obtained while higher velocities give maximum throughput
with node density parameter. At velocity 10m/sec
The simulation is carried out in the network simulator NS-2
throughput is minimum whereas at velocity 60 m/sec
(version-2.31) over LINUX (FEDORA 8) environment [3]
throughput is higher in comparison to other velocities. This
[4].The experiment has been performed for a set of 5
is because of the nature of DSR on-demand routing protocol.
random network topologies having number of nodes
Once the route is formed the possibility of (link-breakage) is
40,60,80 and 100 respectively. In each topology all nodes
lesser as compared to the former case where the probability
are fixed except the source node. For a given topology two
of route formation is lower. Hence it is clear from figure
nodes are selected to be source-sink nodes and the
after node density 60 per square km throughput is generally
observations are made for five different velocities 10, 20, 40
increasing in nature with different node densities due to less
and 60 m/sec. The experiment was repeated by selecting
link breakage in route formation. It is obtained that When
another pairs of source-sink nodes chosen randomly. Five
the throughput of received packets is maximum the number
such observations were taken for each of the following three
of overheads reduces in the network.
performance parameters and their averages are obtained to
plot graphs: 6000
nv 10
1. Bytes Delivery Fraction: The ratio of the number of nv 20
nv 40
5500
data sends (in bytes) successfully delivered to the nv 60

destinations to those are generated by TCP agents/sources. 5000

Thus, Bytes Delivery Fraction = (Received data /Total sent
Throughput

data) 4500

2. Average Number of Dropped Packets: This parameter

is worth mentioning while taking the effectiveness of this 4000

routing protocol. These are the average number of dropped

3500
packets in simulation study of our scenario.
3. Average Simulation End to End Delay: The delivery 3000
40 50 60 70 80 90 100
delay for the data is the interval between when it is Node Density(per square km)
generated at a sensor node and when it is collected by a sink
Figure3.Throughput (of received packets) vs. node density
[2]. It does not depend on the time for collecting data from
the network. These are the possible delays caused by
The packet delivery fraction vs. node density behavior is
buffering during route discovery latency, queuing at the
given in figure 4. We have seen the effect of delivery ratio
interface queue, retransmission delays at the MAC and
with node density on varying the velocities 10, 20, 40 and
propagation and transfer times.
60 m/sec. The delivery fraction graph is almost similar to
Simulation Parameters: The parameter values for the throughput vs. node density figure. On varying node
simulation are given in table 1. density the delivery fraction is increased for all the velocities
except at node density 60 per square km.
 64 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

[3] The Network Simulator NS-2 homepage,

0.8
http://www.isi.edu/nsnam/ns
[4] The Network Simulator NS-2 tutorial homepage
http://www.isi.edu/nsnam/ns/tutorial/index.html
[5] E. M. Royer, and C. K. Toh. “A review of current
0.75
routing protocols for ad hoc mobile wireless
Deliv ery Ratio

networks”,IEEE Personal Communications, (April),

1999.
[6] J. Rao, and S.Biswas, “Data Harvesting In Sensor
0.7 Networks Using Mobile Sinks”, IEEE Wireless
nv10 Communication, pp.1536-1284, 2008.
nv20 [7] M Tamilarasi, V.R.S.Sunder, “Scalability Improved
nv40
nv60 DSR Protocol for MANETs”, ICCIMA-pp.283-287,
0.65 2007.
40 50 60 70 80 90 100
Node Density(per square km)

Figure4. Delivery Ratio vs. node density Authors Profile

The average simulation end to end delay is given in figure5.
Khushboo Tripathi is a D.Phil. student
According to the topology we found a vibrating nature of delay
in the Department of Electronics and
in graph. But generally it decreases with the node density. An Communication of J.K.Institute of
exceptions is attained at node density = 60 nodes/km2.When the Applied Physics and Technology
delay is high i.e. high node density in the network and the route ,University of Allahabad, Allahabad She
finding is difficult in that scenario whereas in case of less delay obtained her master’s degree in
the possibility of finding routes is more. From above graph it is mathematics in 2005 from University of
evident that for the higher velocity the throughput is maximum Allahabad. Also she received the MCA
and corresponding delay is less as in figure5 and vice versa. degree from UPRTOU, Allahabad. Her research is in the
area of Wireless Communications and networking,
0.41 particularly in MANETs, WSN and coding, cryptography.
nv10
0.4 nv20
nv40 Tulika Agarwal is a M.Tech. student in
Avg.Sim ulation End to E nd Delay(sec.)

0.39 nv60 the Department of Electronics and

0.38
Communication of J.K.Institute of
Applied Physics and Technology,
0.37 University of Allahabad, Allahabad. She
obtained her B.Tech. in Electronics and
0.36
Communication Engineering from
0.35 BBSCET, Allahabad in 2007. She is
carrying out her M.Tech. project on “A Comparative Study
0.34
of Routing Protocols in Ad Hoc Networks”. She has also
0.33 interest in “Digital Electronics” and “Wireless
Communication”.
0.32
40 50 60 70 80 90 100
Node Density(per square km)
S.D.Dixit is a professor in the
Figure5. Avg.simulation end2end delay vs. node density Department of Electronics and
Communication Engineering, University
of Allahabad at Allahabad. He has
teaching experience of 33 years in the
5. Conclusion field of Electronics and Communication
In the given scenario with increase in velocity throughput of Engineering at J.K.Institute of Applied
received packets is higher which reduces the overheads in Physics and Technology, University of
the network. Packet delivery ratio is maximum as Allahabad, Allahabad. He obtained M.Tech. and D.Phil.
throughput of received packet is high except at node density degrees in 1975 and 1981 respectively. He has published a
60/km2.When the number of dropped packet is considered number of research papers in National and International
Journals. He is the fellow of IETE, India. His current
the average End to End delay is decreases except at 80
interests are in the field of Wireless Communications,
nodes/km2 and so throughput is high. Therefore DSR Coding and Cryptography.
performs better in this scenario and reduces minimum
overhead.

References
[1] Anna Hac, “Wireless Sensor Network Designs”, John
Wiley and Sons, 2003.
[2] M.K. Marina, S. R. Das, “Performance of Route Caching
Strategies in Dynamic Source Routing,” ICDCS, pp.
425-432,2001.
 (IJCNS) International Journal of Computer and Network Security, 65
Vol. 2, No. 5, May 2010

Route Information Update in Mobile Ad Hoc

Networks
Seungjin Park1 and Seong-Moo Yoo2
1
University of Southern Indiana, Department of Management, MIS & CS,
8600 University Blvd., Evansville, IN 47712, USA
park@usi.edu
2
The University of Alabama in Huntsville, Electrical & Computer Engineering Department,
301 Sparkman Dr., Huntsville, AL 35899, USA
yoos@eng.uah.edu
found, S sends data along the path in the data delivery
Abstract: A mobile ad hoc network (MANET) is composed of
moving wireless hosts that, within range of each other, form
phase. In the route discovery phase S broadcasts a control
wireless networks. For communication to occur between hosts packet called a route request packet (REQ) to all its
that are not within each other’s range, routes involving neighbor nodes. If a neighbor node does not have route
intermediate nodes should be established. However, due to the information to D, it appends its information to the REQ and
mobility of the nodes, finding routes in a MANET is quite a relays the REQ to all its neighbor nodes, and so on until the
challenging task, and takes up a lot of system resources such as REQ reaches D. (This forwarding of requests from one
bandwidth and battery power. Although it would be wise to take node to all its neighbors is called “flooding” and consumes a
full advantage of the already discovered paths, many algorithms lot of time and bandwidth.) If the REQ reaches either D or
simply discard the paths if their topology has changed. To a node that contains the path information to D, the node
overcome this shortcoming, we have proposed a routing table sends a control packet called route reply packet (RPY) back
maintenance algorithm that responds to the changes in network
to S by reversing the path stored in the REQ.
topology promptly, and adjusts the paths so that their lifetimes
could be maximized.
On receiving the RPY, S knows that the path to D has been
Keywords: mobile ad hoc network, routing, routing table, established, and S sends its data packet along the path. For
unicast example, suppose node A in Figure 1 tries to find a path to
G that is not found yet. As a first step, A prepares an REQ
1. Introduction with its ID appended to it, and sends it to its neighbors
A mobile ad hoc network (MANET) consists of wireless including B. If B does not have any information about G, B
mobile hosts without any centralized control point or fixed appends its ID to the REQ and retransmits it to its neighbors
infrastructure. Since a MANET does not require any prior including C, and so on until it reaches G. The snapshot of
arrangement, it can be quickly deployed and used in such the REQ at E would contain the nodes it passed through
applications including fast establishment of military thus far, which is A, B, C, D. When the REQ reaches the
communication and rescue missions where an established destination node G, G prepares a RPY which is transmitted
network is neither feasible nor available [1]. to A along the path stored in the REQ.

Discovering a path from the source node to the destination A D

node is quite challenging in a MANET due to the mobility
of hosts. A node implementing a proactive algorithm [2]
maintains a routing table that contains the routes to all
nodes in the network by periodical exchanges of local
information with other nodes. However, this information
exchange takes up a lot of bandwidth and battery power.
On the other hand, nodes in reactive algorithms [3, 4, 5] do
not maintain any routing information. Instead, a node
dynamically discovers a path to the destination on demand.
Therefore, it may reduce considerable overhead to enhance
the network throughput.
Although other routing algorithm could be used, this paper
is based on DSR [5]. DSR is reactive and has been proved it
could be as competitive as any other algorithms in MANET
[7, 8, 9]. Like many reactive routing protocol, DSR consists
of two phases: route discovery phase and data delivery
phase. When node S has a data for node D, S starts finding
a path to D in the route discovery phase. Once a route is
C E
B G
F
I
H
Figure 1. There are two paths, P(A, B, C, D, E, F, G) and
P(H, I, C, D, E, F, G) in the networks.
For example, on receiving the RPY, node A in Figure 1
stores the path (A, B, C, D, E, F, G), and node C stores the
path (C, D, E, F, G) in their routing tables so that they can
use it later. As mentioned before, it takes a lot of resources
to discover a path in a MANET. Therefore, it would be
wise to take advantage of it as much as possible. However,
most routing protocols do not efficiently utilize the existing
paths, because they do not respond properly and promptly to
the changes in the network topology. In this paper, we
propose a strategy that tries to use the existing paths
66
efficiently by modifying route table as quickly as possible.
This paper is organized as follows. Previous work is
mentioned in section 2. Terminology needed to understand
this paper is explained in section 3. Proposed algorithm is
presented in section 4, followed by the theoretical analysis
in section 5. Section 6 presents the conclusion.
2. Previous Works
Wu et al. [6] proposed an algorithm that tries to fix some
problems in paths such as finding shorter paths and fixing
broken links. However, nodes in their algorithm stores only
the destination, next hop, hop count, etc. for each path so
that the coverage of the algorithm is limited. Ko and
Vaidya [10] proposed a location-aided routing scheme
where a route request is flooded in the direction of the target
node. Here, a node’s response to the route request depends
on whether or not it is in the region approaching the
destination.
Liao et al. [11] proposed a routing scheme called GRID.
This scheme imposes a grid system on the earth’s surface
and selects a leader in each grid to act as the gateway. Data
packets are then forwarded to their destination grid by grid.
This protocol confirms that the availability of location
information improves the routing performance.
Lee and Gerla [12] proposed AODV-BR algorithm that
utilizes a mesh structure to provide multiple alternate paths
to existing on-demand routing protocols without producing
additional control messages. In their algorithm, data
packets are delivered through the primary route unless there
is a route disconnection. When a node detects a link break,
data packets can be delivered through one or more alternate
routes and are not dropped. Route maintenance is executed
utilizing alternate paths. Here, the route is maintained only
when a link is broken.
Li and Mohapatra [13] introduced LAKER, a location-aided
knowledge extraction routing. This scheme reduces the
flooding overhead in route discovery by extracting
knowledge of the nodal density distribution of the network
and remembering the series of locations along the route
where there are many nodes around. However, this scheme
does not deal with route maintenance.
Stoimenovic [14] reviewed many position-based routings in
ad hoc networks. It is likely that only position-based
approaches provide satisfactory performance for large
networks. Greedy mode routing was shown to nearly
guarantee delivery for dense graphs, but to fail frequently
for sparse graphs since the destination is also moving and it
is not clear where to send message. The routing process is
converted from the greedy mode to recovery mode at a node
where greedy mode fails to advance a message toward the
destination
Park et al. [15] proposed an anticipated route maintenance
protocol with two extensions to route discovery based
routing scheme: extending a route by inserting a common
neighbor of two nodes in a route when the two nodes move
apart from each other that may cause link breakage, and
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
shrinking a route when a node discovers a shorter path than
the existing one. By utilizing only local geographic
information, a host can anticipate its neighbor’s departure
and, if other hosts are available, choose a host to bridge the
gap, keeping the path connected. The benefits are that this
reduces the need to find new routes and prevents
interruptions in service.
Chou et al. [16] presented a dynamic route maintenance
algorithm for beacon-based geographic routing. In this
approach, the mobile nodes dynamically adjust their beacon
intervals based on their speed of movement. Moreover the
routing information could be well managed using the
mobility prediction.
Park et al. [17] proposed an algorithm that is a preliminary
version of the paper. However, the algorithm presented in
this paper considers more cases that may save more battery
power and bandwidth.
3. Terminology
A Graph can be represented by G(V, E), where V is the set
of nodes and E is the set of links in the graph G. In a
wireless network, if a node B is within the transmission
range of node A, then it is said that there is a link from A to
B which is denoted as (A, B). A is an upstream node of B,
and B is the downstream node of A. B is also called as a
neighbor of A. If (A, B) implies (B, A), then the link is
called bidirectional, and unidirectional otherwise.
A path P(v1 , v2, . . . , vn-1, vn), where vi ∈ V, 1 ≤ i ≤ n,
consists of a set of links (v1 , v2), (v2, v3), … , (vn-1, vn). A
subpath of P(v1 , v2, . . . , vn-1, vn) is the path that is a part of
P. Function rem (P, vi) returns the portion of path P starting
at node vi to the end of P. For example, both P(B, C, D, E)
and P(A, B) are subpaths of P(A, B, C, D, E), and rem (P(A,
B, C, D, E), B) is P(B, C, D, E). |P| denotes the number of
the links in path P. For example, |P(A, B, C, D, E)| is four,
since P consists of four links.
Note that if the links are bidirectional (unidirectional, resp.),
then the paths are also bidirectional (unidirectional, resp.).
Although links assumed to be bidirectional in this paper,
paths are assumed to be unidirectional for simplicity in
explanation. It can be easily extended to cover the handling
of unidirectional paths.
When a new path is found, each node in the path may
contain the path information in its Routing Table for
possible later use. This can be done without any extra
expense at each node by simply storing the path information
when the RPY passes through the node. For example in
Figure 1, nodes A, B, C, D, E, and F in the path P(A, B, C,
D, E, F, G) store the path information when the RPY passes
through them.
4. Proposed Routing Table Maintenance
Scheme
Based on the assumption that all links are bidirectional, the
proposed scheme is hybrid, since it discovers new paths
 (IJCNS) International Journal of Computer and Network Security, 67
Vol. 2, No. 5, May 2010

reactively but the maintenance of the paths is proactive. In
other words, a new path from A to B is discovered only
when 1) A has a data for B, and 2) there is no existing path
from A to B. On the other hand, any change of the existing
paths is updated immediately regardless of the usage of the
path. This strategy makes sense, because path discovery
takes a considerable amount of resources due to the flooding
of REQ packets, whereas maintenance requires unicasts
only.
In this paper, only the routing table maintenance scheme is
presented, since route discovery algorithms can be found in
many routing protocols. The proposed routing table
maintenance scheme is explained below for different cases.
substitutes (A, B, C, D, E, F, G) with P(A, B, C, F, G) and
P(A, B, C, D, E) in its routing table. Note that the upstream
nodes of C in the path need not be informed this finding.
Instead, if a packet from either A or B destined to either F or
G is arrived at C, C directs the packet to F using P(A, B, C,
F, G), not to D. If the packet is destined to D or E, then the
packet is transferred along P(A, B, C, D, E).
Figure 2 (b) shows another case where a shorter path is
discovered by a node in the same path that is at the other
end, F. In this case, path P(A, B, C, D, E, F) is split into
two paths, P(A, B, C) and P(A, F, E, D). Note that this is
possible since A has all nodes in the path, and therefore
knows exactly where to split the path.

Case 1) Link breakage is discovered. Consider two paths, Figure 2 (c) shows the case where a shorter path is
P(A, B, C, D, E, F, G) and P(H, I, C, D, E, F, G), in the discovered by other newly found path. For example,
network as shown in Figure 1. Suppose when node C suppose P(A, B, C, D, E, F, G) is the existing path, and P(H,
sends a data packet to node G along P(A, B, C, D, E, F, G), C, J, F, K) is just discovered. Then, P(A, B, C, D, E, F, G)
C found out that the link (E, F) is broken. Then, this will be replaced by P(A, B, C, J, F, G) and P(A, B, C, D, E).
information is disseminated to all upstream nodes of C in Note that, as was done in Case 2 (a), this new information is
both paths. On receiving the link breakage information not informed to the upstream nodes of C.
nodes A, B, H, and I modify the paths by removing subpath
P(F, G). Otherwise, attempting to use this broken link may So far the cases for shorter paths have been considered. In
cause delay. the following, the cases when extension of the existing path
. is found will be discussed.
A
C F G Case 3) An extension of an existing path is found by the
B
other path. In this case, the extended portion is
disseminated to existing path. For example, suppose P1 =
P(A, B, C, D, E) in Figure 3 is the existing path, and P(F,
D E G, C, H, J) is discovered later. Then, P(A, B, C, D, E) has
(a) an extended portion of P(C, H, J) at node C, which will be
propagated to upstream nodes of C. Therefore, P(A, B, C, H,
F J) will be added to routing tables at A, B, C. Likewise, P(F,
A
G, C, D, E) will be added to routing tables at C, F and G.
E
B J
H
C A D
D
C E
(b) B

D
H
E
G
C

F
F G
A J
B Figure 3. As a result of the discovery of another path P(F,
K G, C, H, J), nodes A, B, C add P(A, B, C, H, J) to
(c) their routing tables. Likewise, P(F, G, C, D, E)
will be added to routing tables of F, G, and C.
Figure 2. (a) A shorter path P(A, B, C, F, G) is found in
the same path. (b) A shorter path is found by the Case 4) A more general form of Case 3 can be found in
node at the other end of the same path. (c) A Figure 4, where an existing path and a newly found path
shorter path P(A, B, C, J, F, G) is found by other have more than one common node. In this case, the shorter
path P(H, C, J, F, K). path connecting any adjacent two common nodes is chosen
and used. For example, refer to Figure 4, where P(A, B, C,
Case 2) A shorter path is found. There are two possible
D, E, F) is the existing path and P (G, B, H, J, K, E, L) is
cases. First, C in Figure 2 (a) found a shorter path (A, B, C,
the newly found path, and they share both B and E. In this
F, G) in the existing path (A, B, C, D, E, F, G), when F
case, subpaths P(B, C, D, E) and P(B, H, J, K, E) are
moves into C’s transmission range. In this case, C
compared and P(B, C, D, E) is chosen since it is shorter of
68
the two. Then, both existing and newly found path will be
replaced using the shorter path. For example, P(A, B, C,
D, E, F) is replaced by P(A, B, H, J, K, E, F), P(A, B, C, D)
and P(A, B, H, J, K, E, L). Also, two routes, P(G, B, C, D)
and P(G, B, H, J, K, E, F) are added to the routing table at
G.
Summary of the proposed algorithm is presented in Figure
5.
G D
C
B E
A H K common nodes
J
Figure 4. As a result of the discovery of another path P(F,
G, H, E, J, K), existing path P(A, B, C, D, E) is
extended to P(A, B, C, D, E, J, K) and this
information will be stored at nodes A, B, C, D,
and E.
Algorithm RoutingTable_Maintenance
At current node V1
Case when a broken link is found
if V1 found out that an existing path P is broken between
node A and B then for each path P in the Routing Table that
contains a link (A, B)
P – rem(P, B)
Sends a Broken Link Message to its upstream
nodes, if any, of all paths that contain (A, B).
if V1 received a broken link message from its downstream
node in an existing path,
then for each path P in the Routing Table that contains a
link (A, B)
P – rem(P, B)
Sends a Broken Link Message to its upstream
nodes, if any, of all paths that contain (A, B).
Case when there is a shorter path
// On receiving a RPY, V1 check if there is a shorter path for
// each path in the routing table using the path in the RPY.
Let the path in the RPY be P(v1, v2 , . . . , vn-1, vn). Note that
the current node is v1, since the first node in the RPY is v1 .
If there is a path P(u1, u2, . . . , um-1, um) in the routing table
such that 1) v1 = ui and vn = uj, and 2) |P(v1, v2, . . . , vn-1 ,
vn)| ≤ |P(ui, ui+1, . . . , uj-1, uj)|, then replace the path with
P(u1, u2, . . , ui-1, v1, v2, . . . , vn-1, vn, uj+1, . . . , um-1 , um), and
add new path P(u1, u2 , . . , ui-1, ui, . . . ., uj).
Case when the existing path and newly found path have one
common node
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
// On receiving a RPY, call check_modification_path (path
in // the RPY), where each path in the routing table is
checked if // the path should be further modified due to the
path in the
// RPY.
Let the newly found path be P(v1, v2, . . . , vn-1, vn).
// For the cases shown in Figure 3
If there is a path P(u1, u2 , . . . , um-1, um) in the routing table
such that vi = uj, add P(u1, u2 , . . . , ui-1 , ui, vj+1, . . . , vn-1, vn )
to the routing tables of u1, u2, . . . , ui-1, ui. Also add P(v1,
L
v2, . . . , vj, ui+1, ui+2, . . . , um) to routing tables of v1, v2, . . . ,
vj.
F
Case when the existing path and newly found path have two
Let the newly found path be P(v1, v2, . . . , vn-1, vn).
// For the cases shown in Figure 4
If there is a path P(u1, u2 , . . . , um-1, um) in the routing table
such that vi = uj, vk = ul, and |P(uj, uj+1 , . . . , ul-1 , ul)| > P(vi,
vi+1, . . . , vk-1, vk), then replace P(u1, u2 , . . . , um-1 , um) with
P(u1, u2 , . . . , ul-1, ul) and P(u1, u2 , . . . , uj-1 , vj, vj+1, . . . , vk,
ul+1 , um) to routing tables of nodes u1 , u2, . . . , uj-1.
Further, add P(v1, v2, . . . , uj, uj+1, . . . , ul) and P(v1, v2 , . . .
, vk, ul+1, . . . , um) to routing tables of nodes v1 , v2, . . . , vi-1,
vi.
Figure 5. Pseudo-code for the proposed algorithm.
5. Analysis of the proposed Algorithm
This section presents the time complexity of the proposed
algorithm in section 4.1, and the impact on the routing in
section 4.2.
5.1 Time complexity of routing table maintenance
Let us consider the time complexity of the routing table
maintenance at each node. If a node A detects a new
neighbor node B, then A checks if B is in any path in its
routing table so that A can adjust the existing paths. This
searching process takes O(nm), where n is the number of
paths in the routing table and m is the average path length.
The searching time would be improved to O(1), if the nodes
in the paths are stored in a hash table using additional
storage. Note that communication takes much longer time
and more battery power than computation. Therefore, it
would worth the computation if it results in reducing any
amount of path length.
Once the path is found, actual adjustment of the routing
table would take O(1). If the path falls into one of the Case
3, then the new discovery is transmitted to the source node.
Note that the transmission is unicast which is much cheaper
than broadcast in wireless communication, because there is
a single designated destination node within the source’s
transmission range in unicast, whereas all nodes are the
destinations in broadcasting.
 (IJCNS) International Journal of Computer and Network Security, 69
5.2 Impact on route discovery
This section presents the theoretical analysis of the proposed
algorithm to show how our proposed algorithm can reduce
communication overhead and battery power by modifying
routing tables on discovering a new path.
Suppose N is the average number of neighbors of each node
in the network. If an algorithm abandons the path on its
topological changes, then it would take at least
N 0 ⋅ N 1 ⋅ ... ⋅ N L−1 ≈ N2L REQ transmissions to find a
path of length L all over again by using broadcast. On the
other hand, if an algorithm can adjust to the minor
topological changes of paths like our proposed algorithm,
then it would take only L transmissions for the path of
length L. Therefore, the advantage of having the proposed
algorithm is obvious. This analysis would be true, if the
changed path is used, otherwise, the adjustment would be a
waste.
6. Conclusion
Routing in a wireless mobile ad hoc network is quite a
challenging task, because it takes up a lot of network
resources. Therefore, although it would be wise to take full
advantage of the already discovered paths, many algorithms
simply discard the paths when their topology is changed.
To overcome this shortcoming, we have proposed a routing
table maintenance algorithm that responds to the changes in
network topology as promptly yet efficiently as possible to
fully utilize existing paths by modifying them.
Our simple analysis shows that the proposed algorithm
reduces number of packet transmissions, and therefore saves
resources such as bandwidth and battery power.
References
[1] C. Perkins, “Ad Hoc Networking,” Addison-Wesley,
2001.
[2] C. Perkins, and P. Bhagwat, “Highly-dynamic
destination-sequenced distance-vector routing (DSDV)
for mobile computers,” pp. 234-244, SIGCOM’94,
1994.
[3] C. Perkins, E. Royer and S. Das, “Ad hoc on demand
distance vector (AODV) routing,” Internet draft, IETF,
Oct. 1999.
[4] V. Park and M. Corson, “A highly adaptive distributed
routing algorithm for mobile wireless networks,” In
Proc. IEEE Infocom, pp.1405-1413, Apr. 1997
[5] D. Johnson and D. Maltz, “Dynamic source routing in
ad-hoc wireless networking,” in Mobile Computing, T.
Imielinski and H. Korth, editors Kluwer Academic
Publishing, 1996.
[6] S. Wu, S. Ni, Y. Tseng, and J. Sheu, “Route
maintenance in a wireless mobile ad hoc network,”
Proc. 33 Hawaii International Conference on System
Sciences, 2000.
[7] J. Broch, D. Maltz, D. Johnson, Y. Hu, and J. Jetcheva,
“Performance comparison of multi-hop wireless ad hoc
network routing protocols,” MobiCom’98, 1998.
Vol. 2, No. 5, May 2010
[8] A. Rahman, S. Islam, and A. Talevski, “Performance
measurement of various routing protocols in ad-hoc
network,” Proc. International Multiconference of
Engineers and Computer Scientists, 2009.
[9] G. Jayakumar and G. Gopinath, “Performance
Comparison of two on-demand routing protocols for
ad-choc networks based on random way point mobility
model,” American Journal of Applied Sciences, 5 (6),
pp. 659 – 664, 2008.
[10] Y. B. Ko and N.H. Vaidya, “Location-aided routng
(LAR) in mobile ad hoc networks,” Annual
International Conference on Mobile Computing and
Networking (MobiCom), Oct. 1998, pp. 66-75.
[11] W.H. Liao, Y.C. Tseng, and J.P. Sheu, “GRID: a fully
location-aware routing protocol for mobile ad hoc
networks,” Telcommunication Systems, 18(13), 2001,
pp. 37-60.
[12] S.J. Lee and M. Gerla, "AODV-BR: backup routing in
ad hoc networks", IEEE Wireless Communications and
Networking Conference (WCNC), Sept. 2000, pp.
1311–1316.
[13] Jian Li and P. Mohapatra, "LAKER: location aided
knowledge extraction routing for mobile ad hoc
networks", IEEE Wireless Communications and
Networking (WCNC), March 2003, pp. 1180–1184.
[14] I. Stojmenovic, “Position-based routing in ad hoc
networks,” IEEE Communications Magazine, July
2002, pp. 2-8.
[15] S. Park, S.M. Yoo, M. Al-Shurman, B. VanVoorst,
and C.H. Jo, “ARM: anticipated route maintenance
scheme in location-aided mobile ad hoc networks,”
Journal of Communications and Networks, vol. 7, no.
3, pp. 325-336, Sep. 2005.
[16] C.H. Chou, K.F. Ssu, and H.C. Jiau, “Dynamic route
maintenance for geographic forwarding in mobile ad
hoc networks,” Computer Networks, vol. 52, issue 2,
pp. 418-431, Feb. 2008.
[17] S. Park and S. Yoo, Routing table Maintenance in
Mobile Ad Hoc Networks, The 12th International
Conference on Advanced Communication Technology,
pp. 1321 - 1325, Feb., 2010.
Authors Profile
Seungjin Park received the BE degree in civil engineering from
Hanyang University, Korea, in 1973, MS degree in computer
science from University of Texas at Arlington, in 1986, and PhD
degree in computer science from Oregon State University, in 1993.
He is currently an assistant professor in the Department of MNGT,
MIS, and CS, University of Southern Indiana. His research
interests include parallel algorithms, wireless ad hoc networks, and
sensor networks.
Seong-Moo Yoo is an Associate Professor of Electrical and
Computer Engineering at the University of Alabama in Huntsville
(UAH), USA. Before joining UAH, he was an Assistant Professor
at Columbus State University, Georgia – USA. He earned MS and
PhD degrees in Computer Science at the University of Texas at
Arlington. He was a Fulbright Scholar, Kazakhstan 2008-2009.
His research interests include computer network security, wireless
network routing, and parallel computer architecture. He has co-
authored over 70 scientific articles in refereed journals and
international conferences.
70 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

An Improved Proxy Ring Signature Scheme with

Revocable Anonymity
Binayak Kar1, Pritam Prava Sahoo2 and Ashok Kumar Das3
1
Department of Computer Science and Engineering
International Institute of Information Technology, Bhubaneswar 751 013, India
binayakdilu@rediffmail.com
2
Department of Computer Science and Engineering
International Institute of Information Technology, Bhubaneswar 751 013, India
sahoo_pritam@rediffmail.com
3
Department of Computer Science and Engineering
International Institute of Information Technology, Bhubaneswar 751 013, India
iitkgp_akdas2006@yahoo.co.in

examples, [2, 7, 10]). Since it is proposed, the proxy

Abstract: In this paper, we review Hu et al.'s proxy ring
signature with revocable anonymity. We then briefly review
Zhang and Yang's attacks on Hu et al.'s proxy ring signature
scheme. In order to eliminate those security flaws in Hu et al.'s
proxy ring signature scheme, we propose an improvement of Hu
et al.'s proxy ring signature with revocable anonymity. We show
that our proposed scheme provides better security as compared
to that for Hu et al.'s scheme.
Keywords: Proxy signature, ring signature, revocable
anonymity, security.
signature schemes have been suggested for use in many
applications [6, 8] particularly in distributed computing
where delegation of right is quite common. Examples
discussed in in the literature include distributed systems,
grid computing, mobile agent applications, distributed
shared object systems, global distribution networks and
mobile communications. To adapt to different requirements,
many variants of ring signature [5] were put forward, such
as ring blind signature, proxy ring signature, etc.

1. Introduction In this paper, we propose an improvement of Hu et al.’s

In Mobile Ad hoc Networks in order to ensure the service
available to the customers distributed in the whole networks,
the server needs to delegate his/her rights to some other
parties in the systems, for example to mobile agents. The
way of realizing this delegation is based on the notion of
proxy signature.
Digital signature is a cryptographic means through
which the authority, data integrity and signer’s non-
repudiation can be verified. The proxy signature is a kind of
digital signature scheme. In the proxy signature scheme, one
user called the original signer, can delegate his/her signing
capability to another user called the proxy signer. This is
similar to a person delegating his/her seal to another person
in the real world.
proxy ring signature scheme [1] in order to eliminate Zhang
and Yang’s attacks [9] on Hu et al.’s proxy ring signature
scheme. The remainder of this paper is organized as follows.
In Section 2, we give a brief overview of Hu et al.’s proxy
ring signature scheme with revocable anonymity and then
we discuss Zhang and Yang’s attacks on Hu et al.’s proxy
ring signature scheme. In Section 3, we propose an
improved version of Hu et al.’s proxy ring signature scheme
in order to withstand Zhang and Yang’s attacks in their
scheme. In Section 4, we discuss the security aspects of our
proposed improved scheme and then in Section 5, we
compare the performances of our scheme with those for Hu
et al.’s proxy ring signature scheme. Finally, we conclude
the paper in Section 6.

The notion of proxy signature was first introduced by
Mambo, Usuda and Okamoto in 1996 [3]. After that the
concept of ring signature was formalized in 2001 by Rivest,
Shamir and Tauman [4]. In the ring signature an
anonymous signature allows an user to anonymously sign on
behalf of a group, where no one can know which the actual
signer is. When verifying the verifier only knows that the
signature has been produced by some member of this ring,
but he/she has no information about who is the actual of the
signature.
Proxy signature can be combined with other special
signatures to obtain some new types of proxy signatures (for
2. Overview and Attacks on Hu et al’s Proxy
Ring Signature Scheme
In this section, we first discuss in brief Hu et al.’s proxy
ring signature scheme. We then describe Zhang and Yang’s
attacks on Hu et al.’s proxy ring signature scheme.
2.1 Overview of Hu et al.'s Proxy Ring
Signature Scheme
Hu et.al introduced a new type of signature scheme called
the proxy ring signature scheme with revocable anonymity
which combines ring signature with proxy signature.
 (IJCNS) International Journal of Computer and Network Security, 71
Vol. 2, No. 5, May 2010

In the following, we briefly review their scheme.

2.1.4 Verification Phase
2.1.1 System Parameters
For verification of the signature δ = (m, A, zN , c N , r N , V )
The following system parameters are used in describing the on message m , the verifier executes the followings:
scheme. 1. Computes h = H (m) .
• p, q : two large prime numbers, q | p − 1 . 2. For i = 1,2,⋅ ⋅ ⋅, n , computes δ i = Ai ,
*
• g : an element of Z whose order is q .
p bi = h zi δ ici ,
• x0, x1 ,⋅ ⋅ ⋅, xn : the original signer A0 's secret key and a i = g zi ( y i y 0 r i ) i
c

proxy signers U i 's secret key, where i = 1,⋅ ⋅ ⋅, n . −1

3. If ri = y0 , rejects the signature.
• y0 = g 0 mod p : A0 's public key.
H (m, a N , bN ,V) = ∑ci
x

4. Otherwise, checks whether

• yi = g i mod p : U i 's public key.
x
i∈B
• H : a hash function, H : {0,1} → Z q .
* is satisfied or not.

If the verification is valid, then the verifier accepts δ as

2.1.2 Proxy Phase a valid proxy ring signature of message m .

1. Commission Generation: For the user U i , the original 2.1.5 Open Phase
Signer A0 randomly chooses ki ∈ Z q and then To open a signature and reveal the actual identity of the
signer, the original signer checks the following.
computes sˆ i = x 0 g ki + k i mod q and rˆi = g
mod q . ki
rˆ −1

Then A0 sends (sˆi , rˆi ) secretly to U i and keeps For i = 1 to n , verifies whether g i = V 0 i .
r y i rˆ
ki secret.
2. Proxy Verification: Each user U i checks If for some i , the verification phase , it indicates that U i is
the actual signer.
whether g i = yoi rˆi mod q . If it holds, then U i
sˆ rˆ

computes s i = xi + sˆi mod q . si is his/her proxy signing

key. 2.2 Attacks on Hu et al.'s Proxy Ring Signature
Scheme
2.1.3 Signing Phase In the following, we briefly review the attacks due to Zhang
and Yang on Hu et al.’s scheme.
Let the user U i be the real signer and the ring be
B = (U 1 ,⋅ ⋅ ⋅,U n ) . On input a group size n ∈ Z , a 2.2.1 Attack on Unforgeability
message m and a public key set yN = ( y1 ,⋅ ⋅ ⋅, yn ) , the This attack consists of the following steps.
signer U i does the followings: 1. Assume that the message m is a forged message.
We then compute h = H (m) .
1. Selects d ∈R Z q and randomly computes the followings:
2. Randomly choose a number a ∈R Z q to compute
h = H (m) and δ i = h si −d .Then set A = δ i1 / i .
a i = g , and bi = h .
a a

2. Randomly chooses wi ∈ Z q and computes a i = g i

w
3. Randomly choose a number l ∈R Z q to compute
and bi = h i .
w
A= h . l

3. For all j ≠ i , picks up at random z j , c j , r j ∈ Z q and 4. For all j ≠ i , randomly select z j , c j , r j ∈ Z q and
r j ≠ y0−1 , then computes: a j = g j y j y0 r j j ,
if
z
( )
c −1
if r j ≠ y0 then compute the followings:
δ j = A j , and b j = h j δ j j
z c
aj = g
zj
(y j y0 r j ) j ,
c

4. Let a N = (a1 ,⋅ ⋅ ⋅, a n ) , bN = (b1 ,⋅ ⋅ ⋅, bn ) . U i then δ j = A j , and b j = h j δ j j .

z c

and c = H (m, a N , bN , V ) . 5. Set a N = (a 1 ,⋅ ⋅ ⋅, a n ) and bN = (b1 ,⋅ ⋅ ⋅, bn ) .

g−d
computes V = g
n 6. Randomly choose V ∈ Z p to compute
5. Computes the following ci = c − ∑c
j =1, j ≠ i
j and
H (m, a N , bN ,V ) = c .
zi = wi − ci s i + ci d . n
rˆ −1 −d
6. Finally computes ri = y0i rˆi g .
7. Set ci = c − ∑c
j =1, j ≠ i
j , and then set

zi = a − i ⋅ l ⋅ ci mod q .
Let z N = (z1 ,⋅ ⋅ ⋅, zn ) , c N = (c1 ,⋅ ⋅ ⋅, c n ) , and i ⋅l −1 −1
8. After this, compute ri = g yi y0 , and then set
rN = (r1 ,⋅ ⋅ ⋅, rn ) .
The resultant proxy ring signature on message m is rN = (r1 ,⋅ ⋅ ⋅, rn ) .
δ = (m, A, zN , c N , r N ,V ) . 9. Finally, the resultant proxy ring signature on
message m becomes δ = (m, A, zN , c N , r N , V ) .
72 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

By verifying the following three equations:

a j = g j ( y j y0 r j ) j ,
z c

δ j = A j , and 3.1 Proxy Phase

bj = h jδ j j
z c
This phase consist of the followings.

Zhang and Yang showed that the forged proxy ring 1. Commission Generation: For the user U i , the
signature is valid signature.
2.2.2 Attack on Revocable Anonymity original signer A0 randomly chooses ki ∈ Z q and

In the following, Zhang and Yang showed how to produce a then computes sˆ i = x 0 g ki mod q and
proxy ring signature in which the anonymity of the
dishonest proxy signer’s identity is not revoked.
rˆi = g ki mod q , where x0 is the original signer
A0 ’s secret (private) key. After that A0 sends
Let the user U i be the real signer and the ring be (sˆi , rˆi ) secretly to U i and keeps ki as secret.
B = (U 1 ,⋅ ⋅ ⋅,U n ) . On inputs, a group size n ∈ Z , a 2. Proxy Verification: Each user U i checks whether
message m and the public key set yN = ( y1 ,⋅ ⋅ ⋅, yn ) , the x0 ⋅ g ki
g =g sˆi
mod q . Here y0 is the original
signer U i performs the following:
signer A0 ’s public key. Then U i computes
1. Here, for j = 1 to n , the generations of
si = xi + sˆi mod q as his/her proxy signing key.
a j ,δ j , b j and A remain same as those for Hu et
al.’s scheme.
2. Sets a N = (a1 ,⋅ ⋅ ⋅, a n ) , bN = (b1 ,⋅ ⋅ ⋅, bn ) and
3.2 Singing Phase
Assume that the user U i be the real signer. Let the ring be
randomly chooses V ∈ Z p to compute
B = (U 1 ,⋅ ⋅ ⋅,U n ) . Inputs in this phases are (i ) a group size
H (m, a N , b N , V ) = c . Here we note that
generation of V is different from that of Hu et al.’s
n ∈ Z , (ii ) a message m and (iii ) a public set
scheme and V is a random number. yN = ( y1 ,⋅ ⋅ ⋅, yn ) . The signer U i then executes the
n following steps:
3. Computes ci = c − ∑c
j =1, j ≠ i
j and
Step-1. Selects d ∈R Z q randomly and then computes
zi = wi − ci si + ci d . h = H (m) and δ i = h si −d .
4. Then computes ri = y0rˆi −1 rˆi g − d . After computing these, U i sets A = δ i1 / i .
Let z N = (z1 ,⋅ ⋅ ⋅, zn ) , c N = (c1 ,⋅ ⋅ ⋅, c n ) , and Step-2. Randomly choose wi ∈ Z q and then compute
rN = (r1 ,⋅ ⋅ ⋅, rn ) .
a i = yi bi = h wi .
wi
and
Finally, the resultant proxy ring signature on message m Step-3. For all j ≠ i, selects at random
is δ = (m, A, z N , c N , r N , V ) . From the above generation
z j , c j , r j ∈ Z q such that r j ≠ y0−1 . Then computes
process of proxy ring signature, the original signer cannot
z j + x jc j
a j = yj
c c
revoke the identity of proxy signer, because of the fact that y0 j r j j ,
V in the signature δ = (m, A, z N , c N , rN ,V )is a random
δ j = A j , and b j = h j δ j j .
z c
number and the value of i can not be retrieved by using the
above open algorithm. As a result, the original signer can Step-4. Let a N = (a1 ,⋅ ⋅ ⋅, a n ) , and bN = (b1 ,⋅ ⋅ ⋅, bn ) .
not revoke the anonymity of the proxy signer’s identity. −d
y0−1
U i then computes V = g yi and
c = H (m, a N , bN ,V ) .
3. Improved Proxy Ring Signature Scheme
n
In this section, we describe our improved signature scheme
in order to eliminate the security flaws discussed in Section
Step-5. Computes ci = c − ∑c
j =1, j ≠ i
j and

2.2. We use the same set of system parameters as used in Hu zi = wi − ci si + ci d .

et al.’s proxy ring signature scheme given in Section 2.1.1.
Step-6. Computes ri = y0xi rˆi −1 yi− d .
The different phases of our improved scheme are given
in the following subsections.
Let z N = (z1 ,⋅ ⋅ ⋅, zn ) , c N = (c1 ,⋅ ⋅ ⋅, c n ) , and
 (IJCNS) International Journal of Computer and Network Security, 73
Vol. 2, No. 5, May 2010

rN = (r1 ,⋅ ⋅ ⋅, rn ) . a i = yizi yixi ci y0ci ri ci = yiwi −ci si +ci d yixi ci y0ci y0xi rˆi −1 yi−d [ ]ci

Finally, the resultant proxy ring signature on message m is

δ = (m, A, zN , c N , r N ,V ) . = yiwi yi−ci si yi
ci d
yixi ci y0ci y0xi ⋅rˆi ⋅ci y0−ci yi−ci d
= yiwi yi−ci [xi + sˆi ] yixi ci y0xi ⋅rˆi ⋅ci

3.3 Verification Phase

= yiwi yi−ci xi yi−ci sˆi yixi ci y0xi ⋅rˆi ⋅ci
To verify the signature δ = (m, A, z N , c N , rN ,V ) on = yiwi yi−ci [x0 rˆi ] y0xi ⋅rˆi ⋅ci
message m , the verifier needs to execute the following = yiwi yi−ci x0rˆi yici x0rˆi = yiwi
steps.

Step-1. Computes h = H (m) . Thus if the proxy ring signature is generated by a valid
Step-2. For i = 1,⋅ ⋅ ⋅, n , computes member in the ring, the verification of the equation

a i = yi
zi + xi ci
y0ci ri ci , H (m, a N , bN ,V) = ∑ci passes.
i∈B
δ i = Ai , and
bi = h zi δ ici . 4.2 Correctness of Open Phase

Step-3. If ri = y0−1 , then the verifier immediately rejects We prove the correctness of open algorithm as follows:
ˆ −1
yi − d y 0x i r i y 0− 1 y i− d
ˆ
y 0x i r i
the signature. g ri
= g = g
ri ≠ yo−1 , the verifier checks the
( )
Step-4. Otherwise, if ˆ
y 0x i r i
y 0− 1 y i− d
x rˆi x 0 x i rˆi
= g = V = V
H (m, a N , bN ,V) = ∑ci
y0 i g

validity of the equation: x 0 rˆi sˆ i

i∈B = V yi
=V yi

If the above equation is valid , then the verifier accepts δ

as a valid proxy ring signature on message m .
4.3 Security Analysis
3.4 Open Phase
In this section, we show that our scheme preserves the
In order to open a signature and reveal the actual identity of unforgeability as well as revocable anonymity properties
the signer, the original signer can verify the following
equation: • Unforgeability: The improved scheme satisfies the
sˆ unforgeability property as follows.
For i = 1 to n , checks whether g = V ri yi i
.
Assume that B denotes the ring. We prove that
If the verification passes for some i , then the user U i is the any adversary A∉ B can not forge the valid ring
actual signer. signature. Suppose A can repeat the proxy phase
and also query the original signer for ŝ k and rˆk ,
4. Analysis of Our Improved Scheme where k ∈ B in order to receive some proxy
In this section, we give correctness proofs for our signature key which does not belong the ring
verification phase as well as open phase. We then describe B . To forge a ring signature, A needs to execute
the security analysis of our scheme. the signing phase. Let A select randomly
zi , ci , ri ∈ Z q , for all i ∈ B and compute
4.1 Correctness of Verification Phase
c = H (m, a N , bN ,V ) . However, this becomes
We prove the correctness of verification process as follows: impossible because in order to compute
For the signer i , δi = A = δ i
( ),
1/ i i
i a i = yi
zi + xi ci ci ci
y r , A needs the private key xi
0 i

of the proxy signer U i . Computation of the

bi = h zi δ ici = h wi −ci si + ci d δ ici = h wi h ci d −ci si δ ici
private key xi from the public key yi is
= h wi h −ci ( si − d )δ ici = h wi δ i− ci δ ici = h wi computationally infeasible, since it is based on the
discrete logarithm problem (DLP) for a large
prime q. As a result, except U i no one can
calculate a i and hence, our scheme satisfies the
unforgeability property.
74
• Revocable anonymity: Let an untrustworthy proxy
signer can produce a proxy ring signature which
can pass the verification phase. But then the
original signer can revoke the anonymity of his
identity. Now, to open the signature and reveal the
actual identity of the signer the verification of the
sˆ i
equation g ri = V yi , where i = 1,⋅ ⋅ ⋅, n is
needed. If for some i , the check is valid, then U i
becomes the actual signer. This is impossible again
because V in the signature
δ = (m, A, zN , c N , r N ,V ) itself is not a random
number. It depends upon the public key of the
original signer and the proxy key of proxy signer as
−d
y0−1
V = g yi .Thus, our scheme also preserves the
property of revocable anonymity.
5. Performance Comparison of Our Scheme
with Hu et al.’s Proxy Ring Signature Scheme
In this section, we compare the computational costs of
different phases of our scheme with those for Hu et al’s
scheme.
Table 1: Performance comparison between our scheme and
Hu et al.’s proxy ring signature scheme.
Notes: TM: time taken for a modular multiplication, TE:
time taken for an exponential operation, TH: time taken for a
one-way hash function.
We have compared the computational costs of different
phases of our scheme with those for Hu et al’s scheme in
Table 1. From this table, we see that our scheme requires
one modular multiplication less than Hu et al’s scheme.
Thus, the computational costs of our scheme are also
comparable with Hu et al’s signature scheme.
6. Conclusion
In this paper, we have proposed an improved proxy ring
sig-nature with revocable anonymity to eliminate the
security flaws in Hu et al's scheme. The proposed scheme
allows the original signer to know exactly who the signer
is. We have given correctness proofs of our scheme and
analyze the security aspects of our scheme. The security
of the proposed scheme is based on the security of the
DLP problem. We have also shown that our scheme
preserves the properties of unforgeability as well as
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
revocable anonymity.
References
[1] C. Hu, P. Liu, and D. Li. New Type of Proxy Ring Sig-
nature Scheme with Revocable Anonymity and No Info
Leaked. In Multimedia Content Analysis and Mining
(MCAM 2007), LNCS, volume 4577, pages 262–266,
2007.
[2] W.D. Lin and J.K. Jan. A security personal learning tools
using a proxy blind signature scheme. In Proceedings of
International Conference on Chinese Language
Computing, Illinois,USA, pages 273–277, July 2000.
[3] M. Mambo, K. Usuda, and E. Okamot. Proxy
signatures: delegation of the power to sign message .
IEICE Transaction Functional, E79-A(9):1338–1353,
1996.
[4] R.L. Rivest, A. Shamir, and Y. Tauman. How to leak a
secret. In Advances in Cryptology-Asiacrypt 2001,
LNCS, volume 2248, pages 552–565. Springer-Verlag,
2001.
[5] Kyung-Ah Shim. An Identity-based Proxy Signature
Scheme from Pairings. In ICICS 2006, LNCS, volume
4307, pages 60–71. Springer-Verlag, 2006.
[6] J. Xu, Z. Zhang, and D. Feng. ID-Based Proxy Signature
Using Bilinear Pairings. In ISPA Workshops, LNCS,
volume 3759, pages 359–367. Springer-Verlag, 2005.
[7] L. Yi, G. Bai, and G. Xiao. Proxy multi-signature
scheme: a new type of proxy signature scheme.
Electronics Letters, 36(6):527–528, 2000.
[8] F. Zhang and K. Kim. Efficient ID-based blind
signature and proxy signature from pairings. In
Information Security and Privacy, volume 2727, pages
218– 219. Springer Berlin, 2003.
[9] J. Zhang and Y. Yang. On the Security of a Proxy
Ring Signature with Revocable Anonymity. In 2009
International Conference on Multimedia Information
Networking and Security (MINES), volume 1, pages
205–209, 2009.
[10] K. Zhang. Threshold proxy signature schemes. In 1997
Information Security Workshop, Japan, volume 1396,
pages 282–290. Springer Berlin, September, 1997.
 (IJCNS) International Journal of Computer and Network Security, 75
Vol. 2, No. 5, May 2010

Coloring of Triangle Free Girth Graphs with

Minimal k-Colors
Dr. Mohammad Malkawi1 and Dr. Mohammad Al-Hajj Hasan2
1
Middle East University, Amman, Jordan, mmalkawi@aimws.com
2
Middle East University, Amman, Jordan hajhasan100@yahoo.com

Abstract: It is known that there exist triangle free graphs with
arbitrary high chromatic number. A recursive construction of
such graphs was first described by Blanches Descrates and
Mycielski. They have shown that for any positive integer k, there
exists a k-chromatic graph containing no triangle. In this paper,
we introduce an algorithm which detects the chromatic number
of an arbitrarily large girth-base graph and assigns colors to its
nodes using the minimal number of colors, namely with exactly
k colors.. The algorithm complexity is O(Lg(n)), where n is the
number of nodes of the graph.
that, while the number of colors grows linearly, the number
of nodes grows exponentially. Just like any other graph,
girth-base graphs once constructed are difficult to color with
the minimum number of colors, especially when the number
of nodes is relatively large.
Lemma 1: If a graph is a Girth-g triangle free graph with n
nodes, then there exists a k-coloring of the graph, where k =
lg [(n+1)/3] + 2

Keywords: Graphs, Graph Girth, Graph Coloring, Chromatic

Number, Triangle-Free Graphs, Coloring Algorithm.

1. Introduction
There exist triangle free graphs with arbitrary high v1
chromatic number [4,5,6,7]. A recursive construction of g- v2
girth graphs is described in [1,2,3]. Blanches Descrates and
Mycielski have shown that for any positive integer k, there v5
exists a k-chromatic graph containing no triangle [9]. In this v3
paper, we will introduce an algorithm, which colors such
graphs with exactly k colors. First, we show how the k- v4
chromatic graphs are constructed using girth graphs.
A girth g-graph is a triangle free graph with the
shortest cycle containing g nodes. Blanches Descrates and Level 0, Girth 5, 3 Colors Graph
Mycielski [9] showed how to construct a k-chromatic graph
using a 5-girth graph (Figure 1, level 0).
v1
Call the original 5-node graph a level (0) graph. Add 6
new nodes to the graph (u1, u 2, u 3, u 4, u 5) and u.
Connect node ui to the neighbors of node vi . Obviously u1 v2
nodes ui and vi can be colored with the same color, since
they are not interconnected. Now connect u to (u1 , u 2, u 3 ,
v5 u5 u2
u 4 , u 5). Node u must be colored with a new color. The
resulting graph is the Grötzch graph with 11 nodes, no u
triangles, girth 4, and exactly 4 colors. Call this graph a
level (1) graph. In the same manner, we can construct a 5-
u4 u3
chromatic graph from the 11-node Grötzch graph. We add
12 new nodes (11+1). We connect each of the newly added
11 nodes to the neighbors of one and only one of the nodes v3
in the original 11-node graph. The newly added nodes will
be colored with the same colors as the original 11 nodes. v4
The 12th node connects to the newly added 11 nodes and
requires a new color. The total number of colors used is 5
colors. Call this graph a level (2) graph. By repeating this Level 1, 4 Colors Graph
process, we can obtain a triangle free graph with any
chromatic number k= 3+i, where (i) is the level of the Figure 1: Girth g Graph Construction and Coloring
graph.

The number of nodes in a graph constructed in this

manner is n = (5*2i)+( 2i) – 1 = 6*2i– 1 = 3*2k-2 – 1. Note
76
This paper introduces an algorithm which detects the
chromatic number of an arbitrarily large girth-base graph
and assigns colors to its nodes using the minimal number of
colors. The algorithm complexity is O(Lg(n)). A formal
definition of the algorithm is provided in the next section.
2. Coloring of k-chromatic Girth Graphs
2.1 Algorithm Color Girth Triangle Free Graphs
Step 1: Sort the nodes in the graph based on the degree of
the nodes (largest first)
Step 2:
• Select the first node in the list
• Color the node with the smallest available color (color 1
for the first node)
• Check the neighbors of this node; Color all the
uncolored nodes in the list of neighbors
• Move to the next node in the main list
Step 3: Repeat Step 2 until all nodes are colored
Sort the nodes in descending order based on the nodes
degree. Note that node 23 (with the highest degree of 11) is
the last node added at level 2 and node 11 (with the next
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
The idea behind the algorithm was first observed when
constructing an exam scheduling algorithm with special
constraints [8].
Theorem 1: Algorithm “Color Girth triangle Free Graphs”
will color all k-chromatic girth-g graphs with exactly k
colors.
First, we will show an example of how the algorithm can
color the constructed graphs with minimal number of colors.
The example will provide a ground for proof by induction.
We will also provide a formal proof for the theorem.
Example 1: Consider a level 2, 5-chromatic girth-4 graph.
The number of nodes in the graph is 23. The graph is
illustrated in a matrix format in Table 1. For this graph the
chromatic number is k=lg[(23+1)/3]+2 = 3+2 = 5.
The graph is initially constructed with nodes 1-5 (level 0).
At level 1, nodes 6-10 are added; node 6 connects to
neighbors of 1, node 7 connects to neighbors of 2, and so
on. Then node 11 is added and connected to nodes 6-10. At
level 2, we add nodes 12-22. Node 12 connects to neighbors
of nodes 1 through 11 respectively. Then node 23 is added
and connected to nodes 12-22. The degree of each node is
shown in the table.
highest degree of 10) is the last node added at level 1. Let’s
call nodes 23 and 11 the roots of level 2 and 1 respectively.
According to the algorithm, we start with node 23. We color
node 23 with the first available color, say the red color.
 (IJCNS) International Journal of Computer and Network Security, 77
Vol. 2, No. 5, May 2010

Node 23 has 11 neighbors. All the neighbors can be colored c. 3rd stage: 3 nodes (23+1/8). Number of colors is 1
with the same color, since by definition they are not d. 4th and 5th stages: 2 nodes. Number of colors is 1
interconnected with each other. So we color nodes 12-22 2. The total number of steps required to color the graph is
with blue. The number of nodes colored at this step is 5 = Lg 23
(1+11) = 12. Actually this is equal to the number of nodes
added at level 2, i.e., (N+1)/2.
2.2 Proof:
We now present a formal proof of the theorem. By
Next we move down the list to the root node at level 1,
construction, a girth graph consists of two major parts; the
i.e., node 11 with degree 10. By definition, node 11 is not
foundation, which is the 5-node cycle, and a multi-level
connected to node 23. Thus it may reuse the red color. Half
structure of nodes. Each level adds twice the number of
of the neighbors of node 11 are also neighbors of node 23
nodes to the previous level plus one root. The root at each
since they were added at the second level, and thus they
level is connected only to the newly added nodes, and is
must have been colored at the previous step with the blue
isolated from all the nodes added at the previous level(s).
color (nodes 19-22). The other half (nodes 6-10) were added
The largest degree node is the root of the last level in the
at level 1. Each of these nodes is connected to at least one of
graph. The degree of this node is d=(N-1)/2. The next
the nodes that have been colored at the previous step. A new
largest degree node is the root of the next level down, with
color, say color green, is required to color nodes 6-10. Note
degree = d-1, and so on.
that nodes 6-10 are not connected within each other, so they
The algorithm sorts the nodes of the graph in a
all can use the same color. At the end of this step, all nodes
descending order based on the degree of nodes. By
added at level one (5+1) are colored, and only one new color
definition, the largest degree node is the root of the last
is added.
(highest) level.
Assume that we construct a girth graph with (i) levels.
The next node in the list is node 5. Note that at this
For (i=0), the graph has 5 nodes. For (i=1), the graph has 11
stage, we have arrived at the original 5 nodes cycle (1
nodes. For (i=5), the graph has 191 nodes. At the (ith) level,
through 5). All of these nodes have the same degree. So we
the number of nodes is N=(5*2i)+( 2i)-1. The node with the
can start with any of them and get the same result. Node 5
largest degree is the last node added to the graph, i.e., the
has 8 neighbors. 6 out of 8 nodes have already been colored
root at level (i) and is connected to (N-1)/2-1 nodes; so the
with colors green and blue. Node 5 can take the color red.
degree at level (i) di = (3*2i-1). The next largest degree node
Note that color red is only used by the root nodes (11 and
is the root node added at the previous level (i-1). This node
23) at levels 1 and 2. These nodes do not connect to any of
is connected to nodes at the current level (i-1) and nodes at
the inner most 5-nodes cycle. The neighbors of 5 (nodes 1
the upper level (i). The number of nodes connected at the
and 4) require a new color, say the yellow color. The total
current level (i-1) = (3*2i-1-1). The number of nodes
number of nodes colored at this stage is (2+1). It is worth
connected to the root at the upper level (i) is = (3*2i-1 -1). So
noting that the inner 5-node cycle requires at most 3 colors.
the degree at level (i-1), di-1 =2*(3*2i-1-1) = 3*2i-2. The
One of the already used colors can certainly be reused. This
next largest degree node is the root at level (i-2). The root is
is the color given to the root nodes added at each of the
connected to (3*2i-2-1) nodes at level (i-2). It is connected to
subsequent levels of the graph construction. So, when the
same number of nodes at level (i-1). It is also connected to
algorithm reaches the last 5 nodes in the graph, only 2 new
2*(3*2i-2-1) at level (i). So the degree at level (i-2) di-2 =
colors will be required.
(3*2i-2-1) + (3*2i-2-1) + 2*(3*2i-2-1) = (3*2i-2-1)*22. In
The next node in the list is node 1. Node 1 has already been
general, the degree of a root node at level (i-k), where k =
colored with the yellow color. All the neighbors of node 1
0,1, 2, … (i-2) is given by di-k= (3*2i-k-1)* 2k.
have been colored except for node 2, which can take the
For example, let (i=5). The total number of nodes in
color red. The next node in the list is node 2. Node 2 has
the graph is (6*25-1) = 191. d5= (3*25-1) = 95; d4= 2*(3*24 -
already been colored. All of its neighbors are colored except
1) = 94; d3= 22*(3*23 -1) = 92; d2= 23 *(3*22-1) = 88.
for node 3. Node 3 requires a new color, say the color pink.
The algorithm sorts the nodes based on node degrees
in a descending order. Thus, the algorithm will color the
At this stage all the nodes of the graph have been
nodes starting with the roots at each level.
colored with five colors, and the algorithm terminates. The
At level (i), the algorithm will color the root node and
termination criterion of the algorithm is simple. The
all its adjacent nodes. Since the adjacent nodes are not
algorithm maintains a list of all the nodes that have not been
interconnected, they can assume the same color. The root
colored. Every time a node is colored, the node will be
will have a different color. So step one, will consume two
removed from the list. The algorithm terminates when the
colors. The next step will take care of the nodes at the next
list becomes empty.
lower level. Since the root at this level is not connected to
We observe the followings from the process of coloring the
the root at the upper level, it can assume the color of the
graph.
root at the upper level; so no new color is consumed. The
1. The number of nodes and colors covered at each stage
neighbors of the root at this level are split in two sets. One
are as follows:
set belongs to the upper level, so this set has already been
a. First stage: 12 nodes = ((23+1)/2). Number of colors
colored. The second set is not inter-connected, so the nodes
used is 2
of this set can all be colored with the same color. A new
b. 2nd stage: 6 nodes = ((23+1)/4). Number of colors is
color will be consumed. Similarly, at the next level, the root
1
78 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

can assume the same colors of the previous roots, since they
are not interconnected. The neighbors of this node are split Authors Profile
in 3 sets. Two sets belong to the upper 2 levels and one set Dr. Mohammad Malkawi received his
belongs to the current level. The nodes, which belong to the Ph.D. degree in computer engineering
upper 2 levels, have been colored at the previous steps. The from the University of Illinois at Urbana-
nodes in the current level are not interconnected, so they can Champaign in 1986. He received his BS
assume one color. Continuing in this manner, the algorithm and MS in computer engineering in 1980
and 1983. He was a member of the High
will consume one color at each level until it reaches the
Productivity Computing Systems team at
lowest 5 nodes in the graph, where it will consume 2 new
SUN Microsystems, and the high availability platform team at
colors. Motorola. Currently, he is associate professor at Middle East
Thus, the algorithm detects the structure of the graph University in Jordan.
and colors the nodes with the minimal number of
Mohammad M Al-Haj Hassan obtained
colors 2 + ∑1 . The algorithm also uses minimum number
i +1
his B.Sc. and Master degrees from The
University of Jordan in 1973 and 1977
of steps because it colors the roots and their neighbors first. respectively, and his Ph.D. degree in
Computer Science from Clarkson
University at NY / USA. His main field
3. Conclusions of Specialization is Computer Algorithms
This paper presents a coloring algorithm which colors with specific concentration on Graph
a girth-g, triangle free graphs with arbitrarily large k- Algorithms & Their Applications. Other
chromatic number. The algorithm detects the structure of a fields of research are: Machine Learning,
girth-g graph and colors the graph with exactly k colors. Parallel Computations and Algorithms,
The algorithm terminates in O(Lg(n)) steps. Distance Learning and Web-Based Courses Design. Professor M.
Al-Haj Hassan was the Dean of several Faculties in several
universities. He is a member in the Editorial Board and/or referee
of several journals. He has been the Secretary General of The
International Arab Conference on Information Technology (ACIT)
References
for 4 years. Currently, he is a professor of Computer Science and
[1] Bollobas,B., Sauer,N.: ”Uniquely Colorable Graphs
the Vice President at Middle East University in Jordan.
with Large Girth,” Canad. J. Math. 28 (1976), no.
6, 1340-1344.
[2] Emden-Weinert,T., Hougardy,S., Kreuter,B.:
”Uniquely Colorable Graphs and the Hardness of
Coloring Graphs of Large Girth,” Com-bin.
Probab. Comput. 7 (1998), no. 4, 375-386.
[3] Erdos,P.: ”Graph Theory and Probability,” Canad.
J. Math 11 (1959), 34-38.
[4] Muller,V.: ”On Colorings of Graphs Without Short
Cycles,” Discrete Mathematics 26 (1979), 165-176.
[5] Nesetril,J., Zhu,X.: ”Construction of Sparse Graphs
with Prescribed Circular Colorings,” Discrete
Mathematics 233 (2001), 277-291.
[6] Steffen,E., Zhu,X.: ”Star Chromatic Numbers of
Graphs,” Combinatoria 16 (1996), no. 3, 439-448.
[7] Zhu,X.: ”Uniquely H-colorable Graphs with Large
Girth”, J. Graph Theory 23 (1996), no. 1, 33-41.
[8] Mohammad Malkawi, Mohammad Al-Haj Hassan,
and Osama Al-Haj Hassan, “A New Exam
Scheduling Algorithm Using Graph Coloring”, The
International Arab Journal of Information
Technology, Vol. 5, No. 1, 2008; pp. 80-86.
[9] Soifer, Alexander “Mathematics of Coloring and
the Colorful Life of its Creators”, Springer; pp, 83-
86; ISBN: 978-0-387-74640-1
 (IJCNS) International Journal of Computer and Network Security, 79
Vol. 2, No. 5, May 2010

An Efficient Solution for Privacy Preserving

Association Rule Mining
Chirag N. Modi1, Udai Pratap Rao1 and Dhiren R. Patel2
1
Department of Computer Engineering, Sardar Vallabhbhai National Institute of Technology,
Ichchhanath, Surat, India-395 007
cnmodi.956@gmail.com
upr@coed.svnit.ac.in
2
Department of Computer Science & Engineering, Indian Institute of Technology Gandinagar,
Ahmedabad, Gujarat, India-382 424
dhiren@iitgn.ac.in

suppose the Lipton Tea supplier misuse the database and

Abstract: The recent advances of data mining techniques
played an important role in many areas for various applications.
However the misuse of these techniques may disclose the private
or sensitive information which database owners do not want
reveal to others. To preserve privacy for sensitive association
rules in statistical databases, many approaches have been found.
In this paper, we propose two heuristic blocking based
algorithms named ISARC (Increase Support of common
Antecedent of Rule Clusters) and DSCRC (Decrease Support of
common Consequent of Rule Clusters) to preserve privacy for
sensitive association rules. Proposed algorithms cluster the
sensitive rules based on some criteria and hide them in fewer
selected transactions by using unknowns (“?”). They preserve
certain privacy for sensitive rules in database, while
maintaining knowledge discovery. An example illustrating
proposed algorithms is given. We also discuss the performance
results of both the algorithms.
Keywords: Data Mining, Association Rule Hiding,
Knowledge Hiding, Clustering, Sensitivity, Database Security and
Privacy.
1. Introduction
Recently privacy preserving data mining (PPDM) become a
hot directive to prevent disclosure of sensitive knowledge or
rules by data mining techniques. M.Attallah et al. [1] was
the first to present disclosure limitation of sensitive
knowledge by data mining algorithms and proposed
heuristic algorithms to prevent disclosure of sensitive
knowledge. The authors in [2] presented some suggestions
for defining and measuring privacy preservation and showed
how these relate to both privacy policy and practice in wider
community and to techniques in privacy preserving data
mining. Oliveria and Saygin et al. [4] presented taxonomy
of attacks against sensitive knowledge. Clifton et al. [7] was
the first to show the importance of association rule hiding
and also discussed security and privacy implication of data
mining.
1.1 Motivation
Consider, as a tea reseller we are purchasing tea at low price
from two companies, Tata Tea Ltd. and Lipton Tea Ltd.,
while granting them to access our customer database. Now,
mines association rules related to the Tata Tea, saying that
most of the customers who buy bread also buy Tata tea.
Lipton tea supplier now runs a coupon scheme that offers
some discount on bread with purchase of Lipton Tea. So, the
amount of sales on Tata Tea is down rapidly and Tata Tea
supplier cannot offer tea at low price to us as before and
Lipton monopolizes the tea market and is not offering tea at
low price to us as before. As a result, we may start losing
business to our competitors. So, releasing database with
sensitive knowledge is bad for us. This scenario leads to the
research of sensitive knowledge (or rule) hiding in database.
The knowledge privacy problem, in context of association
rule mining which is proved NP hard in [1]. There are
several heuristic techniques have been found to solve above
problem. M.Attallah et al. [1] were the first to propose
heuristic algorithms which decrease significance of the rules
for hiding the sensitive rules in databases. Verykios et al. [5]
proposed five algorithms which reduce the support or
confidence of sensitive rules by removing or inserting some
items in selected transactions. These algorithms hide only
one sensitive rule or sensitive item at a time and provide
certain privacy. Y-H Wu et al. [10] proposed a method to
reduce the side effects produced by sanitization, which
formulates set of constraints related to the possible side
effects and allows item modification based on more
significant constraint. K.Duraiswamy et al. [9] proposed an
efficient clustering based approach that clusters the sensitive
rules based on common item in R.H.S. of the sensitive rules
and hides the R.H.S. items in each cluster by reducing
support of it.
Y.Saygin et al. [8][11] were the first proposed blocking
approach to increase or decrease the support of item by
placing unknowns (“?”) in place of 1’s or 0’s .So, It is
difficult for an adversary to know the value behind
unknowns (“?”). The safety margin is also used to show how
much below the minimum threshold new significance of a
rule should. This approach is effective and provides certain
privacy. Wang and Jafari [6] proposed more efficient
approaches than other approaches as in [8][11]. They
require less number of database scans and prune more
80
number of rules, while hiding many rules at a time. But they
produce undesirable side effects.
In this paper, we propose two heuristic blocking based
algorithms named ISARC (Increase Support of common
Antecedent of Rule Clusters) and DSCRC (Decrease
Support of common Consequent of Rule Clusters) to
preserve privacy for sensitive association rules in database.
To hide sensitive rules, proposed algorithms replace the 1’s
or 0’s by unknowns (“?”) in fewer selected transactions for
decreasing their confidence. So, it is difficult for an
adversary to know the value behind unknowns (“?”). To
decrease the confidence of specified rules, first algorithm
increases the support of rule antecedent, while another
decreases the support of rule consequent. They can hide
many rules at a time in rule clusters. So, they are more
efficient than other heuristic approaches. Moreover they
maintain data quality in database.
The rest of this paper is organized as follows: In section
2, we discuss some related background and concepts, while
a detailed description of proposed algorithms (ISARC and
DSCRC) is given in section 3. An example illustrating
ISARC and DSCRC is given in section 4. In section 5 we
discuss and analyze the performance results of both the
proposed algorithm. Finally in section 6, we conclude
proposed algorithms by defining some future trends.
2. Theoretical Background and
Preliminaries
The standard definition of association rule is given in [5],
which is as follows: Let I={i1,…,im}be a set of items.
Database D={T1,….,Tn}is a set of transactions, where
Ti⊆I(1≤i≤m). Each transaction T is an itemset such that
T⊆I. We say that a transaction T supports X, a set of items
in I, if X⊆I. The association rule is an implication formula
like X⇒Y, where X⊂I, Y⊂I and X∩Y=∅. The itemset
X∪Y called generating itemset which lead to the generation
of an association rule which is consists of two parts: left
hand side (L.H.S.) item (X) called rule antecedent and right
hand side (R.H.S) item (Y) called rule consequent. The rule
with support s and confidence c is called, if |X∪Y|/|D| ≥ s
and |X∪Y|/|X| ≥ c. Because of interestingness, we consider
user specified thresholds for support and confidence, MST
(minimum support threshold) and MCT (minimum
confidence threshold). A detailed overview of association
rule mining algorithms are presented in [3].
2.1 Problem Description
Let given dataset D, a set of association rules R over D is
given and also Rh⊆R, Rh is specified as sensitive rules set.
Now, the problem is to find sanitized database D’ such that
there exist only a set of rules R-Rh, can be mined. It should
achieve one of the following goals: (1) All the sensitive
association rules must be hidden in sanitized database. (2)
All the rules that are not specified as sensitive can be mined
from sanitized database. (3) No new rule that was not
previously found in original database can be mined from
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
sanitized database. First goal considers privacy issue.
Second goal is related to the usefulness of sanitized dataset.
Third goal is related to the side effect of the sanitization
process.
2.2 Basic Concepts
2.2.1 Rule Hiding by Reducing Confidence
To hide an association rule like X ⇒Y, we decrease its
confidence (|X∪Y|/|X|) to be smaller than pre-specified
minimum confidence threshold (MCT). To decrease the
confidence of a rule, two strategies as in [5] can be
considered. The first one is to increase the support of rule
antecedent (X), but not support of generating itemset
(X∪Y). The second one is to decrease the support of
generating itemset (X∪Y). To decrease support of
generating itemset, we decrease the support of rule
consequent Y only in transactions containing both X and Y.
So, it can reduce the rule confidence faster than reducing the
support of X.
2.2.2 Increasing or Decreasing Support of item by
using Unknowns
Instead of distorting transactions, we increase or decrease
support of one item at a time in a selected transaction by
replacing from 0 to unknowns (“?”) or 1 to unknowns (“?”)
respectively. A detailed overview of unknown technique to
decrease or increase support of item is given in [8].
2.2.3 Clustering
We cluster the sensitive rules based on common antecedent
or common consequent to hide many rules at a time by
replacing fewer 1’s or 0’s to unknowns (“?”) in selected
transactions.
2.2.4 Sensitivity
We use sensitive characteristics of sensitive rules to select
transactions for modification. A detailed overview of
sensitivities is given in [13]. Following sensitivities are used
in our proposed algorithms: (1) Item Sensitivity: is the
frequency of data item exists in the number of the sensitive
association rules containing sensitive item. It is used to
measure rule sensitivity. (2) Rule Sensitivity: is the sum of
the sensitivities of all items containing that association rule.
(3) Cluster Sensitivity: is the sum of the sensitivities of all
association rules in cluster. Cluster sensitivity defines the
rule cluster which is most affecting to the privacy. (4)
Sensitive Transaction: is the transaction in given database
which contains sensitive item. (5) Transaction sensitivity: is
the sum of sensitivities of sensitive items contained in the
transaction.
3. ISARC and DSCRC - Proposed Privacy
Preserving Association Rule Mining
Algorithms
3.1 Proposed Framework
The proposed common framework for ISARC and DSCRC
is shown in Fig. 1. As shown in Fig. 1, First both the
 (IJCNS) International Journal of Computer and Network Security, 81
Vol. 2, No. 5, May 2010

algorithms mine the association rule from given database D
by using association rule mining algorithms e.g. Apriori
algorithm [3]. Then some rules from mined rules are
specified as sensitive. Now the processes of proposed
framework are different for both the algorithms. ISARC
algorithm groups sensitive rules based on common L.H.S.
item (rule antecedent) of specified rules, where DSCRC
algorithm groups sensitive rules based on common R.H.S.
item (rule consequent) of specified rules. For each rule
cluster, ISARC indexes the transactions which are not
supporting to common antecedent of cluster, where DSCRC
indexes the fully supporting transaction to cluster. Then
both the algorithms calculate all the sensitivities as
explained in section 2.2.4.
Now, both the algorithms sort the clusters based on
decreasing order of their sensitivities. For each sorted
clusters, ISARC sorts the indexed transaction in ascending
order of their sensitivities, where DSCRC sorts the indexed
transaction in descending order of their sensitivities. Then
rule hiding procedure of ISARC increases the support of
common antecedent item of each cluster in sorted
transactions, where rule hiding procedure of DSCRC
decreases the support of common consequent item of each
cluster in sorted transaction, until all the rules are not
hidden in each cluster. Rule hiding procedure increases or
decreases the support of item based on strategy mentioned in
section 2.2.1 and section 2.2.2. After hiding process, both
the algorithms update the modified transaction into original
database and produce the final privacy aware database for
releasing.
Figure 1. Proposed common framework for ISARC and
DSCRC algorithms
3.2 ISARC Algorithm
The proposed ISARC algorithm is shown in Fig. 2. A source
database D, minimum confidence threshold MCT, minimum
support threshold MST and safety margin SM is given as
input to ISARC. Safety margin used to show how much
below the new minimum confidence or minimum support of
rule should [8]. According to proposed framework, ISARC
algorithm first generates number of possible association
rules satisfying MST and MCT from the given database D.
Then it selects a set of sensitive rules RH with single
antecedent and consequent for hiding and clusters them
based on common antecedent. Then it finds sensitivity of
each cluster and sorts them in decreasing order of their
sensitivity.
Now, for each cluster, it indexes transactions which are
not fully supporting to common antecedent of cluster and
sorts transactions for the first cluster in ascending order of
their sensitivity. The hiding process of ISARC tries to hide
all the sensitive rules by placing an unknown (“?”) for the
place common antecedent item of the rule clusters, from the
sensitive transactions. While loop continues until all the
rules are not hidden in cluster. A detailed overview of
maxsupport, minsupport, maxconfidence and minconfidence
is given in [8]. Every time in while loop it updates the
sensitivity of modified transactions in other cluster and sorts
them for other clusters. Finally algorithm updates all the
modified transactions in original database and produces
sanitized database D’.
INPUT: Source database D, Minimum Confidence Threshold
(MCT), Minimum support threshold (MST), SM.
OUTPUT: The sanitized database D’.
1. Begin
2. Generate association rules from the database D.
3. Selecting the Sensitive rule set Rh with single antecedent
and consequent e.g. x⇒y.
4. Clustering-based on common item in L.H.S. of the
selected rules.
5. Find the sensitivity of each cluster.
6. For each cluster, index the transactions that are not fully
support the common antecedent item of cluster.
7. Sort generated clusters in decreasing order of their
sensitivity.
8. For the first cluster, sort selected transaction in
ascending order of their sensitivity.
9. For each cluster c∈ C
10. {
11. While(sensitive rule set Rh ∈ c is not empty)
12. {
13. Take first transaction of cluster c.
14. Place an unknown (“?”) in the transaction for the
place of common antecedent of c.
15. Update the sensitivity of modified transaction for
other cluster and sort it.
16. For i = 1 to no. of rules∈c
17. {
18. Update maxsupport of antecedent of r∈Rh∈c
and minconfidence of the rule r.
19. If(minconfidence of r < MCT-SM)
20. {
21. Remove rule r from Rh.
22. }
23. }
24. Take next transaction of cluster c.
25. }
26. End while
27. }
28. End for
29. Update the modified transactions in D.
Figure 2. Proposed ISARC (Increase Support of common
Antecedent of Rule Cluster) algorithm
82 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

3.3 DSCRC Algorithm Both the proposed ISARC and DSCRC algorithms hide
The proposed DSCRC algorithm is shown in Fig. 3. It many rules in an iteration of hiding process and modify
clusters the sensitive rules based on common R.H.S. item of fewer transactions in database. But DSCRC hides rules
rules and sorts sensitive transaction in decreasing order of faster than ISARC, because it reduces the confidence of
their sensitivities. sensitive rule faster than ISARC. The safety margin is also
The hiding process of DSCRC tries to hide all the used in both the algorithms to provide better privacy.
sensitive rules by placing an unknown (“?”) for the place
common consequent of the rule clusters, from the sensitive
transactions. While loop continues until all the rules are not
4. Example
hidden in cluster c. Every time in while loop it updates the A sample transaction database D taken from [8] is shown in
sensitivity of modified transaction in other cluster and sorts Table 1. TID shows unique transaction number. Binary
them for other clusters. Finally algorithm updates all the valued item shows whether an item is present or absent in
modified transactions in original database and produces that transaction. Suppose MST, MCT and SM are selected
sanitized database D’, in which most of the sensitive rules 50%, 70% and 0 respectively. Table 2 shows frequent
are hidden. itemsets satisfying MST, generated from sample database D.
In following, the possible number of association rules
INPUT: Source database D, Minimum Confidence Threshold satisfying MST and MCT, generated by Apriori algorithm
(MCT), Minimum support threshold (MST).
[2] are given: 0⇒1, 1⇒0, 0⇒3, 3⇒0. Suppose the rules
OUTPUT: The sanitized database D’. 0⇒1 and 0⇒3 are specified as sensitive and should be
hidden in sanitized database. There are one different L.H.S.
1. Begin item in selected rules, named “0” and two different R.H.S.
2. Generate association rules from the database D.
items in selected rules, named “1” and “3”. As shown in
Selecting the Sensitive rule set Rh with single antecedent
and consequent e.g. x⇒y.
Table 3(a), ISARC algorithm generates one cluster based on
4. Clustering based on common item in R.H.S. of the common L.H.S. item of the selected rules, where Table 3(b)
selected rules. shows rule clusters based on common R.H.S. item of the
5. Find the sensitivity of each cluster. selected rules, generated by DSCRC algorithm.
For each cluster, index the transactions that are fully
support the rules of cluster. Table 1: Sample database D
7. Sort generated clusters in decreasing order of their
sensitivity. Item Items (Binary
8. For the first cluster, sort selected transaction in TID
s Form)
decreasing order of their sensitivity
9. For each cluster c∈ C 0 013 1101
10. { 1 1 0100
11. While(sensitive rule set Rh ∈ c is not empty) 2 023 1011
12. {
13. Take first transaction of cluster c. 3 01 1100
14. Place an unknown (“?”) in the transaction for the 4 013 1101
place of common consequent of c.
15. Update the sensitivity of modified transaction for
Table 2: Frequent itemsets in sample database D
other cluster and sort it.
16. For i = 1 to no. of rules∈c
Frequent Itemsets with
17. {
Support(%)
18. Update minsupport and minconfidence of the
rule r∈Rh∈c. 0:80, 1:80, 3:60, 01:60, 03:60
19. If(minconfidence of r < MCT-SM or
minsupport of r < MST-SM) For ISARC algorithm, cluster 1 includes sensitive rules
20. {
namely 0⇒1 and 0⇒3. For cluster 1 sensitivities of items 0,
21. Remove rule r from Rh.
22. } 1, and 3 have 2, 1 and 1 respectively. Total sensitivity for
23. } cluster-1 is 6. For DSCRC algorithm, cluster 1 includes
24. Take next transaction of cluster c. sensitive rule namely 0⇒1, where cluster 2 includes 0⇒3.
25. } For cluster 1 sensitivities of items 0 and 1 have 1 and 1
26. End while respectively, where sensitivities for items 0 and 3 in cluster
27. } 2 have 1and 1 respectively. Total sensitivity for cluster-1
28. End for and cluster-2 is 2 and 2 respectively.
29. Update the modified transactions in D.

Figure 3. Proposed DSCRC (Decrease Support of

common Consequent of Rule Cluster) algorithm
 (IJCNS) International Journal of Computer and Network Security, 83
Vol. 2, No. 5, May 2010

Table 3: Clusters generated by both the algorithms modification, it updates the sensitivity of modified
transactions for other clusters. Table 5(b) shows final
Cluster-1(based on item 0), Rules {0⇒1, sanitized databases produced by DSCRC algorithm.
0⇒3}
Item Sensitivity
Table 5: Final sanitized databases produced by both the
0 2 algorithms
1 1
TID Items (Binary
3 1
Form)
Total cluster sensitivity 6 0 1101
(a). Rule clusters generated by ISARC algorithm 1 ?100
2 1011
Cluster-1(based on item 1), Cluster-2(based on item 3), 3 1100
Rule {0⇒1} Rule {0⇒3} 4 1101
Item Sensitivity Item Sensitivity (a). Final sanitized database produced by ISARC algorithm
0 1 0 1
1 1 3 1 TID Items (Binary
Total cluster Total cluster Form)
sensitivity 2 sensitivity 2 0 1?0?
(b). Rule clusters generated by DSCRC algorithm 1 0100

For each cluster, sensitive transactions are indexed. 2 1011

Indexed transactions with their sensitivity for both the 3 1100
algorithms are shown in Table 4. Clusters are sorted based 4 1101
on their sensitivity. Indexed transactions for ISARC (b). Final sanitized database produced by DSCRC algorithm
algorithm are shown in Table 4(a), where indexed Now, if we consider 1 for unknowns (“?”) in sanitized
transactions for DSCRC are shown in Table 4(b). For the database produced by ISARC algorithm and 0 for unknowns
first cluster, ISARC algorithm sorts transactions in (“?”) in sanitized database produced by DSCRC algorithm
increasing order of their sensitivity, where DSCRC and mine association rules from both the sanitized database,
algorithm sorts transactions in decreasing order of their we can see that most of the specified sensitive rules are
sensitivity. successfully hidden using less iterations and modifications
in transactions.
Table 4: Indexed transactions for modification for both the
algorithms
5. Experimental Results and Discussion
Cluster-1 For performance results of ISARC and DSCRC algorithms,
TID Sensitivity we used both algorithm to sanitize first 5k transactions of
1 1 real dataset given in [14] and applied Apriori algorithm on
both the produced sanitized database to mine association
(a). Transactions indexed by ISARC algorithm
rules. In our experiment, we selected 5 sensitive association
rules (namely 32⇒41, 38⇒41, 48⇒38, 38⇒170 and
Cluster-1 Cluster-2 170⇒38). The applied dataset has 39 association rules with
TID Sensitivity TID Sensitivity support ≥ 4% and confidence ≥ 20%. Then we evaluated the
0 2 0 2 performance metrics for both the algorithms. The
performance metrics for privacy preserving association rule
3 2 2 2
mining algorithms are given in [12].
4 2 4 2
(b). Transactions indexed by DSCRC algorithm Table 6: Performance results of ISARC and DSCRC
Now, hiding process of ISARC algorithm modifies algorithms
selected transactions of the cluster by placing unknowns
Parameters ISARC DSCRC
(“?”) for the place of common antecedent of cluster until all
HF 20.00 20.00%
the rules in cluster are not hidden. Table 5(a) shows final %
sanitized database produced by ISARC algorithm, where the MC 5.88% 32.35%
hiding process of DSCRC algorithm modifies selected AP 62.50 0%
%
transactions of each cluster by placing unknowns (“?”) for DISS(D,D’ 4.56% 0.42%
)
the place of common consequent of clusters until all the SEF 1.59 29.41
rules in cluster are not hidden and every time of
84
As shown in Table 6, performance results of ISARC and
DSRRC are better in terms of hiding failure (HF), misses
cost (MC), artifactual patterns (AP), dissimilarity (DISS),
and side effect factor (SEF). They provide certain privacy
and knowledge discovery. Here we selected 0.1% safety
margin. If we select higher safety margin, both the
algorithms provide more privacy to sensitive rules.
Performance results of both the algorithms depend on
database to modify and specified sensitive rules.
We can see that both the proposed algorithms hide many
sensitive rules successfully, while maintaining knowledge
discovery in terms of data quality By analyzing
performance results,. The number of modifications in
database is very less because of clustering and use of
sensitivities. Efficiency is improved by clustering which
helps to hide many rules at a time by modifying less
sensitive transactions.
Suppose an adversary is more intelligent and he/she
knows that the original database is modified by some safety
margin and thresholds. If he/she mines association rules
satisfying support and confidence below the MST-SM and
MCT-SM respectively. Then he/she can derive sensitive
rules from the sanitized database. Therefore, if random
safety margin for each rule or each rule cluster is used in
sanitization process, it can provide more privacy to sensitive
rules.
6. Conclusion and Future Scope
In this paper, we proposed two algorithms named ISARC
and DSCRC to hide sensitive rules in database for privacy
issue. Both algorithms provide certain privacy to sensitive
rules by using unknowns and maintain data quality for
knowledge discovery. Through an example we have
demonstrated proposed algorithms and analyzed
performance results. Both algorithms can hide rules which
have single antecedent or consequent and they are more
efficient than other heuristic approaches.
Proposed algorithms viz; ISARC and DSCRC can be
modified to hide sensitive rules which have different number
of L.H.S. items or R.H.S. items. In future, a parallelization
can be also applied to improve efficiency of proposed
algorithms.
References
[1] M. Atallah, E. Bertino, A.K. Elmagarmid, M. Ibrahim,
V.S. Verykios. “Disclosure Limitation of Sensitive
Rules”. In Proceedings of the 1999 IEEE Knowledge
and Data Engineering Exchange Workshop (KDEX
1999), pp. 45-52, 1999.
[2] C. Clifton, M. Kantarcioglu, J. Vaidya. “Defining
Privacy for Data Mining”. In Proceedings US Nat'l
Science Foundation Workshop on Next Generation
Data Mining, pp. 126-133, 2002.
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
[3] J. Han, M. Kamber. Data Mining: Concepts and
Techniques, Morgan Kaufmann Publishers, San
Francisco, pp. 227–245, 2001.
[4] S.R.M. Oliveira, O.R. Zaïane, Y. Saygın. “Secure
Association Rule Sharing”. In Proceedings of the 8th
Pacific-Asia Conference on PAKDD2004, Sydney,
Australia, pp. 74–85, 2004.
[5] V.S. Verykios, A.K. Elmagarmid, E. Bertino, Y.
Saygin, E. Dasseni. “Association Rule Hiding”. IEEE
Transactions on Knowledge and Data Engineering,
Vol. 16, no. 4, pp. 434–447, 2004.
[6] S.L. Wang, A. Jafari. “Using Unknowns for Hiding
Sensitive Predictive Association Rules”. In
Proceedings IEEE International Conference on
Information Reuse and Integration (IRI 2005), pp.
223–228, 2005.
[7] C. Clifton, D. Marks. “Security and Privacy
Implications of Data Mining”. In Proceedings ACM
SIGMOD International Conference on Management of
Data (SIGMOD’96), pp. 15–19, 1996.
[8] Y. Saygin, V.S. Verykios, C. Clifton. “Using
Unknowns to Prevent Discovery of Association Rules”.
ACM SIGMOD, Vol. 30, no. 4, pp. 45–54, 2001.
[9] K. Duraiswamy, D. Manjula. “Advanced Approach in
Sensitive Rule Hiding”. Modern Applied Science, Vol.
3, no. 2, 2009.
[10] Y.H. Wu, C.M. Chiang, A.L.P. Chen. “Hiding
Sensitive Association Rules with Limited Side
Effects”. IEEE Transactions on Knowledge and Data
Engineering, Vol. 19, no. 1, pp. 29–42, 2007.
[11] Y. Saygin, V.S. Verykios, A.K. Elmagarmid. “Privacy
Preserving Association Rule Mining”. In Proceedings
International Workshop on Research Issues in Data
Engineering (RIDE 2002), pp.151–163, 2002.
[12] C.C. Aggarwal, P.S. Yu. Privacy-Preserving Data
Mining: Models and Algorithms, Springer, Heidelberg,
pp. 267–286, 2008.
[13] S. Wu, H. Wang. “Research On The Privacy Preserving
Algorithm Of Association Rule Mining In Centralized
Database”. International Symposiums on Information
Processing (ISIP), pp. 131 – 134, 2008.
[14] T. Brijis, G. Swinnen, K.Vanhoof and G. Wets, “The
use of association rules for product assortment
decisions: a case study,” In Proceedings of the Fifth
International Conference on Knowledge Discovery and
Data Mining, San Diego (USA), pp. 254-260, 1999.
Authors Profile
Chirag N. Modi was born in Patan, Gujarat,
India, on 3rd Feb 1987. He received his B.E.
degree in Computer Engineering from S.P.
University Anand (Gujarat)-India in 2008. He
is Pursuing M.Tech degree in Computer
Engineering at S. V. National Institute of
Technology Surat (Gujarat)-India. His
research interests include Data Mining and
Database security.
 (IJCNS) International Journal of Computer and Network Security, 85
Vol. 2, No. 5, May 2010

Udai Pratap Rao received the B.E. degree in

Computer Science and Engineering in 2002 &
M.Tech degree in Computer Science and
Engineering in 2006, and currently working as
Assistant Professor in the Department of
Computer Engineering at S. V. National
Institute of Technology Surat (Gujarat)-
INDIA. His research interests include Data
Mining, Database security, Information Security, and distributed
systems.

Dr. Dhiren R. Patel is currently a visiting

Professor of Computer Science &
Engineering at IIT Gandhinagar,
Ahmedabad, India (on leave from NIT Surat,
India). He carries 20 years of experience in
Academics, Research & Development and
Secure ICT Infrastructure Design. His
research interests cover Security and
Encryption Systems, Web Services & Programming, SOA and
Cloud Computing, Digital Identity Management, e-Voting,
Advanced Computer Architecture etc. Besides numerous journal
and conference articles, Prof. Dhiren has authored a book
"Information Security: Theory & Practice" published by Prentice
Hall of India (PHI) in 2008. He is actively involved in Indo-UK,
Indo-French, and Indo-UK security research collaborations.
86 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

A Modified Location-Dependent Data Encryption

for Mobile Information System with Interlacing, key
dependent Permutation and rotation
Prasad Reddy. P.V.G.D1, K.R.Sudha2 , P Sanyasi Naidu3
1
Department of Computer Science and Systems Engineering, Andhra University, Visakhapatnam, India,
prasadreddy.vizag@gmail.com
2
Department of Electrical Engineering, Andhra University, Visakhapatnam, India,
arsudhaa@gmail.com
3
Department of Computer Science and Systems Engineering, GITAM University, Visakhapatnam, India
snpasala@gitam.edu,

Abstract: The wide spread use of WLAN (Wireless LAN) and
the popularity of mobile devices increases the frequency of data
transmission among mobile users. In such scenario, a need for
Secure Communication arises. Secure communication is
possible through encryption of data. A lot of encryption
techniques have evolved over time. However, most of the data
encryption techniques are location-independent. Data encrypted
with such techniques can be decrypted anywhere. The
encryption technology cannot restrict the location of data
decryption. GPS-based encryption (or geo-encryption) is an
innovative technique that uses GPS-technology to encode
location information into the encryption keys to provide location
based security. In this paper a location-dependent approach is
proposed for mobile information system. The mobile client
transmits a target latitude/longitude coordinate and an LDEA
key is obtained for data encryption to information server. The
client can only decrypt the ciphertext when the coordinate
acquired form GPS receiver matches with the target coordinate.
For improved security, a random key (R-key) is incorporated in
addition to the LDEA key. The cipher text is obtained by
interlacing , permuting and rotating based on the R-key.
Keywords: data security, location-based key, mobile security,
random generator, permutation
1. Introduction
The dominant trend in telecommunications in recent years
is towards mobile communication. The next generation
network will extend today’s voice-only mobile networks to
multi-service networks, able to carry data and video services
alongside the traditional voice services. Wireless
communication is the fastest growing segment of
communication industry. Wireless became a commercial
success in early 1980’s with the introduction of cellular
systems. Today wireless has become a critical business tool
and a part of everyday life in most developed countries.
Applications of wireless range from common appliances that
are used everyday, such as cordless phones, pagers, to high
frequency applications such as cellular phones. The
widespread deployment of cellular phones based on the
frequency reuse principle has clearly indicated the need for
mobility and convenience. The concept of mobility in
application is not only limited to voice transfer over the
wireless media, but also data transfer in the form of text ,
alpha numeric characters and images which include the
transfer of credit card information, financial details and
other important documents.
The basic goal of most cryptographic system is to transmit
some data, termed the plaintext, in such a way that it cannot
be decoded by unauthorized agents[5][6][7][8][9]. This is
done by using a cryptographic key and algorithm to convert
the plaintext into encrypted data or cipher text. Only
authorized agents should be able to convert the cipher text
back to the plaintext.
GPS-based encryption (or geo-encryption) is an innovative
technique that uses GPS-technology to encode location
information into the encryption keys to provide location
based security[12][13][14][15]. GPS-based encryption adds
another layer of security on top of existing encryption
methods by restricting the decryption of a message to a
particular location. It can be used with both fixed and
mobile.
The terms location-based encryption or geo-encryption are
used to refer to any method of encryption in which the
encrypted information, called cipher text, can be decrypted
only at a specified location. If, someone attempts to decrypt
the data at another location, the decryption process fails and
reveals no details about the original plaintext information.
The device performing the decryption determines its
location using some type of location sensor such as a GPS
receiver. Location-based encryption can be used to ensure
that data cannot be decrypted outside a particular facility -
for example, the headquarters of a government agency or
corporation or an individual's office or home. Alternatively,
it may be used to confine access to a broad geographic
region. Time as well as space constraints can be placed on
the decryption location.
Adding security to transmissions uses location-based
encryption to limit the area inside which the intended
recipient can decrypt messages. The latitude/longitude
coordinate of node B is used as the key for the data
encryption in LDEA. When the target coordinate is
determined, using GPS receiver, for data encryption, the
ciphertext can only be decrypted at the expected location. A
toleration distance(TD) is designed to overcome the
inaccuracy and inconsistent problem of GPS receiver. The
sender can also determine the TD and the receiver can
decrypt the ciphertext within the range of TD. Denning’s
model is effective when the sender of a message knows the
 (IJCNS) International Journal of Computer and Network Security, 87
Vol. 2, No. 5, May 2010

recipient’s location L and the time that the recipient will be {

there, and can be applied especially effectively in situations read LDEA-key,R-key,n,C
where the recipient remains stationary in a well-known for i=1 to n
location. {
The mobile client transmits a target latitude/longitude Rrotate(C)
coordinate and an LDEA key is obtained for data X=permute(C)
encryption to information server. The client can only Interlace(X)
decrypt the ciphertext when the coordinate acquired form p= X LDEA key
GPS receiver matches with the target coordinate. For P=convert(p)
improved security, a random key (R-key) is incorporated in }
addition to the LDEA key. In the present paper the objective Permute(P)
is to modify the cipher by introducing the concept of key write P;
dependent circular rotation. In this the bits are rotated }
depending upon the R- key after whitening with the LDEA 4 Illustration of the Cipher:
key using the Exclusive – OR operation. Encryption :
2. Random number generator using quadruple vector: “The distance between every pair of points in the universe is
For the generation of the random numbers a quadruple negligible by virtue of communication facilities. Let us
vector is used[7][10]. The quadruple vector T is generated reach each point in the sky. This is the wish of scientists.”
for 44 values i.e for 0-255 ASCII values. ASCII equivalent is obtained
T=[0 0 0 0 0 0 0 0 1 1 ……………… 0 0 0 0 1 1 1 1 2 LDEA- key:2334719
2…………………….. 0 1 2 3 0 1 2 3 0 1 X=P xor LDEA key
……………………..3] After interlacing ,permuting and rotating with the
The recurrence matrix[1][2][3] [4] key,cipher text is obtained as shown in fig 1
0 1 0  Decryption: cipher text C after interlacing ,permuting and
A = 1 1 0  rotating with the key is as shown in Fig 2
0 0 1  P=X xor LDEA key,the required plain text is obtained.
is used to generate the random sequence for the 0-255
ASCII characters by multiplying r=[ A] *[T] and
considering the values to mod 4. The random sequence
generated using the formula [40 41 42]*r is generated.[10]
3. Development of the cipher:
Consider a plain text represented by P which is represented
in the form
P=[Pij] where i=1to n and j=1 to n ---1
Let the R-key matrix be defined by
K=[Kij] where i=1 to n and j=1 to n ---2
Let the cipher text be denoted by C=[ Cij] where i=1to n
and j=1 to n corresponding to the plain text (1)
(a)
For the sake of convenience the matrices P,K and C are
represented as
P=[p1 p2 ……pn2]
K=[k1 k2 ……kn2]
C=[c1 c2 ……cn2]
3.1 Algorithm for generation of R-key and LDEA key:
Algorithm:
Algorithm for Encryption:
{
read n,K,P,r
Permute(P)
For i=1 to n
{
p=convert(P);
X=p LDEA key (b)
Interlace(X) Figure 1. Prototype for Encryption
C1=Permute(X)
C=LRotate(C1)
}
Write(C)
}
Algorithm for Decryption:
88
(a)
(b)
Figure 2. Prototype for Decryption
Cryptanalysis:
If the latitude and longitude coordinate is simply used as the
key for data encryption; the strength is not strong enough.
That is the reason why a random key is incorporated into
LDEA algorithm. Let us consider the cryptanalysis of the
cipher. In this cipher the length of the key is 8n2 binary bits.
Hence the key space is 28n2 . Due to this fact the cipher
cannot be broken by Brute force attack.
The Cipher cannot be broken with known plain text attack
as there is no direct relation between the plain text and the
cipher text even if the longitude and latitude details are
known.
It is noted that the key dependent permutation plays an
important role in displacing the binary bits at various stages
of iteration, and this induces enormous strength to the
cipher.
Avalanche Effect:
With change in LDEA key from 2334719 to 2334718 . It is
observed that there is a 77 bit change in the new cipher text.
5.Conclusions:
In present paper a cipher is developed by interlacing and
using the LDEA key dependent permutation and rotation as
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
the primary concept. The cryptanalysis is discussed which
indicates that the cipher is strong and cannot be broken by
any cryptanalytic attack since this includes confusion at
every stage which plays a vital role in strengthening the
cipher.
6. Acknowledgements:
This work was supported by grants from the All India
Council for Technical Education (AICTE) project under
RPS Scheme under file No. F.No.8023/BOR/RID/RPS-
114/2008-09.
References
[1] K.R.Sudha, A.Chandra Sekhar and Prasad
Reddy.P.V.G.D “Cryptography protection of digital
signals using some Recurrence relations” IJCSNS
International Journal of Computer Science and
Network Security, VOL.7 No.5, May 2007 pp 203-
207
[2] A.P. Stakhov, ”The ‘‘golden’’ matrices and a new kind
of cryptography”, Chaos, Soltions and Fractals 32 (
(2007) pp1138–1146
[3] A.P. Stakhov. “The golden section and modern
harmony mathematics. Applications of Fibonacci
numbers,” 7,Kluwer Academic Publishers; (1998).
pp393–99.
[4] A.P. Stakhov. “The golden section in the measurement
theory”. Compute Math Appl; 17(1989):pp613–638.
[5] Whitfield Diffie And Martin E. Hellman, New
Directions in Cryptography” IEEE Transactions on
Information Theory, Vol. -22, No. 6, November 1976
,pp 644-654
[6] Whitfield Diffie and Martin E. Hellman “Privacy and
Authentication: An Introduction to Cryptography”
PROCEEDINGS OF THE IEEE, VOL. 67, NO. 3,
MARCH 1979,pp397-427.
[7] A. V. N. Krishna, S. N. N. Pandit, A. Vinaya Babu “A
generalized scheme for data encryption technique
using a randomized matrix key” Journal of Discrete
Mathematical Sciences & Cryptography Vol. 10
(2007), No. 1, pp. 73–81
[8] C. E. SHANNON Communication Theory of Secrecy
Systems The material in this paper appeared in a
confidential report “A Mathematical Theory of
Cryptography” dated Sept.1, 1946, which has now
been declassified.
[9] E. Shannon, A Mathematical Theory of
Communication, Bell System Technical Journal 27
(1948) 379–423, 623–656.
[10] A. Chandra Sekhar , ,K.R.Sudha and Prasad
Reddy.P.V.G.D “Data Encryption Technique Using
Random Number Generator” Granular Computing,
2007. GRC 2007. IEEE International Conference, on
2-4 Nov. 2007 Page(s):573 – 576
[11] V. Tolety, Load Reduction in Ad Hoc Networks using
Mobile Servers. Master’s thesis, Colorado School of
Mines, 1999.
[12]L. Scott, D. Denning, Geo-encryption: Using GPS to
Enhance Data Security, GPS World, April 1 2003.
 (IJCNS) International Journal of Computer and Network Security, 89
Vol. 2, No. 5, May 2010

[13] Geo-encryption protocol for mobile networks A. Al-

Fuqaha, O. Al-Ibrahim / Computer Communications
30 (2007) 2510–25
[14] PrasadReddy.P.V.G.D, K.R.Sudha and S.Krishna Rao
“Data Encryption technique using Location based key
dependent Permutation and circular rotation”
(IJCNS) International Journal of Computer and
Network Security,Vol. 2, No. 3, March 2010 pp46-49
[15]PrasadReddy.P.V.G.D, K.R.Sudha and S.Krishna Rao
Rao “Data Encryption technique using Location
based key dependent circular rotation” Journal of
Advanced Research in Computer Engineering, Vol.
4, No. 1, January-June 2010, pp. 27 – 30

Authors Profile

Dr Prasad Reddy P V G D, is a Professor of

Computer Engineering with Andhra
University, Visakhapatnam, INDIA. He works
in the areas of enterprise/distributed
technologies, XML based object models. He is
specialized in scalable web applications as an
enterprise architect. With over 20 Years of
experience in filed of IT and teaching, Dr
Prasad Reddy has developed a number of
products, and completed several industry projects. He is a regular
speaker in many conferences and contributes technical articles to
international Journals and Magazines with research areas of
interest in Software Engineering, Image Processing, Data
Engineering , Communications & Bio informatics

K.R.Sudha received her B.E. degree in

Electrical Engineering from GITAM, Andhra
University 1991.She did her M.E in Power
Systems 1994. She was awarded her
Doctorate in Electrical Engineering in 2006 by
Andhra University. During 1994-2006, she
worked with GITAM Engineering College and
presently she is working as Professor in the
department of Electrical Engineering, Andhra University,
Visakhapatnam, India.

P Sanyasi Naidu is currently working as

Associate Professor in the Department of
Computer Science and systems Engg., GITAM
University, Visakhapatnam. He is pursuing his
PhD in Andhra University under the guidance
of Prof.Prasad Reddy.
90 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

Factors Identification of Software Process Model –

Using a Questionnaire
Abdur Rahshid Khan1, Zia Ur Rehman2
1
Institute of Computing & Information Technology,
Gomal Univeristy Dera Ismail Khan, Pakistan
Email: drarashid.khan09@gmail.com
2
Institute of Information Technology,
Kohat University of Science and Technology (KUST),
Kohat, Pakistan
Email: ziagulzia@gmail.com

Building and running corporate-wide knowledge

Abstract: this work is an attempt to identify factors that are
critical to software process model for a software project. Diverse
types of software engineers can get benefit to select a software
process model during development of software project. A
questionnaire was developed with the help of domain experts,
cited literature and rules of thumbs. This was sent to 100
domain experts with highest qualification and a vast experience
of the fields of various institutions in Pakistan. Experts’
opinions from 34 domain experts were received with remarkable
comments and suggestions. This questionnaire will be used as a
Knowledge Acquisition tool for ESPMS (Expert System for
Software Process Model Selection); which will integrate the
knowledge of experts of various domains, like Software
Engineering, Artificial Intelligence and Project Management.
This questionnaire will lead to become a standard document of
knowledge acquisition for selection of an appropriate software
process model for a particular project. Further work is required
to verify this tool through experts of various organizations
(public and private) software houses within country and abroad.
Keywords: Software Engineering, Project Management,
Artificial Intelligence, Expert System, Knowledge Acquisition
Tool, Questionnaire
1. Study Background
Explicit models of software evaluation date back to the
earliest projects developing large software systems in the
1950’s and 1960’s [1]-[2]. The main purpose of early life
cycle models was to provide a conceptual scheme for
managing the development of software system. Since the
1960’s descriptions of the classic software life cycle have
appeared e.g., [1, 2, 3, 4 & 5].
If software is not a product, then software development is
not a product. It cannot be. If the true product is not the
software but the knowledge contained in the software, then
software development can only be a knowledge acquisition
activity. Translating that Knowledge into a specific
language form known as “code” [7]. Knowledge is crucial
resource of each organization and therefore it will be
carefully managed. Knowledge is intensive in business
areas. Knowledge is a mandatory source of business success
[6].
management systems in the domain of Software
Engineering is an idea that was initiated about 20 years ago
[8]. There are two types of knowledge, tacit knowledge and
explicit knowledge. Tacit knowledge is personal knowledge
embedded in personal experience and is shared and
exchanged through direct, face-to-face contact. Explicit
knowledge is formal or semi-formal knowledge that can be
packaged as information [6].
In article [6], author defined “Knowledge Management”
(KM) as human resource management activities, such as
hiring new staff or to train staff in order to increase the
company’s capacities. The goal of KM is to improve
organizational skills of all levels of the organization
(individual, group, division, organization) by better usage of
the resource knowledge. It is a management activity, and as
such is goal oriented, planned and monitored.
The three laws for process are: (1) Process only allows us
to do things we already know how to do. (2) We can only
define software processes at two levels: too vague or too
confining. (3) The very last type of knowledge to be
considered as a candidate for implementation into an
executable software system is the knowledge of how to
implement knowledge into an executable software system.
Author also focuses attention, that there are several
problems in defining process in industry [7].
Comperative Study of Software Process Models
Software process is a framework for the tasks that are
required to build high quality software [9].
A descriptive model describes the history of how a
particular software system was developed [4 & 13].
Prescriptive models are used as guidelines or frameworks
to organize and structure how software development
activities should be performed, and in what order.
Prescriptive models are also used to package the
development tasks and techniques for using a given set of
software engineering tools or environment during a
development project. Descriptive life cycle models,
characterize how particular software systems are actually
developed in specific settings [3 & 4].
 (IJCNS) International Journal of Computer and Network Security, 91
In order to ensure that cost-effective, quality systems are
developed which address an organization’s business needs,
developers employ some kind of system development
Process Model to direct the project’s lifecycle [10]. To solve
actual problems in an industry setting, a software engineer
or a team of engineers must incorporate a development
strategy that encompasses the process, methods, and tools
layers and the generic phases [9].
Software process models often represent a networked
sequence of activities, objects, transformations, and events
that embody strategies for accomplishing software evolution.
Software process networks can be viewed as representing
multiple interconnected task chains. Task chains can be
employed to characterize either prescriptive or descriptive
action sequences. Prescriptive task chains are idealized
plans of what actions should be accomplished, and in what
order [11].
Some of the important models have been described in the
following paragraphs.
The Linear Sequential Model, The linear sequential
model, sometimes called the Classic or Waterfall model,
proposed by Winsten Royce [11]. The linear sequential
model suggests a systematic, sequential approach to
software development that begins at the system level and
progresses through analysis, design, coding, testing, and
support [9].
Waterfall Model has some criticisms as:
• Real projects rarely follow the sequential flow that the
model proposes.
• At the beginning of most projects there is often a great
deal of uncertainty about requirements and goals, and it is
therefore difficult for customers to identify these criteria on
a detailed level. The model does not accommodate this
natural uncertainty very well.
• Blocking state: means in this model some project team
members must wait for other members of the team to
complete dependant tasks.
• Developing a system using the Waterfall Model can be
a long, painstaking process that does not yield a working
version of the system until late in the process [10].
The Prototyping Model: A customer defines a set of
general objectives for software but does not identify detailed
input, processing, or output requirements. In these, and
many other situations, a prototyping paradigm may offer the
best approach [9 &11].
There are a few different approaches that may be followed
when using the Prototyping Model [10]:
• Creation of the major user interfaces without any
substantive coding in the background in order to give the
users a “feel” for what the system will look like,
• Development of an abbreviated version of the system
that performs a limited subset of functions; development of a
paper system.
• Use of an existing system or system components to
demonstrate some functions that will be included in the
developed system.
Vol. 2, No. 5, May 2010
Criticisms of the Prototyping Model generally fall into the
following categories [10]:
• Prototyping can lead to false expectations.
• Prototyping can lead to poorly designed systems.
The RAD Model: Rapid Application Development
(RAD) is an incremental software development process
model that emphasizes an extremely short development life
cycle. It is a component-based construction, but high speed
linear sequential. If requirements are well understood and
project scopes constrained, the RAD process enables a
development team to create “a fully functional system” with
in very short time periods (e.g., 60 to 90 days) [9 & 11].
The RAD approach has drawbacks [9]:
• Large and scalable projects require sufficient human
resources.
• RAD requires developers and customers who are
committed to the rapid-fire activities necessary to get a
system complete in specified time. If commitment is lacking
from either constituency, RAD project will fail.
• If a system cannot be properly modularized, building
the components necessary for RAD will be problematic.
• If high performance is issue and performance is to be
achieved through tuning the interfaces to system
components, the RAD approach may not work.
• RAD is not appropriate when technical risks are
high. This occurs when a new application makes heavy use
of new technology or when the new software requires a high
degree of interoperability whit existing computer programs.
Evolutionary Software Process Model: Evolutionary
models are iterative, which is one of the modern software
development processes. Evolutionary models enable
software engineers to develop increasingly more complete
versions of the software [9 & 11].
Incremental Models: It combines elements of the linear
sequential model applied respectively with the iterative
philosophy of prototyping. First increment is often a core
product. It is iterative in nature [5 & 9].
The Spiral Model: The spiral model, proposed by [3], is
an evolutionary software process model that couples the
iterative nature of prototyping with controlled and
systematic aspects of the linear sequential model. A spiral
model is divided into a number of framework activities, also
called tasks or regions [9 & 11].
The Win-Win Spiral Model: The Customer wins by
getting the system or product that satisfy the majority of the
customer’s needs and the developer wins by working to
realistic and achievable budgets and deadlines. Boehm’s
WINWIN spiral model [12] defines a set of negotiation
activities at the beginning of each pass around the spiral.
Criticism of a spiral model: The risk assessment
component of the Spiral Model provides both developers
and customers with a measuring tool that earlier Process
Models do not have. The measurement of risk is a feature
that occurs every day in real-life situations but
(unfortunately) not as often in the system development
industry. The practical nature of this tool helps to make the
92
Spiral Model a more realistic Process Model than some of
its predecessors.
The Concurrent Development Model: The concurrent
development model is also called concurrent engineering,
[12]. The concurrent process model defines a series of
events that will trigger transitions from state to state for
each of the software engineering activities. A system and
component activities occur simultaneously and can be
modeled using the state-oriented approach described
previously. Each activity on the network exists
simultaneously with other activities [5, 9 & 11].
Component-Based Development: Object-Oriented
technologies provide the technical framework for a
component-based process model for Software Engineering
[5, 9 & 11]. A general criticism of the Component-Based
Model is that it is limited for use in object-oriented
development environments. Although this environment is
rapidly growing in popularity, it is currently used in only a
minority of system development applications [10].
1.1 Creating and Combining Models:
In many cases, parts and procedures from various Process
Models are integrated to support system development. This
occurs because most models were designed to provide a
framework for achieving success only under a certain set of
circumstances. It is sometimes necessary to alter the existing
model to accommodate the change in circumstances, or
adopt or combine different models to accommodate the new
circumstances. The selection of an appropriate Process
Model hinges primarily on two factors: organizational
environment and the nature of the application. Frank Land,
from the London School of Economics, suggests that
suitable approaches to system analysis, design, development,
and implementation be based on the relationship between
the information system and its organizational environment.
Four categories of relationships are identified: The
Unchanging Environment, the Turbulent Environment, the
Uncertain Environment, and the Adaptive Environment
[10].
1.2 Expert Systems Application
For decision making and for problem solving Expert
System tools are used, which is Artificial Intelligence tool,
they have ability to imitate the reasoning of human expert
with in a specialized area [14]. Now in various areas such as
computer science, agriculture, engineering, geology, space
technology, medicine, etc expert systems are being used and
developed (Durken 1990). The knowledgebase component of
expert system is consists of human experts’ knowledge. The
most difficult task is to acquire knowledge from experts and
identification of problem in the development of expert
systems [7, 8]. Interview, observation analysis, case study
analysis, questionnaire, protocol analysis are the some
knowledge acquisition techniques to acquiring knowledge
and expertise in domain experts [14, 15, 16 & 17].
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
1.3 Knowledge Acquistion
Capturing knowledge source from expertise is known as
knowledge acquisition technique. There are different
sources from which knowledge can be acquired such as
human experts, books, journals, reports, databases and other
computer system [14]. The process through which
knowledge engineers extract knowledge called Knowledge
Acquisition [18]. Knowledge acquisition process is the first
phase to develop an expert system [18]. There are some
difficulties in during knowledge acquisition process i.e.
sometime problem is not very clear to knowledge engineers,
knowledge engineers have lack of knowledge in particular
domain, sometime experts are not available in particular
domain [16 & 18].
There are several reasons for employing an expert system
such as [14]:-
§ Replacement of human expert
§ Assistant to human expert
§ Transfer of expertise to novice
From literature study it is revealed that there are a
number of knowledge acquisition techniques exist, such as
questionnaire, protocol analysis, Interview (structured and
unstructured interview), case study analysis, observation
analysis, simulation and prototyping [14 & 15).
Depending upon the nature of the problem, need,
situation and domain experts, different acquisition
techniques can be used accordingly. If the domain experts
can easily be accessed, an interview is the best technique. If
the expert are far-away from access of knowledge engineers
or experts are much busy, then questionnaire is the most
suitable tool. Similarly, if the target problem is complex,
multiple techniques can be adopted.
2. Problem Domain
Knowledge acquisition process was started from
identification of critical factors that affect directly or
indirectly the process models, being critical in selection of
appropriate software process model to software engineers.
This tool will be used in ESPMS (Expert System for
Software Process Model Selection), which is available free
of cost.
Literature study reveals that research has been conducted
on different phases of software process models, such as
requirement analysis and specification, planning of software
project, System Design and Architectural Design etc but
there is no proper guidance for novice software designer and
engineer, how to select a process model for a particular
situation.
Ref [7] described the three laws of software process and
also highlighted that there are several problems in defining
a process in industry. Author also highlights rule of process
bifurcation (Software process rules that are stated in terms
of two levels: a general statement of the rule, and a specific
detailed example), and the dual hypothesis of knowledge
discovery (we can only “discover” knowledge in an
environment that contains that knowledge). Ref [6] stated
overall objective of Learning Software Organization (LSO)
 (IJCNS) International Journal of Computer and Network Security, 93
Vol. 2, No. 5, May 2010

is to improve software processes and products according to
the strategic goals of the organization. Knowledge
Management (KM) was defined, because KM is prerequisite
of learning organization. Ref [19] stated that there is still
lacking integration of software process modeling and
software process measurement by software engineers.
Language construct is used to better understand the process
modeling technique. The Attribute Grammar (AG) approach
to realize modeling of software processes modeling in
language construct is based on its specification and
automatic construction of language-based editors. Ref [20]
Build Knowledge Analysis System (KANAL), which relates
pieces of information in process models among themselves
and to the existing Knowledge Base (KB), analyzing how
different pieces of input are put together to achieve some
effect. It builds interdependency models from this analysis
and uses them to find errors and propose fixes. Ref [21]
Recent research in software engineering has moved away
from this linear approach of the life cycle in order to take
account of dependencies between decisions taken in
different parts of the development and to allow methods to
be selected during developments which are appropriate to
address problems that arise. The most advanced approach is
the spiral model proposed by Boehm [3]. The major research
project on KBS methodology is the KADS (Knowledge
universities of Pakistan. 100 domain experts along with
detailed about them were found in 124 universities of
Pakistan. See Table I for detail.
3.2 Design and Distribution of Questionnaire
Extracting experts’ opinion to develop standard criteria
for factors that can affect software process model’s
performance is the main purpose of this questionnaire.
Questionnaire was the tool through which we could solicit
experts’ opinion. A total of 99 questions (parameters) were
identified, which were grouped into 11 main factors. Five
fuzzy variables scales were defined, like, 5=Strongly Agree
(Critical to process’s performance), 4 = Agree (Important to
process’s performance), 3= Neutral (Helpful to process’s
performance), 2= Disagree (Minimally affects process’s
performance), 1= Strongly Disagree (Do not affect process’s
performance).
We designed the questionnaire in such a way that the
respondent have to rank the importance of factor regarding
to its objective. Among 100 domain experts we received 34
experts’ opinions with valuable and remarkable suggestions.
Table 1: Experts Detail with their Responses
Experts’ Subject
Charac.
Gender Designation Degree
Specialty
Demographic

Acquisition and Design System). KADS proposes that Software

Male: 30 Professor: 4 PhD: 4 Engineeri Federal: 10
knowledge modelling should pass through five stages: ng: 20
(Data, Conceptual Model, Design Model, Detailed Design Associate M.Phil/M
Artificial
Categorical Female:4 Intelligenc Sindh:0
Model, and Implementation). The Conceptual Model is Responses
Professor: 8 S:12
e: 8
further divided into a four layer knowledge model where Software
Assistant MCS/ Project
- Punjab:4
each successive layer interprets the description of the lower Professor: 9 MSc: 12 Managem
ent: 6
layer: static domain knowledge, knowledge sources, task, BCS
- - Lecturers: 7 - NWFP:10
and strategic knowledge. The major difficulty with advice (Hons): 6
Baluchistan:
on maintenance is that there is very little data or published - - Teachers: 6 - -
10
Azad Jammu
experience on maintaining KBS. Two main properties - - - - -
& Kashmir: 0
directly influence the maintainability of a KBS. (1) The - - - - -
Total: 34 34 34 34 34 34
homogeneity of the maintained representation: (2) the
predictability of the representation.

3. Approach 3.3 Analysis and Findings

In this research we are introducing the tool (i.e. a SPSS was used to analyze the collected facts. Table 2
questionnaire) for knowledge acquisition process to solve a shows the summary of questionnaire distribution.
selected problem. During development of this Knowledge
Acquisition tool, following steps were adopted: Table 2: Questionnaire Distribution
A). Selection of Domain Experts Medium Questionnaire Responses %age
Distributed
(B), Design and Distribution of Questionnaire and Through Email 60 20 33.33
(C) Analysis and Findings. Through Post 28 9 32.14
3.1 Selection of Domain Experts Through By hand 12 4 33.33

Knowledge Base (collection of expert knowledge) is Total 100 34 34

critical and important component of the expert system [22].
The questionnaire was properly evaluated by domain
experts, which will lead to become a standard document for
selection of appropriate software process model.
In Pakistan SE and AI are used mostly for research in
Education, that’s why this questionnaire was distributed in
various Higher Education Commission (HEC) recognized
The questionnaire was resulted in 11 main groups with 84
questions after repeatedly contact with the domain experts.
Statistical analysis showed that System Integration &
Testing, User Community & Unit Development, Software
Integration & Testing were the most important factors
among rest of the main factors. See Table 3 for detail.
94 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

The ranking of factors were also helpful in determining model but it is just an approach, guideline and helping tool
the weights of the various factors, which influence the for software engineer in selection of process model. This
decision making process. questionnaire may become a standard Knowledge
Acquisition tool for Expert System development.
Table 3: Response Summary for Affect on Proces’
Performance
(Total Reponses N: 34, 5= Critical Affects, 1= Do not Affect) Domain Experts
Std. (Books, Journals, Databases, Reports, Internet etc)
Main Groups of Factors Mean
Deviation
System Integration and Testing 3.8941 1.1494 Reference Knowledge
User Community 3.4510 1.1441
Unit Development, Software Integration &
Testing 3.7426 1.0890 Characteristics of Software Process Models

Implementation & Maintenance 3.7904 1.0881

Project Type and Risk Management 3.7941 1.0184
Validation and Verification 3.8168 0.9981
Quality Assurance and Quality Control 3.8333 0.9774 Factors Identification
Requirement Analysis and Specification 3.8407 0.9762
System Design and Architectural Design 3.8848 0.9613
Analyzed Facts
Project Team 3.8445 0.9565 (Knowledge Elicitation)
Documentation 4.1823 0.8577

Questionnaire
Table 4 depicts weights assigned to the main factors. In
weight criteria documentation, system integration and
testing and validation and verification were the most Figure 1: Knowledge Elicitation Process of a Software
important factors among rest of the main factors. Process Model Selection
Table 4: Weight Assignment to main factors
Weigh References
S.No Main Groups of Factors
ts
1 Documentation 0.1041 [1] Hoiser, W.A., “Pitfalls and Safegaurds in Rea-Time
2 System Integration and Testing 0.0971 Digital Systems with Emphasis on programming,” IRE
3 Validation and verification 0.0969 Trans. Engineering Management, EM-8, June 1961.
4 System Design and Architectural Design 0.0956 [2] Royce, W. W., “Managing the Development of Large
5 Quality Assurance and Quality Control 0.0952 Software Systems,” Proc. 9th. Intern. Conf.Software
6 Project Team 0.0923 Engineering, ,IEEE Computer Society, 1987 ,328-338
7 Implementation & Maintenance 0.0915 Originally published in Proc. WESCON, 1970.
Requirements Analysis And [3] Boehm, B., “ A Spiral Model for Software
8 0.0914
Specification Development and Enhancement,” Computer, Vol. 21,
9 Project Type and Risk Management 0.0849 no. 5, May 1988, pp. 61-72
Unit Development, Software Integration
10 0.0793 [4] Scacch i W., Process Models in Software
& Testing
Engineering, Institute for Software Research,
11 User Community 0.0718
Total weight: 1.0000 University of California, Irvine February 2001.
[5] Sommerville I., Software Engineering 5th Edition, 2000
After calculation of mean and standard deviation of main [6] Ruhe G. (2000), “ Learning Software Organizations”,
factors, we calculated mean and standard deviation of each Fraunhofer Institute for Experimental Software
and every sub-factor (questions). Engineering (IESE).
[7] Caesar & Cleopatra (2004), “Chapter 1,The Nature of
Software and the Laws of Software Process” www.ism-
4. Conclusions and Future Recommendation
journal.com/ITToday/AU1489_C01.pdf, CRC Press,
Each and every factor was properly analyzed, means and LLC.
standard deviations were calculated of each factors with the [8] Basili V.R, Zelkowitz, M. V. McGarry, F. Page, S.
help of SPSS. Weights of the main factors were evaluated. Waligora, R. Pajerski. SEL’s Software Process
Overall score will be calculated at run time through Improvement Program, IEEE Software, vol. 12,
summation function, which will be used for ranking in the November, pp 83-87, 1995.
selection of a process model. (See my research thesis for [9] Pressman R. S, Software Engineering a Practitioner ’s
Approach, Fifth Edition, 2001.
detail).
[10] SUNY 1988 , “ A Sur vey of System Development
It is further recommended that this questionnaire is not in
Pr ocess Models ” , CTG.MFA – 00, Cen ter for
final shape which will be distributed in local (Pakistan) and
Techn ology in Gover nment, Uni ver sity at
international software houses. This questionnaire is not final Alban y.
standard tool for selection of an appropriate software process
 (IJCNS) International Journal of Computer and Network Security, 95
Vol. 2, No. 5, May 2010

[11] Reddy A. R. M, Govindarajulu P., Naidu M., A [27] Liu & Perry (2004), “On the Meaning of Software
Process Model for Software Architecture, IJCSNS, Architecture, Interview Questionnaire, (July, 2004),
VOL.7 No.4, April 2007. Version 1.2
[12] Davis, A. and Sitaram P., “ A Concurrent Process
Model for Software Development, ” Software Authors Profile
Engineering Notes, ACM Press, vol. 19, n0. 2, April
1994, pp. 38-51 Abdur Rashid Khan is presently working as an Associate
[13] Curtis, B., H. Krasner, and N. Iscoe, A Field Study of Professor at ICIT, Gomal University D.I.Khan, Pakistan. He has
received his PhD degree from Kyrgyz Republic in 2004. His
the Software Design Process for Large Systems, research interest includes AI, Software Engineering, MIS and DSS.
Communications ACM, 31, 11, 1268-1287, November,
1988. Zia Ur Rehman is currently pursuing his MS degree in Computer
[14] Awad, E.M. “Building Experts Systems: Principals, Science from the Institute of Information Technology, Kohat
Procedures, and Applications,” New York: West University of Science & Technology (KUST), Kohat, Pakistan. His
area of interest includes Software Engineering, Artificial
Publishing Company. 1996. Intelligence/Expert System, Databases and Data Mining.
[15] Durkin, J. “Application of Expert Systems in the
Sciences,” OHIO J. SCI. Vol. 90 (5), pp. 171-179,
1990.
[16] Medsker, L. and Liebowits, J. “Design and
Development of Experts Systems and Neural
Networks,” New York: Macmillan Publishing
Company, 1987.
[17] Sagheb-Tehrani, M.“The design process of expert
systems development: some concerns, Expert system,”
Vol. 23 (2), pp. 116-125, 2006.
[18] Wagner, W.P., Najdawi, M.K., Chung, Q.B “Selection
of Knowledge Acquisition techniques based upon the
problem domain characteristics of the production and
operations management expert systems,” Expert
system, vol. 18 (2), pp. 76-87, 2001.
[19] Atan, R. Abd, A. A. Ghani, M. H.S. & Mahmod, R. “
Software Process Modelling using Attribute Grammar”
University Putra Malaysia, Serdang, Selangor Darul
Ehsan, Malaysia, IJCSNS International Journal of
Computer Science and Network Security, VOL.7 No.8,
August 2007.
[20] Kim, J. & Gil, Y. “Knowledge Analysis on Process
Models”, Information Sciences Institute University of
Southern California, International Joint Conference on
Artificial Intelligence (IJCAI-2001), Seattle,
Washington, USA, August 2001.
[21] Wilson, “Knowledge Acquisition: The current
position”, Science and Engineering Research Council
Rutherford Appleton Laboratory.
[22] Lightfoot,J.M.” Expert knowledge acquisition and the
unwilling experts: a knowledge engineering
perspective,” Expert system, vol. 16(3), pp.141-147,
1999.
[23] Futrell R. T., Shafer L. I, Shafer D. F, “Quality
Software Project Management”, Low Price Edition,
2004.
[24] Anderson S., & Felici M., “Requirement Engineering
Questionnaire”, Version 1.0, Laboratory for
Foundation of Computer Science, Edinburgh EH9 3JZ,
Scotland, UK (2001).
[25] Skonicki M., “QA/QC Questionnaire for Software
Suppliers”, January 2006.
[26] Energy U. S. D., “Project Planning Questionnaire”,
www.cio.energy.gov/Plnquest.pdf (last access, 04
August 2009)
96 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

Appendix: (Questionnaire)
A SURVEY ON FACTORS AFFECTING TO SELECT A SOFTWARE DEVELOPMENT PROCESS MODEL
Respondent’s Name: _______________ Qualification: _________ Designation: ______________ Address/University: __________________________

INSTRUCTIONS
Different software process model exist, to select one for a particular software depends upon the nature of project, problem structure, risk involve, budget
estimation, goals & objectives of the project etc. As an EXPERT in this area, you are requested to examine each item in terms of suitability and then to explain
the degree of your agreement to each item whether, in your opinion, it would measure the factors affecting the selection of software development process model
and to what extent. You may recommend new & delete unnecessary items from the existing scale. Your in-time response will highly be appreciated. Please, use
the scale below to mark (P) your responses in the area provided.
5= Strongly Agree 4= Agree 3=Neutral 2= Disagree 1=Strongly Disagree
Critical to Process’s Important to Process’s Helpful to Process’s Minimally affects Process’s Do not affect Process’s performance
Performance Performance performance performance
ATTRIBUTES LEVELS
5 4 3 2 1
1 Requirements Analysis And Specification

1.1 Are the requirements precisely defined, well defined and well known?
1.2 Are requirements sufficient to help you understand the problem structure?
1.3 Are the requirements defined early in the cycle [23]?
1.4 Are the requirements will change often in the cycle [23]?
1.5 Is there need to demonstrate capability proof of concept [23]?
1.6 Does the requirement indicate a complex system/simple system [23]?
1.7 Is there appropriate procedure prepared for cost/benefit analysis [24]?
1.8 Does the requirements phase meet the requirements methodology [24]?
1.9 Are you sensible to organizational and political factors which influence requirements sources while eliciting requirements [24]?
1.10 Have you reuse requirements from other systems which have been developed in the same application area [24]?
1.11 How do you handle continuous requirements change [24]?
1.12 Does any risk analysis are performed on requirements [24]?
1.13 Is the requirements document easy to change [24]?
1.14 Do you define Cost-effectiveness criteria [24]?
1.15 Do you identify accuracy and completeness risks on the requirements analysis [24]?
1.16 Are rules established on handling inaccurate and incomplete data [24]?
1.17 Do you cross-check operational and functional requirements against safety (or security, availability, etc.) requirements [24]?
1.18 Are specifications kept up-to-date and controlled [25]?
1.19 Are your software-product versions reviewed for conformity to customer’s specifications [25]?
1.20 Are the versions reviewed for compliance to quality requirements before submission for customer approval [25]?
2 Project Team
2.1 Does the majority of team members new to the problem domain [23]?
2.2 Does the majority of team members new to the technology domain [23]?
2.3 Is there need to train to project team [23]?
2.4 Is the team more comfortable with structure than flexibility [23]?
2.5 Will the project manager closely track the team’s progress [23]?
2.6 Is ease of resource allocation important [23]?
2.7 Does the team accept peer reviews and inspections, management/customer reviews, and mile-stones [23]?
3 User Community
3.1 Will the availability of the user representatives be restricted or limited during the life cycle [23]?
3.2 Are the user experts in the problem domain [23]?
3.3 Do the users want to be involved in all phases of the life cycle [23]?
4 Project Type and Risk
4.1 Dose the customer want to track project progress [23]?
4.2 Does the project identify a new product direction for the organization [23]?
4.3 Is the project a system integration project [23]?
4.4 Is the project an enhancement to an existing system [23]?
4.5 Is the funding for the project expected to be stable throughout the life cycle [23]?
4.6 Is the product expected to have a long life in the organization [23]?
4.7 Is high reliability a must [23]?
4.8 Is the system expected to be modified perhaps in ways not anticipated, post-deployment [23]?
4.9 Is the schedule constrained [23]?
4.10 Are the modules interfaces clean [23]?
4.11 Are reusable components available [23]?
 (IJCNS) International Journal of Computer and Network Security, 97
Vol. 2, No. 5, May 2010

4.12 Are resources (time, money, tools, and people) scarce [23]?
4.13 What measurements will be used to track this project (Cost, Schedule Effort, LOC, Defects, Function Points Other ______) [26]?
5 System Design and Architectural Design
5.1 Does a design phase schedule exists which identifies tasks, people, budgets and costs?
5.2 Do you develop complementary system models [24]?
5.3 Do you model the system’s environment [24]?
5.4 Do you model the system architecture [24]?
5.5 Do you use structured methods for system modeling [24]?
5.6 Do you define operational processes to reveal process requirements and requirements constraints [24]?
5.7 Do you use a data dictionary [24]?
5.8 Do you document the links between stakeholder requirements and system [24]?
5.9 Do you specify systems using formal specifications [24]?
5.10 Can the data required by the application be collected with the desired degree of reliability [24]?
5.11 Do you identify non functional requirements (e.g., usability, quality, cognitive workload, etc.) for a system [24]?
5.12 Are significant software changes expected during the life of the project [24]?
6 Unit Development, Software Integration & Testing
6.1 Will the application system be run in multiple locations [24]?
6.2 If an on-line application, will different types of terminal be used [24]?
6.3 Is the proposed solution dependent on specific hardware [24]?
6.4 Have the portability requirements been documented [24]?
7 Quality Assurance and Quality Control
7.1 Have you written Quality Assurance plan? Or written record for Quality Assurance program [25]?
7.2 Have you performed internal quality audits on software design processes and configuration management functions [25]?
7.3 Is there any implementation of a formal correction action process for customer complaints [25]?
8 System Integration and Testing
8.1 Are software depends on Operating Environment(s)? (MS Windows, OS/2, Unix/AIX, Windows N/T, Macintosh, Novell, CICS,
DOS, Sun, Web Browser (specify), other ______________) [26]?
8.2 Are constraints effects the targeted operating environment(s)? (Firewalls, Transmission speed, Server capacity, Security,
Workstation capacity, Remote access capability, Widely dispersed user community without appropriate inter-operability Other
[26]?
8.3 Which programming languages you will be, or are considering used in project? (Cobol, Java, C++, Visual Basic, FoxPro, Paradox,
HTML, Delphi, Visual C++, Power Builder, Others ____________) [26]?
8.4 Are the software’s operations (System/Application) so essential that data must be immediately available at all times or
recoverable? (Data recovery, Backups, Fault Tolerance, System Performance, Mirroring/Imaging Disaster Recovery) [26]?
8.5 Which individual, team, and organizational information systems development and management practices will be implemented on
this project? Project Planning, Requirements Management, Configuration Management, Project Tracking and Oversight, Quality
Assurance, Sub-Contractor Management, Risk Assessment, Peer Reviews, Training Program Software Product Engineering, Inter-
group Coordination, Integrated Software Management, Organizational Process Definition, Organizational Process Focus Defect
Tracking, Other_________________________ [26]?
9 Validation and verification
9.1 Is there any implementation of formal process for functional requirement review [25]?
9.2 Is there any implementation of formal process for system design review [25]?
9.3 Is there any implementation of formal process for code testing [25]?
9.4 Is there any implementation of formal process for integration testing [25]?
10 Documentation
10.1 Are written test plans, instructions, product specifications, and system design documents available to support the tests [25]?
10.2 Are functional requirements properly documented [25]?
10.3 Have you created system programmers documentation [25]?
10.4 Have you documented functional specification [25]?
10.5 Have you documented the system design [25]?
11 Implementation & Maintenance
11.1 Do you implement a formal set of code testing procedures?
11.2 Do you implement a formal set of integration testing procedures?
11.3 Have you implemented a formal process for acceptance testing by the User / customer?
11.4 Are you usually concerned with reusability while designing [27]?
11.5 Do you involve external (from the project) reviewers in the validation process [24]?
11.6 Has the expected life of the project been defined?
11.7 Has the importance of keeping the system up to date functionally been defined [24]?
11.8 Which medium will be used for the project/system documentation? Web Site, Hardcopy, CD Rom, Online Help, Video Quick
Reference Card Diskette, Other [26]?

Comments:___________________________________________________________________________________________________________________________________
Date: ____________________ Signature: ______________
98 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

MOBINA: A Novel DBA Method Based on Second

Price Auction in Ethernet Passive Optical Network
AR. Hedayati1, M.N Feshaaraki1 ,K. Badie2 , A. Khademzadeh2 and M. Davari3
1
Department of Computer Engineering, Science and Research Branch, Islamic Azad University(IAU)
Hesarak ,poonak SQ, Tehran, Iran.
ar.hedayati@srbiau.ac.ir
2
Iran Telecom Research Center
North kargar str, Tehran, Iran
3
Department of Computer Engineering, Central Tehran Branch, Islamic Azad University
Hamila Blv, poonak SQ, Tehran, Iran.

Abstract: Ethernet Passive Optical Network(EPON) is one of
the best solutions for implementing the next generation
access networks. One of the key issue in these network is
bandwidth allocation to provide the quality of service for the
End users.In this paper, we present a Novel Dynamic
Bandwidth Allocation (DBA) method based on Second Price
Auction in Ethernet Passive Optical Network. Results have
shown improvements of the packet loss rate, throughput
rate, and utilization line of our method when compared
with other methods.
Keywords: Ethernet Passive Optical Network, Dynamic
Bandwidth Allocation, Second price auction, Quality of Services.
1. Introduction
Ethernet Passive Optical Network is one of the most
important solutions for implementing the next generation of
access networks. High capacity and low cost of
implementing are among the characteristics of this
technology, which can provide the required quality service
in spite of expansion the quantity of user and their requests
to use different multimedia services [1].
An EPON system is a point-to-multipoint fiber optical
network with no active elements in the transmission path
from its source, i.e., an optical line terminal (OLT), to a
destination, i.e., an optical network unit (ONU). It can use
different multipoint topologies, such as bus, ring, tree, and
different network architectures [2]. The most typical EPON
architecture is based on a tree topology and consists of an
OLT, a 1:N passive star coupler (or splitter/combiner), and
multiple ONUs, as shown in Figure (1). The OLT resides in
a central office (CO) that connects the access network to a
metropolitan area network (MAN) or a wide area network
(WAN), and is connected to the passive star coupler through
a single optical fiber [2]- [5].
Figure 1. EPON’s Architecture
One of the main concerns in this network is bandwidth
allocation to provide end users quality of service for using
multimedia service. To solve this problem, there have
suggested different static and dynamic bandwidth allocation
methods [2], [4]. In this paper, along with studying
bandwidth allocation methods in EPON, we present a novel
method named MOBINA, based on Second Price Auction.
The MOBINA can allocate the users required bandwidth
righteously and with high profitability.
The rest of this paper is organized as follows: In Section 2,
we introduce related work of Bandwidth allocation methods
in EPON. In section 3, we explain MOBINA method in
details. In section 4, we present simulation results. In
section 5, we conclude this paper.
2. Related work
Bandwidth allocation methods are discussed as one of the
most important QOS’s parameters in EPON [6]. In
bandwidth allocating for each network unit, optical line
terminal needs to provide some method based on the
received bandwidth demands of optical network units and
some allocation policies or service level agreements are in
this content [7]. At present many methods are
recommended, in general they classified in two general
types: Static Bandwidth Allocation and Dynamic Bandwidth
Allocation.
2.1 Static Bandwidth Allocation methods
Each ONU is assigned with a constant bandwidth regardless
to its actual bandwidth demand. As a result, the allocated
bandwidth might either not be fully utilized, in the case of
ONUs with light traffic requirement or not be adequate to
accommodate heavy traffic load ONUs, resulting in both
scenarios to inefficient bandwidth utilization rate. In
contrast a dynamic MAC protocol could more successfully
allocate bandwidth according to ONUs’ instantaneous buffer
queue status increasing the bandwidth utilization rate [8].
Research initiatives in MAC protocol development over the
years have attempted to enhance further bandwidth
allocation of standardized PON topologies, tracing back to
PONs and more intensively to more recent and lately
deployed EPONs . In common to all these topologies, the
feeder pathway is shared in the time domain, and as a result
each ONU can utilize the whole upstream optical carrier
 (IJCNS) International Journal of Computer and Network Security, 99
capacity for defined, according to network penetration, and
in most cases flexible duration time-slots, assigned by the
OLT [7].
2.2 Dynamic Bandwidth Allocation methods
To increase bandwidth utilization, the OLT must
dynamically allocate a variable timeslot to each ONU based
on the instantaneous bandwidth demand of the ONUs.
Given that QoS is the main concern in EPONs, these
methods classified into DBA with QoS support, DBA
without QoS support, and describe their characteristics and
performances [11].
2.2.1 DBA without differentiated QoS support
There are several DBA methods proposed for EPON which
do not support differentiated services, such as IPACT [9], it
employs a resource negotiation process to facilitate queue
report and bandwidth allocation. The OLT polls ONUs and
grants timeslots to each ONU in a round robin fashion. The
timeslot granted to an ONU is determined by the queue
status reported from that ONU. Therefore, the OLT is able
to know the dynamic traffic load of each ONU and allocate
the upstream bandwidth in accordance with the bandwidth
demand of each ONU. So demanded bandwidth can be
granted efficiently also there exists three different ways for
the OLT to determine the granted efficiently also there
exists three different ways for the OLT to determine the
granted window size to allocate bandwidth so they
introduced Fixed service, Gated service, Limited service as
IPACT service discipline to demonstrate bandwidth
allocation in this method . Development and improvement
of this method is considered in IPACT GE [10], Estimation-
based DBA [11] and IPACT with SARF [2].
Bandwidth guaranteed polling (BGP) is another DBA
method proposed for providing bandwidth guarantees in
EPONs [12]. In BGP, all ONUs are divided into two groups:
bandwidth guaranteed and bandwidth non-guaranteed. The
OLT performs bandwidth allocation through using couple of
polling tables. The number of bandwidth units allocated to
an ONU is determined by the bandwidth demand of that
ONU, which is given by its SLA with a service provider. A
bandwidth guaranteed ONU with more than one entry in the
poling table has its entries spread through the table. This
can reduce the average queuing delay because the ONU is
polled more frequently. However, this leads to more grants
in a cycle and thus requires more guard times between
grants, which reduces channel utilization. Methods have
been introduced for all best effort traffic is suitable.
2.2.2 DBA with differentiated QoS support
An EPON system is expected to deliver not only the best
effort data traffic, but also real-time data traffic (e.g., voice
and video) that have strict bandwidth, packet delay, and
delay jitter requirements. In this subsection present several
DBA methods that can provide differentiated QoS support
for different types of data traffic in EPON. One of the most
important methods in field is fair sharing with dual SLAs
(FSD-SLA) proposed a fair sharing with dual SLAs (FSD-
SLA) method, which employs dual SLAs in IPACT to
manage the fairness for both subscribers and service
Vol. 2, No. 5, May 2010
providers [13]. The primary SLA specifies those services
whose minimum requirements must be guaranteed with a
high priority. The secondary SLA describes the service
requirements with a lower priority. This method first
allocates timeslots to those services with the primary SLA to
guarantee their upstream transmissions. After the services
with the primary SLA are guaranteed, the next round is to
accommodate the secondary SLA services. If the bandwidth
is not sufficient to accommodate the secondary SLA
services, the max-min policy is adopted to allocate the
bandwidth with fairness. If there is excessive bandwidth,
FSD-SLA will allocate the bandwidth to the primary SLA
entities first and then to the secondary SLA entities, both by
using max-min fair allocation.
Another optimizing method for dynamic bandwidth
allocation with differentiated QoS support is LSTP .during
the waiting period and thus more accurately grants
bandwidth to each ONU. For each class of traffic, LSTP
estimates the data that arrive during the waiting period
based on the data of this class that actually arrived in
previous transmission cycles by using a linear predictor
[14].
The bandwidth demand of an ONU is thus the reported
queue length plus the estimation. The OLT arbitrates the
upstream bandwidth using this estimation and reserves a
portion of the upstream bandwidth for transmitting the
estimated data in the earliest transmission cycle, thus
reducing packet delay and loss. Other methods proposed for
dynamic bandwidth allocation with differentiated QoS
support are: COPS, HGP, DBAM and so forth [2].
3. The MOBINA DBA Method
The MOBINA is dynamic bandwidth allocation based on
Second Price Auction. In this method, all bandwidth
requests are sent to OLT by ONUs. OLT can make a
dynamic and fair decision about bandwidth allocation by
auction process. The steps of auction process and
bandwidth allocation in the suggested method are as
follows:
Step 1: Kick-off the auction by OLT
Step 2: Submit the bid by ONUs
Step 3: Assess received bids by OLT
Step 4: Schedule and allocate bandwidth to winners
Step5: Update and go to next round of auction
Figure (2) represents the block diagram of the MOBINA
method.
100 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

send the result to ONUs. In step 4, ONU will allocate the

requested bandwidth for each user according to the timing
of auction winners, considering the maximum accessible
bandwidth. Equation (4) shows bandwidth allocation by
OLT.

 BRQ onu (t ) (ONU win )

 i i
onu 
BW (t ) =  (4)
i

 0 (ONU loss )
 i

In step 5, OLT updates the LSTionu value of all the ONUs

that have received bandwidth service in time t and will send
the respective ONU. Equation (5) shows the value of
LSTionu .

0 (ONU i win )

LSTi onu
(t ) =  (5)
1 (ONU
Figure 2.Dialgue diagram of the MOBINA method  i loss )

Finally, after sending LST ionu values, OLT holds for a time
As shown in Figure (2), in Step 1 a signal is sent from OLT
to ONUs in order to ONUs send their requested bandwidth in order to allocated bandwidth be released and the
allocation to OLT for participation in an auction. In step 2 conditions for holding auction by OLT is provided.
based on the MOBINA method, ONUs calculate their Bid
amount based on two parameters of the requested bandwidth 4. Computer Simulations
and the last time of bandwidth allocation by OLT to ONU, To measure the performance of each bandwidth allocation
and send it to OLT. Equation (1) shows the sent Bid amount methods we designed an event-driven C++ based EPON
by ONUi to OLT. simulator. NS-2 and C# software is used as a unique system
to compare simulation results. In this article we use EPON
(
Bid i (t ) = β BRQionu (t ), LSTionu (t − 1) ) (1) structure, as shown in Figure (1), to measure QoS
parameters. The parameters considered for simulation are as
illustrated in Table (1):
In Equation (1), BRQionu is the total bandwidth amount that
has been requested by the users from ONUi and LST ionu Is Table 1: Simulation Parameters
Parameters Symbol Value
the Boolean parameter that shown the last time of ONUi
Number of ONUs N 16
received bandwidth service. Equation (2) shows ONUi
Bit Rate Λ 5 to 57.5 MBit/s
requested bandwidth. Two way delay fiber Tfiber 200 µs
Processing Time Tproc 35 µs
K

∑ BW
Packet Size B 15000Byte(30
BRQionu (t ) = J
R eq
(t ) (2) Packets)
J =1 Ethernet Overhead Beth 38 Byte
Request Message Size Breq 570 Bit
In Equation (2), BWJR eq is the bandwidth required by users Upstream Bandwidth Ru 1 GBit/s
connected to ONUi. Equation (3) shows the required Maximum Transition Pmax 10 Packets
Window
bandwidth of each user.
Guard Time Tg 5 µs
Max cycling time TMax 2 ms
BWJReq (t ) = mJP × LJP × PJ (3) Buffer capacity BQmax 10 Mbyte
Guarantied Bandwidth BAga 60 MBit/s
Interval traffic randomly Interval 50 µs-100 µs
In Equation (3), m JP is the number of requested packets to
Traffic Type CBR
send for user j. LPJ is the length of requested packets to
send, and PJ is the priority of the jth user. In MOBINA All the provided results in this section which will be
introduced, are measured the QoS based on proposed
suggested method, the priority of all users linked to an ONU
approach in comparison with other methods, will be
is the same and is equal to (1/j).
analyzed by some different charts with their quantities such
In step 3, after receiving all the Bids based on second price
auction, OLT will select ONUs that win in auction and will
 (IJCNS) International Journal of Computer and Network Security, 101
as delay, jitter, Throughput, Packet loss and Utilization rate
.
Figure (3) shows traffic packet delay. The proposed
approach has less delay in comparison with the methods of
bandwidth allocation without QoS, although in the previous
methods of bandwidth allocation with different QoS, at the
first place in this approach delay is more because of
providing situation awareness for ONUs, nevertheless in the
other times delay is much better.
Figure 3.The charts for average of packets Delay
Also, Figure (4), shows the vector of The Average of Traffic
Packet Jitter. This chart we see that BGP and FSD-SLA
jitter range is higher than other methods. Albeit with
increasing traffic load and good behavior of IPACT
methods, they have good Jitter rather than the other methods
. At the point of view in jitter charts, which is implied much
better jitter ratio of proposed approach compared to the
other methods.
Figure 4.The Charts for Average of Packet Jitter
In addition, The chart related to throughput and loss rate of
this traffic Loads is shown in Figure (5) and (6). As a result
of analyzing these charts that implied better and more
sufficient throughput compared to the other methods that we
mentioned completely. Also we can realize that packet lost
is reduced by increasing traffic load.
Vol. 2, No. 5, May 2010
Figure 5.The Charts for Throughput Rate
Figure 6 . The Charts for Packet Loss Rate
Finally, Figure (7) depicts the rate of utilization in EPON
systems. This Fig demonstrates that increasing of traffic
load has anonymous utilization compared to the best
methods. MOBINA approach has the most rate of network
Utilization. Albeit this subject, the behavior of this
Approach was predictable. By increasing the traffic load to
network, Utilization rate in other methods didn’t have any
difference.
Figure 7.The Charts for Line Utilization rate
5. Conclusion
Bandwidth allocation is a critical issue in the design of an
EPON system. Since multiple ONUs share a common
upstream channel, an EPON system must efficiently utilize
the limited upstream bandwidth in order to meet the
bandwidth demands and quality of service requirements of
end users. In this article, MOBINA method optimizes the
bandwidth limitation problem. The results of the
simulations showed that in comparison to other methods,
the proposed method offers better performance in terms of
average packet loss ratio and throughput ratio. And as a
result, we improve the overall network performance.
102
References
[1] G. Kramer and G. Pesavento, "Ethernet passive optical
network (EPON):building a next-generation optical
access network," IEEE Communications Magazine,
pp. 66-73, 2002.
[2] J. Zhenga, T. Hussein, ”A survey of dynamic
bandwidth allocation methods for Ethernet Passive
Optical Networks ”, Elsevier Journal Optical
Switching and Networking, pp. 151-162, 2009.
[3] A. Xiaofeng ,s. Abdallah , “On the fairness of dynamic
bandwidth allocation schemes in Ethernet passive
optical networks”, Elsevier journal of computer
communications, pp. 2123-2135, 2008.
[4] H. Naser, H. Mouftah, “A joint-ONU interval-based
dynamic scheduling algorithm for Ethernet passive
optical networks”, IEEE/ACM Transactions on
Networking ,pp. 889-899 , 2006.
[5] Z. J., "Efficient Bandwidth Allocation Method for
Ethernet Passive Optical Networks," in Proceedings
of IEE Communications, pp.464-468, 2006.
[6] X. Bai, A. Shami, C. Assi, Statistical Bandwidth
Multiplexing in Ethernet Passive Optical Networks,
in Proceedings of IEEE Global Telecommunications
Conference, 2005.
[7] J. Zheng, H. T. Mouftah, "An Adaptive MAC Polling
Protocol for Ethernet Passive Optical Networks", in
Proceedings of IEEE ICC, pp. 1874-1878, 2005.
[8] B. Moon, “Emergency handling in Ethernet passive
optical networks using priority-based dynamic
bandwidth allocation”, in Proceeding of IEEE
INFOCOM'08, pp. 1319_1327, 2008. .
[9] S. Bhatia, R. Bartos, IPACT with smallest available
report first: A new DBA method for EPON, in
Proceeding. of IEEE ICC'07, pp. 2168_2173, 2007.
[10] H. Song, B.-W. Kim, B. Mukherjee, Multi-thread
polling: A dynamic bandwidth distribution scheme in
long-reach PON, IEEE Journal on Selected Areas in
Communications ,) pp. 134-142, 2009.
[11] Y. Zhu, M. Ma, “IPACT with grant estimation
(IPACT-GE) scheme for Ethernet passive optical
networks”, IEEE/OSA Journal of Lightwave
Technology, pp. 2055-2063, 2008.
[12] M. Ma, Y. Zhu, T. Cheng, “A bandwidth guaranteed
polling MAC protocol for Ethernet passive optical
Networks”, in Proceeding of IEEE INFOCOM'03, pp.
22_31, 2003.
[13] A. Banerjee, G. Kramer, B. Mukherjee, “Fair sharing
using dual service-level agreements to achieve open
access in a passive optical network”, IEEE Journal on
Selected Areas in Communications, pp. 32_44, 2006.
[14] Y. Luo, N. Ansari, “Limited sharing with traffic
prediction for dynamic bandwidth allocation and QoS
provisioning over EPONs”, OSA Journal of Optical
Networking, pp.561_572, 2005 .
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
Authors Profile
Alireza Hedayati:. received his B.Sc. and
M.Sc. in Computer Hardware Engineering
from Azad University of Tehran Central
Branch and Tehran science and research
branch in 2000 and 2003, respectively. He is
currently studying Computer Hardware in his
Ph.D. degree. His research interests include
Optical Networks, Next Generation Network, Network
Management and QoS.
Mehdi N. Feshaaraki:. received his Ph.D.
in computer Engineering from NSW
university of Australia. He is currently
associate professor in computer Engineering
Department of IAU. . His research interests
include Computer Networks, information and
knowledge architecture.
Kambiz Badie:. received his B.Sc., M.,Sc, and
Ph.D. in electronic engineering from the Tokyo
Institute of Technology, Japan, majoring in
Pattern Recognition & Arlificia1lntelligance. His
major research interests are Machine Learning,
Cognitive Modeling, and Systematic Knowledge
Processing in general, and Analogical
Knowledge Processing, Experience-Based
Modeling and Interpretative Modeling in particular with emphasis
on ides and technique generation .He is currently associate
professor in Iran Telecom Research Center(ITRC).
Ahmad Khadem zadeh : He received the B.Sc.
degree in applied physics from Ferdowsi
University, Meshed, Iran, in1969 and the M.Sc.,
Ph.D. degrees respectively in Digital
Communication and Information Theory & Error
Control Coding from the University of Kent,
Canterbury, U. K .He is currently associate professor in Iran
Telecom Research Center and also He is the Head of Education &
National Scientific and Informational Scientific Cooperation
Department at Iran Telecom Research Center (ITRC).
Masoud Davari:. received his B.Sc. in
Computer Engineering from Azad University
of Tehran Central Branch 2010. His research
interests include Computer network, Network
simulation and Software Engineering
 (IJCNS) International Journal of Computer and Network Security, 103
Vol. 2, No. 5, May 2010

An Efficient Key Management Scheme for Mobile

Ad hoc Networks with Authentication
N.Suganthi1, Dr. V.Sumathy2
1
Asst.Professor, Dept. of Information Technology
Kumaraguru College of Technology,
Coimbatore, Tamil Nadu.
suganthiduraisamy@yahoo.co.in
2
Asst.Professor, Dept. of ECE
Government College of Technology,
Coimbatore, Tamil Nadu.
sumi_gct2001@yahoo.co.in

Abstract: mobile ad hoc networks (MANETs) are dynamically
reconfigured networks in which security is a major concern.
MANETs face serious security problems due to their unique
characteristics such as mobility, dynamic topology and lack of
central infrastructure support. Key management is crucial part
of security, this issue is even bigger in MANETs. The
distribution of encryption keys in an authenticated manner is a
difficult task. Because of dynamic nature of MANETs, when a
node leaves or joins it need to generate new session key to
maintain forward and backward secrecy. In this paper we divide
the network into clusters. Cluster head will maintain the group
key, it will also update the group key whenever there is a change
in the membership. Here the re-keying process will be
performed only if there is any movement of nodes within the
clusters. So the computation and communication cost will be
reduced. And also we provide authentication between
communicating nodes both in inter and intra cluster. The
network life time will be extended with the help of monitoring
node. The performance results prove the effectiveness of our key
management scheme.
Keywords: Network Security, Key Management, Key Update,
Mobile Networks, clusters
1. Introduction
We encounter new types of security problems in ad hoc
networks because these networks have little or no support of
infrastructure. The network has no base stations, access
points, remote servers etc. All network functions are
performed by the nodes itself. Each node performs the
functionality of host & router. In mobile ad hoc networks,
nodes within their wireless transmission range can
communicate with each other directly, while nodes outside
the range have to rely on some other nodes to relay message.
MANET is autonomous, multihop networks interconnected
via wireless links. Thus a multi-hop scenario occurs, where
the packets sent by the source node are relayed by several
intermediate nodes before reaching the destination. The
success of communication highly depends on the other
nodes’ cooperation. While mobile ad hoc networks can be
quickly and inexpensively setup as needed, security is a
critical issue compared to wire or other wireless
counterparts. Many passive and active security attacks could
be launched from the outside by malicious hosts or from the
inside by compromised hosts. Without the appropriate
security precautions, critical applications for commercial or
military use cannot employ networking technologies.
Key management is an essential cryptographic
primitive upon which other security primitives such as
privacy, authenticity and integrity are built. However, none
of the existing key management schemes are suitable for ad
hoc networks. The major limitation of these schemes is that
most of them rely on a trusted third party (TTP), thus not
fulfilling the self-organization requirement of an ad hoc
network. Special mechanisms and protocols designed
specifically for ad hoc networks are necessary. Key
management deals with key generation, storage,
distribution, updating, and revocation and certificate service
in accordance with security policies.
Due to dynamic behavior of the MANET, secret
key used for communication is need to be updated whenever
any node joins or leaves the network in order to maintain
the forward and backward secrecy with in the network. If
the network is large and also the mobility is higher,
updating of the key will be more frequent. It will consume
more computation power and also communication power of
nodes. So in our proposal we divide the network into
clusters consisting of small group of nodes. Here the re-
keying process will be performed only if there is any
movement of nodes with in the clusters.
The re-keying process will be distributed between
cluster heads. And also using monitoring node the life time
of the network is extended. The monitoring node is
continuously checking the cluster head for its energy. If the
cluster head has less energy next high capability node will
be elected as cluster head.
And also the details contained in the CH will be shifted to
new CH and this process will be informed to all other nodes
by monitoring nodes.
2. Previous Work
Majority of research on security of ad hoc networks
emphasize the secure routing protocols, there are some
proposals on key generation and distribution issues.
Zhou et al [7] proposed a technique to distribute
certificate authority (CA) functionality. In this method, the
104 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

networks includes n servers providing the certificates, out of

which t + 1 are needed for creation of the valid certificate
but t is not enough.
Seung Yi et al [11] proposed an efficient and
effective distributed CA by selecting physically and
computationally more secure nodes as MOCAs(Mobile
Certificate Authority) and they used threshold cryptography
to distribute the CA’s private key among these MOCA
nodes.
Caner Budakoglu et al [12] proposed a modified
form of distributing the certificate authority functionality.
They proposed a hierarchical threshold level, so that it offers
a different level of security to satisfy the needs for a wide
variety of applications.
Bing Wua et al [4] propose a secure and efficient
key management (SEKM) framework for mobile ad hoc
networks. They build a public key infrastructure (PKI) by .
applying a secret sharing scheme and using an underlying Figure 1. Cluster formation
multi-cast server groups. They gave detailed information on
the formation and maintenance of the server groups. Each
server group creates a view of the certificate authority (CA) 3.1 Group Key Generation And Distribution:
and provides certificate update service for all nodes,
including the servers themselves. Whenever a node comes within the radius of any cluster,
In the case of self organized public key first it will send hello message along with its id and public
management, the system does not need any kind of key.
infrastructure to authenticate keys. S. Capkun et al [3] Cluster head receive this message and initiate the group
suggested a method based on the users issuing certificates to key calculation. It uses the entire nodes public key to
each other based on personal acquaintance. These calculate group key as follows.
certificates are used to bind a public key and a node id. CH : grpkey =(α) p1+p2+….pn+ kCH mod p X Srv -----(1)
Every node should collect and maintain an up-to-date where α – primitive root of p
certificate repository. Certificate conflict is just another kCH -secret key of cluster head,
example of a potential problem in this scheme. p1,p2 …pn – public keys of individual nodes within
the cluster,
3. Proposed Scheme p – prime number and
We had divided the networks into clusters and each Srv – secret random value generated every time while
cluster will have 1- hop nodes and cluster head. Some re-keying.
efficient existing algorithm can be used to group the users
into clusters and generate a cluster head for each one. The Algorithm for the group key generation:
users in each cluster are in a flat network topology and the Procedure:
local key management policy is centralized. BEGIN
The users in the group are classified into two types: Precondition:
cluster heads and ordinary users. The cluster head is Node should present within the cluster
responsible for cluster management, membership if CH gets public keys of all nodes
maintenance and group key distribution and updating. calculate group key as follows:
Initially, all nodes are assigned an id, status code (for CH : grpkey =(α) p1+p2+….pn+ kCH mod p X Srv
cluster head differentiation), its private key and public key. endif
Cluster head is selected based on lowest id algorithm. To
cope up with the dynamic nature of the ad hoc nodes //comment: Encryption
security is enhanced by providing re-keying concept. Re- CH à nodes within the cluster : E (grpkey)
keying is done by cluster head when ever any node joins or using RSA algorithm where
leaves the network to ensure backward secrecy (i.e., a new E(grpkey) = ((grpkey)e mod n) in which
member should not know the previous information that was {e,n} are public key pair.
exchanged) and forward secrecy (i.e., an existing member
should not receive the information exchanged after it leaves Nodes: D(grpkey) = (E(grpkey)d mod n) in which
the network) {d,n} are private key pair.
END
 (IJCNS) International Journal of Computer and Network Security, 105
Vol. 2, No. 5, May 2010

The distribution of group key is done using RSA algorithm.

Cluster head having (e,n) public key and every node C = (M)e mod n (e,n) – public key ---(2)
maintains (d,n) private key. When ever any new node joins M = (C)d mod n (d,n) – private key ---(3)
into the cluster, cluster head calculates new group key and
multicast to already existing nodes. And cluster head unicast Suppose if any node leaves the cluster to maintain
the group key to new node along with private key for RSA forward secrecy, cluster head calculate new group key and
algorithm. also it will change key encryption key also.
And using individual node’s public key, key encryption
3.2 New Node Joins key will be transmitted and using that key group key will be
encrypted. Cluster head will unicast this information to all
When ever a node A joins in the network the following the nodes.
messages are exchanged.
CH à all node: E(Pbnode , (E(e,[Ko]) || (d,n) ) )
Node A joins CH already where
existing nodes Pbnode - public key of new node,
{e, n } - Public key pair of RSA algorithm and
(hello msg, {d, n } - Private key pair of RSA algorithm.
ida, pka,N) calculates new
group key 3.4 Providing Authentication

multicast group
key to all nodes
unicast
(grp key,
private key(d,n))
First group key will be encrypted using RSA algorithm and
it is again encrypted using new node public key.
CH à new node: E(Pbnode , (E(e,[Ko]) || (d,n) ) ) where
Pbnode - public key of new node,
{e, n } - Public key pair of RSA algorithm and
{d, n } - Private key pair of RSA algorithm.
Initially the new node gets the RSA private key pair by
decrypting received key information using its own private
key, after then it decrypts using RSA private key to get
group key.
All other already existing nodes receive encrypted
information and decrypt using RSA algorithm to get the
updated group key.
3.3 Existing Node Leaves
Whenever an existing node B leaves the cluster the
following messages are exchanged.
Node B leaves CH already
existing nodes
Within the network, if any two nodes A and B want to
communicate first it will authenticate each other. The
authentication steps are as follows.
1. Node A calculate hash value using its (id, public
key, group key) and transmit the hash value, id
and public key to node B
2. Node B receive hash value and also it calculate new
hash value from A’s id, public key and group key.
3. Node B will check the received values and
calculated value both are equal or not.
4. If the hashed values are equal, it identifies the peer
node as authenticated node.
Algorithm for authentication:
Procedure:
BEGIN
Precondition:
The nodes must wish to communicate with each other.
if
get the peer nodes public key and ID.
calculate the hash value and transmit to the peer one.
A à B : hash (ida ,pka, group key) || ida ,pka
peer node : calcuate the same hash value.
endif
if both hash values are same
both are authenticated nodes
endif
END

(leave msg, idb) calculates 3.5 To maintain network connectivity

new group key
unicast group key to
all individual nodes
(grp key, private key(d,n))
The group key is encrypted first using RSA algorithm and
then using individual nodes public keys.
After clustering, the cluster head has to do more
computation and so its energy may drain fast.
To check this we can use periodic hello message. If the
cluster head is not done any re-keying operation for the
threshold time of 20 s, it has to broadcast hello message.
This network maintenance can be done with the help of
monitoring nodes. This node can be next highest resource
available node after the cluster head.
This node will maintain a timer, it will check for any
communication from CH. If it does not receive any message
(or) rekeying information, it will send hello message to CH
106 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

to check for it's presence and wait for it's reply. Suppose if it ii) With respect to transmission delay:
does not receive reply, it will inform to other nodes to
initiate Cluster head selection process.
The monitoring node will wait for a period T and then it
will send hello message.
T= threshold time + propagation delay --- (4)

4. Implementation and performance

comparison

The simulations are performed using Network simulator

(Ns-2), particularly popular in ad hoc networks. The MAC
layer protocol IEEE 802.11 is used in all simulations. The
Destination Sequence Distance Vector (DSDV) routing
protocol is chosen for the simulations. The simulation
parameters used are summarized in Table 1:

Table 1: Simulation Parameters

Figure 3. graphs showing the transmission delay from
Parameter Value central node to individual node (19) during the distribution
Simulation time 1000 sec of group key.
Topology size 750m X 750m
No. of nodes 20
Routing protocol DSDV
Transmission Range 250m
Mobility Model Random Waypoint
MAC IEEE 802.11
Node Mobility 0 to 20 m/sec

Without clusters, the computation time, time delay and

packets transferred from central node are more. Whereas
with the formation of clusters, it is greatly reduced due to
it’s distributed behavior. The performance comparison is as
follows:

i) With respect to communication cost:

Figure 4. graphs showing the transmission delay from
central node to individual node (12) during the distribution
of group key.

5. Conclusion:

The cluster based key management provides the

distribution of computation needed for re-keying process.
Re-keying will be performed within the cluster. And also
using hash function, authentication is provided between
communicating nodes. Inter cluster communication is done
in efficient manner. To maintain network connectivity, a
node called monitoring node is used.

Figure 2. graphs showing the no of packets handled to

distribute the group key during the rekeying process
 (IJCNS) International Journal of Computer and Network Security, 107
Vol. 2, No. 5, May 2010

References Systems, IEEE Transactions on Volume 19, Issue 5,

Page(s):577 – 587, 2008.
[1] Aftab Ahmad, Mona El-Kadi Rizvi, Stephan Olariu [16]Kejie Lu, Yi Qian, Mohsen Guizani, and Hsiao-Hwa
“Common Data Security Network (CDSN)” Chen, “A Framework for a Distributed Key
Q2SWinet’05, Montreal, Quebec, Canada, 2005. Management Scheme in Heterogeneous Wireless Sensor
[2] P.Papadimitratos and Z.J. Haas. "Secure Data Networks”, IEEE Transactions on Wireless
Communication in Mobile Ad Hoc Networks" IEEE Communications, Vol 7, No. 2, Feb’ 2008
Journal on Selected Areas in Communications (JSAC),
Special Issue on Security in Wireless Ad Hoc Networks,
2nd Quarter of 2006.
[3] S.Capkun, L. Butty´an, and J.-P. Hubaux. “Self-
organized public-key management for mobile ad hoc
networks” IEEE Transactions on MobileComputing,
2(1), January-March 2003.
[4] Bing Wu, Jie Wu, Eduardo B. Fernandez, Spyros
Magliveras. “Secure and Efficient Key Management in
Mobile Ad Hoc Networks” Proceedings of the 19th
IEEE International Parallel and Distributed Processing
Symposium (IPDPS’05) IEEE, 2005
[5] Ozkan M.Erdem.” Efficient Distributed Key
Management for Mobile Ad Hoc Networks”0-7803-
8623-W04, IEEE, 2004
[6] YANG Ya-tao, ZENG Ping, FANG Yong, CHI Ya-Ping.
“A Feasible Key Management Scheme in Adhoc
Network” Eighth ACIS International Conference on
Software Engineering, Artificial Intelligence,
Networking, and Parallel/Distributed, IEEE, 2007
[7] Zhou, L. and Z. Haas. “Securing Ad Hoc Networks”,
IEEE Network Magazine, Vol. 13, 1999.
[8] Aldar C-F. Chan, “Distributed Symmetric Key
Management for Mobile Ad hoc Networks”, IEEE
INFOCOM, 2004.
[9] Atef Z. Ghalwash, Aliaa A. A. Youssif, Sherif M.
Hashad and †Robin Doss “Self Adjusted Security
Architecture for Mobile Ad Hoc Networks (MANETs)”
6th IEEE/ACIS International Conference on Computer
and Information Science (ICIS 2007)
[10] Yanchao Zhang, Wei Liu, Wenjing Lou, and Yuguang
Fang, “Securing Mobile Ad Hoc Networks with
Certificateless Public Keys”, IEEE Transactions on
Dependable and Secure Computing, vol. 3, No. 4, Oct-
Dec 2006.
[11] Seung Yi, Robin Kravets, “Key Management for
Heterogeneous Ad Hoc Wireless Networks ”, Proc. of
the 10th IEEE International conference on Network
Protocols (ICNP’02), pages 202-203, Nov. 2002.
[12] Caner Budakoglu and T.Aaron Gulliver, “Hierarchical
Key Management for Mobile Ad Hoc Networks”, 0 -
7803-8521-7/04 IEEE, 2004.
[13]Renuka. A and Shet K.C, “Cluster Based Group Key
Management in Mobile Ad hoc Networks”,
International Journal of Computer Science and Network
Security, Vol 9, No 4, 2009.
[14] Patrick P.C.Lee, John C.S Lui, and David K.Y. Yau,
“Distributed Collaborative Key Agreement and
Authentication Protocols for Dynamic Peer Groups”,
IEEE/ACM Transactions on Networking (TON)
Volume 14, Issue 2 Pages: 263 – 276, 2006.
[15]Lihao Xu, and Cheng Huang, “Computation Efficient
Multicast Key Distribution”, Parallel and Distributed
108 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

CAN Bus, an Efficient Approach for Monitoring

and Controlling
Manufacturing Processes in Industries
Rashmi Jain1, N. G. Bawne2
1
Department of Computer Technology,Yeshwantrao Chavan College of Engineering,
RTM Nagpur University, India
rashmil_jain@rediffmail.com
2
Department of Computer Science,G. H. Raisoni College of Engineering,
RTM Nagpur University, India
narenbawane@rediffmail.com

Abstract: This paper shows a simple and effective approach to
the design and implementation of automated Industrial
manufacturing Systems oriented to control the characteristics of
each individual product manufactured in a production line and
also their manufacturing conditions. The particular products
considered in this work are steel sheets that under go the
galvanizing processes. However, the approach can be modified
to be applicable in other industries, like paper, textile,
aluminum, etc. The system makes use of sensor nodes placed at
different places to monitor and control various parameters of
interest to guarantee and improve the quality of the products
manufactured in many industries. The proposed system uses
CAN bus to send and receive the data from the nodes to the
central controller which provides many advantages over the
currently used distributed control system.
Keywords: Galvanizing processes, CAN bus, multi-master,
Electro magnetic interference (EMI), distributed control systems
(DCS).
thousands of sensors monitor current naval vessel’s
condition and performance [1]. The system can monitor
different quantities like temperatures, speed, torque,
pressure with the help of set of hundreds of sensor nodes at
distributed locations to sense different physical quantities
and send the data to a centralized place to monitor the
processes going on. Currently distributed control systems
(DCS) are used to collect the data from various places and
control the required processes with the help of these control
systems. The structure of a typical DCS is shown in fig 1.
Instead of using distributed control system to carry the data
we use CAN (Controller Area Networks) bus so that the
speed of communication increases and reduce the number of
wire harnesses caused otherwise. Different protocols and
serial communication standards are used for this
communication.

1. Introduction

CAN (Controller Area Network) was originally designed for

robust use in communicating among different control
systems of a vehicle. It works efficiently in the noisy and
continuously moving circumstances in a vehicle. The
robustness of this communication can also be efficient for
industries.
In industries the sensors are essential components of
closed-loop control systems. They are used in all industries
for different types of applications from monitoring a
machine tool in a manufacturing plant to control of a
process in a chemical plant.
Traditionally, using sensors meant that long cables were
needed to connect each sensor back to a centralized
monitoring or control station. Each cable may have
contained multiple wires for both power and sensor data.
When working with a small number of sensors, this may not
seem like a daunting task to wire, but consider the number
of sensors in some practical applications. For example,
dozens of sensors may be used to monitor various thermal
parameters in a manufacturing facility; thousands of sensors
monitor the heat shield tiles on the Space Shuttle; tens of
Figure 1. Distributed Control Systems
Using CAN bus we can connect all these nodes to a single
twisted pair cable with maximum of 1MBPS speed. It was
originally designed for application in vehicle networks
where the work environment is noisy and have more Electro
magnetic interference (EMI) but it is being used in
industries in production line and machine as an internal bus
[7]. The ISO standard 11898 is used in this network. As it is
a multimaster system, any node can initiate communication
if the bus is free. We can use 11 bit or 29 bit arbitration field
to identify the message. In this application we are using 11
bit message identifier in the data frame. CAN transmit data
 (IJCNS) International Journal of Computer and Network Security, 109
through a binary model of "dominant" bits and "recessive"
bits where dominant is a logical 0 and recessive is a logical
1. If one node transmits a dominant bit and another node
transmits a recessive bit then the dominant bit "wins". A
typical CAN node is shown in fig. 2.
Figure 2. CAN Node
In this paper we are using CAN for monitoring and
controlling the process of galvanizing the metal sheet. In
this process the metal sheet has to pass through several
stages of operation, with different temperatures, pH value of
solution maintained at certain value in different stages to
ensure the quality of the product. So our system will monitor
these temperatures & other quantities like speed of
conveyer, pH value, time etc. with the help of set of sensor
nodes.
2. The Galvanizing Process
The process of galvanization is used to save the metal sheets
from oxidation and increase the life of the product
manufactured from this sheet.
2.1 Steps in Galvanization process
The process consist of major four steps:
i. Clinning
ii. Degreasing
iii. Derusting
iv. Galvanizing
In the first step the surface of the sheet is simply clinned
with water. In the next step Degreasing is done with
degreasing solution at 75 to 80ºC for 15 to 20 minutes and
the sheet is rinsed with water. The temperature is sensed
using temperature sensor and controlled using a relay
circuit. Next derusting is done with the solution of H2 SO4 or
HCL with concentration of 10 to 15%. Depending upon the
rust present the pH value of this acid can be chosen from 1
to 5 and the process continues for 15 to 20 minutes. After
derusting the material is again washed in the next water
container. After this degreasing and derusting the material
is ready for galvanization. The galvanizing container
Vol. 2, No. 5, May 2010
contains alkali solution. The container has a conductor rod
which acts as an anode. The pure zinc block is hanged to
this rod. A constant current is supplied through a rectifier.
A chemical reaction between the alkali solution and zinc
takes place which releases the positive ions of zinc. The
material to be galvanized is dipped into the alkali solution
in the tub. The tub itself acts as ground making the material
cathode and the positive ions of zinc gets deposited on the
material, carrying out the galvanization process. The time
for which this process should continue depends upon the
shape and size of the material. If the material has some
inner curved surfaces or with small opening then it takes
more time to complete the process. During this chemical
reaction the pH value of the alkali solution decreases. If this
value goes below a certain set value then the system should
give an alarm indicating that the solution is to be changed.
This can be done by using a pH value sensor and
periodically sending output to the controller.
The material is moved from one container to other
container using a conveyer. With the help of a flexible hook
the material can be connected to the conveyer. A typical
schematic is shown in the following Fig. 3.
1: Degreasing tub, 2: Washing tub, 3: Derusting tub, 4: washing
tub, 5: galvanizing tub with zinc block connected to anode, 6, 7:
Heaters, 8: Metal Sheet, 9: Conveyer, 10: Hook
Figure 3. Process of Galvanization.
2.2 The CAN standard Used
CAN is the leading serial bus system for embedded control.
More than two billion CAN nodes have been sold since the
protocol's development in the early 1980s. CAN is a
mainstream network and was internationally standardized
(ISO 11898–1) in 1993 [4]. The automobile and other
industries have hitherto witnessed the advent of various
electronic control systems & embedded systems that have
been developed in pursuit of safety, comfort, pollution
prevention, and low cost [1]. For physical transmission of
the node data to the controller the CAN standard used is
ISO 11898. It defines all the electrical parameters for the
serial communication. As the data is to be communicated
over a single network, the CAN uses only three layers of the
ISO/OSI reference model, namely, physical, data link and
application layer.[7]
The major design issues considered are as follows:-
• Develop cost effective sensor network using CAN
• Enable message communication in both directions
over CAN bus
110 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

• Flexibility in increasing number of nodes without it with the preset value. If the temperature is less than this
any problem, set point then the heater is switched on with the help of a
• Make system more reliable, as wire harnesses is relay, otherwise it is switched off. This relay operates with
reduced he interrupt signal at pin PD3/INT1. The information
generated i.e. current temperature value and status of relay,
3. The System Architecture by the microcontroller is sent to the centralized controller
through the CAN port for further processing. The CAN
Various nodes with multiple sensors can be deployed in controller used is 8 pin chip MCP2551. It is connected to
various sections to sense the quantity of interest and send it microcontroller at pins PD0/RXD and PD1/TXD. The
over the CAN bus to the base station as given in the block circuit diagram of the node structure or slave unit is shown
diagram in fig 5. Some processing is done on this data so in fig. 6.
that the size of the data to be further communicated to the
server can be reduced and bandwidth utilization will be
done efficiently. At the server end the data gathered will be
analyzed for proper monitoring purpose and taking SLAVE UNIT
appropriate actions during the manufacturing process. The
structure of the Centralized monitoring server is given in
CAN BUS
TEMP SENSOR J2
5
9
U2 4
R1 U1 5 8
U3 Y
9 22 3 3
RESET PC0/SCL D

1
23 6 7
47k PC1/SDA Z
13 24 2
XTAL1 PC2/TCK

+V S
12 25 7 6
XTAL2 PC3/TMS B
26 2 1
PC4/TDO R
2 40 27 8
VOUT PA0/ADC0 PC5/TDI A
39 28 CONN-D9M
PA1/ADC1 PC6/TOSC1
38 29 MCP2551
PA2/ADC2 PC7/TOSC2

GN D
37
PA3/ADC3
36 14
PA4/ADC4 PD0/RXD
35 15
PA5/ADC5 PD1/TXD
3
LM35 34 16
PA6/ADC6 PD2/INT0
33 17
PA7/ADC7 PD3/INT1
18
PD4/OC1B
1 19
PB0/XCK/T0 PD5/OC1A
2 20
PB1/T1 PD6/ICP
3 21
PB2/INT2/AIN0 PD7/OC2
4
PB3/OC0/AIN1
5
PB4/SS
6
PB5/MOSI
7 30
PB6/MISO AVCC

following fig 4.
8
PB7/SCK
ATMEGA16
AREF
32
OUTPUT RELAY
12V RL1
POWER SUPPLY OMI-SH-124L

U4
Figure 4. Centralized Monitoring Server 7805
J3 D1 12V 1 3 VCC
VI VO
1
GND

2 R2 Q1
3 1N4007
D2 C3 BC547
C1 C2 100n 47k
2

CONN-SIL3 2200u 100n

1N4007
GND

Figure 6. Circuit Diagram of Node

Centralized controller may use some keypad with 3 buttons

Figure 5. Node Structure
Different nodes can have the number of sensors attached to
them. Here for demonstration purpose we are using
temperature senor, LM35, at a node. The data from the
sensors is collected and processed by the microcontroller.
The microcontroller used here is AVR ATMEGA16 which
has an inbuilt 8 channel ADC implemented at port PA. The
output of the temperature controller is given to PA0 pin.
Microcontroller converts this to digital value and compares
UP, DOWN, ENTER for entering the new set points by
increasing or decreasing the current set point. A LCD (16 X
2 lines) is used to display value of temperature and the
status of the relay. To receive the information from the
nodes and sending new set point it again uses the CAN port.
ATMEGA16 is used as the controller here. For further use
and storage of the monitored data it is sent to the PC via
serial communication port, MAX232 [7]. It is connected to
the TXD and RXD pins of the microcontroller. The circuit
diagram of the Centralized controller is shown in fig. 7.
Both the circuits use the regulated power supply which will
give 5V output for driving the circuits.
 (IJCNS) International Journal of Computer and Network Security, 111
Vol. 2, No. 5, May 2010

MASTER UNIT
LCD1
LM016L

VDD
V SS
V EE CAN BUS
RW
RS

D0
D1
D2
D3
D4
D5
D6
D7
J2
E

5
1
2
3
4
5
6
7
8
9
10
11
12
13
14
9
U2 4
R1 U1 5 8
Y
9 22 3 3
RESET PC0/SCL D
23 6 7
47k PC1/SDA Z
13 24 2
XTAL1 PC2/TCK
R2 12
XTAL2 PC3/TMS
25 7
B
6
47k 26 2 1
PC4/TDO R
40 27 8
PA0/ADC0 PC5/TDI A
39 28 CONN-D9M
PA1/ADC1 PC6/TOSC1
38 29 mcp2551
PA2/ADC2 PC7/TOSC2
37
PA3/ADC3
36 14
PA4/ADC4 PD0/RXD
35
34
33
PA5/ADC5
PA6/ADC6
PD1/TXD
PD2/INT0
15
16
17
SERIAL COMMUNICATION
PA7/ADC7 PD3/INT1
18
PD4/OC1B
1 19
PB0/XCK/T0 PD5/OC1A VCC
2 20
PB1/T1 PD6/ICP
3 21
PB2/INT2/AIN0 PD7/OC2
4
PB3/OC0/AIN1 C8
5
PB4/SS
6
PB5/MOSI
7
PB6/MISO AVCC
30 C5
8 32 1u 1u
PB7/SCK AREF
1 3
ATMEGA16 J6
1
C1+ C1-
6

POWER SUPPLY 11
12
10
9
T1IN
R1OUT
T2IN U4
T1OUT
R1IN
T2OUT
14
13
7
8
2
7
3
8
U4 R2OUT R2IN
4
7805 MAX232 2 9
J3 D1 VS+
12V 1 3 VCC 6 5
VI VO VS-
1
G ND

2
C2+ C2- C3 CONN-D9F
3 1N4007
D2 C3
C1 C2 100n
C4 100n
2

CONN-SIL3 2200u 100n 4 5 C6

1u
1N4007
GND 1u

Figure 7. Circuit Diagram of Monitoring Node

4. The Software Design

To develop the software to control the operation of the

circuits developed we are using AVR Studio 4 and WinAVR
which gives a complete Integrated Development
Environment, IDE required. This studio is chosen as it is
easier to write the code in C Language and efficient for
execution. The Hex file is created as the output file which
can be downloaded onto the microcontroller.
The flowchart for process of sending data from slave node
to the monitoring and controlling node is given in fig. 8. Figure 8. Flowchart for sensor node operation
The microcontroller periodically reads the data from the
temperature sensor process it through ADC and compares it 5. Advantages
with the set point, here 30 ºC. If the temperature is above 30
ºC then switch off the relay controlling the heater. The proposed system will provide following advantages:
Otherwise if relay is off then switch it on to maintain the • The high-volume applications increase their
temperature. Then this data is sent to the central monitor if efficiencies.
the channel is free. The arbitration logic will allow the • They are lowering the prices and promoting
message with dominant bit (0) to transmit the message first adoption in other devices.
if two nodes start the transmission at the same time. • There is flexibility in increasing number of
The CAN frame uses the 11 bit ID field which allows nodes without any problem.
2048 total number of nodes to be connected in a network[8]. • They make the system more reliable.
If more than this num out of 8 bytes of the data block 3 bytes
are used for sending temperature information and 3 bytes for
sending status of the relay. Two bytes for checksum and one 6. Applications
byte for indicating EOF (end of frame) are used. • The proposed system can be used in different
The flowchart in fig, 9 shows the working of the monitor industries to control manufacturing processes that
node. This node is assigned the highest priority by giving are continuous or batch-oriented, such as oil
the message id “0” to the message transmitted by this node. refining, petrochemicals, central station power
generation, pharmaceuticals, food & beverage
The network is initialized for transmission and reception manufacturing, cement production, steelmaking,
of data. UART is set at 9600 Baud. The data is collected and papermaking [3].
from the nodes periodically and analyzed, if it is reasonable • Multiple sensors and actuators can be added with
then send it to PC for further operations and stored. If it is the different nodes and they can be used to control
not valid then indicate abnormal situation and give alarm. If the flow of material through the plant [6].
some time we have to change the setpoint then using the • A recent trend in the automotive industry has
Keys available it can be increased or decreased and increased the complexity of applications [1]. As a
conveyed to the node accordingly.
112 (IJCNS) International Journal of Computer and Network Security,
result, the number of nodes connected to the CAN
bus has increased.
• Safety and efficiency requirements such as latency,
high data rate, immunity to noise, and error-
detection capability are challenging the current
CAN capabilities.
Figure 9. Flowchart for Controller node operation
7. Conclusion
The system developed with the help of sensor nodes and
CAN Bus improves the performance of the system and
reliability of the system at lower costs as compared to the
presently applied system of distributed control systems using
RS-232 based communication. Multimaster communication
allows any node to initiate communication when bus is free.
The intelligent nodes make proper utilization of the bus.
Automobile industries and other industries are making use
of it for better noise immunity and EMI. Along with the
application in vehicle networks CAN bus is efficiently used
in industries also. In the Galvanization process also the
quality of the finished product will increase if all the phases
are completed in due course of stipulated time and at
controlled temperature and pH value of the solution.
References
[1] W. Xing, H. Chen, H. Ding, “The Application of
Controller Area Network on Vehicle”, Proc. IVEC,
1999, pp. 455 – 458.
Vol. 2, No. 5, May 2010
[2] Krishna Chaitanya Emani, Keong Kam, Maciej
Zawodniok Yahong Rosa Zheng, Jagannathan
Sarangapani, “Improvement of CAN Bus Performance
By Using Error-Correction Codes”, IEEE Region 5
Technical, Professional & Student conference 2007.
[3] Ian F. Akyildiz, Weilian Su, Yogesh
Sankarasubramaniam,“A Survey on Sensor
Networks”, Erdal Cayirci Georgia Instetude of
Technology. IEEE Communication Magazine- August
2002
[4] Kay R¨omer and Friedemann Mattern, “The Design
Space of Wireless Sensor Networks”, IEEE Wireless
Communications, Dec. 2004.
[5] Ping Ran, Baoqiang Wang, Wei Wang, “The Design of
Communication Convertor Based on CAN Bus”,
Proceedings of the 2008 IEEE/ASME International
Conference on Advanced Intelligent Mechatronics,
July 2 - 5, 2008.
[6] William Kaisar and Gregory Pottie, “Principles of
Embedded Network System Design”, First edition,
Cambridge University Press 2005.
[7] Xia Dong, Kedian Wang, Kai Zhao, “Design and
Implementation of an Automatic Weighing System
Based on CAN Bus”,,Proceedings of the 2008
IEEE/ASME International Conference on Advanced
Intelligent Mechatronics July 2 - 5, 2008...O
[8] Javier Diaz, Elias Rodriguez, Luis Hurtado, Hugo
Cacique, Nimrod Vazquez, Agustin Ramirez,,“CAN
Bus Embedded System for Lighting Network
Applications Proceedings of the 2008 IEEE/ASME
International Conference on Advanced Intelligent
Mechatronics, July 2 - 5, 2008.
Authors Profile
Rashmi Jain completed her B. E. (Computer
Technology) from Kavi Kulguru Institute of
Technology and Science, Nagpur University in
1997. Currently completing M.E. in Embedded
systems and Computing from G. H. Raisoni
College of Engineering, Nagpur University.
Meanwhile worked as a lecturer in Comp.
Tech. Dept. at KITS Ramtek, SSGMCE
Shegaon and YCCE Nagpur.ET VOLUME 31
JULY 2008 ISSN 1307-6884 513 © 2008 WASET.ORG
Dr. Narendra G Bawane is Professor and
Former Head of Computer Science and
Engineering department at G. H. Raisoni
college of Engineering, Nagpur
(Maharashtra). Prior to this assignment, he
worked with B.D. College of Engineering
Sewagram and Govt. Polytechnic, Nagpur
for several years. He has total teaching
experience of more than 22 years. He has
completed his B.E. from Nagpur University in 1987 and M. Tech.
in 1992 from IIT, New Delhi. He completed his Ph. D. in 2006 at
VNIT, Nagpur.
 (IJCNS) International Journal of Computer and Network Security, 113
Vol. 2, No. 5, May 2010

A Novel Architecture for QoS Management in

MANET using Mobile Agent
A.K.Vatsa1 and Gaurav Kumar2
1
Assistant Professor
School of Computer Engineering and Information Technology, SHOBHIT UNIVERSITY,
NH-58, MODIPURAM, MEERUT, UTTAR PRADESH, INDIA.
avimanyou@rediffmail.com
2
Bachelor of Technology
School of Computer Engineering and Information Technology, SHOBHIT UNIVERSITY,
NH-58, MODIPURAM, MEERUT, UTTAR PRADESH, INDIA.
kumar.22.gaurav@gmail.com

interest in this field and researches to be carried out for

Abstract: The wide availability of mobile devices equipped with
infrastructure-free wireless communication capabilities together
with the technical possibility to form ad-hoc networks paved the
path for building highly dynamic communities of mobile users.
In many performance oriented applications like co-operative
information sharing, defense applications, disaster
management, mission critical applications and commercial
production applications needs QoS, which is, no doubt, a critical
issue. In this paper, we propose a novel architecture using
hierarchical model to effectively manage the resources on and
attached with Mobile ad-hoc Network and provide robustness
with zero probability of fault. The dynamic clustering and the
criteria for cluster formation causes adaptability to various
situations and enhancement of load bearing capacity as well as
an effective co-operation between several nodes to bear the
responsibilities. The main focus is to minimize the overhead
transmission in the network for which a new approach of co-
operative routing is implemented.
Keywords: MANET, QoS, Mobile Agent, Cluster, Cluster
Head, co-operative Routing, Hierarchical model, Dynamic
Clustering, Hybrid Approach and Host.
1. Introduction
1.1 MANET
Mobile Ad hoc network [9], is infrastructure less, dynamic
multi-hop wireless network, where each node acts as a
mobile router. They are equipped with a wireless transceiver
and by a group of mobile hosts formed on a shared wireless
channel by virtue of their proximity to each other [10].
Routing protocols should adapt to such dynamics and
continue to maintain connection between the
communicating nodes. As a consequence of mobility and
disconnection of mobile hosts, the dynamics of wireless ad
hoc networks, causes a lot of problems to emerge in
designing proper routing algorithms for effective
communication [11]-[12]. Also the time of entering the node
to the network and exiting from it be unpredictable. Work
on unipath routing has been proposed in many literatures
[18]-[22]. There two main classes of routing protocols in
MANET, i.e., table-based or proactive and on demand or
reactive protocols [23]. Any suitable mix may give better
results. Their varied scope and promising features cause
betterment of the network functioning.
1.2 Mobile Agent
Mobile agents are programs that can migrate from host to
host in a network [11], at times and to places of their own
choosing where this is mandatory for the host to posses’
agent hosting capability. The agent code should not have to
be installed on every machine the agent could visit
providing them platform independency [11]. Agents are able
to decide, based on their local knowledge, if, when and
where to migrate in the network. The ability of mobile
agents to replicate themselves into many copies that travel to
different points across the network sounds promising.
During migration, the agent’s local state variables [7] are
transferred to the new host to continue computation there.
Mobile agents are chosen for many applications, due to
several reasons [8], viz. improvements in latency and
bandwidth of client-server applications and reducing
vulnerability to network disconnection.
1.3 Quality of Service
QoS support in MANET’s is a challenging task and it is
required to have a better management over parameters like
minimum bandwidth; minimum delay, jitter guarantee, dis-
connectivity and denial of services and contract based
enforced service agreements. Although, MANET’s are used
for a variety of applications, but, the commercial one’s
mainly multimedia applications require a predefined Quality
of Service [1]. This QoS is in the form of an agreed upon
service level agreement [1]. QoS means a guarantee of some
parameters and their threshold value. We know that
different applications have different requirements of
resources. QoS roughly means providing the applications
with enough network resources, so that they can function
within acceptable performance limits. Since the
requirements of resources vary, the value of QoS parameters
also varies for every application.
1.4 Problem Identification
The problem considered in this paper is to provide a routing
scheme for MANET’s so that the QoS can be provided to
the mobile nodes with decreasing the amount of overhead
transmission and providing such an architecture which
114
properly manages the network with minimum probability of
error occurrence. Though the consumption of resources is a
problem, yet a bigger problem is that even after giving so
much of network capacity, protocols still lack in providing
QoS strictly. Getting in a bit depth there is a strong
requirement to make the updation of the topological
information to a defined level. The reason being frequent
updation may consume considerable amount of network
resources as well as delayed updation may cause delay in
detection of topology changes. Hence, this calls for the
development, which finds a way solving both of these
problems.
1.5 Paper Organization
The rest of the paper is organized in such a way that the
second section describes previous works in the field of
quality of Service management, clustering and formation of
cluster heads. This section also describes the previously
proposed architectures in this field along with a short
description of co-operative routing. The third section consist
of our proposed work with thorough description of how
clustering is to be done and how hierarchical modeling can
help in solving the problems. This provides with a
comprehensive architecture for QoS management
mechanism with detailed description. The fourth section
consists of the future work that can be done in this field.
Finally, the paper is concluded in section five.
2. Related Work
2.1 Quality of Service
The QoS parameters are those properties of the network
which plays important role in the delivery of required
information over the network. The requirement of QoS is
that neither all applications require same network resources
nor from commercial point of view all users pay same for
services. So there is a need to make such distinctions. Now
as many parameters are considered in the QoS related to the
network the transmission will become more effective. There
are some parameters [1], identified to be considered
essential for the transmission viz. bandwidth, delay, jitter
and reliable delivery of data. There is also requirement to
pay attention to the way they are managed, about where and
how. Some architecture effectively managed these in the
form of matrices and calculations are done with defined
formulas, [2]. The QoS matrix which is used to store the
values of parameters of links between every two nodes is
having the node id on both rows and columns. The values
filled in the matrix are the values for every link in the
network and looks like as given below:
the equations are as follows:
bwp(s,d)= min bwe(i,j)>=bwmin …(1)
(i,j)έP(s,d)
delayP(s,d)= ∑ de(i,j) <= D …(2)
(i,j)έP(s,d)
These equations can be interpreted as follows:
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
Equation 1. states that the path with banwidth left on them
denoted by bwe(i,j) which is greater than bwmin are selected
and the one with min. value is decided to be assigned.
Equation 2. states that for delay the summation of all the
delays on all the links is taken and the path where it is
found to be less than D is selected. Similarly other criterion
can be calculated.
2.2 Clusters and Cluster Head
Among several approaches there are techniques which have
considered the network as a whole and others have used
clustering to manage the network effectively. This clustering
is the technique which made it possible to deploy hybrid
approach of protocols in MANET’s. Since, the formation of
clusters is to be done carefully, previous works have done
clustering on three main basis viz. fixed geographical
locations [3], based on numerousity of nodes in a cluster and
depending upon hop counts [4]. The GPS system provides
the information about exact co-ordinate of the node in the
geographical location through which we can form fix size
and fixed boundary clusters with a little dynamics. However,
in the second approach we formed the clusters depending
upon the count of the nodes permissible in a cluster. The
third approach uses the proximity of neighborhood of nodes
in the network and other subjective values. Also, the
determination of cluster head should be done in a way
keeping in mind the proper division of work on the nodes so
that no node would get overloaded with work. This
determination cannot be done with exact calculation and
without error but yet we have much data to calculate that
which node is free most of the time and hence which can
serve as cluster head. Although in current scenario we have
such advance devices which have extremely high processing
capability and hence the selection of cluster head may not be
very strict but yet we try to keep cluster heads away from
relaying the data.
2.3 Previously Defined Architecture
Several architectures have been defined using hybrid of
proactive as well as reactive approach and tried to solve the
problem of MANETS’s. One of the most popular has used
the same hybrid approach [1] and defined how the work is
done in reactive way for path determination and reservation.
(a) Contract enforcer
 (IJCNS) International Journal of Computer and Network Security, 115
(b) Design of QoS management in MANET
Figure 1. The architecture of contract enforcement and path
determination/reservation.
This is sown in figure 1(a) and 1(b). This architecture
divides the work in two main portions viz. contract
enforcement and QoS management. The contract
enforcement is done using four modules viz. shaper module,
Packet Classifier module, packet scheduler module and
feedback module. The QoS management finds a path in
reactive way to the destination following the defined service
level agreement. Another architecture defines how this
traditional Ad-Hoc network can be interfaced with IP
network [5]. This architecture provides a well defined
separation of IP network and MANET. It tries to remove the
misconceptions which most people have about IP links. It
has defined the interaction well. Another model tells about
Integrated Services [13], [14], Differentiated Service [15],
FQMM [6],[13],[16] and SWAN [6],[17]. The Integrated
Services (IntServ) model was the first standardized model
developed for the Internet. The Differentiated Services
(DiffServ) model was designed to overcome the inherent
demerits of the IntServ model by IETF (RFC 1633). A
Flexible QoS Model for MANET (FQMM) has been
proposed which considers the characteristics of MANET’s
and tries to take advantage of both the per-flow service
granularity in IntServ and the service differentiation in
DiffServ. SWAN model was developed by the Comet team
at Columbia University. The model divides the traffic into
real time UDP traffic and best effort TCP traffic.
2.4 Co-operative Routing
Since the motion of mobile agents is pure overhead and if
this could be minimized then it can save a considerable
amount of bandwidth and also the mobile nodes are quiet
sufficient to bear the overhead of merging packets if they
can be an approach of co-operative routing[3] was proposed
which reduces the overhead transmission.
Figure 2. Cooperative Routing
It is shown in figure 2. The message from 22, 23 and 24
arrived 18 at three different times and are destined to 1, 2
and 3. As in figure 1 either they could be individually
Vol. 2, No. 5, May 2010
transmitted to the destination with their individual overhead
or they may be merged in a single packet destined to 15
where they may be reconstructed to form separate packets.
Hence the load on network from 18 to 15 is reduced on the
cost of processing on 18 and 15. Such places where this
merging is carried out are called Rendezvous points. Since it
cannot be assumed for packets to arrive these points at
exactly same time the packets face some waiting time at
these positions, the time for which these packets wait are
called rendezvous period and should be chosen such so that
they may not increase delay to unacceptable level. These
rendezvous points are mainly those from where traffic is
considerably high i.e. in case of MANET’s they are mainly
interfaces between clusters.
3. Proposed Work
3.1 Cluster Formation Criteria and Dynamics in
Clustering
The method of cluster formation in this work depends upon
two conditions. Depending upon previous experiences we
have data about what amount of time some node is using the
transmission medium. The sum of this value should not
exceed a threshold value within a period of time. Hence as
soon as we reach this value we start the formation of new
cluster. Though it may increase the performance but can
cause non-symmetric structure of cluster. To keep it
symmetric we pose a limit on the maximum hops from the
cluster head permissible in the cluster. Now, these are two
criteria’s and whichever is met first can cause the start of
formation of new cluster. Till the dynamics of cluster is
considered the changes in clusters may be triggered due to
two reasons. First, if the usage value exceeds the maximum
threshold value and secondly if any node moves out of
neighbourhood of cluster, causing changes in many clusters.
This is shown in figure 3.
Figure 3. Flow of Control for Cluster formation
Pseudo code for clustering:
Here usage_threshold maximum usage permitted,
hop_threshold maximum hops permitted from the cluster
head, passed as arguments to a function named
form_cluster(), pop() retrieves a node from the
neighbourhood list of head sent as argument and named
head_threshold, sum_cluster_usage is the sum of usage in
the cluster, node_usage is the individual usage of one node,
hop_to_cluster is the hops taken by anode to reach the head,
cluster_head_mamber[] is an array containing the id’s of
nodes added in the cluster, node_id is the identification of
nodes, node_cluster_head is a field with every node pointing
116
to the cluster head of itself, current_cluster_head is the head
of the cluster for which clustering is done presently and
least_usage_value is a function finding the node with least
usage among unassigned nodes kept in a list and passed as
argument called list_unassigned_node.
form_cluster(usage_threshold,hop_threshold)
(1) pop(head_neighbourhood)
(2) if(sum_cluster_usage+node_usage
usage_threshold || hop_to_cluster < hop_threshold)
(3) then: cluster_head_member[next]= node_id;
(4) :node_cluster_head= cluster_head_id;
(5) else current_cluster_head
least_usage_value(list_unassigned_node)
(6) goto step(1);
3.3 Hierarchical Model
The hierarchy in the network means dividing the network in
smaller regions and managing the regions as if they are
small MANET’s. This is like the divide and conquers
approach in the algorithms which we use to divide the
problem in smaller parts. Initially it looks like the same
clustering approach. This gives us several benefits, as this
may divide the load between several nodes to maintain the
topology of the network. This hierarchy will also cause
effective path determination as well as routing and need not
to travel on any path which may not lead to the destination.
This causes the QoS effectively managed. The hierarchy
which we use is rooted to several levels and taking an
abstract view the functioning is similar. The cluster
formation criterions which are previously described can be
used to make upper layer clusters. Now consider a tree
where root is the complete MANET and the leaves are the
lowest and most fundamental clusters. Each node on the n-
1th level is a group of these fundamental clusters. All such
clusters on every level are on equivalent positions on the
network. The formation and splitting criteria are same
except the values which will be different in every level. The
values will also be dynamically generated since there is no
limitation over maximum Value of levels. This can be better
shown by figure below. The hierarchical model is
maintained by a mobile agent instances of which are
travelling in the network inside each cluster and instances
are also running on every level in proactive way updating
the information after every certain time intervals. Also this
mobile agent waits for some time on every node. There are
nodes which act as interfaces between clusters acting as
Rendezvous points where mobile agents co-operate to carry
the information. It is illustrated in figure 4.
Figure 4. Cluster Formation and Hierarchical Model
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
3.2 Architecture
The architecture shown in the figure is just a more detailed
and modified form a previously proposed architectures. This
architecture mainly defines the way in which path
determination is done in case of multilevel hierarchical
model and reservation is done afterwards. The architecture
can be referred in figure 5. This calls for generation of a
< mobile agent on demand hence reactive in nature. The agent
after finding the QoS path to the destination returns via
same path to the source reserving resources at each
intermediate cluster head. If the reservation fails due to
= reservation of resources by some other agent meanwhile,
then again the path determination is done. However the path
determination phase requires contract as well as the
information of network capacity left at that time which is
maintained by the cluster heads.
Figure 5. The Proposed Architecture for QoS management
3.3 Working Principle and Explanation
3.3.1 Hybrid Approach
The previously defined problem of finding a way between
high bandwidth consumption for overhead transmission and
delayed updation can be better solved by the hybrid
approach where at each level the topological information is
updated by mobile agent for every cluster after certain time
and the path determination can be done in reactive way by
another mobile agent on demand. Before coming on to the
actual description of operation of mobile agents we must
discuss that there are several instances of same mobile agent
travelling in the network and hence to distinguish between
them we need to give every instance a unique Id.
3.3.2 Topology Maintaining Mobile Agent
The mode of operation can be described in following steps:
• Every node performs the neighbouhood detection. Each
node must detect the neighbor nodes with which it has a
direct link. Neighbor Discovery Protocol (NDP) is used
to discover the neighbors and update the neighbors
table. It periodically transmits HELLO message at
regular intervals. The reception of NDP beacons
indicates "node found" or "node unreachable"
information, the node will respond as follows:
Node found NDP packet: upon receiving a beacon from a
new neighbor; the neighbor table is updated by adding this
neighbor information as a new record with a zero in the NV
field.
Unreachable NDP packet: NDP packet could be detected as
an unreachable neighbor node, so that neighbor record will
be deleted from the neighbors table.
• If any change is detected, it is communicated to mobile
agent when comes to the node. The motive of
communicating only changes is to minimize the amount
 (IJCNS) International Journal of Computer and Network Security, 117
of load on network again on the cost of extra processing
on the side of nodes.
• This mobile agent moves inside the cluster and collects
all such information as well as updates all such
information on the node. This causes the maintenance
of topology of the cluster on every node. The cluster
head calculates some figures about the paths and the
network capacity left with its cluster. This complete
information with the agent is updated when it reaches
the cluster head. It is later on communicated to the
higher level cluster head by another agent and the
process of updation goes on to maintain the complete
topological information.
This updation in upper level cluster is done in following
way:
• The information of cluster available at lower level is
calculated to generate some information for upper layer
cluster head.
• Every such lower level cluster head is travelled through
interfaces by agent and agent for this level takes
information from cluster heads only.
• This information is updated for upper level cluster head.
An important point to be noted is that the generated
information is not the info about each link but an abstract
knowledge of the bandwidth available through that cluster,
delay through that cluster and similar information about
other parameters. Also if some change in the node
membership is encountered. In this way the info is updated
at the highest cluster head.
3.3.3 Path Determining and Resource Reserving
Mobile Agent
The process of path determination and reservation is done
by another mobile agent who works in the following way:
• Whenever any node needs to transmit the data, it
generates the mobile agent and directs it to the cluster
head.
• At cluster head if the destination node is in the current
cluster then the head tells a QoS path to the agent.
• If the destination is not in the current cluster, the agent is
told the way for upper layer cluster head.
• The process is repeated till the cluster of some level tells
that the node is available in its cluster. From here the
agent takes the path from source to destination
following the lowest level clusters.
• This movement of the mobile agent till it reaches such a
cluster head which has both the source and destination
in its region causes the agent to see from the highest
level the topology of the required region.
• The agent then reaches the destination’s lowest cluster
head following the told path by the cluster head.
At this stage the path to the destination cluster is known. All
the intermediate cluster heads of lowest level are then
travelled in reverse order to reserve the path resources for
the source in the following way:
• The agent itself returns from the lowest level cluster
heads, intermediate in the path.
• At every cluster head it reaches intermediately it reserves
the required amount of resources which should strictly
follow the contract.
Vol. 2, No. 5, May 2010
• The agent reaches back to the source. In this way a
quality of service path is reserved for the source.
The last packet of transmission moves on with setting
resources free. For this we need a special field value of
which causes continuation of reservation or setting resources
free.
3.3.4 QoS Path Determination in detail
This process is done by reaching every cluster head and
collecting the information for various parameters. The
formulas in equation 1 and 2 are as it is followed in our
work and give a clear picture about how the calculations are
done to determine the paths fulfilling the criteria. However
from the side of cluster head the responsibilities are to
provide access to the data on the node about its cluster to the
mobile agent. The rest is on the agent. Whenever the
destination node is found in the cluster the head is
responsible to tell the lowest level QoS path. In case of
failure of path determination an alternative path is chosen,
which in our work, we try to prepare for, if it happens to
occur. For this we maintain several such paths and choose
alternative on the spot. However, choosing such alternative
path may constrain our future selections. If all such paths
are not found then we move forward for second execution of
path determination. Another point which should be
considered is there are some QoS parameters which
themselves depend upon other parameters such as delay or
congestion, jitter and reliability on mobility of nodes etc.
The guarantee for control of jitter is given by cashing of
packets i.e. in case a data packet is delayed due to excessive
traffic load or caused due to new path selection then the
packet already cashed in the memory is used to make up the
delivery gap. Another thing to be considered is the handoff
process. If handoff occurs frequently, it has adverse effect on
every aspect of QoS. Hence several factors which could
cause potential problems during handoff are also managed
under QoS. This management of QoS during handoff is
considered differently. During handoff process, the data
available at the cluster head is to be transmitted to the new
cluster head. This transmission of data is always inter-
cluster in cases when the handoff is due to head moving out
of the cluster, since the indication of need of handoff is
received only at the motion of head out of the cluster. Also
the handoff may occur if the load on the node is so increased
that it can no more bear the responsibility of cluster head in
that case the handoff is intra-cluster. To manage this
handoff process we divide the information available at the
cluster head to be divided into two parts:
1) Readily available
2) Calculated information
This readily available information is the one which can be
gathered inside the cluster as and when it is required. The
calculated info is very concentrated and calculated from a
long time in the past. Now since the handoff process
requires some transfer of information about the cluster if the
cluster head is leaving the cluster this transferred info would
be only the calculated info. The will be gathered by a
request.
118
4. Conclusion
The architecture proposed and the model used solves the
problem of management of topology and providing QoS in a
better way than any other work proposed in this field
however the practicality and implementation are no doubt
matter of molding the proposals and to change them the way
they fit. Higher load of processing is imposed on the nodes
having ample amount of processing capacity and the load of
transmission is reduced thereby saving bandwidth.
Reliability of the routing protocol depends on its
performance within highly dynamic networks. Scalability is
determined by the protocol's performance at crowded
networks. QoS is expressed by the protocol resistance to the
network changes. Applying divide and conquer approach
used in this design helps to effectively manage the network
in highly dynamic situations and decreases the probability of
fault on the network.
5. Future Scope
The architecture in view of its functioning provides a strict
set of rules for effective management of QoS in MANET’s
yet it deserves thorough study and simulation on some test
bed to evaluate its practical results and feasibility. Still the
work can be done to minimize the overhead transmission for
topology maintenance. If feasible to deploy in practical
environment GPS tracking can also be used to reduce load
on bandwidth. Also fixed geographical clustering has its
own advantages and hence can be modified in similar way
to reduce load. Consider agent's size, especially in scalable
networks. Minimizing agent size may benefit from the idea
of MPRs. Regulate TtM value depending on the network
state to save the unnecessary fast updating. Adjust the agent
population to maintain the optimal node-agent ratio in the
network.
References
[1] M.Sulleman Memon, Manzoor Hashmani and Niaz
A.Memon,A Framework for QoS Management and
Contract Enforced Services in MANETs for Prioritized
Traffic Environment, IJCSNS International Journal of
Computer Science and Network Security, VOL.9 No.4,
April 2009.
[2] S.S.Manvi and P.Venkataram, Mobile agent based
approach for QoS routing, The Institution of Engineering
and Technology 2007.
[3] Parama Bhaumik, Somprokash Banyopadhyay, Mobile
Agent Based Message Communication in Large Ad hoc
Networks through Co-operative Routing using Inter-
Agent Negotiation at Rendezvous Points.
[4] S.A.Hosseini-Seno, B.Pahlevanzadeh, T.C.Wan, R.
Budiarto, M.M.Kadhum, A DISTRIBUTED
RESOURCE-MANAGEMENT APPROACH IN
MANETS, Journal of Engineering Science and
Technology Vol. 4, No. 2 (2009) 142 – 153,2009.
[5] Thomas Heide Clausen, A MANET Architecture Model,
INSTITUT NATIONAL DE RECHERCHE EN
INFORMATIQUE ET EN AUTOMATIQUE, January
2007.
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
[6] John B. Oladosu, Akinwale O. Alao, Matthew O.
Adigun, Justice O. Emuoyibofarhe, Design and
Simulation of a Quality of Service (QoS) Guaranteed
Model for Mobile Ad Hoc Networks.
[7] David Kotz and Robert S. Gray, Mobile Agents and the
Future of the Internet, ACM Operating Systems Review,
August 1999, pp. 7-13
[8] Danny B. Lange and Mitsuru Oshima. Seven good
reasons for mobile agents Communications of the ACM,
42(3):88– 89, March 1999.
[9] Corson, S., J.Macker, “Mobile ad hoc networking
(MANET): routing protocol performance issues and
evaluation considerations,” RFC 2501, Jan.1999.
[10] Hakim Badis and Khaldoun Al Agha, An Efficient
QOLSR Extension Protocol For QoS in Ad hoc
Networks.
[11] Manal Abdullah, Helen Bakhsh, Agent-based Dynamic
Routing System for MANETs, ICGST-CNIR Journal,
Volume 9, Issue 1, July 2009.
[12] K. A. H. Amin, “resource efficient and scalable routing
using,” university of north texas, 2003.
[13] Hannan Xiao, Winston K.G. Seah, Anthony L O and
Kee Chaing Chua, A flexible Quality of service Model
For Mobile Ad-Hoc Networks, IEEE, 2000
[14] R. Braden, D. Clark, and S. Shenker, “Integrated
Services in the Internet Architecture – an Overview,”
IETF RFC1663, June 1994.
[15] K. Nichols, V. Jacobson, and L. Zhang, “A Two-Bit
Differentiated Services Architecture for the Internet,”
RFC2638, ftp://ftp.rfceditororg/in-notes/rfc2638.txt, July
1999.
[16] K. Wu and J. Harms, “QoS Support in Mobile Ad Hoc
Networks,” Crossing Boundaries – an interdisciplinary
journal, vol.1, no. 1, pp. 92-104.
[17] R. Sivakumar, P. Sinha, and V. Bharghavan, “CEDAR:
A Core-Extraction Distributed Ad Hoc Routing
Algorithm,” IEEE Journal on Selected Areas in
Communication, vol. 17, no.8, pp. 1454-1465, August
1999.
[18] S. Murthy and G. L. Aceves, “An Efficient Routing
Protocol for Wireless Networks,” Mobile Networks and
Applications, 1996, Vol. 1, No. 2.
[19] C. E. Perkins and P. Bhagwat, “Highly Dynamic
Destination-Sequenced Distance-Vector Routing
(DSDV) for Mobile Computers,” Proceedings of ACM
SIGCOMM, 1994, pp. 234-244.
[20] C. E. Perkins and E. M. Royer, “Ad-hoc On-Demand
Distance Vector Routing,” Proceedings of the 2nd IEEE
Workshop on Mobile Computing Systems and
Applications, 1999.
[21] D. B. Johnson and D.A. Maltz, “Dynamic Source
Routing in Ad Hoc Wireless Networks,” Mobile
Computing, 1996, pp. 153-181.
[22] S. Mueller, R. P. Tsang and D. Ghosal, “Multipath
Routing in Mobile Ad Hoc Networks: Issues and
Challenges,” In Performance Tools and Applications to
Networked Systems, Vol. 2965 of LNCS, 2004.
 (IJCNS) International Journal of Computer and Network Security, 119
Vol. 2, No. 5, May 2010

[23] Muhammad Sulleman Memon, Manzoor Hashmani

and Niaz A. Memon, 2008, A review for uniqueness and
variations in throughput due to performance parameters
of simulators on MANETs routing protocols, presented
in 7th International Conference on EHAC’08, University
of Cambridge, UK Cambridge, 20-22 Feb, 2008, PP:
202-208.

Authors Profile

A.K.Vatsa is working as Assistant Professor

in the School of Computer Engineering and
Information Technology at Shobhit University,
Meerut (U.P.). He obtained his M-Tech(C.E.)
from Shobhit University and B-Tech(I.T.)
from V.B.S. Purvanchal University, Jaunpur
(U.P.). He has worked as software engineer in
software industry. He has been in teaching for
the past one decade. He has been member of
several academic and administrative bodies. During his teaching
he has coordinated several Technical fests and National
Conferences at Institute and University Level. He has attended
several seminars, workshops and conferences at various levels. His
several papers are published in various national and international
conferences across India. His area of research includes MANET
(Mobile Ad-Hoc network), Congestion Control and VOIP-SIP
(Voice over IP).

Gaurav Kumar is pursuing his Bachelor of

Technology in Computer Science and
Engineering from Shobhit University, Meerut
(Formally Shobhit Institute of Engineering
and Technology, Meerut) affiliated to Uttar
Pradesh Technical University and is likely to
complete in the June, 2010. He has published
several papers in National Conferences. He
has good organizing and leadership skills and has been a part of
several societies. He has organized several technical events at Inter
and Intra University and Institute level. He has keen interest in
research area related to MANET (Mobile Ad-Hoc Network) and
WSN (Wireless Sensor Network).
120 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

A Critical Evaluation of Relational Database

Systems for OLAP Applications
Sami M. Halawani1, Ibrahim A. Albidewi2, Jahangir Alam3 and Zubair Khan4
1, 2,
Faculty of Computing and Information Technology,
Rabigh Campus,
King Abdulaziz University
Saudi Arabia
halawani@kau.edu.sa, zubair.762001@gmail.com
3
University Women’s Polytechnic, Faculty of Engineering & Technology
Aligarh Muslim University, Aligarh 202002, India
jahangir.uk786@yahoo.co.uk
4
Department of Computer Science and Engineering IIET, Bareilly, India
zubair.762001@gmail.com

Abstract: This paper analyses in detail various aspects of
currently available Relational Database Management Systems
(RDBMS) and investigates why the performance of RDBMS
suffers even on the most powerful hardware when they support
OLAP, Data Mining and Decision Support Systems. A large
number of data models and algorithms have been presented
during the past decade to improve RDBMS performance under
heavy workloads. We argue if the problems presented here will
be taken care of in the future architectures of relational
databases then RDBMS would be at least the moderate platform
for OLAP, Data mining and Decision Support Systems.
Keywords: RDBMS, OLAP, Data Mining, Decision Support
Systems
1. Introduction
Commercial server applications such as database services,
file services, media and email services are the dominant
applications run on the server machines. Database
applications occupied around 57% of the server volume and
the share is increasing [8]. Corporate profits and
organizational efficiency are becoming increasingly
dependent upon their database server systems, and those
systems are becoming so complex and diverse that they are
difficult to maintain and control. With the advent of modern
Internet based tools for accessing the remote and local
databases, more emphasis is being placed on improving the
performance of Computer Systems along with their
functionality[7][12]. To monitor, predict and improve the
performance of databases has always been a challenge for
database administrators.
A database server performs only database
functions. In terms of workload it performs only
transactions. When a SELECT/ UPDATE statement is
executed a database server interprets it as a series of reads/
writes. Considering that atomic level of anything is smallest
part i.e. it could be said that the atomic level of a transaction
consists of the reads or writes it generates. If broken down to
this level a database server processes I/Os. So, the database
systems can be classified based on the type of transactions
they handle and the subsequent I/Os. This classification
leads to the following two types of database systems:
(i) Online Transaction Processing (OLTP) Systems.
(ii) Decision Support Transaction Processing (DSS)
Systems.
1.1 OLTP Transactions
An OLTP transaction is a unit of work that is usually
expected to run in a very short duration of time because it
deals with the database in real time or online mode. In other
words these transactions constantly update the database
based on the most current information available so, the next
user can rely on that information being the most current. An
example of this kind of system will be a Library Information
System. In this case, all the information pertaining to the
system is kept in tables spread across a disk system, and the
database is online. Any user in the community will have
access to that information.
1.2 Decision Support Transactions:
A different type of application system that is currently in
demand is the DSS (Decision Support System). This type of
system is generally used to provide information to
management so decisions can be made about issues such as
business growth, levels of stock on hand etc. The challenge
is in deriving answers to business questions from the
available data, so that decision makers at all levels can
respond quickly to changes in the business climate. While a
standard transactional query might ask, "When did order
84305 ship?" a typical decision support might ask, "How do
sales in the Southwestern region for this quarter compare
with sales a year ago? What can we predict for sales next
quarter? What factors can we alter to improve the sales
forecast?"
In the OLTP system the rate of the throughput of
transaction is commonly measured in transactions per
second (TPS) or transactions per minute (TPS). With DSS,
throughput is usually measured in queries per hour (QPH).
 (IJCNS) International Journal of Computer and Network Security, 121
Vol. 2, No. 5, May 2010

This unit itself indicates that these queries are of extreme Relatively
size and overwhelm the machine’s resource until they are standardized and
Often complex queries
complete. In almost all cases, the ratio between OLTP and Queries simple queries,
involving aggregations.
DSS transactions equals thousands (sometimes tens of Returning relatively
thousands) of OLTP transactions to one DSS transaction [6]. few records.
An OLTP system, even a large one, is usually not much Depends on the
more than 300 GB, whereas a large DSS system can be 2-5 amount of data
involved, batch data
TB in size. Examples of Decision Support Systems are On
Processing refreshers and complex
Line Analytical Processing (OLAP) and Data Mining. Typically very fast.
Speed queries may take many
1.2.1 On Line Analytical Processing: hours, Query speed
may be improved by
OLAP may be defined as “the interactive process of creating indexes
creating, managing, analyzing and reporting on data”. The Larger due to the
Can be relatively
first point is that analytical processing invariably requires Space existence of
small if historical
some kind of data aggregation, usually in many different Requirements aggregation structures
data is archived
ways (i.e. according to many different groupings. In fact, and history data
one of the fundamental problems of analytical processing is Typically de-
that the number of possible groupings becomes very large Database Highly normalized normalized with fewer
very quickly, and yet users need to consider all or most of Design with many tables tables for faster
processing
them. Relational languages do support such aggregations
but each individual query in such a language produces just Instead of regular
backups, some
one table as its result (and all rows in that table are of the Backup Religiously,
environments may
same form and have the same kind of interpretation). Thus, Backup and Operational data is
consider simply
to obtain n distinct groupings require n distinct queries and Recovery critical to run the
reloading the OLTP
produces n distinct result tables. In RDBMS the drawbacks business
data as a recovery
to approach is obvious. Formulating so many similar but method
distinct queries is a tedious job for the user. Executing those
1.2.2 Data Mining:
queries means passing over the same data over and over
again – is likely to be expensive in execution time. Thus the Data mining can be described as “exploratory data
challenges with OLAP are: analysis”. The aim is to look for interesting patterns in data,
patterns that can be used to set business strategy or to
 Requesting several levels of aggregation in a single
identify unusual behavior (e.g. a sudden change in credit
query.
card activity could mean a card has been stolen). Data
 Offering the implementation the opportunity to compute mining tools apply statistical techniques to large quantities
all those aggregations more efficiently (probably in a of stored data in order to look for such patterns. Data
single pass). mining databases are often very large. Data Mining is also
known as discovering knowledge form a very large database
Special SQL statements are provided to take up the above i.e. getting the interesting patterns from very large
challenges in RDBMS. Examples are the use of GROUPING databases. The process certainly involves thorough analysis
SETS, ROLLUP, and CUBE options available with the of data hence executing one Data Mining query means
GROUP BY clause. But, as far as efficiency is concerned, execution of several OLAP queries. Thus Data Mining may
almost all commercially available RDBMS exhibit poor also be referred as “repeated OLAP”. RDBMS don’t support
performance under OLAP workloads [1][5]. OLAP queries efficiently. Executing repeated OLAP on an
Following table [9] summarizes the major differences RDBMS means to bring the system on its knees.
between OLTP and OLAP systems.
2. Performance Issues of Relational Databases
Table 1: Differences between OLTP and OLAP
OLTP System OLAP System Most of the relational databases currently in use were
Operational data: Consolidation Data: developed during 1990s. Since then their architecture has
Source of OLTPs are the OLAP data comes rarely been revised. The design decisions made at that time
Data original Source of the from the various OLTP still influence the way RDBMS processes the transactions.
Data. Databases. For example designers at that time were not aware of the
To control and run To help with planning, exponential growth of Internet and nobody could expect that
Purpose of
fundamental business problem solving and one day databases would serve as the backbone of large
Data
tasks. decision support. Content Management and Decision Support Systems.
A snapshot of ongoing
Multi-dimensional The current RDBMS architecture was developed keeping in
What the
views of various kinds mind the OLTP, i.e. the architecture was optimized for the
Data Reveals business processes.
of business activities. Hardware, available at that time, to efficiently support
Short and fast inserts Periodic long-running OLTP. In the following sections we discuss some problems
Inserts and
and updates initiated batch jobs refresh the which affect the performance, efficiency and usage of
Updates
by end users. data. RDBMS for OLAP and Data Mining applications.
 122
2.1 Difficult/ Inefficient to Formulate the OLAP Queries:
Consider the following MANUFACTURER_PART (MP)
table:
Table 2: MANUFACTURER_PART (MP)
M_ID Item_No QTY
M101 A11 1000
M101 A12 500
M102 A11 800
M102 A12 900
M103 A12 700
M104 A12 200
Now, consider the following OLAP queries:
1: Get the total item quantity.
2: Get total item quantities by manufacturer.
3: Get total item quantities by item.
4: Get total item quantities by manufacturer and item.
Following are the psudoSQL formulations of these queries:
1. SELECT SUM(QTY) FROM MP GROUP BY ( );
2. SELECT M_ID, SUM(QTY) FROM MP GROUP BY
(M_ID);
3. SELECT Item_No, SUM (QTY) FROM MP GROUP
BY (Item_No);
4. SELECT M_ID, Item_No, SUM (QTY) FROM MP
GROUP BY (M_ID,Item_No);
Drawbacks to this approach are obvious: Formulating
so many similar but distinct queries is tedious for end user.
Also, their execution requires passing over the same data
over and over again which is likely to be quite expensive in
execution time.
Some special SQL options on the GROUP BY clause
may ease life, but up to some extent, e.g. GROUP SETS
option allows the user to specify exactly which particular
groupings are to be performed. The following SQL
statement represents a combination of Queries 2 and 3:
SELECT M_ID,Item_No,SUM(QTY) FROM MP
GROUP BY GROUPING SETS (M_ID,Item_No);
Solution seems to be acceptable but, firstly it requires
a lot of analytical thinking at end user’s part and secondly
options like GROUP SETS are not universally supported
across all SQL implementations. Another problem is that
every SQL Query returns a relation as query result that
forces the user into a kind of row-at-a-time thinking. OLAP
products often display query results not as SQL-style table
but as cross tabulations or graphs.
2.2 RDBMS Storage and Access Methods
Currently available RDBMS implementations store data in
the form of rows and columns (table). This approach favors
OLTP queries. The storage scheme of these relational
systems is based on the Flattened Storage Model (FSM) or
Normalized Storage Model (NSM). Both of these models
store the data as a consecutive byte sequence. The complete
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
table (rows and columns) is stored in a single database file
as a sequence of bytes, representing each row, with some
free space between two consecutive bytes (figure 1). The
space is left to accommodate future updates.
Table 3: Storage Scheme for Table 2
M101 A11 1000 M101 A12 500 ----
Now consider the following query:
SELECT * FROM MP WHERE M_ID = ‘M101’;
We observe that all data of one row is consecutive
and thus can be found in the same disk block, hence a
minimum of I/O activity (i.e. one block read/ write) is
necessary to execute a typical OLTP query.
Now consider the following query:
“Draw a histogram representing total quantities of all items
manufactured by different Manufacturers”
To answer the query the RDBMS needs to analyze all
rows of which only one column (QTY) is used. As a portion
of all rows is required, the RDBMS has to analyze the
complete table to execute such queries. It implies that a
heavy load in terms of I/O is generated.
In the large RDBMS applications like data warehousing
and Data Mining tables may contain several columns and
billions of rows. When an OLAP query is executed on such
tables it scans each row of the table and generates a heavy
load of I/Os. Only few columns from the table are used but
the process takes significant amount of time. Clearly, most
of the part of the table scan processes is a wasted effort. This
phenomenon is the main reason for the poor performance of
RDBMS products under heavy workloads.
Someone may argue that the problem of wasted efforts
done in the scanning of large database files may be solved
using some index structures like bit map, but OLAP and
Data Mining queries have low selectivity. The real benefits
of any index structure come out a in selection process
because an indexing technique provides a list of selected
record identifiers quickly. In OLAP and Data Mining
applications selection is not the end of the query. The
RDBMS has to perform some action on the selected tuples.
So, indexing is not the solution to this problem.
2.3 Trends in Hardware Technology
Figures 1 – 9 show the advancements in the Hardware
technology during the past decade. The size and bandwidth
of memory, size and bandwidth of hard disk, Network
Bandwidth and processing power of computer have
increased tremendously. Exceptions to these trends are I/O
efficiency and memory latency. These factors are affecting
the performance of all Software including the RDBMS. In
the following sections we shall consider each of them
individually and shall analyze its effect on RDBMS
performance.
 (IJCNS) International Journal of Computer and Network Security, 123
Vol. 2, No. 5, May 2010

CPU Speed Increments During Last Decade

Disk RPM Increments

4000
18000

3500
16000

3000
14000

2500
CP U Sp ee d (M h z )

12000

R o te tio n s P er M in u t e
2000

10000
1500

8000
1000

6000
500

0
4000
1995 1996 1997 1998 1999 2000 2001 2001 2002 2002 2003 2003 2004 2005
Years
2000

Figure 1. CPU Speed Increments 0

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
Years

Disk Bandwidth Increments

40
Figure 4. Disk RPM Increments
35

30
Disk Bandwidth (MB/s)

25 Disk Size Increments

180500
20

160500
15

140500
10

120500
5
Disk Size (MB)

100500

0
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
80500
Years

60500

Figure 2. Disk Bandwidth Increments 40500

20500

500
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
Years

Figure 5. Disk Size Increments

I/O Efficiency Improvements during Past decade

10

9
Memory Bandwidth Increments
800
8

700
7
Disk Latency (ns)

6 600

5
Memory Bandwidth

500

4
400

3
300

2
200

100
0
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
0
Years
1995 1996 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
Years

Figure 3. I/O Efficiency Improvements Figure 6. Memory Bandwidth Increments

124 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

Memory Latency Decrements 2.3.1 I/O Efficiency

160
From Fig 2 it is obvious that the I/O efficiency has not
changed much during the past decade. It means that the
time to access the data has stayed about the same. I/O
150
efficiency is an unglamours and often overlooked area of
RDBMS technology. In the vast majority of commercial
140 applications RDBMS performance is more dependent on I/O
M e m o r y L a t e n c y (n s )

than on any other computing resource, because the

130 performance cost of I/O outweighs other costs by orders of
magnitude. The most important item to consider is whether
120
the I/O subsystem of a given RDBMS will support sustained
performance as time passes. One of the most common
problems with databases is “Saw - Toothed” performance
110
(Figure 10) where, upon first install and after RDBMS
reorganizations, performance is excellent [3].
100
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
Years

Figure 7. Memory Latency Decrements

Memory Size Increments

3000
2900
2800
2700
2600
2500
2400
2300
2200
2100
2000
1900
1800
M e m o ry S ize ( M B )

1700
1600
1500
1400
1300
1200
1100
1000
900
800 Figure 10. Saw Tooth Performance of RDBMS
700
600
500
400 Performance will typically degrade over time due to
300
200 database fragmentations. Systems that fragment the database
100
0 do not support sustained performance because they are
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
Years
unable to perform following two tasks:
 Online reclamation of deleted space
Figure 8. Memory Size Increments  Online reorganization of data.
Benchmarks typically do not detect “Saw - Toothed”
performance because they are not typically run for a
sufficient amount of time to test the anti-fragmentation
Network Bandwidth Growth
techniques implemented in a given system.
400
Another aspect that also contributes towards low I/O
efficiency is the product’s ability to explicitly cluster data on
350 the disk with other data that would typically be accessed at
300
the same time. One solution to this problem may be to add
additional database volumes as the amount of data stored in
250 it grows over time. For better I/O efficiency it should be
B a n d w i d t h (M b p s )

possible to place those volumes on separate disks (RAID).

200
This way many queries can be executed in parallel on
150 different disks that improves performance despite low I/O
efficiency.
100
As the database grows larger, the potential for database “hot
50 spots” increases. A hot spot is a volume within a database
that is accessed with great regularity, while other volumes
0
1994 1996 1998 2000 2002 2004 2006 are accessed much less frequently. Hot spots can become a
Years
bottleneck to database access and thus affect performance.
Solution to this problem may be installing additional cache
Figure 9. Network Bandwidth Increments on the disk itself. This solution may bring performance
improvements if the I/O access pattern exhibits locality of
 (IJCNS) International Journal of Computer and Network Security, 125
reference and that is not the case with OLAP and Data
Mining queries.
2.3.2 Network Latency:
Figure 11. Database CPU Demand
Low I/O efficiency is not the only form of I/O one should be
concerned with. Network Latency is also equally important
particularly in a Client – Server environment. An RDBMS
should be able to move groups of objects (fields) to and from
the Client when it is appropriate for the application. To
optimally use Network Bandwidth, only objects that are
requested by the Client should be returned. But what the
RDBMS does is that, it passes all objects that reside together
on a physical disk block to the Client, whether they are
needed or not. This is really a serious problem with current
RDBMS architecture.
2.3.3 Disk and Memory Bandwidth:
From Figure 1 and 5 it is obvious that Disk and Memory
bandwidths have increased tremendously during the past
decade. In section 3.3.1 we have pointed out that I/O
efficiency can be improved by installing RAID devices, but
it may create another serious problem. It leads to poor
performance if the total bandwidth of all RAID disks
exceeds the memory bandwidth. As there is no way to scale
the memory bandwidth, it can become a hurdle in
installations with RAID and may contribute to poor RDBMS
performance.
2.3.4 Memory Latency:
Figure 7 shows the improvements in the Memory Latency.
Clearly we can store more (Figure 5) and process faster but
the time the memory takes to access the data has stayed
about the same. Memory latency is the most urgent reason
for low performance of RDBMS under heavy workloads.
The only solution to this problem provided by vendors has
been to install more cache memory. An application that has
a memory access pattern that achieves a low cache hit rate
can’t take the advantage of this solution. RDBMS
applications that have this property are hash join and sorting
[2], which are common operations in OLAP and Data
Mining Queries.
Vol. 2, No. 5, May 2010
2.3.5 CPU Utilization:
Most of the researches relating database performance to
processor have been carried out on multiprocessor platforms
using OLTP workloads [10] [13]. Few studies evaluate
database performance on
multiprocessor platforms using both types of loads (i.e.
OLTP and OLAP) [11]. All of the studies agree that the
RDBMS behavior depends upon the nature of workloads
(DSS or OLTP).
With the development of www and emergence of DSS
and CMS (Content Management Systems), RDBMS demand
for CPU has doubled every 9-12 months according to Greg’s
Law. At the same time processor speed has doubled every 18
months based on Moor’s Law. As shown in Figure 11, the
gap between RDBMS demand and CPU speed is increasing
[1].
Figure 12 shows CPU Utilization under OLTP workloads. A
response time versus CPU utilization curve can be obtained
using the following formula [6]:
RESPONSE_TIME = ∑ Response_TIMES =
((QUEUE_LENGTH * SERVICE_TIME) +
SERVICE_TIME)
Figure 13 shows CPU Utilization Percentage versus
response time. From figure 12 it is clear that at about 75%
utilization, the growth of the queue length shifts from linear
to asymptotic. The curve in the queue length graph also
exists in response time graph. This is why we never want to
run our CPUs in a steady state over 75%.
We are seeing techniques such as speculative pre-fetch and
multithreading in the processors but the RDBMS demand
for CPU is still increasing. The main reason is that modern
CPUs tend to be stalled for the majority of their time
achieving the low utilization of their true power [5]. These
CPU stalls are because of the memory latency. One solution
to get the most out of modern CPUs may be that the
performance intensive parts of program must contain only
processor dependent instructions. This is a hard requirement
and RDBMS applications can’t be designed using this
principle. RDBMS instructions are more memory or I/O
dependent, so RDBMS can’t take full advantage of modern
CPUs.
Queue Length Vs. Processor Utilization
16000
14000
12000
10000
Queue Length
8000
6000
4000
2000
0
0 20 40 60
Utilization (% Values)
80 100 120
Figure 12. CPU Utilization under OLTP workloads.
126
Response Time Vs. Processor Utilization
1.2
1
0.8
ime(Sec.)
nseT
0.6
Respo
0.4
0.2
0
0 10 20 30 40
CPU Utilization (%)
Figure 13. CPU Utilization Percentage versus response
time.
We are seeing techniques such as speculative pre-fetch and
multithreading in the processors but the RDBMS demand
for CPU is still increasing. The main reason is that modern
CPUs tend to be stalled for the majority of their time
achieving the low utilization of their true power [5]. These
CPU stalls are because of the memory latency. One solution
to get the most out of modern CPUs may be that the
performance intensive parts of program must contain only
processor dependent instructions. This is a hard requirement
and RDBMS applications can’t be designed using this
principle. RDBMS instructions are more memory or I/O
dependent, so RDBMS can’t take full advantage of modern
CPUs.
3. Inability to Support Other Data Models
In the RDBMS world the relational model and its query
language (SQL) are most popular. Every entity is expressed
in the form of table and the attributes of entity are the
columns of the table. Relations between tables (Entities) are
established by including the prime attribute (or a set of
attributes which defines the primary key) of one table into
another. This approach restricts the user to think in terms of
tables, primary key, foreign key and other related
terminologies. Clearly the thinking of an RDBMS database
designer is narrow when it comes on expressing the data.
Other data models like Object Oriented Model, Object –
Relational Model and Multidimensional Data Model are
becoming increasingly popular as they perform better under
heavy workloads generated by DSS queries. We argue that if
the RDBMS architecture is revised to support other data
models too, then performance issues may be resolved up to
some extent.
4. Conclusions and Future Work
Relational Databases are the backbones of several
Decision Support Systems and Content Management System
in use. Despite the performance optimization found in
today’s relational database systems, they are still not able to
take full advantage of many improvements in the computing
technology during the last decade. In this paper we have
examined various aspects which contribute to the poor
performance of relational databases when they support
OLAP, Data Mining and Decision Support Systems. Several
available solutions to address the performance issues and
why they are insufficient have been discussed at length
throughout the paper. We have also shown that expressing
the queries related to these applications is a tough task in
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
the native language of relational databases and that the
RDBMS tends to restrict the designer’s thinking around the
row-column concept which shows the inability of relational
databases to support any other data model.
The results from our analysis suggest that hardware
engineers must pay more attention to improve I/O, memory
and network efficiencies. Memory latency has been
identified as the most urgent reason contributing to the poor
performance of RDBMS under heavy workloads. This is
50
60 70 80 90 100
also responsible for long CPU stalls and can’t be cured by
simply installing more caches. SQL must be revised and
must be implemented to support other data models and
should not bind the designer’s thinking around rows
columns. Also it must be revised form the point of view of
expressing the OLAP, Data Mining and DSS queries in an
easy way. The RDBMS kernel must be rewritten to take the
full advantage of the hardware trends seen during the last
decade.
References
[1] X. CUI, “A Capacity Planning Study of DBMS With
OLAP Workloads”, Masters Thesis, School of
Computing, Queen’s University, Ontario, Canada
October 2003.
[2] S. Manegold, P. Boncz and M. Kersten “Optimizing
Join on Modern Hardware” IEEE TKDE, 14(4), July
2002.
[3] H. Zawawy “Capacity Planning for DBMS Using
Analytical Modeling”, Master Thesis, School of
Computing, Queen’s University, Ontario, Canada,
December 2001.
[4] S. Manegold, P. Boncz and M. Kersten “What Happens
During a Join? Dissecting CPU and Memory
Optimization Effects” In Proceedings of VLDB
Conference, Cairo, Egypt, July 2000.
[5] Anastassia Ailamaki, D.J. DeWitt, M.D. Hill, and D.A.
Wood “DBMSs on a modern processor: Where Does
Time Go?” In Proceedings of the 25th VLDB
Conference, Edinburg, Scotland, 1999.
[6] E. Whalen “Teach Yourself Oracle 8 in 21 Days”,
Macmillan Computer Publishing, 1998, USA.
[7] J. Vijavan “Capacity Planning More Vital Than Ever”,
Computer World, pp 1, February 1999.
[8] PC Quest Magazine www.pcquest.com.
[9] Rain Maker Group Whitepapers and Tools Library
available from www.rainmakerworks.com.
[10] K. Keeton, D.A. Patterson, Y.Q. He, R.C. Raphael and
W.E. Baker “Performance Characterization of a Quad
Pentium Pro SMP using OLTP Workloads.” In
Proceedings of the 25th International Symposium on
Computer Architecture, pages 15 – 26, Barcelona,
Spain, June 1998.
[11] P. Ranganathan, K. Gharachorloo, S. Adve and L.
Barroso “Performance of Database Workloads on
Shared Memory Systems with Out-of-Order
 (IJCNS) International Journal of Computer and Network Security, 127
Vol. 2, No. 5, May 2010

Processors.” In Proceeding of the 8th International
Conference on Architectural Support for Programming
Languages and Operating Systems, San Jose,
California, October 1998.
[12] D.A. Menasce, V.A.F. Almeida “Challenges in Scaling
e-Business Sites”, In Proceedings of Computer
Measurement Group Conference, 2000.
[13] L.Lo, L.A. Barroso, S.J. Eggers, K.Gharachorloo, H.M.
Levy and S.S. Parekh, “An Analysis of Database
Workload Performance on Simulated Multithreaded
Processors.” In Proceedings of the 25th International
Symposium on Computer Architecture, pages 39 – 50,
Barcelona, Spain, June1998.
Authors Profile
Zubair Khan received his Bachelor Degree in
Science and Master of Computer Application
Degree from MJP Rohilkhand University
Bareilly, India in 1996 and 2001 respectively
and also Master in Technology (M.Tech)
Degree in Computer Science and Engineering
from Uttar Pardesh Technical University in the
year 2008. He is currently pursuing his P.hd in
computer science and Information Technology from MJP
Rohilkhand University Bareilly, UP India. He also worked as a
senior lecturer in JAZAN University Kingdom of Saudi Arbia. He
is also servicing as Reader in the Department Of Computer
Science and Engineering Invertis Institute of Technology Bareilly,
India. His areas of interest include data mining and warehousing,
parallel systems and computer communication networks. He is an
author/ co-author of more than 15 international and national
publications in journals and conference proceedings.

Dr. Sami M. Halawani Received the M.S.

degree in Computer Science from the University
of Miami, USA, in 1987. He received the
Professional Applied Engineering Certificate
from The George Washington University, USA,
in 1992. He earned the Ph.D. degree in
Information Technology from the George Mason
University, USA in 1996. He is a faculty
member of the College of Computing and Information Technology,
King Abdul aziz University, Jeddah, Saudi Arabia. He is currently
working as the Dean for Graduate Studies and Research. He has
authored/ co-authored many publications in journals/conference
proceedings.

Dr. Ibrahim Albidewi is an Associate Professor

at the department of Information Systems,
Faculty of Computing and Information
Technology, King Abdualziz University, Jeddah.
He was the director of Computer Center at King
Abdulaziz University. His experience and
publications are in the area of image processing,
new vision of programming, expert system, computer education,
and management of information centers.
1986: Bachelor degree from Department of Islamic Economic,
College of Islamic Studies, University of Umm Al-Qura,
1989: Master degree from Department of Computer Science,
College of Sciences, swansea University,
1993: Doctorate degree from Electrical and Electronics Eng,
College of Engineering, Swansea University.

Jahangir Alam graduated in science from Meerut

University, Meerut in the year 1992 and received
Master degree in Computer Science and
Applications from Aligarh Muslim University,
Aligarh in the year 1995 and Master in Technology
(M.Tech) Degree in Computer Engineering from
JRN Rajasthan Vidyapeeth University, Udaipur, Rajasthan, in the
year 2008. He is currently working towards his Ph.D. in Computer
Engineering from Thapar University, Patiala. He is also serving as
an Assistant Professor in University Women’s Polytechnic, Faculty
of Engineering and Technology at Aligarh Muslim University,
Aligarh. His areas of interest include Interconnection Networks,
Scheduling and Load Balancing, Computer Communication
Networks and Databases. He has authored/ co-authored over 12
publications in journals/conference proceedings
128 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

Cluster Based Hierarchical Routing Protocol For

Wireless Sensor Network
Md. Golam Rashed1, M. Hasnat Kabir2, Muhammad Sajjadur Rahim3 , Shaikh Enayet Ullah4

1
Department of Electronic and Telecommunication Engineering (ETE),
University of Development Alternative (UODA), Bangladesh .

2,4
Department of Information and Communication Engineering
University of Rajshahi, Rajshahi-6205, Bangladesh.
3
School of Information Engineering, University of Padova, Italy.

Abstract: The efficient use of energy source in a sensor node is
most desirable criteria for prolong the life time of wireless
sensor network. In this paper, we propose a two layer
hierarchical routing protocol called Cluster Based Hierarchical
Routing Protocol (CBHRP). We introduce a new concept called
head-set, consists of one active cluster head and some other
associate cluster heads within a cluster. The head-set members
are responsible for control and management of the network.
Results show that this protocol reduces energy consumption
quite significantly and prolongs the life time of sensor network
as compared to LEACH.
Keywords: WSN, Cluster, Routing Protocols
1. Introduction
Wireless Sensor Networks (WSN) consists with huge tiny
sensors that are arranged in spatially distributed terrain. Due
to a large number of applications such as security,
agriculture, automation and monitoring, WSN has been
identified as one of the pioneer technique in 21st century.
The deployment and maintenance of the small sensor nodes
should be straightforward and scalable when they are
useless. But the small sensor nodes are usually inaccessible
to the user, and thus replacement of the energy source is not
feasible. There are some technical constrains also in
microsensor node such as low energy in battery and
processing speed etc. This limitation should be overcome as
much as possible to use the node in wireless sensor network.
The source of energy in sensor is a main constrain among
them [1]. It can be minimized by increasing the density of
energy in conventional energy sources [2],[3]. One of the
elegant ways to prolong the lifetime of the network is to be
reduced energy consumption in sensor nodes. This idea has
been applied many research [4],[5] to increase the life time
of the network. Cluster based Low-Energy Adaptive
Clustering Hierarchy (LEACH) is an attractive routing
protocol that disperses the energy load among the sensor
nodes [6],[7],[8]. The improvements of LEACH have been
presented elsewhere [9],[10],[11]. Proposed protocol is
another approach to improve of LEACH protocol by
reducing energy consumption and prolong the life time of
the network.
2. The Proposed Approach (CBHRP)
In this paper an optimum energy efficient cluster based
hierarchical routing protocol for wireless sensor network is
proposed, which is a two layer protocol where a number of
cluster cover the whole region. Proposed protocol introduces
a concept of head-set instead of a cluster head. At one time,
only one member of head-set is active and the remaining are
in sleep mode. Several states of a node are found in this
protocol such as- candidate state, non-candidate state,
active state, associate state, and passive associate state.
This protocol divides the network into a few real clusters
including an active cluster head and some associate cluster
heads. For a given number of data collecting nodes, the
head-set members are systematically adjusted to reduce the
energy consumption, which increases the network life.
3. Architecture of CBHRP
In the proposed model, the number of clusters k and nodes n
are pre-determined for the wireless sensor network. Iteration
consists of two stages: an election phase, and a data transfer
phase. At the beginning of election phase, a set of cluster
heads are chosen on random basis. These cluster heads send
a short range advertisement broadcast message. The sensor
nodes receive the advertisements and choose their cluster
heads based on the signal strength of the advertisement
messages. Each sensor node sends an acknowledgment
message to its cluster head. Moreover, in each iteration, the
cluster heads choose a set of associate heads based on the
signal strength of the acknowledgments. A head-set consists
of a cluster head and the associates. The head-set member is
responsible to send messages to the base station. Each data
transfer phase consists of several epochs. Each member of
head-set becomes a cluster head once during an epoch. A
round consists of several iterations. In one round, each
sensor node becomes a member of head-set for one time. All
the head-set members share the same time slot to transmit
their frames.
 (IJCNS) International Journal of Computer and Network Security, 129
Figure 1. Different states of a sensor node in CBHRP
protocol.
Different states of a sensor node in wireless sensor network
are shown in Figure 1. The damaged or malfunctioning
sensor states are not considered. Each sensor node joins the
network as a candidate. At the start of each iteration, a fixed
number of sensor nodes are chosen as cluster heads; these
chosen cluster heads acquire the active state. By the end of
election phase, a few nodes are selected as members of the
head-sets within a cluster; these nodes acquire associate state
where one of them is in active state and the remaining is in
associate state. In an epoch of a data transfer stage, the
active sensor node transmits a frame to the base station and
goes to the passive associate state. At that time the next
associate member acquires the active state. Therefore, during
an epoch, the head-set members are distributed as follows:
one member is in active state, a few members are in
associate state, and remaining are in passive associate state.
At the time of last frame transmission of an epoch, one
member is active and the remaining are passive associates;
there is no member in an associate state. For the start of next
epoch, one head-set member acquires active state and the
remaining are associate. By completing iteration, all the
head-set members acquire the non-candidate state. The
members in non-candidate state are not eligible to become a
member of a head-set. For starting a new round, all non-
candidate sensor nodes acquire candidate state.
4. Result and Discussion:
4.1 Energy consumption
The energy consumption for specific number of frames and
different head set-sizes with respect to the variation of
cluster number, network diameter size, and distance of the
base station from the network are examined.
Vol. 2, No. 5, May 2010
1 LEACH
0.8
He ad-set size =2
Einit(J)
0.6
0.4 He ad-se t size=4
0.2
0
60
40 200
150
20 100
50
0 0
# Of Clusters Network Diameter (m)
Figure 2. Consumed energy per round with respect to
number of cluster and distance of base station.
Figure 2. shows that the variation of energy consumption per
node with respect to the number of clusters and network
diameter. Figure shows that energy consumption is reduced
when the number of cluster are increased. It also shows that
the optimum range of cluster between 20 and 60. It is
interesting to point out that the number of head-sets reduces
the consumption of energy as compared to LEACH because
it reduces the election process.
Figure 3. illustrates the difference of the energy consumed
per round with respect to the head set size and network
diameter. It is seen that energy consumption is reduced
when the head set size is increased. It can be concluded from
above two figures that the head-set members play significant
role to reduce the energy consumption than that of LEACH.
Figure 3. Consumed energy consumed per round wrt
network diameter and head-set size when base station is at
fixed location.
4.2 Iteration time and frames
The estimated time for one iteration with respect to the
network diameter considering the percentage of head-set size
is shown in figure 4. The iteration time is proportional to the
initial energy E init and the network diameter found in this
figure. The network will be alive for a longest period of time
with initial energy when the head-set size is 50% of the
cluster size. However, it is more or less with respect to the
head-set size.
130 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

Data transmission in a wireless network is one of the most

important issues. It is a measuring tool of a network
suitability and perfectness. Our results show that the
proposed method is able to transmit higher data frame in
contrast to LECAH for the same dimension and distance of
BS. While, head-set size is responsible to increase the data
frame transmission. On the other hand, it is clearly found
that the transmission rate slightly decreases with respect to
the distance of BS. Figure 6 shows the number of frames
transmitted with respect to network diameter and distance
from BS.

Figure 4. Time for iteration with respect to network 5. Conclusion

diameter and head-set size.
An effective routing protocol is designed and tested it
performances to overcome some present limitation of WSN.
Introducing head-set concepts instead of only one cluster
head within a cluster, our results show the better
performance than that of LEACH in context of energy
consumption, frame transmition and the life time of the
sensor network.

Figure 5. Time for iteration with respect to number of
clusters and network diameter.
Figure 5 shows a graph that illustrates the estimated time for
one iteration with respect to the number of clusters and
network diameter. The number of cluster does not affect on
iteration time for a particular network diameter. However,
the head-set size plays a significant role to increase the
iteration time. In contrast to LEACH, iteration time is
increased by several times for a given number of head-set.
x 10
4 microsensor networks”, Proc. International Conference
5
Head set size=5
4 Head set size=3
# of Frames
References
[1] K. Pister, “On the limits and applications of MEMS
sensor networks”, Defense Science Study Group report,
Institute for Defense Analysis, Alexandria, VA.
[2] J. M. Rabaey, M. J. Ammer, J. L. da Silva, D. Patel, S.
Roundy, “PicoRadio supports ad hoc ultra-low power
wireless networking, IEEE Computer, July 2000, pp.
42-48.
[3] D. Timmermann, “Smart environments”, Invited Paper,
Ladenburger Diskurs, Ladenburg, January 2001.
[4] L. Zhong, R. Shah, C. Guo, J. Rabaey, “An ultra low
power and distributed access protocol for broadband
wireless sensor networks”, IEEE Broadband Wireless
Summit, Las Vegas, May 2001.
[5] V. Rodoplu, T. H. Meng, “Minimum energy mobile
wireless networks”, IEEE Jour. Selected Areas Comm.,
August 1999, pp. 1333-1344.
[6] W. Heinzelman, A. Sinha, A. Wang, A. Chandrakasan,
“Energyscalable algorithms and protocols for wireless
on Acoustics, Speech, and Signal Processing (ICASSP
'00), June 2000.
[7] W. Heinzelman, A. Chandrakasan, and H.Balakrishnan,

3
“Energy-efficient communication protocol for wireless
2 microsensor Networks”, Proceedings of the 33rd
1 International Conference on System Sciences (HICSS
0 ’00), January 2000.
200
150 [8] S. Lindsey, C. Raghavendra, K. Sivalingam, “Data
180
LEACH 140
130 gathering in sensor networks using the energy*delay
160 120
110 metric”, Proc. of the IPDPS Workshop on Issues in
140 100
Distance Diameter Wireless Networks and Mobile Computing, April 2001.
[9] E. B. Heinzelman, A. P. Chandrakasan, and H.
Figure 6. Number of frames transmission per iteration Balakrishnan, “An Application-Specific Protocol
with respect to distance of BS and network diameter. Architecture for Wireless Microsensor Networks”, IEEE
Transactions On Woreless Communications, Vol. 1, No.
4, October 2002.
 (IJCNS) International Journal of Computer and Network Security, 131
Vol. 2, No. 5, May 2010

[10] A. Manjeshwar and D.P. Agrawal, "APTEEN: A

Hybrid Protocol for Efficient Routing and
Comprehensive Information Retrieval in Wireless
Sensor Networks," in the Proceeding of the 2nd
International Workshop on Parallel and Distributed
Computing Issues in Wireless Networks and Mobile
Computing, Ft.Lauderdale, FL, April 2002.
[11] Wang Wei, “Study on Low Energy Grade Routing
Protocols of Wireless Sensor Networks”, Dissertation,
Hang Zhou, Zhe Jiang university,2006.

Authors Profile

Md. Golam Rashed received the B.Sc. and M.Sc. degree in

Information and Communication Engineering from University of
Rajshahi, Bangladesh in 2008 and 2009, respectively. He is
currently working as a Lecturer in the department of Electronic and
Telecommunication Engineering(ETE), University of Development
Alternative (UODA), Bangladesh .

M. Hasnat Kabir received the B.Sc. and M.Sc. degree in Applied

Physics and Electronics from University of Rajshahi, Bangladesh in
1995 and 1996, respectively. He has completed his PhD from
Kochi University of Technology, Japan in 2007. Now he is working
as an Assistant Professor in the dept. of Information and
Communication Engineering, University of Rajshahi, Bangladesh.

M. Sajjadur Rahim received his B.Sc. (Engg.) degree in

Electrical and Electronic Engineering from Bangladesh University
of Engineering and Technology, Bangladesh and Master of
Engineering degree in Information Science and Systems
Engineering from Ritsumeikan University, Japan in 1999 and
2006, respectively. He is working as an Assistant Professor in the
dept. of Information and Communication Engineering, University
of Rajshahi, Bangladesh. At present, he is pursuing his PhD degree
in the School of Information Engineering at the University of
Padova, Italy on study leave.
Shaikh Enayet Ullah received the B.Sc. and M.Sc. degree in
Applied Physics and Electronics from University of Rajshahi,
Bangladesh. He has completed his PhD from Jahangirnagor
University, Bangladesh. He is currently working as a Professor and
Chairman of the dept. of Information and Communication
Engineering, University of Rajshahi, Bangladesh.
132
Analysis of Subband Speech Enhancement
Technique for Digital Hearing Aids
D. Deepa1, Dr. A. Shanmugam2
Bannari Amman Institute o Technology, Sathyamangalam, Tamilnadu, India
1
deepa_dhanaskodi@yahoo.co.in, 2 dras@yahoo.co.in
Abstract: Hearing impairment is the number one chronic
disability affecting people in the world. Many people have great
difficulty in Understanding speech with background noise.
Speech Enhancement plays a vital role in such situations.
Random noise will disturb the signal between speaker and the
listener. Therefore the background noise has to be removed
from the noisy speech signal to increase the signal intelligibility
and to reduce the listener fatigue. The proposed approach is a
speech enhancement method based on the preprocessed sub
band spectral subtraction method, and the preprocessing is done
by using partial differential equation. This method provides a
greater degree of flexibility and control on the noise subtraction
levels that reduces artifacts in the enhanced speech, resulting in
improved speech quality and intelligibility.
Key words- Sub band Spectral Subtraction, Partial differential
equation, Adaptive noise estimation, Signal to Noise ratio, IS
distance.
1. Introduction
Understanding speech with background noise is difficult
even for normal hearing people, for the people with hearing
disability it is one of the major problem which is to be
reduced. This is especially true for a large number of elderly
people and the sensorineural impaired persons. Several
investigations on speech intelligibility have demonstrated
that subjects with sensorineural loss may need a 5-15 dB
higher signal-to-noise ratio than the normal hearing
subjects. Recent statistics of the hearing impaired patients
applying for a hearing aid reveal that 20% of the cases are
due to conductive losses, more than 50% are due to
sensorineural losses, and the rest 30% of the cases are of
mixed origin. Presenting speech to the hearing impaired in
an intelligible form remains a major challenge in hearing-
aid research today. In single channel system, speech
enhancement is a challenging on because reference noise
signal will not be available for enhancement. The clean
speech cannot be processed prior to being affected by the
noise. This is one of the most difficult situations in speech
enhancement for a single channel system. The conventional
power spectral subtraction method for single channel speech
enhancement substantially reduces the noise levels in the
noisy speech but this will introduce musical noise.
The proposed method is a preprocessed sub band
spectral subtraction method of speech enhancement and the
preprocessing technique is using partial differential
equation. In this method first the input noisy speech signal
is enhanced using PDE by taking the adjacent samples and
calculating the gradient, influencing coefficients and then
the output of this process is applied to the input for sub band
spectral subtraction method. The noise estimate in spectral
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
subtraction is updated by averaging the noise speech
spectrum using a time and frequency dependant smoothing
factor, which is adjusted based on signal presence
probability in sub bands. Signal presence is determined by
computing the ratio of the noisy speech power spectrum to
its local minimum, which is computed by averaging past
values of the noisy speech power spectra with a look-ahead
factor. This local minimum estimation algorithm adapts
very quickly to highly non stationary noise environments.
By using this noise estimation algorithm in this method
outperforms the standard power spectral subtraction method
resulting in superior speech quality and largely reduced
musical noise in single channel system for both stationary
and non stationary noise environments.
2. Partial Differential Equation Technique
Input noisy speech signal affected by additive background
noise can be enhanced by this method. First step in speech
enhancement using PDE is to obtain the gradient (g) of each
sample in noisy speech signal using the samples before and
after the current sample.
g f = S ( x − ∆x, t ) − S ( x, t )
--------- (2.1)
g b = S ( x + ∆x, t ) − S ( x, t )
Where, S(x,t) is the noisy speech signal, Δx is the
sampling rate. After the gradient is calculated the
influencing coefficients in each directions of the current
sample are computed.
1
IC f = 2
g 
1 +  f 
 k  ---------- (2.2)
1
IC b =
2
g 
1+  b 
 k 
Where, ICf is the Forward influencing coefficient and ICb
is the backward influencing coefficient. In the equation (2)
‘k’ is constant value between 1 and 100. From the above
calculated influencing coefficients and gradients the speech
signal is enhanced using
S ( x, t + ∆t ) = S ( x, t ) + ∆t ( g f IC f + g b ICb ) ------- (2.3)
In the above equations S(x,t) is the input noisy speech
signal, Δt is a coefficient between 0.1 t0 0.6 representing the
step of noise reduction in each iteration. The output of the
signal is again processed by applying into the algorithm of
 (IJCNS) International Journal of Computer and Network Security, 133
Sub band spectral subtraction method to reduce the non
stationary noise.
3. Sub band Spectral Subtraction method
Sub band Spectral subtraction method is the
frequency dependent processing of the Spectral Subtraction
procedure. This method offers better quality [9] of the
enhanced speech with reduced residual noise. This approach
has been justified due to variation in signal to noise ratio
across the speech spectrum. White Gaussian noise has a flat
spectrum, where as the real world noise is not flat. The
noise spectrum does not affect the speech signal uniformly
over the whole spectrum; some frequencies are affected
more adversely than others.
To take into account the fact that colored noise
affects the speech spectrum differently at various
frequencies, we use this approach to spectral subtraction.
The speech spectrum is divided into non-overlapping
bands, and spectral subtraction is performed independently
in each band. The estimate of the clean speech spectrum in
the ith band is obtained by:
2 2 2 updated, else the previous value is maintained as it is.
Ŝi (k) = Y(k) − α δ D̂i (k) ; b ≤ k ≤ e
i i i i
(3.1)
where bi and ei are the beginning and ending frequency
bins of the ith frequency band, a i is the over subtraction
factor and δi is the band subtraction factor. Over subtraction
factor provides a degree of control over the noise subtraction
level in each band, the use of multiple frequency bands and
the use of the δi weights provide an additional degree of
control within each band.
4. Noise estimation
Noise estimation plays an important role in this
work of speech enhancement. For an efficient noise
estimation algorithm the resultant signal estimation will
have great accuracy.
4.1. Adaptive noise-estimation algorithm
Several noise-estimation algorithms have been proposed for
speech enhancement applications [2] [3] [4] [5] [6]. The
main drawback of most of the noise estimation algorithms is
that they are either slow in tracking sudden increases of
noise power or that they are over estimating the noise
energy resulting in speech distortion.
In the adaptive method the smoothed power
spectrum of noisy speech is computed using the following
first-order recursive equation:
2 --------(4.1)
P (α , k ) = gP(α − 1, k) + (1 − g) Y(α ,k)
where P(α,k) is the smoothed power spectrum, α is the frame
index, k is the frequency index, |Y(α,K)| 2 is the short-time
power spectrum of noisy speech and g is a smoothing
constant. The proposed algorithm is summarized in the
following steps.
4.1.1 Speech present and speech absent frames.
In any speech sentence there are pauses between
words which do not contain any speech, those frames will
Vol. 2, No. 5, May 2010
contain only background noise. The noise estimate can be
updated by tracking those noise only frames. To identify
those frames, a simple procedure is used which calculates
the ratio of noisy speech power spectrum to the noise power
spectrum at 3 different frequency bands in each frame and
the sampling frequency respectively. If all the three ratios
are smaller than the threshold that frame is concluded as a
noise only frame, otherwise , if any one or all the ratios are
greater than threshold that frame is considered as speech
present frame.
The noise estimate is updated in speech absent frames
with a constant smoothing factor. In speech present frames
the noise is updated by tracking the local minimum of noisy
speech and the deciding speech presence in each frequency
bin separately using the ration of noisy speech power to its
local minimum.
4.1.2 The minimum of noisy speech
In our method for tracking the minimum of the
noisy speech by continuously averaging past spectral values.
In this algorithm if the value of the noisy speech spectrum
in the present frequency bin is greater than the minimum
value of previous frequency bin then the minimum value is
4.1.3. Detection of Speech-presence frames
The approach taken to determine speech presence
in each frequency bin is similar to the method used in [4].
Let the ratio of noisy speech power spectrum and its local
minimum be defined as
S (α , k) = P( α , k) P -------- (4.2)
min (α , k)
r
This ratio is compared with a frequency dependent
threshold, and if the ratio is found to be greater than the
threshold, it is taken as a speech-present frequency bin else
it is taken as a speech-absent frequency bin. This is based on
the principle that the power spectrum of noisy speech will be
nearly equal to its local minimum when speech is absent.
Hence the smaller the ratio is in (6), the higher the
probability that it will be a noise-only region and vice versa.
Note that in [4], a fixed threshold was used in place of
threshold.
From the above rule, the speech-presence
probability, p(α,K), is updated using the following first-
order recursion:
p(α,k) = a p(α -1,k) + (1- a ) I (α,k) ------- (4.3)
where a is a smoothing constant. Note that the above
recursion implicitly exploits the correlation for speech
presence in adjacent frames. This may result in slight
overestimate of the noise spectrum but will not likely have
much effect on the enhanced speech.
4.1.4. Calculation of Smoothing Constants
Using the above speech-presence probability
estimate, we compute the time–frequency dependent
smoothing factor as follows [4].
a(α,K) = d + ( 1- d) p(α,k) ------------ (4.4)
where d is a constant. Note that a(α,K) takes values in the
range of d ≤ a(α,K) ≤ 1.
4.1.5 Updation of noise spectrum estimate
Finally, after computing the frequency- dependent
smoothing factor a(α,k), the noise spectrum estimate is
updated as
134 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

D(α,k) = a(α,k) D(α -1,k) + (1- a(α,k) ) |Y(α,K)| 2 -----(4.5)

where D(α,k) is the estimate of the noise power spectrum.
Finally the noise spectrum estimate is updated using the
time–frequency dependent smoothing factor.
For each frequency band this estimated noise is Figure 1 b. Noisy Car Signal
subtracted from the input noisy speech signal, finally the
output of each band is combined by using OLA method to
get an estimate of clean speech (Enhanced speech).

5. Objective measures for performance Figure1 c. Enhanced Car Signal

evaluation Sample 2:
Objective measures are based on a mathematical
comparison of the original and processed speech signals. It
is desired that the objective measures be consistent with the
judgement of the human perception of speech [7]. However,
it has been seen that the correlation between the results Figure 2 a. Clean speech
obtained by objective measures are not highly correlated
with those obtained by subjective measures. The signal-to-
noise ratio (SNR) and the Itakura-Saito (IS) measure are
two of the most widely used objective measures.
5.1 Signal-to-noise ratio (SNR)
Figure2 b. White stationary Noisy speech
The SNR is a popular method to measure speech
quality. As the name suggests, it is a calculated as the ratio
of the signal to noise power in decibels. If the summation is
performed over the whole signal length, the operation is
called global SNR.
 2  Figure 2 c. Enhanced speech
 ∑ S (n) 
SNR = 10log  n  ----------(5.1)
[ ]
10  Frequency Domain Analysis:
dB 2
 ∑ S(n) − Ŝ (n)  Power Spectral Density of Reference Clean signal,
n  Noisy Signal and the Enhanced Signals were obtained and
5.2 Itakura - Saito Distance compared
PSD COM PARISON
0.7
NOISY SIGNAL
Itakura - Saito measure is one of the distance measures. 0.6
ENHANCED SIGNAL
data3

Lower the IS distance, better will be the quality of speech 0.5

(i.e. minimum phase difference between clean and enhanced

plitude

0.4
Am

signal). The average Itakura-Saito measure across all speech 0.3

frames of the given sentence will be computed to evaluate 0.2

0.1
the spectral noise subtraction algorithm.
0
0 20 40 60 80 100 120 140
Time

(a − b ) R (a − b )
r
d (a , b ) = ----------(5.2)
Figure 3. PSD plot of Cellular noisy, clean and Enhanced
a r Ra
signal
where a is the vector for the prediction Inference for Fig3: Power spectrum density of the enhanced
coefficients of the clean speech signal, vector R is the signal is close to the power spectrum magnitude of clean
autocorrelation matrix of the clean speech signal and vector signal.
b is the prediction coefficients of the enhanced signal. 0.35
PSD C omparis o n of White Statio nary N o is y s ignal

Nois y Signal
Clean Signal
0.3
Enhanc ed Signal

6. Experimental Results 0.25

0.2
PSD

Test samples are taken from SpEAR (Speech 0.15

Enhancement Assessment Resource) database of CSLU 0.1

(Center for Spoken Language Understanding). 0.05

Time domain Results : 0 20 40 60

Fre que nc y
80 100 120 140

Sample 1 Figure 4. PSD comparisons for White Stationary Noisy

signal
Inference: Figure 4 shows the Power spectral density
comparison for White stationary noisy signal. The output
shows that the stationary noise signal in higher frequency
Figure 1 a. Car Clean Signal
range (i.e. 40 to 130) is minimized
 (IJCNS) International Journal of Computer and Network Security, 135
Vol. 2, No. 5, May 2010

into account the non-uniform effect of non stationary noise

0.35
PSD COMP ARISON

C
OISY SIG
ENH
LEA
ANC E
NA L
D SIG
N SIGNA L
NAL
on the spectrum of speech. Signal to noise ratio will be more
than 6 to 10 dB in this method, and can be used in digital
0.3

0.25

0.2
hearing aids for sensorineural loss patients. The added
S
PD

0.15
computational complexity of the algorithm is minimal and it
0.1

adapts with non stationary noise environments. This

0.05

0
0 20 40 60 80 100 120 140
algorithm can be implemented in real-time on a fixed point
Digital Signal Processor (DSP) platform for evaluation in
Frequency

Figure 5. PSD plots of Colored factory Noisy, clean and real-world conditions. Methods can be developed to preserve
Enhanced signal the transitional regions and unvoiced regions, which contain
Inference for Fig 5: Initial Noise segments are very much low speech levels.
reduced in this method for colored factory noisy signal an
dthe spectrum is close to the clean signal. References
SNR O btaine d for different num be r of bands

50 [1] S. Boll, “Suppression of acoustic noise in speech using

40
30 spectral subtraction,” IEEE Transactions on Acoustics,
20
10 Speech, Signal Processing vol.27, pp. 113-120, Apr.
0
-10 1979.
Single band
2 bands
[2] Y. Ephraim and D. Malah, “Speech enhancement using
4 bands a minimum mean-square error short-term spectral
8 bands
amplitude estimator,” IEEE Transactions on Acoustics,
Figure 6. Improvement in SNR while Speech, Signal Processing vol.ASSP-32, No.6, pp.
increasing number of bands 1109-1121, Dec.1984.
SNR comparison with other methods [3] Cohen, I., 2003. Noise spectrum estimation in adverse
50

40
environments: improved minima controlled recursive
30 averaging. IEEE Transactions on Speech Audio
SNR

20

10
Processing. 11 (5), 466–475.
0
[4] Martin, R., 2001. Noise power spectral density
t)
l

l
l
l

l
a

a
a
a

l
a

na
a

al

e
n

n
n
n

kv
n
ig
ig

ig
g
g

ig
ig

ig
si
si

o
S

s
S

(y
y

-10
sy
y

)
y

y
is

le
is

is
is

estimation based on optimal smoothing and minimum

is

e1
is
oi
no

a
o

no

o
n
n

m
n
n

pl
n
db

b
b

am
b

b
d

b
d

6d

(f
0d

3d
0

d
0

s
ry

ry

is
ry
ry
ng

ng

d
r
a

la

statistics. IEEE Transactions on Speech Audio

a
na

e
n

on

Single band
ti

ti

rd
llu
o

rs

rs

tio

r
ti

co
e
bu
u
at

ti
a

C
a

4 bands
a
b
st

st

e
st

st
ite

ite

R
ite
ite

SS method
h

in

in
h

Processing (5), 504–512.(2), 1085–1099.

W

Typa of Noi se
DEKF Algorihm

Figure 7. SNR comparison with other methods [5] Rangachari, S., Loizou, P., Hu, Y., 2004. A noise
estimation algorithm with rapid adaptation for highly
Table 1: Comparison of IS distance measure for different nonstationary environments. Proceedings IEEE
noisy inputs with conventional approach International Conference on Acoustics, Speech Signal
IS Distance for IS distance for single Processing. I, 305–308.
Signal proposed method band SS method [6] Sohn, J., Kim, N., 1999. Statistical model-based voice
Cellular Noise 1.53 3.04
activity detection. IEEE Signal Processing. Lett. 6 (1), 1–
White burst noise 1.256 2.1722
3.
[7] S. Quackenbush, T. Barnwell, and M. Clements,
White stat noise 0.9978 2.067 “Objective Measures for Speech Quality Testing,”
Pink stat noise 0.791 1.0246 Prentice-Hall, 1988.
[8] J. Li and M. Akagi, “ A noise Reduction system
Car noise 0.1569 1.989
based on hybrid Noise estimation technique and post
Recorded speech 1.54 0 Filtering in arbitrary noise
environments”,Speech Communication, vol.48 pp 111-
Recorded speech 1.2564 0
126, 2006.
We obtained the SNR improvement when the number of
[9] D. Deepa and A. Shanmugam “Time And Frequency
bands is increased is shown in the Fig 6 for various noisy
Domain Analysis Of Subband Spectral Subtraction
signals, Comparison of SNR with other methods is give in
Method Of Speech Enhancement Using Adaptive Noise
Fig 7 and IS distance for the proposed method compared
Estimation Algorithm ” International Journal of Engg.
with conventional SS method in Table 1.
Research & Indu. Appls. (IJERIA)., ISSN 0974-1518,
Vol.2, No. VII (2009), pp 57-72
7. Conclusion [10] D. Deepa and A. Shanmugam “Spectral Subtraction
Method Of Speech Enhancement Using Adaptive Noise
The frequency dependent preprocessed sub band Estimation Algorithm with PDE method as a Pre
spectral subtraction method provides a definite improvement processing technique” ICTACT Journal on
over the conventional power spectral subtraction method Communication Technology (IJCT). ISSN 0976-0091,
and does not suffer from musical noise. The improvement Vol.1, Issue 1, March 2010, pp 1-6
can be attributed to the fact that the sub-band approach takes
136 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

An improved low voltage, low power class AB

operational transconductance amplifier for mobile
Applications
M.SanthanaLakshmi1 and Dr. P.T. Vanathi 2
1
PSG College of Technology,
Peelamedu, Coimbatore, India
ptvani@yahoo.com
2
PSG College of Technology,
Peelamedu, Coimbatore, India
santhanalakshmisekar@yahoo.co.in

Gm-C filters. The active RC filter [2] has limited DC gain

Abstract: This paper presents a 1.1V Class AB pseudo
differential operational transconductance amplifier (OTA) for
mobile applications. The circuit design consists of a Class AB
pseudo differential cascoded input stage which acts as a voltage
to current converter and a current multiplier to provide tuning
ability to the transconductor. Simulation results show that the
proposed OTA attains a high transconductance value of 580µS
and is used to realize a second- order Butterworth low pass
filter. All the simulations are carried out using TSMC 0.35µm
CMOS Technology in H-spice, Synopsis tool. The filter
implemented consumes very low power of 0.75mW with a tuning
range from 155 KHz to 3.9MHz. This makes it available for
various wireless applications like GSM, Bluetooth, and CDMA.
Keywords: operational transconductance amplifier, current
multiplier, low pass filter, transconductor.
1. Introduction
Low pass filter is an important component for direct
conversion receiver for mobile applications. Low pass
filter is used to select the required information under
desired channel band width. It acts between the variable
gain amplifier and the base band detection block [1] as
shown in Fig.1. A large tuning range of the low pass
filters would be required for various mobile applications
like European Global System for Mobile Communications
(GSM), Blue tooth, CDMA2000 and WCDMA.
I
VGA LPF
Base band
LNA
Detection
Q
VGA LPF
PLL 90 0
Figure 1. Basic diagram of the direct conversion receiver
Various topologies are presented in literature to implement
the low pass filter such as active RC filters, passive RLC
filters, MOSFET-C filters, Switched capacitor filters and
and bandwidth thus providing a hindrance to operate in
high frequency region. Passive RLC filters have high
potential to operate at high frequency range applications but
it needs large inductor sizes for integrated circuit design[3].
The switched capacitor filters are popular due to its accuracy
in frequency response, good linearity and dynamic range [4]
but requires sampling in time domain. Clock cycle needs to
be at least twice the maximum frequency of the signal to
prevent aliasing. Thus, its ability to operate at high
frequency is limited. MOSFET-C filter [5] is almost similar
to active-RC filter with the exception that the MOS
transistors are operating in triode region. The operating
frequency range of MOSFET-C filter is lower than Gm-C
filter because of the Miller-integrator and op-amp structure
in MOSFET-C filter. So, among all these filter topologies,
Gm-C filter topology has high potential to operate at high
frequency ranges, good linearity and good controllability.
Operational Transconductance Amplifier (OTA) is an
important building block in the design of Gm-C filter
topology [6]. The design of OTA requires an input stage to
perform voltage to current conversion (V-I). Here, a Class
AB pseudo differential input stage is preferred as it
consumes very low power and can be applied to very high
speed switched capacitor applications. Conventional class
AB structures have high power dissipation due to large
parasitic capacitance. So, a cascoded flipped voltage
follower (CASFVF) is used in the Pseudo-Differential pair
since it excels in its performance in terms of signal swing,
slew rate, bandwidth, power dissipation and the lowest
output impedance [7]. The output from a V-I converter is
given to a current multiplier which linearise the output and
provide a wide tuning range. Thus the proposed
transconductor consists of a Class AB pseudo differential
with cascoded FVF as voltage to current converter and a
current multiplier.
In this paper, the design of proposed OTA is discussed in
Section 2. Section 3 explains the implementation of a second
order Butterworth low pass filter with the proposed OTA.
Performance comparison of different filters and their
simulation results are reported in Section 4. Conclusions are
summarised in Section 5.
 (IJCNS) International Journal of Computer and Network Security, 137
Vol. 2, No. 5, May 2010

2. Proposed operational transconductance

amplifier
An Operational Transconductance Amplifier (OTA) is
Ib
the basic analog building block of Gm-C topology. It
B
provides high open loop DC gain, large gain-bandwidth
product (GBW) and high slew rate and consists of an
Vin Vb
M1 M3
input stage, current multiplier and common mode feed
back (CMFB) circuit. The conventional OTA structure Vout

is shown in Fig. 2[1]. M2

Id
VDD

M7 M14 Figure 3. Cascoded Flipped Voltage Follower (CASFVF)

M8 M13 MF9 MF10

Itune By varying the current source Id, the quiescent voltage at

M18 M17 VB1
the drain of M1 is set to close to Vdd. So the CASFVF
M15
M16 M11 M12 MF7
MF8 provides maximal signal swings and very low output
Vo-
impedance compared with FVF. A class AB pseudo
Ib differential input stage employing CASFVF structure is
Vo+ shown in Fig.4.
M3 M4

M9 M10 MF3 MF5

MF6
Vcm MF4
Vi+ M6 Vi-
M1 M2 Vref

M19 MF11
MF1 MF2

M5

Figure 2. Conventional Operational Transconductance

Amplifier [1]

The conventional structure also utilizes a class AB pseudo

differential stage along with a flipped voltage follower
circuit but has limited voltage swing, low slew rate and
larger power dissipation compared to Class AB structure
with a cascoded FVF. Also a wide bandwidth is obtained
with low supply voltage using a cascoded FVF circuit which
makes it suitable for low voltage, low power applications
[7]. A high linearity current multiplier is used to linearise
the current from a V-I converter to attain a better noise
performance. A switched capacitor common mode feedback
circuit (CMFB) is used in order to maintain the stability of
the pseudo differential structure.
2.1 Class AB Pseudo Differential Input Stage
The cascoded Flipped Voltage Follower (CASFVF) is
shown in Fig 3. In this circuit, a PMOS cascoding transistor
M3 is connected between the gate of M2 and the drain of M1 .
This transistor provides additional gain to the negative
feedback loop and leads to an extremely low output
resistance [8]. The output resistance is given by equation
(1).
Figure 4. Voltage-to-Current (V-I) converter
The MOS transistors M3, M4 and M5 and the current
source Ib form a Cascoded Flipped Voltage Follower
(CASFVF). The CASFVF provides very low impedance at
the node X and the voltage at X approximately equal to Vcm-
VGS3 . The transistors M1 and M2 are source coupled and are
operated in saturation region. So this pair resembles the
working of a “Pseudo-differential” structure [8]. When a
large balanced differential input voltage is applied, the
output current variations will be large at node X. This
quiescent current is controlled by Ib and the transistor M4
suppresses the variations at the node X. So, the circuit
operates in a class-AB fashion. Thus this structure which
employs a CASFVF at the input stage performs a voltage
to current conversion.
2.2. Current Multiplier
Current multiplier are broadly classified into current
mode multipliers and voltage mode multipliers In order to
achieve wider band width, greater linearity, wider dynamic
range and simple circuitry, current mode multipliers are
138 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

preferred as compared to voltage-mode multipliers [9]. So, Also M11 and M12 transistors in the conventional OTA are
current multipliers are best suited as a programming replaced with two diode connected transistors M13 and M14
element in systems such as filters, neural networks, mixers as in Fig.6, to linearise the output current and to reduce the
and modulators. The linearity multiplier tries to linearise the noise effect in the transconductor.
VDD
output from the V-I Converter and provides a wide tuning
VCMF
range. The conventional current multiplier [1] is as shown M9 M10
M15 M16
in Fig. 5.
VDD
M7 M8
I2
I2 M13 M14
M7 M8 M5 M6

Vb1
Vop Von
I1+ii I1- ii
IO IO
M11 M12
I1 + ii I1 - ii
Voltage to current
i0 i0
converter Itune
M9 M10
vss
Voltage-to-Current
Converter
2I2 Figure 6. Linearity current multiplier

Figure 5. Conventional Current Multiplier 2.3. Common Mode Feed Back Circuit

The transistors in the multiplier work in linear region to The switched capacitor circuits have become extremely
enhance the linearity performance of the OTA. The current popular because they accurate frequency response, good
equation for a MOSFET operating in the linear region is linearity and dynamic range. In switched capacitor circuits
given by equation (2) the accurate discrete time frequency responses are obtained
due to filter coefficients and are determined by capacitance
ratios. This capacitance ratios can be set quite precisely (in
ordre od 0.1%) in integrated circuits. This accuracy is much
better than that which occurs for integrated RC time
Where W and L are the width and the length of the constants (on order of 20%). The basic building blocks in
transistor, respectively, is the reverse saturation current, switched capacitor circuits are
n is the sub threshold slope factor, and is the thermal • Capacitors.
voltage. From the V-I conversion circuit [1], it is found that • Switches.
(3) • Non over lapping clocks.
Therefore
= The common mode fedd back (CMFB) circuit using
= Switched capacitor technique [10] is shown in Fig. 7. The
switched capacitor CMFB circuits are area efficiant and
does not need an additional amplifier.
In equation (4) n is the sub threshold slope factor and is V ref

equal to where and are the gate

P2
and depletion capacitance per unit area respectively. The P1 P2 P1
output current equation for the differential pair of transistors
V op V on
M9 and M10 is as in equation (5) and (6). MF1 MF2 MF5 MF6

Itune
= ) CS CS
Ca Ca

VCMF
= MF3 MF4 MF7 MF8 MF9

VSS

Thus the output current is given by equation (7)

Figure 7. Switched capacitor CMFB.

The switched capacitor CMFB uses capacitive voltage

The conventional current multiplier is modified by adding division to control the output common mode voltage.
two diode connected transistors M9 and M10 to increase the The circuit consists of two capacitors Ca and CS to
output impedance and enhance the performance of the filter. average the output voltages and it adds the bias voltage
 (IJCNS) International Journal of Computer and Network Security, 139
Vol. 2, No. 5, May 2010

to control the output common mode voltage. Vop and
Von from the transconductor circuit is given as inputs to
CMFB circuit. Suppose if the common mode level of
the outputs Vop and Von increases, then the Ca capacitor
average voltage increases to control the output common
mode level. P1 and P2 are two non overlapping signals
applied to the MOS transistors in the CMFB circuit
which act as switches. By varying the Itune, the tuning
range of the OTA is varied to accommodate wide
frequency range in the low pass filter. The proposed
OTA circuit is shown in Fig. 8 and its circuit
parameters are listed in Table 1.
VDD
modevoltage is not equal to the reference voltage, a
corrected current would be mirrored by transistor MF9 to the
load transconductor, and then the output common mode
voltege is adjusted to the desired voltage. Itune is used to
provide wide tuning trange to the filter.
3. Filter Architecture
The second-order Butterworth low-pass Gm-C filter used
by Uwe Stehr et al. [11] is shown in Fig. 9. The low pass
filter structure consists of four operational transconductance
amplifiers gm1, gm2, gm3 and gm4.

VCMF 2*C 1
2*C 2
M9 M10
Vref
M15 M16 out-

M7 M8
p1 p2 p2 p1
vop von
in-
+ + + +
gm1 gm2 gm3 gm4
M6 M13 MF1 MF2 MF5 MF6 + + +
M5 M14 I in+
CS Ca Ca CS tun
Vb1 e
p2 p2 p1
p1
VCMF
vop von out+
MF3 MF4 MF7 MF8
2*C1 2*C2

Ib
MF9
M11 M12
Vcm Figure. 9 Gm- C realization of 2nd order low-pass Filter
Vi+
M4 Vi-
M1 M2
M17 The transfer function of the above filter structure [11] is
VB given as equation (8)

H(s)= (8)
M3
So, the transfer function of the 2nd order Butterworth filter is
VSS

H(s)= (9)
Figure 8. Complete OTA circuit
Recognizing the common transfer function of a 2nd order
Table 1 Circuit parameters of the Proposed OTA low-pass filter as
H(s) = (10)
Components Value
(W/L)1,2 250µm/0.35µm
And comparing (8) and (10)
(W/L)3,17 200µm/0.35µm
(11)
(W/L)4 1µm/0.35µm
(W/L)5,6,7,8 40µm/0.35µm
Maintaining and
(W/L)9,10 50µm/0.35µm
leads to (12)
(W/L)11,12 20µm/0.35µm

(W/L)13,14,15,16 15µm/0.35µm Where (13)

Ib 2µA
Itune 10µA Now the transconductance of the 2nd order filter from
(8) and (9) is

(14)
The ouput’s of the transconductor Vop and Von are
compared with reference voltage Vref. If the output common The transconductance gm1, gm3 and gm4 of the filter are kept
equal to gm and transconductance gm2 is kept 1.848 times of
140 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

gm to get the 2nd order Butterworth low pass filter response.

This low pass filter is used in a direct conversion receiver to
receive the mobile standards.

4. Simulation Results

4.1. Operational transconductance amplifier

Simulations of the proposed OTA and the filter are
carried out with TSMC 0.35µm CMOS technology using H-
Spice in synopsis. The implemented OTA and its open- loop
AC frequency response are as shown in Fig. 10 (a) and (b)
respectively.

Figure 11. Step transient response of the OTA

The variation in transconductance with reference to

frequency is shown in Fig. 12. Simulation shows that the
filter attains a transconductance value of 580µS.

Figure 10. (a) H-Spice Schematic of Implemented OTA

circuit

Figure 12. Transconductance curve of the OTA

Figure 13. (a) Implemented 2nd order

Figure 10. (b) Frequency response of OTA

The step response of the OTA measures the slew rate of the
filter to be 130 V/µs and is shown in Fig. 11.
4.2. 2nd order Butterworth low pass filter
The implemented circuit of 2nd order Butterworth low
pass filter and its unity gain frequency response are shown
in Fig. 13 (a) and (b) respectively. The maximum cut-off
frequency of the filter is 3.9 MHz.
 (IJCNS) International Journal of Computer and Network Security, 141
Vol. 2, No. 5, May 2010

The filter proposed by Uwe Stehr et al. [11] utilizes

conventional folded cascoded OTA for realizing the second
order filter. This filter consumes large amount of power with
poor noise performance though it has got a good slew rate.
Also the OTA attains a very low transconductance value. In
[10], a new class AB structure is introduced to increase the
slewing behavior and transconductance value f OTA but
with a high power consumption. In the transconductor
proposed by Tien-Yu Lo et al.[1], a class AB structure
utilizing FVF circuit act as a voltage to current converter
along with a high linearity multiplier to minimize the third
order harmonic distortion. This there forth reduces the
power consumption of the filter but with poor the slew rate
Figure 13. (b) Frequency response of the filter and transconductance value. So, in this paper an improved
Butterworth Low Pass Filter circuit class AB structure which employs a cascoded FVF circuit to
increase the slew rate and transconductance value of OTA is
A performance comparison of the 2nd order Butterworth low presented. The table illustrates that the filter with the
pass filter based on proposed OTA with other filters proposed OTA outperforms the other filters in terms of low
presented in literature recently is tabulated in Table 2. supply voltage, low power, reduced noise and wide tuning
range.
Table 2 Performance Comparison Table
5. Conclusion
Uwe Wongnamk Tien-Yu Proposed
An improved operational transconductance amplifier is
Parameter Stehr et um et al. proposed with a class AB pseudo differential structure with
Lo et al. Work cascoded FVF. A 2nd order butter worth low pass filter used
al. [11] [10] in a direct conversion receiver is implemented using
[1]
0.35µm, CMOS process with under 1.1V supply voltage.
Simulation results show that the OTA attains a maximum
CMOS- 0.35-µm 0.35-µm 0.35-µm 0.35-µm transconductance value of 580µS as compared to other
Tech transconductor. Also the power consumption of the filter is
0.75mW at its maximum cutoff frequency. There forth the
Supply 2.5-V 1.5-V 2.5-V 1.1-V filter is suitable for low voltage, low power analog
voltage processing units.

MOS 120 92 120 108 References

transistors
[1] Tien-Yu Lo and Chung-Chih Hung (2008), “Multi
Power 500-mW 172-mW 1.9-mW 0.75-mW
mode Gm-C channel selections filter for mobile
applications in 1-V supply voltage”, IEEE transactions
on circuits and systems-II: express briefs, Vol.55.no.4,
-35dB -34dB -34dB -41dB April.2008.
(0.6VPP (0.4VPP@ (0.4VPP@ (0.4VPP@
[2] J. Lim, Y. Jung and K. Jung, “A Wide-Band Active RC
THD @ 2.5MHz) 2.5MHz) 2.5MHz)
filter with a fast tuning scheme for wireless
2.5MHz)
communication receivers”, Proc. IEEE CICC, Sep 2005
-39dB -41dB -50dB -47dB , pp. 637-640.
[3] A. C. M. de Queiroz and L. P. Caloba, “Passive
HD3 (0.6VPP (0.4VPP @ (0.4VPP @ (0.4VPP symmetrical RLC filters suitable for active
@ @2.5MH simulation”, IEEE International Symposium on
2.5MHz) 2.5MHz) z) Circuits and Systems, vol.3, Jun 1988, pp.2411-2414.
2.5MHz) [4] B. Manai and P. Loumeau,“A Wide-band CMOS
Switched Capacitor for UMTS direct conversion
Slew rate 216 V/µs 185 V/µs 13 V/µs 130 V/µs receiver”, Proc. IEEE MWSCAS, vol.1, 2001, pp.377-
of OTA 380.
[5] Y. Tsividis, M. Banu and J. Khoury, “Continuous-Time
Gm 50µS 330µS 25µS 580µS MOSFET-C filter in VLSI”, IEEE Trans. Circ. Syst.-II,
Vol. CAS-33, No.2, 1886, pp.125–140.
Tuning 200KHz- 50KHz- 120KHz- 155KHz- [6] Sanchez-Sinencio, E, and Silva-Martinez J, “CMOS
range 2.5MHz 5.7MHz 2.25MHz transconductance amplifiers, architectures and active
3.9.MHz filters: a tutorial,” IEEE Proc. On Circuits, Devices
and Systems, vol.147, no.1, Feb 2000, pp.3-12.
142 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

[7] Ramon Gonzalez Carvaja, Jaime Ramirez-Angulol, Authors Profile

Antonio J.Lopez-Martin, Antonio Torralba , Juan
Antonio Gomez Galan, Alfonso Carlosena and
Fernando Chavero, “The Flipped Voltage Follower : A
Useful Cell for Low-Voltage Low-Power Circuit M.santhanalakshmi is currently working as
Design”, IEEE transactions on circuits and systems-I: Senior Lecturer in Electronics and
Regular papers, Vol.52.no.7, July 2005. Communication Department, PSG College of
[8] J. Ramirez-Angulo, S. G. Ivan Padilla, R. G. Carvajal, Technology, Coimbatore, India. She has
A. Torralba, M. Jimenez, F. Munoz and Antonio about 9 years of teaching experience. Her
Lopez-Martin, “Comparison of Conventional and New research areas are Low power VLSI Design,
Flipped Voltage Structures With Increased Analog VLSI Design and Mixed Signal VLSI
Design.
Input/output Signal Swing and Current
Sourcing/Sinking Capabilities”, IEEE International
Midwest Symposium on Circuits and Systems, Aug.
2005, pp. 1151-1154.
[9] Vlademir J.S Oliveira and Nobuo oki, “Low Voltage Dr.P.T.Vanathi is currently working as an
Four-Quadrant Current Multiplier: An improved Assistant Professor in Electronics and
Topology for n-well CMOS process”, IEEE, Trans. Communication department, PSG College of
Design & Technology of Integrated Systems in Technology, Coimbatore India. She has about
Nanoscale Era, DTIS, Sept. 2007, pp.52-55. 20 years of teaching experience. Her research
areas are Speech Signal Processing, Image
[10] S. Wongnamkum and A. Thanachayanont , “New Class-
Processing and VLSI Design.
AB Operational Transconductance Amplifier for High-
Speed Switched-Capacitor Circuits”, IEEE
International Symposium on Communications and
Information Technology, vol.1, Oct 2004, pp. 531-
535.
[11] Uwe Stehr, Frank Henkel, Lutz Dallige and Peter
Waldow (2003), “A Fully Differential CMOS integrated
4th order reconfigurable GM-C Low Pass filter for
mobile communication”, Proc. IEEE ICECS, Vol.1, Dec
2003, pp. 144-147.
 (IJCNS) International Journal of Computer and Network Security, 143
Vol. 2, No. 5, May 2010

Graph Matching Measure for Document Retrieval

Amina Ghardallou Lasmar1, Anis Kricha2 and Najoua Essoukri Ben Amara3
1
Higher Institute of Technological Studies of Sousse,
BP 135, Sousse Erriadh, 4023 Sousse Tunisia
Aminaghardallou@yahoo.fr
2
National Engineering School of Monastir
Avenue Ibn ElJazzar, 5019 Monastir, Tunisia
Anis.Kricha@topnet.tn
3
National Engineering School of Sousse
Avenue 18 janvier Babjdid Sousse 4000, Tunisia
Najoua.Benamara@eniso.rnu.tn

segmentation [5], [7], document retrieval by content [10]

Abstract: In this paper, we propose an algorithm for graph
matching in a pattern recognition context. Graphs used for
mapping are formulated in terms of Attributed Relational Graph
(ARG) to represent document structure. ARG nodes represent
regions composing the structure and edges define different
relations between them. A graph matching algorithm is
presented to compare different document structure, based on
clique finding in an association graph. The originality of this
algorithm resides on the way of defining an association graph,
which allows us to define a similarity coefficient. The
coefficient measures the resemblance degree of two comparing
structures involving the best matching.
Keywords: graph matching, structural representation, graph
retrieval.
1. Introduction
In document retrieval applications, it is essential to define
some description of the document based on a set of features.
These descriptions are then used to search and determine
which documents satisfy the query selection criteria. The
effectiveness of a document retrieval system ultimately
depends on the type of representation used to describe a
document. In pattern recognition, the document
representation can be broadly divided into statistical and
structural methods [6]. In the former, the document is
represented by a feature vector; and in the latter, a data
structure (e.g. graphs or trees) is used to describe objects and
their relationships in the document. In the last decades,
many structural approaches have been developed which deal
especially with graph-based representations of graphic
document. These approaches have different purposes like
and pattern recognition [2], [6].
ARG has become commonly used as an adequate
representation for documents owing to their representational
power. According to this representation, the nodes represent
entities to be matched and edges define relations operating
between them. If the structural descriptions of document
images are represented by graphs, different documents can
be matched by performing some kind of graph matching.
Graph matching is the process of finding a correspondence
between nodes and edges of two graphs that satisfies some
constraints ensuring that similar substructures of one graph
are mapped to substructures of the other. Two drawbacks
can be stated for the use of graph matching in computer
vision. The first is its computational complexity which can
be solved by using approximate algorithms. The second
drawback is dealing with noise and distortion. Encoding
entity from a document by an ARG may not be perfect due
to noise and errors introduced in low-level stages. In such
situations, graphs will be distorted by different attribute
values, missing or added vertices or edges, etc. This fact
makes exact graph matching useless in many applications.
The matching must incorporate an error model able to
identify the distortion results in distorted graphs. A
matching between two graphs involving error models is
referred to as inexact graph matching. Several techniques
have been put forward to solve the graph matching problem.
The main approaches are summarized in table 1 with the
most representative references in the field of computer
vision.

Table 1: Comparison between classical graph matching

Exact matching Inexact Optimal Key references
matching
Backtrack tree search Yes No Yes [17], [1]
Decision tree Yes No Yes [11], [15]
Graph edition Yes Yes Yes [10], [3], [12], [14]
Association graph Yes No Yes [13]

In this paper, we address the problem of indexation of

historical Arabic document, which aims to represent the
structure of a document by a simple model and allows a
144
further process like document or object retrieval. Therefore,
we propose an ARG-based representation for the document
physical structure. A graph matching method is then
developed based on clique finding in an association graph to
compare document structures. Results are then ranked
according to coefficients which convey the degree of
similarity between a query document and database
documents.
The remainder of the paper is organized as follows: In the
next section, we propose the graph matching algorithm and
we also introduce a similarity coefficient between two
graphs. Experimentation is discussed in section 3. The final
part concerns the conclusion and future works.
2. Proposed approach
Even thought we are not interested in the segmentation step,
we should consider error introduced by this step knowing
that we operate with ancient documents, which includes
much degradation. Thus, we will adapt an exact graph
matching algorithm to our application to overcome
distortion and errors introduced in precedent stages. The
developed approach is based on a maximal clique finding in
a modified association graph. First, we define a graph-based
structure representation; the document structure is organized
in an ARG whose nodes represent the different regions
composing the document and the graph edges defining the
spatial relations between these regions. The second step
consists of constructing the association graph which is
modified compared to the proposed definition in literature
[8], [15] and adapted to our application. Finally, we define
a similarity coefficient by examining the characteristics of
the maximal clique in the AG.
2.1 Definition of the ARG
ARG is a powerful means of giving structural
descriptions of visual patterns. Within an ARG, we can
distinguish a syntactic part, made of the layout of unlabeled
nodes and edges respectively identifying the structural
primitive components of the pattern and their relations, and
a semantic part consisting of the attributes associated to
nodes and edges.
Formally an ARG graph is defined as a
quadruple G = (V, E , LV , LE ) , where V is the set of vertices,
E ⊆ V × V is the set of edges, and LV and LE are two labeling
functions associating attributes to vertices and edges
respectively.
A document and its corresponding ARG structure are
illustrated in Figure 1. In this figure, we denote the node
labels (text or graphics), and near the connecting lines we
denote edge labels (Upper; on the Right; Included; and
Upper Right). We have excluded other spatial relations to
avoid redundancy since the existence of a relation in a
direction implies that it is complementary in the other one.
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
I1 (type: graphic)
R:in the right
R
I:included
I2 I3 R
I2
I1 (type: graphic)
I
I3 (type:text)
Figure 1. A document image and the corresponding ARG
representation
2.2 A modified association graph for structure
retrieval of the ARG
The proposed algorithm for graph matching is based on
clique finding in an association graph, which is one of the
most popular algorithms that can achieve an exact
matching.
For the two graphs G1 = (V1, E1, LV1 , LE1 )
and G 2 = (V 2, E 2, LV 2 , LE 2 ) , the association graph is a
vertex labeled, undirected graph which can be created by the
two step process [8], [15]:
1. For each correct vertex mapping from graph G1 to G2,
insert a vertex in the association graph. This vertex is
labeled with the vertex mapping between the vertices of G1
and G2.
2. For each pair of vertices va and wa in the association
graph, insert the edge (va ,wa) if the vertices mapped from
G1 have the same edge characteristics as the vertices
mapped from G2.
Our approach is based on the construction of an
association graph but with modification of the second step.
We will insert edges not only when we retrieve all
characteristics from vertices mapped from G1 and G2 but
also when some characteristics are existing. In this way, the
edges of the association graph will be level-headed with
weights.
We note (I1, I2) tow vertices from G1and (J1, J2) from
G2, where (I1, J1) and (I2, J2) present two correct mappings
from graph G1 to G2. These weights are affected to AG
edges as follows:
(va , wa ) = ((I1, J 1),(I2 , J 2 ))
0 if (I1 = I 2 ) or (J 1 = J 2 )

1 if LE1(I1, I 2 ) = LE2 (J1, J 2 ) (1)

= 0.2 if LE1(I1, I2 ) ≠ LE2 (J1, J 2 )
2 if (L (I , I ) ⊂ {U, R} and L (J , J ) = {UR})
3 E1 1 2 E2 1 2

 or inversly
In Figure 2, we illustrate an example of the way that we
create an association graph from two graphs G1 (figure 2.a)
and G2 (figure 2.b), candidates for comparison. Vertices of
the AG are created from tow correspondent vertices from G1
and G2 (tow vertices are correspondent when they have
similar attributes). Then AG edges are introduced between
tow vertices with different weights (figure 2.c).
Let us remember that a clique is a complete sub-graph so
that every pair of vertices is joined by an edge. Each clique
within the association graph represents a set of vertices
which have the same mutual relationships in G1 and G2;
i.e., the cliques represent sub-graphs common to G1 and G2.
In order to determine the number of possible common sub-
graphs between two graphs, it is sufficient to examine the
 (IJCNS) International Journal of Computer and Network Security, 145
Vol. 2, No. 5, May 2010

characteristics of cliques in an unlabeled, undirected graph Figure 2.d illustrates the maximal clique to a given
of the appropriate size. In literature, we find several association graph. If AG contains more than one clique, the
algorithms for clique finding like Messmer method [11] and maximal one is the one having maximal weights.
Belaid algorithm [4]. We have implemented for this goal an
optimal algorithm based on Messmer method.
I1 J1 (I2 ,J 1 )
Gr Gr (I1 ,J 1 ) (I2 ,J 1 )
R I2

Gr
R I
0 .2 1 1
I

T xt T xt
J2
(I3 ,J 2 )
I3 ( I3 ,J 2 )
(a ) G ra p h G 1 (b ) G r a p h G 2 (c ) A s s o c ia tio n G r a p h (d ) M a x im a l C liq u e

Figure 2. Illustration of the graph matching: (a) Graph G1, (b) Graph G2, (c) Association Graph, (d) Maximal clique

structures. Then, we experiment the proposed algorithm on

2.3 Similarity coefficient
It is intuitive to denote that the criteria that can
characterize the degree of resemblance of tow objects is
calculated, generally from the worth of common features in
ratio of all characteristics of two objects [9], [16]. For this
fact, we propose a measure similarity based on the clique
weight and the characteristic of the two candidates G1 and
G2.
We define description of a graph as the number of
vertices and edges like the number of regions and relations
involved in the structure of a document. This definition is
the same for a clique with a substitution number for the
weight of edges.
descr (C max ) − splits
sim(G1, G 2) =
descr (G1) + descr (G 2) − descr (G1 ∩ G 2)
where splits represent the penalty for the over matching of
vertices belonging to the maximal clique. In our case, splits
represent the sum of weights of all relations between
vertices of the clique and the remainder ones of the
association graph.
3. Results and experiments
To evaluate our approach, we have used a synthetic data
base that we generate from 100 different document
a set of 100 ancient documents. We have supposed that
these pages are pre-segmented and we have generated
corresponding graphs to be introduced as candidates for
matching process. We have illustrated bellow (table 2)
results of the application of our approach to queries
documents from synthetic database (query 1) and from
ancient document database (query 2). We conclude that the
similarity score illustrate well the efficiency of our approach
mainly with synthetic query.
4. Conclusion
In this paper, we have presented a new approach to indexing
and retrieval ancient document structure. Indexing process
(2) is performed by an ARG structure. Thus, document search
process is achieved by a graph matching algorithm based on
clique finding in an association graph. Our contribution is
involved in the creation of the AG which is level-headed
with weights, determined by analyzing relations in graphs
candidates for matching. Finally, the degree of similarity is
determined according to the characteristics of maximal
clique toward those of graphs candidates. We can conclude
that this approach offers good results and a nice challenge
for further improvements mainly by introducing other
content features for characterizing the structure.

Table 2: A tabulation of the top four documents matches for query document

I2 I3
I2 I2 I1
I3 I3 I2 I3
I2 I3 I1
I1
I1

Query 1 Sim=1 Sim=0.9136 Sim=0.8965 Sim=0.8586

Query 2 Sim=1 Sim= 0.7097 Sim=0.7097 Sim=0.5818

146 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

[19] J.R. Ullman. “An algorithm for subgraph

References isomorphism,” Journal Association for Computer
Machinery 23, pp.31-42, 1976.
[1] C. Ah-Soon. Analyse de plans architecturaux. Ph.D.
Thesis, Institut National Polytechnique de Lorraine,
1998.
[2] D. Arivault. Apport des graphes dans la reconnaissance
non-contrainte de caractères manuscrits anciens. Ph.D
Thesis, Université de Poitiers, 2006.
[3] R. Ambauen, S. Fisher, and H. Bunke. Graph edit
distance with node splitting and merging, and its
application to diatom identification. IAPR-TC15
Workshop on GbRPR, LNCS 2726, pp. 95-106, 2003.
[6] A. Belaïd, Y. Belaïd, “Reconnaissance des formes:
Méthodes et applications,” InterEditions, 1992.
[7] F.P.G. Bergo, A.X. Falcão, P.A. Miranda, L.M. Rocha.
“Automatic image segmentation by tree pruning,”
Journal of Mathematical Imaging and Vision, 29(2-3),
pp. 141-162, 2007.
[8] H. Bunke, S.Gunter, X.Jiang, “Towards bridging the
gap between statistical and structural pattern
recognition: Two new concepts in graph matching,”
Advances in pattern recognition: ICAPR, Bresil, pp. 1-
11, March 2001.
[9] P.F. Felzenszwalb, D.P. Huttenlocher, “Efficient
graph-based image segmentation,” International
Journal of Computer Vision, 59(2), pp. 167-181,2004.
[10] P. Heroux. “Contribution au problème de la rétro-
conversion des documents structurés,” Ph.D.Thesis,
Université de Rouen, 2001.
[11] D. Lin, An information theoric definition of Similarity.
International Conférence on machines learning, pp.
296-304. Morgan Kaufmann, 1998.
[12] J. Lladós, “Symbol recognition by error-tolerant
subgraph matching between region adjacency graphs,”
IEEE Transactions on Pattern Analysis and Machine
Intelligence, Vol. 23, No. 10, 2001.
[13] B.T Messmer. “Efficient graph matching algorithms
for pre-processed model graphs,” Ph.D Thesis, Institut
für Informatik und angewandte Mathematik,
Université de Bern, Suisse, 1995.
[14] M. Neuhaus, H. Bunke. “Automatic learning of cost
functions for graph edit distance,” Inform.Sci, 177 (1):
pp 239-247, 2007.
[15] M.Pelillo, K.Siddiqi, S.W. Zucker. “Matching
hierarchical structures using association graphs,” IEEE
Trans. Pattern Analysis and Machine Intelligence, vol.
21, no. 11, pp.1105-1120, November 1999.
[16] K. Riesen, and H. Bunke. “Approximate graph edit
distance computation by means of bipartite graph
matching,” Image and Vision Computing, 27 (7), pp.
950-959, 2009.
[17] K. Shearer, H. Bunke, S. Venkatesh. “Video indexing
and similarity retrieval by largest common subgraph
detection using decision trees,” Pattern Recognition 34,
pp.1075-1091, 2001.
[18] S.Sorlin. “Mesurer la similarité de graphes,” Ph. D
Thesis, Université Claude Bernard Lyon I, 24
Novembre 2006.
 (IJCNS) International Journal of Computer and Network Security, 147
Probability of Inquiry Interference in Bluetooth
Discovery: Mathematical Analysis and Modeling
Khaled Morsi 1, Gao Qiang 2 and Xiong Huagang 3
1,2,3
Department of Electronic and Information Engineering
Beihang University (BUAA), Beijing, 100191, China
1
kh_morsi_1974@yahoo.com, 2 gaoqiang@buaa.edu.cn and 3 hgxiong@buaa.edu.cn
Abstract: Bluetooth device discovery process involves nodes
seeking to join neighboring piconets to enter inquiry substate.
Nodes in a scatternet as well enter inquiry substate to discover
other nodes within range. Inquiring nodes periodically send
inquiry packets on a predetermined frequency partition of hop
frequencies used by piconets. Intuitively, potential collisions will
occur between inquiry packets and packets exchanged between
piconet nodes during connection state, which lead to degraded
network performance and longer discovery process. In this
paper, we develop a mathematical model for interference caused
by a single inquiring node, and extend the model to predict the
severe interference levels due to multiple inquirers. We further
validate our mathematical model through a simulation scenario.
Analytical and simulation results proved that the probability of
packets interfered and disrupted during inquiry procedure is
periodic repeating every 3.16 sec and interference caused by a
single inquiring node is about 1.25%. Results also revealed that
five inquirers can disrupt 5.9% of network packets, and with
increased number of inquiring nodes, the interference levels
become more crucial approaching 35.3%.
Keywords: Bluetooth, Inquiry, Device Discovery, Interference
1. Introduction
Device discovery is an essential process to establish and
maintain successful Bluetooth scatternets. This process is
based on completing two phases; inquiry and paging. The
inquiry phase is used to search and locate other Bluetooth
devices within range. Nodes in a scatternet enter inquiry
substate to discover other neighboring nodes. Nodes arriving
in the vicinity as well enter inquiry substate seeking to join a
pico- or scatternet. The inquiry procedure involves sending
periodically inquiry packets on a predefined 32-frequency
subset of 79 hop frequencies used by scatternet packets in
normal connection state. Thus, interference is likely to occur
between inquiry packets and connection state packets if they
share the same frequency. Such interference will degrade
network throughput, preclude device discovery process, and
may result in undiscovered devices even within radio range.
The discovery process was discussed in previous studies.
The authors in [1] developed an analytical model for time
required by a master to discover one slave and proved that
discovery time is affected by presence of multiple potential
slaves. The work in [2] investigated optimization parameters
needed to improve the performance of Bluetooth connection
time. The authors in [5] analyzed the delay in discovery
process for different situations and proposed means to
decrease it. The proposed protocol in [6] does not separate
inquiry and page phases to decrease interference, provide
Vol. 2, No. 5, May 2010
faster links, and consume less power. In [7], the authors
evaluated the factors affecting discovery time for multiple
neighboring devices and introduced a modified inquiry
scheme to accelerate discovery process. Through analysis
and simulation, the authors in [8] proposed two ways to
reduce delay in discovery process. The work in [9] analyzed
the frequency-matching delay and proposed three schemes
to speed up discovery procedure in Bluetooth.
The previous studies provided different mechanisms and
new protocols to accelerate discovery process. However, the
role of inquiry packets interference during discovery process
was under estimated in most previous researches. We
believe that reducing delay in discovery process, especially
for multiple inquirers, depends mainly on avoiding such
interference. This interference is not only generated when
inquiry packets share same frequency with connection state
packets. Both packets can also be indirectly disrupted if the
header of the preceding packet is directly disrupted, even if
they do not share the same frequency with inquiry packets.
Thus, direct and indirect disruption will aggravate the
inquiry interference problem.
The rest of this paper is organized as follows: In section
2, the Bluetooth inquiry substate is summarized. The
Bluetooth connection state is illustrated in section 3. In
section 4, the developed mathematical model for single
inquiring node interference is presented. The extended
model of multiple inquiring node interference is deduced in
section 5. Final conclusions are given in section 6.
2. The Bluetooth Inquiry Substate
A node enters the inquiry substate and acts as a master to
discover other potential slave nodes to form a piconet. In
contrast, a node in the inquiry scan substate acts as a slave
to be discovered by potential masters in the inquiry substate.
Inquiring nodes use their resident CLK and General Inquiry
Access Code (GIAC) address, (9E8B33)16, to generate a
frequency hopping sequence among a predetermined 32-
frequency subset of 79 Bluetooth frequencies [3]. The set of
32 inquiry frequencies is further segmented during inquiry
into two 16-frequency trains, A and B.
An inquirer chooses A or B train for initial transmission
and repeats this train for 256 times (256 x 625 usec x 16 =
2.56 sec), before switching to the other train [3]. The A-
train or B-train is chosen when the parameter KOFFSET is set
up to 24 or 8, respectively. The change of train is made by
adding 16 to the current value of KOFFSET modulo 32. Three
148
changes have to be done within inquiry procedure length,
which corresponds to 10.24 sec each 60 sec, as specified in
[3]. As shown in Fig.1, the frequencies in these trains
change over time such that each train always has 8
frequencies in the 54-69 index range (Lower Range) and 8
frequencies in the 70-79 and 1-6 index range (Upper
Range).
Figure 1. Shifting and Selection of Inquiry Frequencies
An inquiring device transmits a 68-bit inquiry packet
during an even time slot, when its two least significant clock
bits are zero (CLK1,0=002). The frequency used is pseudo-
randomly selected from lower range frequencies (FL) of the
current train, A or B. Another upper range frequency (FU)
is selected from the same train and used to transmit the
second inquiry packet 312.5 usec later when (CLK1,0 =012)
as shown in Fig.2. During odd time slots, the inquirer
listens for responses from neighboring devices in inquiry
scan substate. Responses are transmitted 625 usec after an
inquiry packet is received and transmitted on same
frequencies, FL and FU [3]. The inquirer then selects new
frequencies, FL and FU, and repeats the process, collecting
data from as many neigh-boring nodes as possible during
inquiry substate duration.
Figure 2. Bluetooth Inquiry, DH1, DH3, and DH5 Packets
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
3. The Bluetooth Connection State
All members of a Bluetooth piconet hop together among
79 frequencies with a sequence that is a function of master's
28 bit free-running clock and first 28 bits of the master's 48
bit address [4]. The hop sequence cycle covers about 23.3
hours, spanning about 64 MHz of spectrum, spreading
transmissions over 80% of the available 79 MHz band, and
visiting all frequencies with equal probability [4]. The FH
channel is based on Time Division Duplexing (TDD) using
625 usec time slots. Master packets are transmitted on even
time slots and slaves respond on odd slots. Each time slot
uses different hop frequency with hopping rate 1600 hops/s.
To increase data rate, and intuitively reduce hopping rate,
multi-slot packets are defined to cover 1, 3, or 5 time slots,
named DH1, DH3, or DH5, respectively. As shown in Fig.2,
all packets contain a 72-bit Access Code to identify piconet
and 54-bit Packet header containing the 3-bit slave address.
Multi-slot packets will continue using same hop channel as
being decided by first time slot, then, the system tunes back
to the carrier frequency sequence as if only a single slot
packet had been transmitted. Using this way, the hop
frequency selection is not affected by packet lengths and all
nodes can still hop synchronously.
During connection state, the instantaneous probability of
a frequency being selected is based on 32-frequency
partitions of the Bluetooth frequency band. The hop
generator begins with a group of 32 frequencies that we call
Master Frequencies Cycle (MFC) determined by master of
piconet. The probability that a MFC is the same 32-
frequency subset as that used by the inquiry substate is 1/79.
The master transmits on a pseudo-randomly chosen
frequency from MFC. The slave subsequently transmits on a
frequency selected from a 32-frequency partition
immediately right to the MFC which we term Slave
Frequencies Cycle (SFC). Once each of the 32-frequencies
in the MFC and SFC have been used exactly once (taking 64
time slots), both intervals shift right by 16 frequencies
modulo 79 and frequency selection begins again as shown in
Fig.3. The MFC/SFC shifts such that it uses all 79
MFC/SFC combinations before repeating a MFC/SFC such
that the frequency selection is uniform across 79
frequencies. Within each MFC/SFC, the hopping sequence
is pseudo-random, based on the master’s CLK and address
[4].
Figure 3. MFC and SFC in Connection State
4. Single Inquiry Node Interference
The time delay (DL) between inquiry packets and
piconets master and slave packets is significant, since it
illustrates, not only difference between direct and indirect
interference, but also it determines which packet (master,
 (IJCNS) International Journal of Computer and Network Security, 149
slave, or both) is subjected to disruption. As shown
previously in Fig.2, both piconet master and inquiring node
have coincided even time slots, thus, lower inquiry packet
FL is a potential direct interferer for master packet if both
share same frequency.
During odd time slots, slave packets are transmitted while
the inquiring node is in the inquiry scan substate receiving
responses from other nodes, then, slave packets can never be
directly disrupted. However, a slave packet can be indirectly
interfered if its preceding master packet’s Access Code (AC)
or Packet Header (PH) is directly disrupted because a slave
node can not determine whether it was addressed or not and
thus will not respond to the master. Similarly, if piconet’s
even slots coincided with inquiring node odd slots, slave /
master packets are susceptible to direct / indirect
interference, respectively.
Table 1: Effect of Inquiry Delay on Packet Disruption
The value of DL is uniformly distributed between 0 and
1250 usec, and can be divided into nine intervals (T1 to T9)
as depicted in Table 1, in which MIL / MIU denotes whether
a master packet is directly interfered by lower / upper range
frequency, and SIL / SIU denotes whether a slave packet is
directly interfered by lower / upper range frequency,
respectively. Since master / slave packet indirect
interference depends on preceding slave / master packet’s
AC or PH direct disruption, it is also important to divide the
value of DL into five intervals (T1 to T5) as shown in Table
2, in which MHIL / MHIU represents whether a master
header is directly interfered by lower / upper range
frequency, and SHIL / SHIU represents whether a slave
header is directly interfered by lower / upper range
frequency, respectively.
Table 2: Effect of Inquiry Delay on Header Disruption
Considering Fig.4 as an example, when DL1 = 250 usec,
which belongs to intervals T3 in Table 1 and T2 in Table 2,
it is clear that the master packet f(k), is subjected to direct
interference if it shares same frequency with FL (MIL=1),
while (MIU=0) since the master packet is not subjected to
direct interference even if it shares same frequency with FU.
Conversely, for a slave packet f(k+ 1), (SIL=0) since it is not
subjected to direct interference even if it shares same
frequency with FL and (SIU=1) since it is subjected to direct
Vol. 2, No. 5, May 2010
interference if it shares same frequency with FU. It is also
important to note that, while (SHIL=0) since slave header is
not subjected to FL, the slave packet is not only threatened
by FU (SIU=1) but also it is subjected to slave header direct
interference if it shares same frequency with FU (SHIU=1),
which will lead to indirect interference for the next master
packet f(k+ 2). Since the master header f(k) is not subjected
to direct interference from either FL or FU, hence,
(MHIL=MHIU=0).
Another example in Fig.4, when DL2 = 765 usec, which
belongs to interval T6 in Table 1, the master packet f(k)
does not suffer from FL or FU, thus, (MIL=MIU=0). The
slave packet is only threatened by direct interference from
FL (SIL=1) but not suffering from FU (SIU=0) even if it
shares same frequency with FU due to 259 usec guard time.
It is obvious that this delay value, DL2 = 765 usec, does not
belong to any of five intervals in Table 2, but belongs to the
range of DL values that will cause no interference with
either MH or SH, hence, they are not even mentioned in
Table 2.
Figure 4. Offset Between Inquiry and Master/Slave Packets
Finally, from Fig.4, when DL3 = 1220 usec, which
belongs to intervals T9 in Table 1 and T5 in Table 2, both
the slave packet f(k+ 1) and slave header are not subjected,
therefore, (SIL=SIU=SHIL=SHIU=0). The master packet
f(k+ 2) is threatened by both FL and FU (MIL=MIU=1)
while MH is only subjected to FL (MHIL=1, MHIU=0).
If DL is unknown, the expected value E [MIL, MIU, SIL,
SIU] can be used instead, which is calculated by averaging
and dividing the range of values of DL which cause MIL,
MIU, SIL, and SIU to equal 1 by the total range of DL, thus:
190 + 119 + 120
E [MIL, MIU , SIL, SIU ] = = 0.344 (1)
1250
Similarly, the expected value E [MHIL, MHIU, SHIL,
SHIU] can be calculated by averaging and dividing the
range of values of DL which cause MHIL, MHIU, SHIL,
and SHIU to equal 1 by the total range of DL, hence:
E [MHIL, MHIU , SHIL, SHIU ] =
192 (2)
= 0.154
1250
The first frequency in each MFC/SFC determines how
many frequencies are shared between a MFC/SFC and both
upper and lower range frequencies. Since the starting
position of MFC/SFC is periodic repeating every 3.16 sec
(64 time slots x 625 usec x 79), hence the number of
overlapped frequencies is periodic, which means that the
probability of intersection (interference) between MFC/SFC
150 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

and both upper and lower range is also periodic and changes inquirer is 1/2 since only 8 of 16 Upper Range frequencies
according to shift in MFC/SFC, thus: are in each train. Finally, if the selected frequency is in the
used train, there is only a 1/8 chance of interference since
f (m) = mod(16m,79) + 1 0 ≤ m ≤ 79 (3) any of the 8 frequencies in the used train in the Upper
Range can be selected. A similar analysis applies to the
Where f(m) marks index of the first frequency in each
Lower Range as well. With DL unknown but uniformly
MFC and m denotes all possible shifts to start a new MFC.
distributed, E [MIL, MIU, SIL, SIU] can be used which was
Similarly, f(s) marks index of the first frequency in each
determined earlier to be 0.344. Thus, the number of
SFC and s denotes all possible shifts to start a new SFC,
frequencies shared between a MFC and all inquiry
hence:
frequencies is:
f ( s ) = mod((16 s + 32),79) + 1 0 ≤ s ≤ 79 (4)
M f ( m) = M f ( m), L + M f ( m),U (9)
Thus, the number of shared frequencies between a MFC
or a SFC and lower / upper ranges of inquiry frequencies is: The probability of Direct interference between a master
packet and an inquiry packet in a MFC starting at frequency
0 . 5 × (2414 − freq ( f ( m)) ) 2402 ≤ freq( f (m)) ≤ 2412 of index f(m) is:
 
 0 2414 ≤ freq( f (m)) ≤ 2476
0 . 5 × ( freq ( f ( m)) − 2476 ) 2478 ≤ freq( f (m)) ≤ 2480 (5) 0.344
M =  P ( DM , I ( f (m))) = M f ( m) (10)
0 .5 × ( freq ( f ( m)) − 2397 ) 2403 ≤ freq( f (m)) ≤ 2427
f ( m ),U
512
 16 2429 ≤ freq( f (m)) ≤ 2461
  Similarly, for slave packets with unknown uniformly
 0 .5 × (2493 − freq ( f ( m)) ) 2463 ≤ freq( f (m)) ≤ 2479
distributed DL, the number of frequencies shared between a
Where Mf(m),U is the number of frequencies shared SFC and all inquiry frequencies is:
between a MFC and upper range of inquiry frequencies.
S f ( s ) = S f ( s ), L + S f ( s ),U (11)
 0 2402 ≤ freq( f ( m)) ≤ 2444
 
0 . 5 × ( freq ( f ( m)) − 2444 ) 2446 ≤ freq( f ( m)) ≤ 2474 The probability of Direct interference between a slave
 16 2476 ≤ freq( f ( m)) ≤ 2480 (6) packet and an inquiry packet in a SFC starting at a
M f ( m), L = 
 16 2403 ≤ freq( f ( m)) ≤ 2429 frequency of index f(s) is:
 0 .5 × (2461 − freq ( f ( m)) ) 2431 ≤ freq ( f (m)) ≤ 2459 
 
 0 2461 ≤ freq ( f (m)) ≤ 2479  P ( DS, I ( f ( s))) =
0.344
S f (s) (12)
512
Where Mf(m),L is the number of frequencies shared between
a MFC and lower range of inquiry frequencies. One MFC/SFC is used each 40 msec (625 usec x 64 time
slots = 40 msec), half of the packets are Master packets and
 0 .5 × (2414 − freq ( f ( s )) ) 2402 ≤ freq( f (s )) ≤ 2412
the other half are Slave ones. Thus, the total probability of
  Direct interference between a master/slave packet and an
 0 2414 ≤ freq( f (s )) ≤ 2476
 0 .5 × ( freq ( f ( s )) − 2476 ) 2478 ≤ freq ( f ( s )) ≤ 2480 (7) inquiry packet in a MFC/SFC starting at index f(m)/f(s) is:
S f ( s ),U = 
 0. 5 × ( freq ( f ( s )) − 2397 ) 2403 ≤ freq( f (s )) ≤ 2427
 2429 ≤ freq( f (s )) ≤ 2461
P ( DM , I ( f (m))) + P ( DS , I ( f ( s )))

16
 P ( DI ( f (m), f ( s))) = (13)
 0 .5 × (2493 − freq ( f ( s )) ) 2463 ≤ freq( f (s )) ≤ 2479 2

Where Sf(s),U is the number of frequencies shared between The Direct interference probability is periodic, repeating
a SFC and upper range of inquiry frequencies. every (3.16 sec). One complete period of P(DM,I (f(m))),
P(DS,I (f(s))), and P(DI (f(m),f(s))) is shown in Fig.5. Note
 0 2402 ≤ freq( f (s )) ≤ 2444 that the direct interference probability of a Master or Slave
 
0 .5 × ( freq ( f ( s )) − 2444 ) 2446 ≤ freq( f (s )) ≤ 2474 packet P(DI (f(m),f(s))) is more than 1% for about 49.37% of
 16 2476 ≤ freq( f (s )) ≤ 2480 (8) the time (1.56 sec) even though its mean value is 0.0086.
S f ( s ), L = 
 16 2403 ≤ freq( f (s )) ≤ 2429 Master Packet

 0 .5 × (2461 − freq ( f ( s )) ) 2431 ≤ freq ( f ( s )) ≤ 2459  0.02

 
 2461 ≤ freq ( f ( s )) ≤ 2479 
0.01
0
Direct Interference Probability

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3 3.2
Where Sf(s),L is the number of frequencies shared between Slave Packet

a SFC and lower range of inquiry frequencies. 0.02

0.01

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3 3.2
Direct Interference Probability Master or Slave Packet

0.01
Generally, any frequency from 32 frequencies in a MFC
0.005
has a 1/32 chance of being selected. If a MFC overlaps the
0
Upper Range by Mf(m),U frequencies, hence, probability of a 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Time ( sec )
2 2.2 2.4 2.6 2.8 3 3.2

Master frequency being selected from Upper Range is

Mf(m),U/32. If a Master frequency is chosen from Upper Figure 5. Direct Interference Probability versus Time
Range, probability that it is in train (A or B) being used by
 (IJCNS) International Journal of Computer and Network Security, 151
Vol. 2, No. 5, May 2010

Master Packet
Indirect Interference Probability
0.02

Direct / Indirect Interference Probability

The Indirect interference probability expression can be 0.01

derived similarly regarding that a packet indirect disruption 0

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3 3.2
is caused by the opposite type packet direct disruption. If a Slave Packet

Master/Slave packet is not disrupted, the next Slave/Master 0.02

packets can not be indirectly disrupted. Thus, probability of 0.01

0
Indirect interference between a master packet and an inquiry 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3 3.2
Master or Slave Packet
packet due to frequency collision in a SFC starting at a 0.02
frequency of index f(s) is: 0.01

0
0.154 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3 3.2
P ( DM , I ( f ( s ))) = S f (s ) (14) Time ( sec )
512
Figure 6. Direct/Indirect Interference Probability vs. Time
The probability of Indirect interference between a slave
packet and an inquiry packet due to frequency collision in a
MFC starting at a frequency of index f(m) is: 5. Multiple Inquiry Nodes Interference
Bluetooth scatternets will encounter more than one
0.154
P ( DS , I ( f ( m))) = M f ( m) (15) inquiring node simultaneously. Thus, all Direct and Indirect
512 interference probabilities will be as shown below.
Note that E [MHIL,MHIU,SHIL,SHIU] is used, which
was determined earlier to be 0.154. Direct Interference Probability with n Inquirers
The probability of Direct interference between a master
Total Interference Probability packet and n inquiry packets in a MFC starting at a
frequency of index f(m) is:
The total probability of a packet being disrupted is the
sum of both its probability of direct and indirect  M f (m), L n
 n M f (m),U n
 n  
interference, taking into account that if the packet is
n
P ( DM , I ( f (m))) = 
 32
∑  x(0.344)
x=1
n
+
32 ∑  n − x(0.344) 
x=1
n

interfered indirectly, it cannot be directly disrupted. Hence,  x  x 

the total probability of Direct/Indirect interference between a ∑
×   (0.5)x 1 − (7 8) y
 y=0  y 
( ) (19)
master packet and an inquiry packet in a MFC/SFC starting  
at frequencies of index f(m)/f(s), respectively, is:
The probability of Direct interference between a slave
P (M D, D , I ( f (m), f ( s ))) = P ( D M , I ( f ( m))) + P ( D M , I ( f ( s ))) packet and n inquiry packets in a SFC starting at a
(16) frequency of index f(s) is:
− P ( D M , I ( f (m))) × P ( D M , I ( f ( s )))

 S f ( s ),L n
n n
 n  
∑  x(0.344) ∑  n − x(0.344) 
The total probability of Direct/Indirect interference S f ( s ),U
P ( DSn,I ( f (s ))) =  n
+ n

between a slave packet and an inquiry packet in a MFC/SFC  32 x=1

32 x=1
starting at frequencies of index f(m)/f(s), respectively, is:  x  x 
∑
×   (0.5)x 1 − (7 8) y
 y=0  y 
( ) 

(20)
P (S D , D, I ( f (m), f (s ))) = P ( D S , I ( f ( s ))) + P ( D S , I ( f (m)))  
(17)
− P ( D S, I ( f ( s ))) × P ( D S , I ( f (m))) Where Mf(m),L/32 and Sf(s),L/32 are the probabilities that
Master/Slave packet is using one of the lower range inquiry
Since there are equal chances for a packet to be Master or
frequencies in a MFC/SFC starting at frequencies of index
Slave, the total probability of Direct/Indirect interference
f(m)/f(s), respectively. Similarly, Mf(m),U/32 and Sf(s),U/32 are
between a master/slave packet and an inquiry packet in a
the probabilities that Master/Slave packet is using one of the
MFC/SFC starting at frequencies of index f(m)/f(s), is:
upper range inquiry frequencies in a MFC/SFC starting at
P (M D, D , I ( f (m), f ( s))) + P ( SD, D, I ( f (m), f ( s))) frequencies of index f(m)/f(s), respectively.
P ( I D, D ( f (m), f ( s))) = (18)
2
For n inquiry nodes, the probabilities of a master/slave
One complete period (3.16 sec) of the direct/indirect packet being subjected to Direct interference in the
interference probabilities for a Master packet, a Slave lower/upper range by x/n-x out of n inquiry packets with
packet, and Master or Slave packet, is shown in Fig.6. Note
are   (0 . 344 )n and  n 
n
that the direct/indirect interference probability of a Master
unknown delays  n − x  (0 . 344 )n
 x  
or Slave packet approaches 1.5% approximately 46.84% of respectively.
the time (1.48 sec) even though its mean value is 0.0125.
x
 y  (0 . 5 )
x
is the probability that y of x inquiry nodes are
 
using the train which contains the same master/slave
frequency.
152 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

Finally, the probability that at least one inquiry node is The direct and direct/indirect interference probabilities
using the same master/slave frequency, assuming that for n= 5 of a Master/Slave packet, are shown in Fig.7,
piconet packet is subjected and in the same train and range respectively. Note that the direct interference probability of
is (1 − (7 8 )y ). a Master or Slave packet approaches 5% approximately
44.3% of the time (1.4 sec) even though its mean value is
The total probability of Direct interference between 0.0411. Whereas the direct/indirect interference probability
master/slave packet and n inquiry packets in a MFC/SFC of a Master or Slave packet is above 7% approximately
starting at frequencies of index f(m)/f(s), respectively, is: 46.84% of the time (1.48 sec) even though its mean value is
0.0594.
P ( DMn , I ( f ( m))) + P ( DSn, I ( f ( s )))
P ( D In ( f ( m), f ( s ))) = (21)
0.06
Master or Slave Packet
2

Inte rfe rence Pro bability

0.04

Direct
Indirect Interference Probability with n Inquirers 0.02

The probability of Indirect interference between a master 0

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3 3.2
packet and n inquiry packets due to frequency collision in a Master or Slave Packet
SFC starting at a frequency of index f(s) is: 0.08

Interferenc e Probability
Direct/Indirect
0.06

 S f ( s ), L n
 n  S f ( s ),U n
n 
∑ (0.154 )n + ∑   (0.154 )n 
0.04
P ( D Mn , I ( f ( s ))) =  
 32 x=1  n − x  32 x =1  x   0.02

  x 
( )
x 0

∑
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3 3.2
×   (0.5)x 1 − (7 8) y (22) T ime ( sec )
 y= 0  y 
 
Figure 7. Interference Probabilities of 5 Inquirers vs. Time
The probability of Indirect interference between a slave
packet and n inquiry packets due to frequency collision in a A MATLAB simulation was conducted to verify our
MFC starting at a frequency of index f(m) is: mathematical models. The scenario involved multiple
piconets, with random clock and address values, generating
 M f (m), L n
 n  M f (m),U n
 n 
P ( DSn,I ( f (m))) = 
 32
∑  n − x (0.154)
x=1
n
+
32 ∑  x(0.154) 
x=1
n both master and slave packets as MFC/SFC achieves all 79
possible shifts. Up to 60 Inquiring nodes arrive to piconets
 x  x  neighborhood with random DL between 0 and 1250 usec.
∑
×   (0.5)x 1 − (7 8) y
 y=0  y 
( )


(23)
The mean Direct/Indirect collisions between piconets
  packets and inquiry packets were recorded every 3.16 sec.
Analytical and simulation results are shown in Fig.8, from
Where the probabilities of a master/slave packet being
which it is clear that as number of inquiring nodes
subjected to Indirect interference in the lower/upper range
increases, interference levels become more crucial.
by n-x/x out of n inquiry packets with unknown delays are
Considering only direct interference, probability approaches
 n  n
 n − x  (0 . 154 )n and  x  (0 . 154 ) respectively.
n
30.7%. Including direct and indirect interference, it
   
approaches 35.3% since Master/Slave packets transmitted in
Upper or Lower Range are disrupted and indirectly cause
Total Interference Probability with n Inquiry Nodes next Slave/Master packets to be interfered even if the next
The total probability of Direct/Indirect interference Slave/Master frequencies do not overlap the Upper/Lower
between a master packet and n inquiry packets in a frequency Ranges.
MFC/SFC starting at frequencies of index f(m)/f(s), is: 0.4

0.35
P ( M Dn , D , I ( f ( m), f ( s ))) = P ( D M
n
, I ( f ( m))) + P ( D M , I ( f ( s )))
n
(24)
Total Interference Probability

0.3
− P (D M
n
, I ( f ( m))) × P ( D M , I ( f ( s )))
n
0.25

The total probability of Direct/Indirect interference 0.2

between a slave packet and n inquiry packets in a MFC/SFC
0.15
starting at frequencies of index f(m)/f(s), respectively, is:
0.1
Direct Interference Probability (Derived)
P ( S Dn , D , I ( f (m), f ( s ))) = P ( D Sn , I ( f ( s ))) + P ( D Sn, I ( f ( m)))
(25) 0.05
Direct Interference Probability (Simulated)
Direct/Indirect Interference Probability (Derived)
− P ( D Sn, I ( f ( s ))) × P ( D Sn, I ( f ( m))) Direct/Indirect Interference Probability (Simulated)
0
0 5 10 15 20 25 30 35 40 45 50 55 60
Number of Inquiring Nodes
Finally, total probability of Direct/Indirect interference
between a master/slave packet and n inquiry packets in a Figure 8. Interference Probabilities for Multiple Inquirers
MFC/SFC starting at frequencies of index f(m)/f(s), is:

P ( M Dn , D , I ( f ( m), f ( s ))) + P ( SDn , D , I ( f ( m), f ( s ))) (26)

P ( I Dn , D ( f ( m), f ( s ))) =
2
 (IJCNS) International Journal of Computer and Network Security, 153
Vol. 2, No. 5, May 2010

6. Conclusions
In this paper, we showed that during discovery process,
the potential collisions between both inquiry and scatternet
packets in connection state are significant. Scatternet nodes
and inquiring nodes will be affected by such interference
that degrades network throughput and elongate discovery
time. We further developed a mathematical model for
interference probability caused by a single inquirer, and
extended the model to predict the severe interference levels
due to multiple inquiring nodes. Results revealed that
interference probability is periodic, repeating every 3.16 sec
over all possible MFC/SFC. The interference caused by a
single inquirer is 1.25%. Five inquiring nodes result in
interference probability of more than 7% for approximately
1.48 sec even though the mean probability of interference is
only 0.059. However, as number of inquirers increase, the
interference levels become more crucial. Considering only
direct interference, results proved that probability reaches
30.7%. Including both direct and indirect interference, it
approaches 35.3% since Master/Slave packets transmitted in
Upper or Lower Range are always disrupted and indirectly
cause the next Slave/Master packets to be interfered.

References
[1] D. Chakraborty, G. Chakraborty, K. Naik, N. Shiratori,
“Analysis of the Bluetooth device discovery protocol”,
Springer Wireless Networks Journal, February 2010,
Vol. 16, pp.421-436.
[2] T. Thamrin, S. Sahib, “The Inquiry and Page
Procedure in Bluetooth Connection,” IEEE Inter.
Conf. SOCPAR09, Dec.2009, pp.218-222.
[3] www.bluetooth.com, “Bluetooth Specification
v4.0”,Dec.17,2009
[4] M. Khaled, H. Xiong, Q. Gao, “Performance
Estimation and Evaluation of Bluetooth Frequency
Hopping Selection Kernel,” IEEE International
Conference ICPCA09, Dec.2009, pp.461-466.
[5] D. Chakraborty, G. Chakraborty, S. Naik, N. Shiratori,
“Discovery and Delay Analysis of Bluetooth Devices”,
IEEE International Conference MDM06, May 2006,
pp.113-117.
[6] S. Z. Kardos, A. Vidacs, “Performance of a new device
discovery and link establishment protocol for
Bluetooth”, IEEE International Conference
GLOBECOM05, January 2006, Vol.6, pp.3522-3527.
[7] X. Zhang, G.F. Riley, “Evaluation and accelerating
Bluetooth device discovery,” Radio and Wireless
Symposium, Jan. 2006, pp.467-470.
[8] G. Chakraborty, K. Naik, D. Chakraborty, N. Shiratori,
& D. Wei, , “Delay analysis and improvement of the
device discovery protocol in Bluetooth”, IEEE
Vehicular Tech. Conf., 2004, Los Angeles, USA.
[9] J. Jiang, B. Lin, & Y. Tseng, “Analysis of Bluetooth
device discovery and some speedup mechanisms”,
Journal of Institute of Electrical Engineering, 2004,
11(4), 301–310.
154 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

Color Image Segmentation Using Double Markov

Random Field (DMRF) Model with Edge Penalty
Function
Sami M. Halawani1 , Ibrahim A. Albidewi2, M. M. Sani3 and M. Z. Khan3
1,2
Faculty of Computing and Information Technology-Rabigh Campus
King Abdulaziz University
Saudi Arabia
Halawani@kau.edu.sa, Ialbidewi@kau.edu.sa
3
Faculty of computer Science and Info. System
Jazan University K.S.A
sani.nitrkl@gmail.com
&
Department of Computer Science and Engineering, IIET, Bareilly India
zubair.762001@gmail.com,

Abstract: In our paper color image segmentation using DMRF
model with edge penalty function as image model and for color
( )
model using RGB and OHTA I1 , I 2 , I 3 color model. In our
paper we are proposing Hybrid algorithm based on DMRF
model and The image segmentation problem is formulated as a
pixel labeling problem and the pixel labels are estimated using
Maximum a Posterior(MAP) criterion. The MAP estimate of the
labels are obtained by the proposed hybrid algorithm converges
to the solution faster than that of Simulated Annealing(SA)
algorithm. The DMRF with edge preserving attribute yields
better segmentation result with local edge boundary. It is also
found that Exponential function improved the result to some
extend. The paper is divided in six section in second section we
explain the model use for Hybrid Algorithm.
Keywords: DMRF, MAP, RGB-Plane, Edge penalty function.
1. Introduction & Background
Image segmentation is an important process in automated
image analysis. In color image segmentation, choice of
proper color model is one of the important factors for
segmentation and different color models such as RGB, HSV,
( )
YIQ, OHTA I1 , I 2 , I 3 , CIE (XYZ, Luv, Lab) are used to
represent different colors [1],[2][3]. Besides color model,
image model does play a crucial role for color image
segmentation. Stochastic models, particularly Markov
Random Field (MRF) models, have been extensively used as
the image model for image restoration and segmentation
problems [4][5]. Often, in stochastic framework, the color
image segmentation problem is formulated as a pixel
labeling problem and the pixel labels are estimated using the
Maximum a Posteriori (MAP) criterion[6][7]. By and large,
the MAP estimates of the pixel labels are obtained using
Simulated Annealing (SA) algorithm [11][8]. MRF model
has also been successfully used as the image model while
addressing the problem of color image segmentation in
supervised and unsupervised framework [9]. Here we use,
color image segmentation problem is formulated in
stochastic framework. Markov Random Field model is used
as the image model and RGB and OHTA model have been
used as the color models. we proposed a Double MRF
(DMRF) model that takes case of intra-plane interactions as
well as the inter-plane interactions[10]. These proposed
models have been tested on different images and also the
results obtained have been compared with the results
obtained using only MRF. Both first order and second order
models have also been studied. As expected, it has been
observed that the use of OHTA model yielded better results
than that of using RGB model We have used line process in
the clique potential function and it has been found that
although the line process could preserve well defined edges
it failed to preserve loosely defined boundaries [11][12]. In
order to preserve loosely defined boundaries, we have
employed the edge penalty model proposed by Qiyao Yu et
al [11][12]. Here, we have used RGB color models with
MRF and DMRF models. Qiyao Yu et al [11][12] have used
exponential edge penalty function preserve the weak edges
together with strong edges. We have proposed the
exponential edge penalty function in the DMRF model to
preserve edge. Here also, the problem is formulated as a
pixel labeling problem and the pixel label estimates are the
MAP estimates.
2. Methodology
Here we are proposing the model for our algorithms. The
image segmentation problem is formulated as a pixel
labeling problem. The labels are estimated using the MAP
estimation criterion usually, the MAP estimates are obtained
by Simulated Annealing (SA) algorithm. It is observed that
SA algorithm is computationally involved and hence we
have proposed a hybrid algorithm. Which is found to be
faster than that of SA algorithm. The proposing models are
as fallow.
2.1 Model 1 (Double Markov Random Field (DMRF)
Model)
The DMRF IS DEFINE AS In order color planes, the a prior
 (IJCNS) International Journal of Computer and Network Security, 155
Vol. 2, No. 5, May 2010

model takes case of intra-plane interactions and inter-plane observed image X which is assumed to be MRF is
interactions as well. This MRF model taking case of intra-plane considered as a degraded version of the label process and
and inter-plane interaction is called Double MRF (DMRF) model. hence consists of Z and W. The model parameter for both
Capturing salient spatial properties of an image leads to the inter and intra color plane processes are selected on a trial
development of image models. MRF theory provides a and error basis. The weak membrane model is considered to
convenient and consistent way to model context dependent the model intra as well as inter plane clique potential
entities for eg. image pixel and correlated features. Through function. For example, the weak membrane model for
the MRF model takes into account the local spatial inter-plane interaction process for two color components,
interactions, it has limitation in modeling natural scenes of
( I1 and I 2 ) (as shown in Fig.1(b) ) can be expressed as
distinct regions. In case of color models, it is known that
 I
( )
there is a correlation among the color components of RGB 2 2
1 I  +  n I2 − n I1 
model. In our formulation, we have de-correlated the color Vc z = α  n ' 2 ' − n '1 '   i' , j' i ', j ' −1  +
components and introduced an interaction process to 

i , j i −1, j   
improve the segmentation accuracy. We have employed 2 2
 n I 2 − n I1   I I  
inter-color plane interaction (OHTA ( I1 , I 2 , I 3 ) color  i ' , j' i ', j ' +1  +  ni'2, j ' − ni'1+1, j '  
    
model) process which reinforces partial correlation among 
different color components. Where α is the MRF field parameter. Superscripts I1 and I 2
We assume all images to be defined on discrete on corresponds to I1 -plane and I 2 -plane respectively. This
rectangular lattice M1 × M 2 . Let Z denote the label process notation is depicted in Fig.1, where the intra plane and inter
corresponding to the segmented image and z is a realization plane interaction are shown. The model parameter both Z
of the label process i.e. the segmented image. The observed and I processes are selected on an Ad hoc manner. The
image X is assumed to be a random field that is assumed to weak membrane model for intra plane as well as inter plane
be the degraded version of the label process Z and this clique potential function.
degradation process is assumed to be Gaussian process W.
The label process Z in a given plane is assume to be MRF.
The prior model takes care of inter as well as intra color
plane interactions. The prior probability distribution of z is
constructed with clique potential function corresponding to
intra-color plane interaction, for example I1 and another
clique potential function within inter plane interactions (for
example between I1 and I 2 ).The of inter color plane
interaction are shows in Figure1.Thus X is a compound
MRF. MRF model taking care of both inter as well as intra
plane interactions. It known that if Z is assumed MRF, then
the prior probability distribution P(Z=z) is Gibb’s
distributed that can be expressed
as P (Z = z | θ ) = 1 e −U (z ,θ ), where Z ' = ∑ z e ( ) is the
−U z, θ Figure 1. (a) RGB Plane Interaction
'
Z
partition function, θ denotes

the clique parameter vector, the exponential term

U(z, θ ) is the energy function and is of the form
U ( z,θ ) = ∑ c ∈ C Vc ( z,θ ) with Vc ( z,θ ) being referred as the
clique potential function. Since the inter-plane process is
viewed to be MRF, we know
 I I I I 
that P  Z 2 = z 2 | Z 1 = z 1 , (k, l ) ≠ (i , j ), ∀(k , l ) ∈ I 
 
 i, j i, j k, l k, l 1

 I2 I I I I 
= P  Z = z 2 | Z 1 = z 1 , (k, l ) ≠ (i , j ), (k , l ) ∈η 1  Where
 i , j i , j k , l k, l i, j 
  Figure 1. (b) Interaction of one plane of R-plane with G-
I1 and I 2 denotes color planes respectively. In other words a plane
pixel in one plane (say for e.g I1 -plane) is assumed to have
2.2 Model 2 (Segmantation in MRF-Map Framework
interaction with pixel of I 2 and I 3 planes, the interaction Using Edge Penalty Function)
process of each color plane is shown in Fig. 1(a). For the Let Y denote the observed data (i.e the original image) and
sake of illustration, the interaction of (i, j )th pixel in I 2 X the desired segmentation. The following optimality
criterion is used
plane with the neighboring pixels of I1 plane for a first
order neighborhood system is also depicted in Fig. 1(b). The
156 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

x = arg max P ( X = x | Y = y)
)
Where U i (s ) is the number of neighbors of s that does not
(1) '
x
Where P ( X = x|Y = y ) the posterior probability is x is the belongs to Ω i . That means the term βU i (s ) is not zero
) '

MAP estimation using Baysian rule. A 2-layer model is only when the site s is at the boundary. Therefore, (4)
constructed from Bayes’ rule. becomes
P (Y = y | X = x )P ( X = x)  [
Y(s) −µi  ]
P ( X = x|Y = y) =
2
n  n
P (Y = y) +β ∑ ∑ Ui(s)
(2) ) '
x= argmin∑ ∑ logσi + (6)
( )
Where P Y = y| X = x is the posterior conditional
Ω1,....Ωni=1s=Ωi 
 2 σ
2
i ( ) i=1s=Ωi
probability of the segmentation given the observed data,
P (Y = y| X = x)P ( X = x) is the conditional probability ∑in=1∑ s=Ωi U i (s ) is approximately proportional to the
'
distribution of the observation conditioned on the length of the obtained boundary. Therefore the role of the
( )
segmentation, P X = x is the prior probability of the region model term actually penalizing the existence of the
segmentation , and P (Y = y ) is the probability is the
boundary by its population.
Instead of penalizing of all boundary pixel, greater penalty
distribution of the observation. Maximization of the can be applied to weak edge pixel and a lesser penalty to
( )
posterior P X = x|Y = y gives a segmentation solution, strong edge pixels, so that local statistics such as edge
and since P (Y = y ) is a constant it is equivalent to strength can be incorporated. Therefore we can replace the

maximizing P (Y = y| X = x)P ( X = x) describes the grey

penalty term with some monotonically decreasing function
( )
of edge strength g ∇ (.) , where ∇ (.) is the gradient
level, texture or any other factors while P ( X = x) can magnitude. The proposed objective function is therefore
incorporate spatial context constraint. Two separate models  [
Y(s) −µi 
2
]
 n
( )
+βi∑=1 s=∑Ω g ∇(Y(s)) (7)
) n
a feature distribution model and a region model are needed x = argmin∑ ∑ logσi +
(
to obtain the analytical expression for the P Y = y| X = x ) Ω1,....Ωn i=1 s=Ωi 
 2 σ( )
i
2
 i

and P ( X = x) respectively. For the region model, a multi-

level logistic model (MLL) is typically used, whose clique
potential is defined as;
β c :if all sites on c equal
Vc ( X ) =  (3)
0 : otherwise
( )
The term P Y = y| X = x is typically assumed to be
Gaussian at each site. Maximizing the posterior is
equivalent to minimizing the objective function of the
equation
n  Y(s) − µi[2
]  Figure 2.Exponential edge penalty function
− βUi (s) (4)
)
x = arg min ∑ ∑ logσi +
( )
The method of penalizing differently on the edge strength
Ω1,....Ωn i=1 s=Ωi  2 σi
2

 has the shortcoming of bias on certain classes since
Whereη is the number of classes Ω ,....., Ω are the obtained boundaries between some classes can be generally weaker
1 n than those of other. However, including edge strength is
Classes, µ i is the mean grey level of class Ω i , σ i is standard advantageous in describing local behaviors, and will also
show next that the bias problem can be alleviated by
deviation of class Ω i ,and U i (s ) is the number of neighbor properly manipulating the edge penalty function g(.).
of s that belongs to Ω i . b. Edge Penalty Functions
The edge penalty function as visualize in figure 2. Where
2.2.1Edge Penalty Based Segmentation. the function g(.) can be any monotonically decreasing
a. Objective Function using Edge Penalty function, so that the greater the edge strength the smaller
the penalty. Suppose the grading ∇ (Y) has been
To estimate the mean and standard deviation for equation
(4), most approaches assume that the number of classes is
known and solve the Gaussian mixture by E-M based normalized. Them, the penalty function can be formulated
method. The estimation uses global statistics and does not as:
−( ∇ (Y) /K )
describe local region well if the image is highly non- 2
stationary. Incorporating local statistics is thus necessary. (
g ∇ (Y) = e) (8)
Consider (4) from another point of view. Maximization of The parameter K in figure 2 defines how fast the edge
(4) is equivalent to the minimization of penalty decays with the increase of edge strength. As K

( ) [
Y(s) − µi ]
 n 2  (figure 2) increases, the penalty difference between weak
' 
− βUi (s) (5)
)
x = argmin ∑ ∑ logσi + and strong edges decreases. When K approaches infinity, all
Ω1,...Ωni=1 s=Ω1
 2 σi
2
( )  edge penalties are equally 1. Therefore, the parameter K

(figure 2) of the edge penalty function g(.) is allowed to
increase during the segmentation, details of which are
described the local feature of edge strength is hence utilized,
but the bias is significantly reduced.
3. Proposed HYBRID Algorithm
A new hybrid algorithm has been proposed to obtain the
MAP estimate. In the hybrid algorithm SA algorithm is first
run for some pre specified amounts of epochs and then ICM
algorithm is run until the stopping criterion has been
satisfied. The combination of global and local concept leads
to a faster converting algorithm. The proposed hybrid
algorithm is described below
1) Initialized the temperature Tin .
2) Compute the energy U of the configuration.
3) Perturb the system slightly with suitable Gaussian
distribution.
' (EPF) while for MRF model. This is mission for many other
4) Compute the new energy U of the perturbed system and
'
evaluate the change in energy ∆U =U −U .
5) If ( )
∆U 〈 0 , accept the perturbed system as the new
configuration Else accept the perturbed system as the
(
new configuration with a probability exp − ∆U / t ,
where t is the cooling schedule.
6) Decrease the temperature according to the cooling
schedule.
7) Repeat steps 2-7 till some pre specified number of epochs
of SA.
8) Compute the energy U of the configuration.
9) Perturb the system slightly with suitable Gaussian
disturbance.
' edge penalty functions. As fig. 4(e) correspond to MRF with
10) Compute the new energy U of the perturbed system
and evaluate the change in energy ∆U = U −U .
( )
11) If ∆U 〈 0 , accept the perturbed system as the new
configuration otherwise retain the original
configuration.
12) Repeat steps 8-12, till the stopping criterion is met. The
(
stopping criterion is the energy U 〈 threshold .
4. Simulation and Results
In this paper we have considered three different image
having strong as well as weak edges. These images are
considered deliberately to validated the weak edge
preserving attributes of the schemes. The first image
considered is a cell the image as shown in fig. 3(a), fig. 3(b)
to fig.
3(g) show the results obtained using MRF, DMRF, and
JSEG methods. Although line field is used to preserved the
edges, the weak edges inside the cell boundary are not being
preserve. As observed from fig. 3 some hazy edge could be
formed because of weak edges, still there are no clear edge
for present inside each cell. Fig.
(e) and Fig. 3(f) shows the result obtained with MRF and
DMRF models with edge penalty function incorporated in
the energy functions. We have also proposed Exponential
edge penalty function. Fig. 3(c) and Fig. 3(d) shows the
(IJCNS) International Journal of Computer and Network Security, 157
Vol. 2, No. 5, May 2010
result obtain MRF and DMRF model. The models
parameters are tabulated in table (II). For exponential edge
penalty function (EPF) the value of K is choose to be σ =5.
It is observed that interior edges could be observed and
percentage of misclassification error is less than that of
using edge penalty functions. The results has been improve
in case of DMRF model as compared to MRF model. As
seen from the table (I), the misclassification error is less in
case of DMRF model as compared to MRF. Use of
Exponential EPF preservation of weak edge. Fig. 3(e) and
Fig.3(c) shows the image of MRF model with exponential
EPF yields better results than that of MRF model. As seen in
Fig. 3(e) and Fig. 3(f) the image of MRF and DMRF model
with the Exponential edge penalty function as shows that
the images of DMRF model with Exponential edge penalty
function yields better results than that of MRF exponential
penalty function, As seen from fig. 3(e) and 3(f) the nearer
the diseased part (toward the right hand top) the inside edge
has been formed using Exponential edge penalty function
cells similar observation can also be made. In this also we
have employed SA and proposed hybrid algorithm. The
convergence of these algorithm are presented in fig. 6
where we observe that the hybrid algorithm converges at
) around 600 iteration where as SA converges at 6000
iterations. It is also observed that Hybrid algorithm is fast
Converges than the SA algorithm.
The image consider in shown in fig. 4, where are some weak
edge seen in the water boat etc. fig. 4 shows the results
obtained using MRF and DMRF. As seen from fig. 4(c) to
(d) the weak edge have not been presented. In case DMRF
there are more misclassification as seen from fig. 4 show the
results obtained using MRF and DMRF with exponential
exponential, observed the weak edge of the bottom left side
'
have been preserved and strong have been formed. This
phenomenon is more prominent in case of second order
DMRF model as shown in fig. 4(f). As found from Table (I)
the percentage of misclassification error is less than that of
using Exponential edge penalty functions. As shows the
) convergence curves for hybrid and SA algorithm converges
at 600 iteration where as SA converges after 6000 iterations.
Similarly observation are also made for the third images
shows in fig. 5(c) and 5(d) shows the bird image and the
corresponding ground truth observed the weak edge could
not be preserved. As seen from fig. 5(e) and 5(f), the MRF
and DMRF model with Exponential edge penalty function
performed better than that of MRF and DMRF model . In
this case also the weak edge would be preserved the model
parameter are tabulated in Table (II). Hence here also hybrid
algorithm converged faster than that of the SA algorithm.
158
Table 1: comparison of %age of misclassification error for
segmentation results of four weak image and scene
images
Table 2: parameters DMRF using gradient exponential
penalty function model
Figure 3.Segmentation of cell image of size 300x200 using
weak membrane model and RGB color model: (a) Original
image (b) Ground Truth (c) MRF using hybrid (d) DMRF
using hybrid (e) MRF using Exponential Penalty Function
image (f) DMRF using Exponential Penalty Function image
(g) JSEG image.
Figure 4.segmentation of water boat image of size 200x300
using weak membrane model and RGB color model: (a)
Original image (b) Ground image (c) MRF using hybrid (d)
DMRF using hybrid (e) MRF using Exponential Penalty
Function image (f) DMRF using Exponential Penalty
Function image (g) JSEG image.
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
Figure 5. Segmentation of Poster image of size 200x248
using weak membrane model and RGB color model: (a)
Original image (b) Ground Truth (c) MRF using hybrid (d)
DMRF using hybrid (e) MRF using Exponential Penalty
Function image (f) DMRF using Exponential Penalty
Function image (g) JSEG image.
Figure 6. Energy Convergence curve comparison for Cell
image: DMRF model with SA and Hybrid algorithm for
whole process
5. Conclusion
In this paper the color image segmentation problem is
addressed using MRF model. We have employed two color
models namely RGB and OHTA model. A new MRF model
called DMRF model is proposed to take care of intra color
plane and inter color plane interactions. In RGB and OHTA
model, the inter-plane correlation are decomposed and
partail correlation has been introduced due to the DMRF
model. The associated model parameters are selected on
trail basis. The segmentation problem is formulated using
MAP estimates. It is observed that the hybrid algorithm
faster than that of using SA algorithm. The algorithm with
different color modeled image model was tested on a wide
variety of images. It is observed that the segmentation result
preserved the strong edge but failed to preserve weak edges.
In order to protect edges, we have introduced the edge
penalty function of Qiyao Yu et al [12] in the clique
potentail function of the a priori model. In this case also we
have used DMRF model with RGB color model. We have
used the exponential edge penalty function of Qiyao Yu et al
and found the results could preserve weak edges. The model
parameters are selected on trail and error basis. We have
proposed Exponential edge penalty function and it is found
that use of DMRF model with proposed function produced
 (IJCNS) International Journal of Computer and Network Security, 159
Vol. 2, No. 5, May 2010

better result than that of using DMRF model. We have
employed hybrid and SA to obtain segmentation and hybrid
algorithm converges faster than that of SA algorithm.
Current work focuses estimation of model parameters and
devising an unsupervised scheme.
Classification and Region Merging,” Proc of the Second
Canadian Conference on Computer and Robot Vision
volume 1(May 2005):pp.579-586.

References

[1] L. Lucchese and S.K. Mitra , “Color Image

Segmentation: A state of the art survey”, Proc of the
Indian National Science Academy , volume 67, No. 2,
(feb 2006): pp. 207-221
[2] Chen H.D., Jiang X.H., Sun Y., and Wang J., “Color
Image Segmentation: advances and prospects,”Pattern
Recognition, volume 34, 2000: pp.2259-2281.
[3] Gonzalez R.C. and Woods R.E. Digital image
processing. Singapore: Pearson Education, 2001.
[4] Stan Z. Li, “Markov Field Modeling in Image analysis,”
Japan: Springer, 2001.
[5] Geman. S and Geman D, “ Stochastic relaxation,
Gibbs distributions, and the Bayesian
Restoration of images, “IEEE Transaction on Pattern
Analysis and Machine Intelligence, volume 6, No. 6,
November 1984: 721-741.
[6] Kato. Z Pong. T. and Lee J.C., “Color Image
Segmentation and Parameter Estimation in a
Markovian framework,” Pattern Recognition Letters,
volume 22, No. 2, 2001: pp. 309-321.
[7] Kato. Z. and Pong. T. , “ A Markov Random Field
Image Segmentation model using Combined Color
and Texture Feature, “ Proceeding of International
Conference on Computer Analysis of Image and
Patterns. (September 2001): pp. 547-554.
[8] Kirkpatrick S., Gelatt C.D., and Vecchi M.P.,
“Optimization by Simulated Annealing,” Science,
volume 220, No. 4598, 1983: pp. 671-679.
[9] Panjawani D.K. and healey G., “Markov Random Field
Models for Unsupervised Segmentation of Texture
Color Images.” IEEE Transaction on Pattern Analysis
and Machine Intelligence, volume 17, No.10,(October
1995): pp. 939-954.
[10] P.K. Nanda, MRF Model Learning and its Application
to Image Restoration and Segmentation
PhD Dessertation, Electrical Engg. Department, IIT
Bombay 1995
[11] Qiyao Yu, David A . Clausi, “IRGS: Image
Segmentation Using Edge Penalties and Region
Growing,” IEEE Transaction on Pattern Analysis and
Machine Intelligence PAMI.
[12] Qiyao Yu, David A. Clausi, “Combination Local and
Global Feature for Image Segmentation Using Iterative
160 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

Performance Evaluation of a Reed-Solomon

Encoded Multi-user Direct Sequence Code Division
Multiple Access (DS-CDMA) Communication
System
Nargis Parvin, Farhana Enam, Shaikh Enayet Ullah and Md. Saifur Rahman
Department of Information and Communication Engineering
Rajshahi University, Rajshahi-6205, Bangladesh.

Abstract: This paper incorporates a comprehensive study on the
performance evaluation of a Reed-Solomon encoded multi-user
Direct Sequence Code Division Multiple Access (DS-CDMA)
communication system in terms of Bit Error Rate (BER). The DS-
CDMA system under present study employing uncorrelated
spreading codes has incorporated BPSK digital modulation over
an Additive White Gaussian Noise (AWGN) and other fading
(Rayleigh and Rician) channels. Computer simulation results on
BER demonstrate that the Reed-Solomon encoded multi-user DS-
CDMA system outperforms in retrieving the transmitted messages
of the individual senders .The system is highly effective to combat
inherent interferences under Rayleigh and Rician fading channels.
It is anticipated from the study that the performance of the
communication system degrades with the lowering of the Signal-
to-Noise Ratio (SNR).
Keywords: Direct Sequence Code Division Multiple Access
(DS- CDMA), Signal-to-Noise Ratio (SNR), Additive White
Gaussian Noise (AWGN), Fading channels.
1. Introduction
Direct Sequence Code Division Multiple Access (DS-
CDMA) is one of the most prominent air-interface
techniques in today’s wireless communication systems [1].
In a DS-CDMA system all users transmit in the same
bandwidth simultaneously [2]. This transmission technique
is a multiple access communication scheme in which each
user is assigned a unique spreading code for modulating the
transmitted symbols. Thus, a receiver with knowledge about
the code of the intended transmitter can select the desired
signal. A DS-CDMA communication system receiver has
generally three main obstacles to overcome such as Multiple
Access Interference (MAI) from other users, Inter-symbol
Interference (ISI) due to a multipath channel and additive
noise at the receiver. In a DS-CDMA system, the transmitter
multiplies each user’s signal by a pseudo-random spreading
sequence [3], [4]. In fact, user’s spreading codes are
designed to be orthogonal (or quasi-orthogonal). However,
due to channel properties like multipath propagation,
orthogonality cannot be maintained at the receiver and the
signals interfere with each other. In practical CDMA
systems the spreading codes among the users are not
perfectly orthogonal. Thus, a CDMA channel with randomly
and independently chosen spreading sequences accurately
models the situation where pseudo-noise sequences span
many symbol periods [5].
Direct Sequence Code Division Multiple Access (DS-
CDMA) has already been established as a strong candidate
for third-generation mobile and personal communication
systems. Now a days, the most prominent applications of the
CDMA are in mobile communication systems like cdmaOne
(IS-95), UMTS or cdma2000. To apply CDMA in a mobile
radio environment, specific additional methods are required
to be implemented in all these systems. Methods such as
power control and soft handover have to be applied to
control the interference by other users and to be able to
separate the users by their respective codes [6], [7].
2. System Description
We assume that a multi-user Direct Sequence Code Division
Multiple Access (DS-CDMA) communication system
depicted in Figure 1 utilizes a Reed-Solomon channel
coding scheme. In transmitting section of the present Reed-
Solomon encoded DS-CDMA system, a typically named b
source information bits of rate rb are grouped and mapped to
one of the M= 2b information symbols. The information
symbols with rate of r s= rb/ log2M ks/s (ksymbols/s) enter a (n,
k, t) Reed–Solomon encoder operating in GF (M) with n, k,
t=[n-k]/2. The output of the Reed–Solomon encoded
symbols is of rate rsc= rs/rc, where, rc= k/n. The coded
symbols are fed into symbol to binary converter and its
output are then converted into bipolar NRZ format and
subsequently fed into BPSK digital modulator. The digitally
modulated signals are then multiplied with the generated
pseudo-random noise for each user. The pseudo randomly
noise contaminated modulated signals for user1 and user2
are then added up and passed through AWGN or fading
channel. The channel output is then multiplied with the
generated pseudo-random noise and demodulated. The
demodulated signal for each user is processed through
integrator. The integrated output is sent up to the decision-
making device and eventually, the information bits
 (IJCNS) International Journal of Computer and Network Security, 161
Vol. 2, No. 5, May 2010

transmitted for each user are recovered [8], [9]. Figure 4 respectively. In Figure 4, it is quite evident that the
amplitude of the signal is maximum around the frequency of
RF Direct sequence 100 KHz. Figure 5 demonstrates the simulation results for
carrier generator for user #1
uncoded multi-user DS-CDMA communication system
under an AWGN channel. It is observable that its value
Mapping into
Binary data Reed- bipolar NRZ BPSK
approaches zero for a SNR value of 5 dB. In Figure 6, the
for user #1 Solomon 1 to +1 modulator BER simulated graphs under AWGN and Rayleigh fading
Encoder 0 to –1
channels are illustrated for user 1 and user 2.
RF Direct sequence
carrier generator
for user # 2

Mapping into
Binary data Reed- bipolar NRZ: BPSK
for user #2 Solomon modulator
1 to +1
Encoder 0 to –1

Direct sequence
generator for user # 1 AWGN/
Sampling Fading
channel
Detection BPSK
Device Integrator Demodulator

Reed-Solomon Figure 2. A segment of original binary bit stream of

Retrieved data RF
Decoder individual user.
carrier
Sampling d
BPSK
Detection Integrator Demodulator
Device

Reed-Solomon Retrieved data RF Direct sequence

Decoder carrier generator for user # 2

Figure 1. Block diagram of a Reed- Solomon Encoded

Multi-user DS-CDMA system

The present study has been made on the consideration of

model parameters shown in Table 1. Figure 3. Received Signal with Additive White Gaussian
noise and Rayleigh fading effects incorporated.

Table 1: parameters used in system simulation

No. of bits used for synthetic data 1,912
Bit rate 1,000 bits/sec
Energy per bit (eb) 0.5
Chip rate 10,000
chips/sec
Channel Coding Reed-
Solomon
Doppler Frequency 100 HZ
Carrier frequency 5 MHz
K-factor used in Rician fading 10
SNR 0 to10 dB
Figure 4. Amplitude Spectrum of noise contaminated
received signal with added Rayleigh fading channel effect.

3. Results and Discussion

Figure 2 shows a segment of original binary bit stream of
individual user. The noise contaminated received signals for
a typical SNR value of 5 dB in Rayleigh fading channel
and its amplitude spectrum are illustrated in Figure 3 and
162 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

Figure 5. BER Performance of the uncoded multiuser DS- Figure 8. BER Performance of the Reed- Solomon
CDMA communication system under AWGN channel. encoded multi-user DS-CDMA communication system
under AWGN channel.

The BER performance of Reed-Solomon encoded multi-user

DS-CDMA communication system under AWGN and
Rayleigh fading channels is shown in Figure 9. Similarly,
the system performance under AWGN and Rician fading
channels is shown in Figure 10. In comparison of BER
values for a typical SNR value of zero dB (Figure 9 and
Figure 10), it is found that the bit error rate for Rayleigh and
Rician fading channels are 0.0455 and 0.011 respectively.
This is due to the fact that in case of Rician fading, the
direct component of transmitted signal is more prominent as
compared to multipath components and as a consequence of
which the bit error rate is reduced with enhancement of
system performance.
Figure 6. BER Performance of the uncoded multiuser DS-
CDMA communication system under AWGN and
Rayleigh fading channels

In Figure 7, the BER simulated graphs under AWGN and

Rician fading channels are illustrated for user 1 and user 2.
The bit error rate approaches zero at 3 dB and 4 dB for user
1 and user 2 respectively. The BER performance of multi-
user DS-CDMA communication system on deployment with
Reed-Solomon channel coding scheme under AWGN
channel is shown in Figure 8. It is observable that the bit Figure 9. BER Performance of the Reed- Solomon
error rate approaches zero at 2 dB for both user 1 and user 2 encoded multi-user DS-CDMA communication system
viz. the system provides better performance as compared to under AWGN and Rayleigh fading channels.
uncoded situation.

Figure 10. BER Performance of the Reed- Solomon

Figure 7. BER Performance of the uncoded multi-user encoded multi-user DS-CDMA communication system
DS-CDMA communication system under AWGN and under AWGN and Rician fading channels.
Rician fading channels.
 (IJCNS) International Journal of Computer and Network Security, 163
Vol. 2, No. 5, May 2010

4. Conclusions
In our present study, we have studied the performance of a
Authors Profile
multi-user Direct Sequence Code Division Multiple Access Nargis Parvin received the B.Sc. and M.Sc. degree in Information
(DS-CDMA) communication system adopting the Reed- and Communication Engineering from University of Rajshahi,
Solomon channel coding and BPSK digital modulation Bangladesh.
scheme. A range of system performance results highlight the
impact of a simplified digital modulation technique and Farhana Enam received the B.Sc. and M.Sc. degree in Computer
different uncorrelated spreading codes. In the context of Science from University of Rajshahi, Bangladesh.. She is currently
system performance, it can be concluded that the working as Assistant Professor of the dept. of Information and
implementation of BPSK digital modulation technique in Communication Engineering, University of Rajshahi, Bangladesh.
Reed-Solomon channel encoded multi-user Direct Sequence
Shaikh Enayet Ullah received the B.Sc. and M.Sc. degree in
Code Division Multiple Access (DS-CDMA)
Applied Physics from University of Rajshahi, Bangladesh.. He has
communication system provides satisfactory performance in completed his PhD from Jahangirnagor University, Bangladesh. He
retrieving the transmitted messages of the individual users. is currently working as a Professor and Chairman of the dept. of
Information and Communication Engineering, University of
Rajshahi, Bangladesh.
References
Saifur Rahman received the B.Sc. and M.Sc. degree in
[1] W. C. Y. Lee, “Overview of cellular CDMA,” IEEE Information and Communication Engineering from University of
Transactions on Vehicular Technology, vol. 40, no. 2, Rajshahi, Bangladesh.
pp. 291–302, 1991.
[2] P. Jung, P. W. Baeir, and A. Steil, “Advantages of
CDMA and spread spectrum techniques over FDMA
and TDMA in cellular mobile radio applications,” IEEE
Transactions on Vehicular Technology, vol. 42, no. 3,
pp. 357–364, 1993.
[3] R L. Pickholtz, D. L. Schilling, and L. B. Milstein,
“Theory of spread-spectrum communications” A
tutorial, IEEE Transactions on Communications, vol.
30, no. 5, pp. 855–884, 1982.
[4] R. L. Pickholtz, L. B. Milstein, and D. L. Schilling,
“Spread spectrum for mobile communications,” IEEE
Transactions on Vehicular Technology, vol. 40, no. 2,
pp. 313–322, 1991.
[5] F. C. M. Lau and C. K. Tse, “Chaos-Based Digital
Communication Systems: Operating Principles,
Analysis Methods, and Performance Evaluation,” First
edition, Springer, 2003.
[6] K.S. Gilhousen et al., “On the Capacity of a Cellular
CDMA System”, IEEE Trans. Vehic. Technol., Vol. 40,
no. 2, pp. 303–312, 1991.
[7] M.G. Jansen and R. Prasad, “Capacity, Throughput, and
Delay Analysis of a Cellular DS CDMA System With
Imperfect Power Control and Imperfect Sectorization”,
IEEE Trans. Vehic. Technol., Vol. 44, no. 1, pp. 67–75,
1995.
[8] A.C. Iossifides and F-N. Pavlidou, ‘’ Performance of
RS-Coded DS/CDMA Micro cellular Systems with
M-ary orthogonal Signaling,” Wireless Personal
Communications 10: pp. 33–52, Netherlands, 1998.
[9] Rappaport, S.Theodore,“Wireless Communications,”
Principle and Practice, Second Edition, Pearson
Educatiob (Singapore) Pte., Ltd.,New Delhi,India,
2004.
164
Phase Information Extraction of Atrial Fibrillation
Using Wavelets
1
V.Bindhu 2G.Ranganathan 3Dr.R.Rangarajan
1
Asst.Professor, PPG Institute of Technology, Affiliated to Anna University Coimbatore, India
2
Associate Professor, RVS School Of Engineering and Technology, Affiliated to Anna University Coimbatore, India
3
Principal, Sri Ramakrishna Engineering College, Affiliated to Anna University Coimbatore, India
bindhurang2002@yahoo.co.in, ranganathang.samy@gmail.com
Abstract: A novel method for characterization of phase
information in atrial fibrillation (AF) is presented. The method
decomposes atrial activity into fundamental and harmonic
components, dividing each component into short blocks for
which the amplitudes, frequencies, and phases are estimated.
The phase delays between the fundamental and each of the
harmonics, here referred to as harmonic phase relationships,
are used as features of f-wave morphology. The estimated waves
are clustered into typical morphologic patterns. The
performance of the method is illustrated by simulated signals,
ECG signals recorded from 36 patients with organized AF, and
an ECG signal recorded during drug loading with flecainide.
The results show that the method can distinguish a wide variety
of f-wave morphologies, and that typical morphologies can be
established for further analysis of AF.
Keywords: Atrial Fibrillation (AF), Electrocardiogram
(ECG), Reverse biorthogonal wavelets, waveform
characterization.
1. Introduction
ATRIAL fibrillation (AF) is a heterogeneous
disease considered to be the resulting end state of several
mechanisms. Initiation of AF typically requires both a
triggering event and an arrhythmia-prone anatomical
substrate [1]–[3]. Suggested triggers include abnormal
activity of the autonomic nervous system and rapidly firing
foci located in the pulmonary veins. The arrhythmia-prone
substrate may result from, e.g., impaired impulse
propagation, changes in tissue structure, or alternations in
atrial electrical properties induced by the arrhythmia itself.
AF has mainly been viewed as an acquired disorder related
to structural heart disease in patients with other
cardiovascular problems, including coronary artery and
mistral valve diseases and hypertension [4]; however, AF is
now recognized to also have a heritable component [5].
Characterization of AF based on ECG recordings calls for
separation of atrial and ventricular activity. Several
approaches to such separation have been suggested based
on, e.g., principal component analysis [6], [7], independent
component analysis [8], and patio temporal modeling [9],
[10]. The most important aspect of AF pattern analysis has,
so far, been to determine the fibrillatory rate. This rate has
potential value for monitoring spontaneous and autonomic
maneuver-induced changes in atrial electrophysiology,
monitoring and prediction of anti arrhythmic drug responses
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
as well as for prediction of atrial defibrillation thresholds
and AF recurrence following cardio version . AF with
slower rates seems more likely to terminate spontaneously
or to respond to anti-arrhythmic drug therapy whereas faster
rates are more often found in persistent and drug- and
cardio version-refractory AF . Recently, methods for
characterizing temporal AF dynamics have been developed
for studying mechanisms and treatment strategies. Using
time-domain methods, the properties of f-waves have been
investigated in terms of the plane of best fit and the spatial
phase properties of different ECG planes . Time–frequency
analysis, combined with frequency alignment of an
adaptively updated spectral profile, has been used to track
variations in AF frequency and f-wave morphology.
Morphologic characterization was based on the harmonics
pattern, expressed as the exponential decay of the
harmonics’ magnitude. The significance of AF harmonics
has been paid little attention in the literature, one reason
being that spectral analysis is almost invariably based on
longer segments (> 30 s). As a Consequence, each segment
contains large temporal variations in AF frequency that
smears the harmonics. The aforementioned time–frequency
analysis is particularly well suited to handle such variations
in AF frequency as spectral alignment is performed,
facilitating the detection and characterization of harmonics
[23]. Several important questions related to the
interpretation of AF patterns in the ECG remain
unanswered, including how the patterns can be used to
determine disease state and predict treatment outcome in
individual patients. Other questions relate to the meaning of
f-wave morphology in terms of intra-atrial activation
patterns and to signal-based identification of AF subgroups.
The use of AF frequency alone does not answer these
questions. In this paper, we introduce a novel approach to f-
wave factorization based on signal phase analysis. The
method has a two-stage structure in which the atrial activity
is first translated into compact sets of features, which are
then clustered so that the typical waveform patterns can be
discerned. The morphologic parameters of the present
method offer information complementary
to the atrial rate and are not intended as a replacement to
rate. The paper is outlined as follows: the method is
presented in Section II, simulated and ECG signals in
Section III, the results are presented in Section IV, and a
discussion of the method and its performance is found in
Section V.
 (IJCNS) International Journal of Computer and Network Security, 165
2. Methodology
The method assumes that an atrial signal is
obtained from the original ECG, produced here by
spatiotemporal QRST cancellation [9]. Subsequent analysis
is divided into two stages of which the first segments the
atrial signal into short blocks and extracts features from
each block that characterize the f-waves. Feature extraction
involves signal decomposition into bandpass components
that are used for estimation of AF frequency, amplitude, and
phase. In the second stage, the resulting features are
clustered into different waveform patterns.
Fourier analysis, using the Fourier transform, is a
powerful tool for analyzing the components of a stationary
signal (a stationary signal is a signal that repeats). For
example, the Fourier transform is a powerful tool for
processing signals that are composed of some combination
of sine and cosine signals.
a) Acquisition system
A total of 56 chest and back leads were acquired
simultaneously for each subject in addition to the standard
limb leads. The predefined site walls are shown in Figure
2.Chest leads (N=40) were arranged as a grid around V1
with an inter-electrode distance of 2.2 cm. (See Figure 2)
while back leads (N=16) were arranged in a similar fashion
around V1post (placed on the back at the same level than
V1). We designed a new belt for specifically attaching the
electrodes in the correct position on the chest and back of
the patient. Perforations are applied to the belt according to
the electrode grid designed. Elastic bands cover the
perforations providing fixation of the electrodes and
applying the needed pressure. The front and back parts of
the belt are joined with stripes of adjustable length, adapting
to the shape of the patient. This attachment system allows
repeatability in the measurements and provides a good
contact of the electrodes with the skin of patient, reducing
movement artefacts.Signals were acquired at a sampling rate
of 2048 Hz, with a resolution of 1 microvolt and a
bandwidth of 500 Hz. Before acquisition, signal quality of
all leads was visually inspected and an ECG recording of 10
minutes was stored for off-line processing.
Figure 1. Predefined sites at the right atrial-free wall
Vol. 2, No. 5, May 2010
Figure 2. Arrangement of the electrodes and belt used for
their attachment to the patient.
b) ECG Signal Processing
ECG signals were processed using Mat lab 7.0.1
(The Math works Inc, The Netherlands). Baseline
wandering was reduced by subtracting the baseline to the
recording, calculated by using a 3rd order low-pass
Chebyshev filter. Signals were low-pass filtered in order to
avoid my electric interference by using a low-pass
Butterworth filter (fc=40 Hz) and down sampled to 512 Hz
in order to speed up the calculations. Only long segments
free from ventricular content were included in the study. For
that purpose, the peaks of QRS complexes were detected in
V1. RR intervals longer than 950 ms were first selected and
a TQ segment was defined as starting 400 ms after the QRS
peak and ending 150 ms before the next QRS peak. All TQ
segments were visually inspected in order to avoid wrong
detections or inclusion of T segments. After this step, only
segments free from ventricular activity and longer than 400
ms were considered. Typically, one to three leads presented
noticeable baseline wandering which was comparable or
greater than the amplitude of the AF signal. In order to
avoid this baseline wandering, AF segments were filtered
again. First, the mean value in each lead was subtracted and
the baseline was again calculated and subtracted. For the
estimation of the baseline, segments were down sampled
(fs=128 Hz) and low-pass filtered (fc=3.3 Hz). Then, the
baseline was up sampled to 512 Hz and subtracted to the
segment. After baseline wandering correction, AF segments
were low-pass filtered (fc=20Hz). All leads in all segments
were visually inspected. Leads presenting noticeable noise
contributions (typically, a transient loss of contact in one
electrode) were discarded and interpolated from its
neighboring electrodes by cubic spline interpolation. No
segment presented more than three discarded leads, and we
considered that the loss of signal content of less than 4
electrodes out of the 56 available was acceptable. Once AF
segments had been isolated and processed, maps for each
time instant belonging to the each segment were
constructed. For that purpose, potentials at a given time
instant were arranged in two matrices: one for the front
(5x8) and one for the back (4x4). In such matrices, signals
recorded from neighboring electrodes are placed together. In
order to obtain a smooth representation of the maps, finer
matrices for the front (50x80) and the back (40x40) were
created. Potentials for unknown positions were interpolated
by cubic spline interpolation. Two display modes were
implemented: the is potential mode and the waterfront
propagation mode. In the is potential mode, voltages at a
166 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

given time instant were represented according to a color both differ greatly or slightly from one activation to another.
scale. Videos of each AF segment were generated. In the By application of the state-space realization algorithm to
wave front propagation mode, only lines connecting points every heart beat with a period of approximately 1 s and 175
on the surface with a voltage equals to zero were ms during atrial fibrillation, it was possible to reproduce the
represented. This display mode allows the visualization of electro gram impulse response which, in turn, makes
both the polarization and re polarization wave fronts possible a quantitative comparison as to the variation and
depending on the time interval selected. The succession of complexity of a set of consecutive heart beats. The wavelet
wave fronts is represented using a color scale according to decomposition of scddvbrk signal is shown in Figure 3.
the time instant in which each wave front appears. We used
a method as previously reported [16]. After removal of a 50-
Hz disturbance component emanating from power supplies,
and after subtraction of the mean signal level of all
recordings, data reduction by means of low-pass filtering,
and resembling at rate of 100 Hz, experimental data from
the 78 electrode array were arranged in impulse response
matrices of dimensions the number being the number of data
points in each recording between two consecutive heart
beats. Subsequent impulse response analysis and simulation
of the resulting state-space realization were applied as
described in next section.

3. Results and Discussion

Maps were generated for all the segments

considered in the study and visually inspected. Single or
Figure 3. wavelet decomposition of scddvbrk signal
multiple simultaneous wave fronts, wave breakages and
disorganized electrical activity could be observed. Figure 2
Relative prediction error versus model order and
shows some selected wave front propagation maps
the singular value sequences indicate a model order to be
representative patterns observed. In panel 1a, a succession of
relevant for the SR and the atrial pacing, whereas a model
wave fronts from patient 3 is observed. A single wave front
order is relevant for atrial fibrillation. As for the temporal
can be observed in both panels 1a and 1b. In panel 1a wave
and channel-dependent autocovariances and cross
fronts are mostly parallel to each other. Panel 1b shows the
covariance of data and residuals, it is apparent that atrial
next succession of wave fronts, which differ from panel
fibrillation data exhibit other temporal covariance than data
1a.Panels 2a and 2b show two consecutive depolarization
of SR and of artificial atrial pacing data. The phase curve
waves from patient 3 which present a similar pattern.
extraction of standard signals like scddvbrk, freqdbrk and
Surface wave fronts rotate along a similar axis in both
cuspamax provides useful information in the atrial
panels. The same pattern was observed in the consecutive
fibrillation. The phase curve extraction using freqdbrk
wave fronts appearing in the same segment (total=5).Panel 3
signal is shown in Figure 4.
shows two consecutive depolarization waves from patient 5.

Depolarization wave shown in panel a is first

shown in the leftmost part on bottom of the grid and spreads
towards the center and top until it gets blocked. A second
depolarization wave starts in the uppermost part on the right
of the grid and propagates towards the bottom and the left
until it gets blocked in roughly the same position where the
previous wave arrived. It can be observed that one wave
front is blocked by the previous propagation contributing to
a chaotic propagation. Panels 4a and 4b show depolarization
waves in patient 6. Multiple waves arise at the same time,
propagate and then disappear in an uncoordinated fashion.

By analyzing the 10-minute recording of each

patient, we observed that the wave front propagation
patterns were, to some extent, repeatable. Patient 3 showed
clear wave fronts with similar patterns in most segments Figure 4. Phase curve extraction using freqdbrk
analyzed. Patients 1, 5 and 6 showed a completely unstable Signal
pattern with multiple simultaneous activation wave fronts
that changed widely even in short time intervals. Patients 2 The phase curve extraction using cuspamax signal
and 4 presented an intermediate degree of repeatability, with is shown in Figure 4.
one or two simultaneous propagation wave fronts that could

Figure 5. Phase curve extraction using freqdbrk
Signal
Modeling: Generally, the purposes of theoretical
models are to pinpoint deficiencies in our knowledge about a
system and to predict properties of the system which
otherwise should not have emerged. In our case we have
studied heart surface properties of impulse propagation as
the impulse response to natural and artificial pacing. Thus,
the nonlinear modeling needed to reproduce the impulse
generation and refractory tissue properties is not included.
As for all behaviorist models, the models obtained here
provide no final or definite answer as to the underlying
physiological mechanisms. Although the linear properties of
the underlying physiological processes might be challenged,
the system identification provides linear approximate
models of finite and low complexity and good prediction
accuracy. As for methodological extensions, the realization
theory applied here lends itself to modeling by linear partial
differential equations by using the spatial information
available in the electro gram data.
The linear models obtained for electro grams of SR,
atrial pacing, and atrial fibrillation have been validated by
means of approximation notions and (preliminary)
extensions of statistical validation properties including cross
validation on data sets not used for system parameter
estimation. The phase curve extraction used above are
applicable to the ECG signal. The raw ECG data is denoised
and applied to phase curve extraction using reverse bi-
orthogonal wavelet is shown below.
1.5
Denois ed E CG s ignal
1 [3] Wavelets and Applications: Wavelet Analysis of
0.5
0
-0.5
-1
0 500 1000
4
3
phas e(rad)
1
0
-1
0 500 1000
Figure 6. Phase curve extraction of ECG signal using
reverse bi-orthogonal wavelet
(IJCNS) International Journal of Computer and Network Security, 167
Vol. 2, No. 5, May 2010
Thus, the models obtained actually provide a means
for prediction of behavior and for classification purposes—
e.g., for distinction among various beat-to-beat behaviors. In
contrast to the hypotheses of Hoekstra et al. [14], no
evidence of chaotic nonlinear dynamics was found. By
application of the state-space realization algorithm to every
heart beat with a period of approximately 1 s and 175 ms,
respectively, it was possible to reproduce the electro gram
impulse response which, in turn, makes possible a
quantitative comparison as to the variation and complexity
of a set of consecutive heart beats (Figs. 2–4). The residual
autocovariance and cross covariance between residuals and
data provide good or excellent model agreement for atrial
fibrillation, artificial pacing rhythm, and SR.
4. Conclusions
A novel method for characterization of phase
information in atrial fibrillation has been presented. Since
phase extraction provides more information about the
signal, we applied it to the various signals and ECG signal.
Different patterns of electrical activation and degrees of
stability of the activation patterns have been observed. The
comparison of this new method for the quantification of the
degree of organization of the atrial electrical activity with
other methods such as spectral analysis and validation by
comparison with invasive mapping data will be of great
interest in future studies. Although observations from the
present study have not been compared to the effectiveness of
any specific treatment, there is a great potential in further
analyzing the activation patterns of the electrical wave
fronts occurring in the atria as observable in body surface
maps.
References
[1] J. Ng, A. V. Sahakian, W. G. Fisher, and S. Swiryn,
“Surface ECG vector characteristics of organized and
disorganized atrial activity during atrial fibrillation,” J.
Electrocardiol., vol. 37, pp. 91–97, 2004.
[2] U. Richter, M. Stridh, A. Bollmann, D. usser, and L.
S¨ornmo, “Spatial characteristics of atrial fibrillation
electrocardiograms,” J lectrocardiol.,
vol. 41, pp. 165–172, 2008.
Asymptotic Signals, Y. Meyer, Ed. New York:
Springer-Verlag, 1991, pp. 13–27.
[4] N. Delprat, B. Escudie, P. Guillemain, and R.
1500 Kronland-Martinet, “Asymptotic wavelet and Gabor
analysis: Extraction of instantaneous
frequencies,” IEEE Trans. Inf. Theory, vol. 38, no. 2, pp.
644–664, Mar. 1992.
[5] G. Erlebacher, M. Hussaini, and L. Jameson, Wavelets:
Theory and Applications. New York: Oxford Univ.
Press, 1996, pp. 115–120.
1500
[6] M. Unser and A. Aldroubi, “A review of wavelets in
biomedical applications,” Proc. IEEE, vol. 84, no. 4,
pp. 626–638, Apr. 1996.
168 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

[7] S. Haddad, J. Karelz, R. Peetersz, R. Westraz, and W.

Serdijny, “Analog complex wavelet filter,” in Proc.
ISCAS, Kobe, Japan, 2005, vol. 4, pp. 3287–3290.
[8] C. Xiangxun, L. Bing, and Z. Bo, “Haar wavelet-based
modified phasor to detect and quantify power quality
disturbances,” Proc. CSEE, vol. 26, no. 24, pp. 37–42,
2006.
[9] X. Zhang and T. Yoshikawa, "Design of orthonormal
IIR wavelet filter banks using allpass filters," Elsevier
Science Signal Processing, vol.78, Nol., pp. 91-100,
1999.
[10] C. Herley and M. Vetterli, "Wavelets and recursive
filter banks,"IEEE Trans. on Signal Processing, vol. 41,
pp. 2536-2555,Aug 1993.

Authors Profile

V.Bindhu received the degree in

Electronics and Communication
Engineering from Govt. College of
Technology, Coimbatore in2002. She is
a research student of Dr.R.Rangarajan.
Currently, she is an Assistant professor
at PPG Institute of
Technology,Coimbatore. Her area of interest includes
Signal processing, and VLSI Design

Prof.G.Ranganathan received the

degree in Electronics and
Communication Engineering from Govt.
College of Technology, Coimbatore in
1998. He is a research student of
Dr.R.Rangarajan. Currently, he is an
Associate Professor at RVS school of
Engineering and Technology. His area
of interest includes Signal processing, Control systems,
Fuzzy logic and Genetic Algorithm.

Dr.R.Rangarajan graduated in
Electronics and communication
Engineering from Madras university in
1972.He took up his academic pursuit in
1975 as a lecturer in coimbatore institute
of technology after completing his post
graduation with second rank from
madras university, he continued to serve at coimbatore
institute of technology and held position as a assistant
professor, professor and HOD of ECE during his tenure. His
area of specialization is VLSI Design and completed his
PhD in Bhartiar university.
 (IJCNS) International Journal of Computer and Network Security, 169
Vol. 2, No. 5, May 2010

Energy efficient wireless networks

1
Dr. Neeraj Tyagi, 2 Awadhesh Kumar,.3 Ajit Kumar, 4 Er. Rati Agarwal
1
Associate Professor, Computer Science & Engg., Mnnit Allahabad (UP)
Email: neeraj@mnnit.ac.in
2
Research Scholar, Computer Science & Engg., Mnnit Allahabad (UP)

Email: awadheshkumar.knit@gmail.com
3
SR.Lecturer, Deptt. Of MCA. SRMSCET, Bareilly (UP)

Email: ajeetkumarsrms@gmail.com
4
Lecturer, Computer Science & Engg., SRMSWCET, Bareilly (UP)

Email: rati1234@gmail.com

Abstract: Several new network types have emerged over the
last couple of years. Many of these networks are being
connected together to provide mobile users with the capability of
staying always connected to the Internet which requires
seamless transitions from one network to another without
causing interruption to on-going connections. To maintain
connectivity during handoff, all the networks that are accessible
to the mobile station need to be known. The Architectural issues
in all-wireless networks are discussed in this regard. The paper
aims to propose different for the energy-efficient network that
enables fast discovery and selection of available access
networks. We suggest three energy efficient techniques for
MANET point of view.
Keywords: energy efficiency, Architecture of wireless
networks, Minimum energy transmission, energy- aware
protocol, power management, power control and topology
control.
1. Introduction
Wireless networking has witnessed an explosion of interest
from consumers in recent years for its applications in mobile
and personal communications. As wireless networks become
an integral component of the modern communication
infrastructure, energy efficiency will be an important design
consideration due to the limited battery life of mobile
terminals. Power conservation techniques are commonly
used in the hardware design of such systems. Since the
network interface is a significant consumer of power,
considerable research has been devoted to low-power design
of the entire network protocol stack of wireless networks in
an effort to enhance energy efficiency. This paper presents a
comprehensive summary of the various approaches and
techniques addressing energy efficient A wireless sensor
network is a network consisting of spatially distributed
autonomous devices using sensors to cooperatively monitor
physical or environmental conditions. As sensor networks
are large in scale, grouping techniques are required to tackle
their scalability issues. The limited energy reserves of the
autonomous devices require grouping techniques to be
energy-aware and strive to extend the life time of the
network. In this paper we propose an energy efficient group
communication technique that aims at reducing the energy
drain of the group leader nodes and the network as a whole.
Energy management remains a critical problem in wire- less
networks since battery technology cannot keep up with
rising communication expectations. Current approaches to
energy conservation reduce the energy consumption of the
wireless interface either for a given communication task or
during idling. However, a complete solution requires
minimizing the energy spent for both communication (i.e.,
for data and control overhead) and idling. This problem can
be expressed as an energy-efficient network design problem,
which is, not surprisingly, NP-hard.
Energy is a critical resource in the design of wireless
networks since wireless devices are usually powered by
batteries. Battery capacity is finite and the progress of
battery technology is very slow, with capacity expected to
make little improvement in the near future. Under these
conditions, many techniques for conserving power have
been proposed to increase battery life. In this dissertation we
consider many approaches for conserving the energy
consumed by a wireless network interface. One techniques
include the use of power saving mode, which allows a node
to power off its wireless network interface (or enter a doze
state) to reduce energy consumption. The other is to use a
technique that suitably varies transmission power to reduce
energy consumption.
2. Architectural Issues in Wireless Networks
The ad hoc wireless networks are quite different from the
cellular systems that have been developed in the commercial
domain. Cellular systems have fixed base stations, which
communicate among themselves using dedicated non-
wireless lines; thus, the only multicast problems that are
new in those systems involve tracking the mobile users.
However, in ad hoc wireless networks it is possible to
establish a wireless link between any pair of nodes, provided
that each has a transceiver available for this purpose and
170
that the signal-to-noise ratio at the receiving node is
sufficiently high. Thus, unlike the case of wired networks,
the set of network links and their capacities are not
determined a priori, but depend on factors such as distance
between nodes, transmitted power, error-control schemes,
other-user interference, and background noise. Thus, even
when the physical locations of the nodes are fixed, many of
the factors that affect network topology are influenced by the
actions of the network nodes (either those directly
participating in a link, or those contributing to the
interference that affects a link). Furthermore, in ad hoc
networks no distinction can be made between uplink and
downlink traffic, thus greatly complicating the interference
environment. In this paper, we focus on wireless networks
in which the node locations are fixed, and the channel
conditions unchanging. The wireless channel is
distinguished by its broadcast nature; when unidirectional
antennas are used, every transmission by a node can be
received by all nodes that lie within its communication
range. Consequently, if the multicast group membership
includes multiple nodes in the immediate communication
vicinity of the transmitting node, a single transmission
suffices for reaching all these receivers. Hence, there is an
incentive to perform a multicast by using high transmitter
power. Of course, doing so results in interference with more
nodes than if reduced power were used. Thus, there is a
trade-off between the long “reach” of a single transmission
and the interference (and/or delay) it creates. Another
undesirable impact of the use of high transmitter power is
that it results in increased energy usage. Since the
propagation loss varies nonlinearly with distance (at
somewhere between the second and fourth power), in
unicast applications it is best (from the perspective of
transmission energy consumption) to transmit at the lowest
possible power level, even though doing so requires multiple
hops to reach the destination. However, in multicast
applications it is not prudent to draw such conclusions
because the use of higher power may permit simultaneous
connectivity to a sufficiently large number of nodes, so that
the total energy required to reach all members of the
multicast group may be actually reduced. Furthermore, even
for unicast applications, the use of lower power (and, hence,
multiple hops) necessitates the complex coordination of
more signals and, therefore, may actually result in higher
total energy expenditure.Figure1 shows an architecture of a
wireless networks.
Figure1. Architecture of a wireless network
2.1 Multicasting in wireless networks
To date, virtually all of the research and development work
on multicasting has centered on tethered, point-to-point
(typically high speed) networks and on methods of
bandwidth efficient maintenance of multicast group
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
addresses and routing trees. There are two basic approaches
to multicast tree construction. The first is the use of Source-
Based Trees (SBT), which are rooted at the sender and
which are designed to minimize the number of
transmissions needed to reach all of the members of the
multicast group. The second is the use of Core-Based Trees
(CBT) [5], under which the same tree is used for all
communication within the multicast group. The Sparse
Mode of the Protocol Independent Multicasting (PIM)
protocol [7] can be used with either SBTs or CBTs, whereas
the PIM Dense Mode is based on the use of SBTs. As
pointed out earlier, the characteristics of the wireless
medium and the ad hoc network architecture may render
multicasting techniques developed for no wireless
applications inappropriate, or at least unable to provide
acceptable performance in some scenarios. There are
numerous and complex issues that must be addressed in
wireless multicasting.
3. Energy Efficient Techniques
In this section we are proposing three energy efficient
techniques for ad hoc network environment. The first
technique minimizes route request message. Second
technique optimizes the transmission power at each node
and third techniques increases network capacity by topology
control mechanism.
3.1. Techniques to Minimize Route Request
Let us consider a multi-hop homogeneous wireless network
in which nodes are randomly deployed over certain
geographical area. In this mobile ad hoc networking
environment each mobile node can access the internet
applications via one or more numbers of gateways. The
mobile nodes communicate with the gateways directly
(single hop) or through multi hop depending upon the
transmission range of the node as shown in the figure 2.
Besides the internet access nodes can also transmit data
among themselves. It is assume that the gateways are
stationary. The geographical area covered by gateway is
partitioned into different logical group with unique group
number assigned by the gateways. Groups are overlapping
and there are some nodes present in the overlapping area.
The partitioned is based upon the number of nodes resent in
it. Nodes are categorized into active node (AN) and common
node (CN). Nodes present in the overlapping area of groups
are called common nodes while nodes belonging to any
particular group called active node of that
group.
Figure2. Single Hop and Multihop Communication with
Gateways
 (IJCNS) International Journal of Computer and Network Security, 171
Vol. 2, No. 5, May 2010

It is assumed that overlapping area of the different group farthest node while latter is the power required to reach the
contains enough number of common nodes (CN) as inter- nearest node. The objective of power control here is to
group communication will takes place through them. When minimize the power consumption of a node. It is assumes
any active node wants to send route request (RREQ) that there may exist some intermediate power level between
message it append its group number in the packet and Pmax and Pmin. Let P(uv) be the power needed to support
broadcast the message. Message will forwarded by the other communication from node u to v, we called
node if they belonging to the same group otherwise message it symmetric if P(uv) = P(vu). The power requirement is
will be dropped. When a CN prepares the RREQ message it called Euclidean if it depends upon Euclidean distance ||uv||.
add one group number from the group it belonging It can be calculated as P(uv)= c+ ||uv|| β where c is a positive
depending upon the share index (SI) calculation. SI is constant real number, and β € [2,5], depends upon
calculated by the CN for the all groups it belongs to and transmission environments. max (u) and Pmin(u) are the
appends the group number to the RREQ based on maximum and minimum power level of node u. When P(uv)
maximization value of SI. The SI is calculated as ≥ Pmin(u) and P(uv) ≤ Pmax(u) then communication
between u and v is possible otherwise it is not possible.
When some intermediate energy efficient path exists
between node u and v via intermediate node w then node u
will transmit with P(uw) rather than P(uv). In the figure 3, if
(1)
P(uv) ≤ [P(uw)+P(wv)] and [P(ux)+P(xv)] then
Where SI =1; when each group contains equal number of
communication from u to v will take place through P(uv)
source node and SI = <1; for other case
otherwise communication will through intermediate node w
Here G is the number of group of the common node and gi
or x by the help of P(uw) or P(ux) respectively.
is the numbers of node present in ith group. The SI
calculations are done for load balancing purpose. When SI
value is one, it indicates groups contain equal number of
nodes. If it is less than one, then groups don’t contain equal
number of nodes. In that case common node will join to that
group which will maximize the SI. The objective of
proposed technique is to reduce the number of route request
by putting restriction on intergroup communication. The
node of one group will not forward the RREQ message to
other group. Only common node will support inter-group
communication to reduce the number of RREQ. The
algorithms for SI calculations and sending procedure are Figure3. Power control technique through intermediate node
given below.

Algorithm ROUTE REQUEST Algorithm POWER CONTROL

Input: (1) Multihop wireless network N
Step1: Calculation SI
Output: power levels p for each node for communicating
to other node
begin
1. for each (u, v) do
(2) 2. calculate Pmin(u) and Pmax(u) /* for node u */
Step2: Procedure SENDING (node) 3. calculate P(uv) by Euclidean distance
4. if P(uw) + P(wv) ≤ P(uv) /* u finds any energy efficient
1. if (node == CN) path to v via w */
2. Calculate SI 5. then transmit with P(uw) for v
3. RREQ = RREQ || gn /* append the group number (gn) 6. else transmit with P(uv) for v
depending upon the max.value of SI */ 7. end if
4. Broadcast (RREQ) End
5. else if ( node == AN)
6. RREQ = RREQ || gn /* RREQ containing group number 3.3. Topology Control Technique to Increase Network
of AN */ Capacity
7. Broadcast(RREQ)
8. end if Topology control can be defined as the techniques by which
9. end if network device takes their own decision regarding their
transmission range, in order to satisfy some network
3.2 Power Control Technique constraints. A network topology can be modeled as a graph
G=(V,E), where V represents the set of nodes and E
In this technique we choose an environment where, nodes represents the set of edges. Cover (vi,vj) means node vj is
are deployed randomly in a two dimensional area. It is within transmission range of vi . Initially all nodes are
assumed that each node has at least two power levels such as transmitting with maximum power Pmax. Depending upon
Pmax and Pmin. Former is the power required to reaches the Pmax value node u, (u€V) has a direct communication
172 (IJCNS) International Journal of Computer and Network Security,
set known as DCS (u). P(uv) denotes the minimum power
required for node u to communicate directly to node v. By
applying topology control we have to get sub graph G'= (V,
E') of G, in G' the node has shorter and fewer numbers of
edges as compare to G. The objective of topology control
here is to remove the energy-inefficient link from the
network. Initially all the node in the network will calculate
their DCS depending upon their maximum transmission
power.
Each node will maintain a node information table
containing information like neighbor_id (NI),
direct_communication_cost(DCC),and
energy_efficient_cost (EEC). Each node has a unique NI.
DCC represents the minimum transmission power required
for a specific neighbor node. DCC of node u to neighbor
node v is represented as DCC(uv) which is same as P(uv)
as described earlier. EEC is initially same as DCC but when
any energy efficient path is obtained through alternate path
ECC is updated. Each node periodically updates their node
information table and broadcast the information to other
node. After a specific time each node will calculate their
DCS by removing energy-inefficient link (if any).
\
Algorithm LINK REMOVAL
Input: (1) Multihop wireless network G=(V,E),
(2) Maximum transmission power
Output: G'= (V, E') , G' has shorter and fewer numbers of
edges as compare to G
Begin
1. Each node broadcast a ―hello" message with its node
information table.
2. If a node u receives the ―hello" message from its
neighbor
3. If u has a power efficient path to node v via k /* k is a
node in alternate path */
4. update (v, P(uv), P(uk) + P(kv) ) into u’s node
information table.
/* energy_efficient_cost =( P(uk) + P(kv )) ≤ P(uv) */
5. Else Insert (v, P(uv), P(uv ) into u’s node information
table./*asenergy_efficient_cost =
direct_communication_cost*/
6. End if.
7. End if.
End
4. Simulation Results
In order to evaluate our power control approach simulation
area of 500mt×500mt were taken. We consider scenario1
with six numbers of nodes and scenario2 with five numbers
of nodes. Nodes are randomly deployed and are in
transmission ranges of each other. In the first scenario
different nodes are placed in different position as shown in
the figure 3. In scenario 1, it is assumed that one hop
communication between sources to destination exists.
It means that source to destination path exist through only
one intermediate node other than direct path. In scenario 2
multi hop communication has been taken into consideration.
In both scenarios node is taken as the source and node 2 is
Vol. 2, No. 5, May 2010
the destination, it is founds that five alternate paths exists in
scenarios 1 and sixteen in scenarios 2. The energy
consumption of each path is measured and it is found that
path through node3 is least. In the scenario 2 five nodes are
placed as shown in the right hand side figure 4. From the
simulation results we measure the energy consumption of all
paths. In both scenarios energy efficient path is the through
node3 although the position of the node3 is different. It is
also observed that in single hop count when the angle of the
intermediate node is more than ninety degree the alternate
path through intermediate node is always energy efficient.
In the first scenario the direct path consumes energy
proportional to 3.200000e+001. and the path through node3
consumes energy proportional to 1.600000e+001.In the
second diagram direct path consume the energy same as first
scenario but the energy efficient path through node3
consumes less which is measured and found proportional to
18. From the both scenario we shows that energy
consumption in intermediate path is better in comparison to
direct path in most of the time.
Figure 4. Energy efficient path through intermediate node
5. Energy Analysis of Routing Protocols
There have been several network routing protocols proposed
for wireless networks that can be examined in the context
of wireless sensor networks. We examine two such
protocols, namely direct communication with the base
station and minimum-energy multi-hop routing using our
sensor network and radio models.
(3)
(4)
In addition, we discuss a conventional clustering approach
to routing and the drawbacks of using such an approach
when the nodes are all energy-constrained using a direct
communication protocol, each sensor sends its data directly
to the base station. If the base station is far away from the
nodes, direct communication will require a large amount of
transmit power from each node (since in Equation 1 is
d
large). This will quickly drain the battery of the nodes and
reduce the system lifetime. However, the only receptions in
this protocol occur at the base station, so if either the base
station is close to the nodes, or the energy required to
receive data is large, this may be an acceptable (and possibly
optimal) method of communication. The second
conventional approach we consider is a “minimum-energy”
routing protocol. There are several power-aware routing

protocols discussed in the literature [6, 9, 10, 1]. In these
protocols, nodes route data destined ultimately for the base
station through intermediate nodes. Thus nodes act as
routers for other nodes’ data in addition to sensing the
environment. These protocols differ in the way the routes
are chosen. Some of these protocols [6, 10, 1], only consider
the energy of the transmitter and neglect the energy
dissipation of the receivers in determining the routes. In this
case, the intermediate nodes are chosen such that the
transmit amplifier energy is minimized; thus node A would
transmit to node C through node B if and only if:
However, for this minimum-transmission-energy (MTE)
routing protocol, rather than just one (high-energy) transmit
of the data, each data message must go through (low n
energy) transmits and receives. Depending on the relative
n
costs of the transmit amplifier and the radio electronics, the
total energy expended in the system might actually be
greater using MTE routing than direct transmission to the
base station. To illustrate this point, consider the linear
network shown in Figure 4, where the distance between the
nodes is r.
Figure 5. Simple linear network.
If we consider the energy expended transmitting a single -
bit message from a node located a distance from the base
nr
station using the direct communication approach and
Equations 1 and 2, we have:
In MTE routing, each node sends a message to the closest
node on the way to the base station. Thus the node located a
distance from the base station would require transmits a
nr n
distance and receives.
r n 1
Therefore, direct communication requires less energy than
MTE routing if:
Using Equations 1 - 6 and the random 100-node network
shown in Figure 3, we simulated transmission of data from
every node to the base station (located 100 m from the
(IJCNS) International Journal of Computer and Network Security, 173
Vol. 2, No. 5, May 2010
closest sensor node, at (x=0, y=-100)) using MATLAB.
Figure 5 shows the total energy expended in the system as
the network diameter increases from 10 m 10 m to 100 m
_ _
100 m and the energy expended in the radio electronics
increases from 10 nJ/bit to 100 nJ/bit, for the scenario where
each node has a 2000-bit data packet to send to the base
station. This shows that, as predicted by our analysis above,
when transmission energy is on the same order as receive
energy, which occurs when transmission distance is short
and/or the radio electronics energy is high, direct
transmission is more energy-efficient on a global scale than
MTE routing. Thus the most energy-efficient protocol to use
depends on the network topology and radio parameters of
the system.
Figure 6. 100node random network.
It is clear that in MTE routing, the nodes closest to the base
station will be used to route a large number of data messages
to the base station. Thus these nodes will die out quickly,
causing the energy required to get the remaining data to the
base station to increase and more nodes to die. This will
create a cascading effect that will shorten system lifetime. In
addition, as nodes close to the base station die, that area of
the environment is no longer being monitored. To prove this
point, we ran simulations using the random 100-node
k
network shown in Figure 4 and had each sensor send a
2000-bit data packet to the base station during each time
step or “round” of the simulation. After the energy
dissipated in a given node reached a set threshold, that node
was considered dead for the remainder of the simulation.
Figure 5 shows the number of sensors that remain alive after
each round for direct transmission and MTE routing with
each node initially given 0.5 J of energy.
Figure 7. Sensors that remain alive (circles) and those that
are dead (dots) after 180 rounds with 0.5 J/node for (a)
direct transmission and (b) MTE routing.
This plot shows that nodes die out quicker using MTE
routing than direct transmission. Figure 6 shows that nodes
closest to the base station are the ones to die out first for
MTE routing, whereas nodes furthest from the base station
174
are the ones to die out first for direct transmission. This is as
expected, since the nodes close to the base station are the
ones most used as “routers” for other sensors’ data in MTE
routing, and the nodes furthest from the base station have
the largest transmit energy in direct communication. A final
conventional protocol for wireless networks is clustering,
where nodes are organized into clusters that communicate
with a local base station, and these local base stations
transmit the data to the global base station, where it is
accessed by the end-user. This greatly reduces the distance
nodes need to transmit their data, as typically the local base
station is close to all the nodes in the cluster.
Thus, clustering appears to be an energy-efficient
communication protocol. However, the local base station is
assumed to be a high-energy node; if the base station is an
energy-constrained node, it would die quickly, as it is being
heavily utilized. Thus, conventional clustering would
perform poorly for our model of micro sensor networks. The
Near Term Digital Radio (NTDR) project [7, 9], an army-
sponsored program, employs an adaptive clustering
approach, similar to our work discussed here. In this work,
cluster-heads change as nodes move in order to keep the
network fully connected. However, the NTDR protocol is
designed for long-range communication, on the order of 10s
of kilometers, and consumes large amounts of power, on the
order of 10s of Watts. Therefore, this protocol also does not
fit our model of sensor networks.
6. Approaches
In this section we discuss several of the multicasting
algorithms we have studied; full descriptions are available
in [10]. In this paper we define the notion of the cost
associated with the support of a multicast tree to be the
power required to reach all destination nodes; thus, it is the
sum of the powers at all transmitting nodes. This is a metric
that is used as the basis of some of our algorithms. However,
performance is always judged by the “yardstick” metric, the
multicast efficiency, and the blocking probability. Each
transmission by a node is characterized by its transmitter
power level, as well as a designation of which (possibly
several) of the nodes receiving this transmission are to
forward it toward which of the ultimate destination nodes.
In all cases, we use greedy algorithms, which attempt to
optimize performance on a “local” (call-by-call) basis.
When the number of transceivers at each node (T) is finite,
it may not be possible to establish minimum-energy trees
(even on a local basis) because of the lack of resources
(transceivers) at one or more nodes. In this case, the greedy
algorithms discussed here are applied to the subset of nodes
that have non-zero residual capacity.
6.1. A unicast-based multicast algorithm
A straightforward approach is the use of multicast trees that
consist of the superposition of the best unicast paths to each
individual destination (see, e.g., [2]). It is assumed that an
underlying unicast algorithm (such as the Bellman–Ford or
Dijkstra algorithm) provides “minimum-distance” paths
from each node to every destination. However, the
minimization of unicast costs does not necessarily minimize
the cost of the multicast tree, as illustrated in figure 5,
which shows a source and two destinations. Figure 5(a)
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
shows the best unicast paths that reach the two destinations,
and figure 5(b) shows the best multicast tree. The use of the
best unicast paths fails to discover the path that reaches a
neighbor of both destinations over a common path, thereby
resulting in lower overall cost. Also, the use of the best
unicast paths fails to incorporate the “multicast advantage,”
which was discussed in section 3. Therefore, the trees
obtained based on unicast information are not expected to
provide optimal multicast performance. Nevertheless, they
do perform reasonably well, and with considerably reduced
complexity as compared to the calculation of
truly optimal multicast trees. Summarizing the above, we
have:
ALGORITHM 1 (Least unicast cost). A minimum-cost
path to each reachable destination is established. The
multicast tree consists of the superposition of the unicast
paths. Paths to all reachable destinations are established,
regardless of the cost required to do so. This algorithm is
scalable.
Figure 9. Unicast-based vs multicast-based trees: (a) best unicast paths,(b) best
multicast tree.
6.2. Algorithms based on pruning MSTs
One approach we have taken in the development of
heuristics for multicasting is the pruning of broadcast
spanning trees. To obtain the multicast tree, the broadcast
tree is pruned by eliminating all transmissions that are not
needed to reach the members of the multicast group. We
noted earlier that, for the case of wired networks, the
determination of minimum-cost broadcast (spanning) trees
is considerably easier than the determination of minimum
cost multicast trees. Nevertheless, the determination of
minimum-cost broadcast trees for wireless networks remains
a difficult problem for which no scalable solutions appear to
be available at this time. In small network examples we
have determined minimum-energy spanning trees by using
the recursive technique of section .3; in moderate to large
networks it is necessary to use heuristics. In this subsection
we discuss the main features of three algorithms that are
based on the technique of pruning.
ALGORITHM 2 (Pruned link-based MST). This algorithm
is based on the use of the standard MST formulation in
which a link cost is associated with each pair of nodes (i.e.,
the power to sustain the link); thus, the “wireless multicast
advantage” is ignored in the construction of the MST. Since
the MST problem is of polynomial complexity, this
algorithm is scalable. To obtain the multicast tree, the MST
is pruned by eliminating all transmissions that are not
needed to reach the members of the multicast group. Once
the MST is constructed in this manner, the evaluation of its
cost (i.e., the total power needed to sustain the broadcast
 (IJCNS) International Journal of Computer and Network Security, 175
tree) does take into consideration the wireless multicast
advantage.
ALGORITHM 3 (Pruned node-based MST). This
algorithm requires the determination of the minimum-
energy spanning tree that is rooted at the Source node.
Unlike algorithm 2, the wireless multicast advantage is
taken into consideration in the determination of the power
needed to sustain the tree.
The recursive algorithm of section .3 can be used to
determine the MST. Thus, this method is not scalable. Once
the MST has been determined in this manner, it is pruned as
in algorithm 2.
ALGORITHM 4 (Pruned node-based spanning tree).
A heuristic is used to determine a suboptimal spanning
tree.10 Once the spanning tree has been determined in this
manner, it is pruned as in algorithm 2. Construction of the
spanning tree begins at the Source node. Its transmission
power is chosen to maximize the following
“n/p” metric:
At the next stage, each of the nodes that has been “covered”
(i.e., the Source node plus all nodes within its
communication range based on the calculation in the first
stage) evaluates the n/p metric for all possible sets of
neighbors (however, in computing this metric, only “new”
nodes, i.e., nodes not previously covered, are included in the
number of destinations). Note that it is possible to increase
the transmission power that was assigned to a node in an
earlier stage. This procedure is repeated until all nodes are
covered.
6.3. Additional algorithms with high complexity
The following algorithms require an exhaustive search, and
are thus not scalable. Nevertheless, they provide a useful
benchmark that permits us to evaluate the performance of
the other algorithms for small network examples.
ALGORITHM 5 (Least multicast cost). As in algorithm 1,
paths to all reachable destinations are established, regardless
of the cost required to do so. An exhaustive search of all
multicast trees that reach all reachable destinations is
performed. The tree with the lowest cost is chosen.
ALGORITHM 6 (Maximum local yardstick). The local
yardstick function yi is computed for each arriving multicast
request i. Multicast trees are formed to all subsets of
intended destinations. The tree that results in the maximum
value of yi is chosen. This tree does not necessarily include
all reachable destinations.
6.4 Energy efficient greedy scheme
The concept of neighbor classification based on node energy
level and their distances has been used in Energy Efficient
Greedy Scheme (EEGS) has been used to cater of the weak
node problem. Some neighbors may be more favorable to
Vol. 2, No. 5, May 2010
choose than the others, not only based on distance, but also
based on energy characteristics. It suggests that a neighbor
selection scheme should avoid the weak nodes. If the
geographic forwarding scheme purely
based on greedy forwarding attempts to minimize the
number of hops by maximizing the geographic distance
covered at each hop, it is likely to incur significant energy
expenditure due to retransmission on the weak nodes. On
the other hand, if the forwarding mechanism attempts to
maximize per hop reliability by forwarding only to close
neighbors with good nodes, it may cover only small
geographic distance at each hop. It would also result in
greater energy expenditure due to the need for more
transmission hops for each packet to reach the destination.
So in both cases energy is not being conserved to increase
the lifetime of the network. Therefore, the strategy used in
the proposed Energy Efficient Greedy Scheme (EEGS) first
calculates the average distance of all the neighbors of
transmitting node and checks their energy levels. Finally, it
selects the neighbor which is alive (i.e. having energy level
above than the set threshold) and having the maximum
energy plus whose distance is equal to or less than the
calculated average distance among its entire neighbors.
Hence, the proposed scheme uses Energy Efficient routing to
select the neighbor that has sufficient energy level and is
closest to the destination for forwarding the query.
7. Conclusion
In this paper we have argued that optimal techniques and
approaches for enhancing the communication and data
exchange between wireless networks. The schemes
discussed in the paper reduce the overall energy, while our
spreading approaches aim at distributing the traffic in a
more balanced way. We note that our basic ideas and
techniques should be able to enhance other routing protocols
as well.
Also In this paper, we reviewed some of the recent work
done in mobile ad hoc network considering energy as the
key issue. It is found that most of the study discusses the
energy issue at data link and network layer. We suggest
three energy efficient techniques for MANET point of view.
Route request minimization technique can be done by
implementing logical grouping; power control techniques
reduce the transmission power of a node while topology
control approach increases the network longevity by
satisfying network constraints. The simulation results
presented in section 4.0, suggests that multi-hop is ideal for
energy point of view but the limitation is the increase
chance of link failure.
References
[1] C.E.Perkins, "Ad Hoc Networking", Addison Wesley,
2001.
[2] IEEE 802.11 Working Group, "Wireless LAN Medium
Access Control (MAC) and Physical Layer (PHY)
Specification",1999.
[3] D.Zhou and.T.H.Lai, " A scalable and adaptive clock
synchronization protocol for IEEE 802.11-based
multihop ad hoc networks," IEEE International
176 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

Conference on Mobile Ad hoc and Sensor Systems
Conference, 2005 , Nov 2005.
[4] S.Singh and C.S.Raghavendra, "PAMAS power aware
multi-access protocol with signaling for ad hoc
networks," ACM Computer Communication Review,
vol. 28(3), pp.5-26, July 1998.
[5] L.M. Freeny, ―Energy efficient communication in ad
hoc networks," Mobile Ad Hoc Networking, Wiley-
IEEE press, pp.301-328, 200.
[6] E.S.Jung and N. H. Vaidya, "An energy efficient MAC
protocol for wireless LANs," IEEE INFOCOM, June
2002.
[7] [E.S.Jung and N.H.Vaidya, "A Power Control MAC
Protocol for Ad Hoc Networks," ACM Intl. Conf.
Mobile Computing and Networking (MOBICOM),
September 2002.
[8] B.Chen,K.Jamieson,H.Balakrishnan and R.Morris,
"Span:An energy efficient coordination algorithm for
topology maintenance in ad hoc wireless networks,"
ACM Wireless Networks Journal, vol. 8(5), pp. 81-9,
September 2002.
[9] P.K.Sahoo,J.P.Sheu and K.Y.Hsieh, ―Power control
based topology construction for the distributed
wireless sensor networks," Science Direct,Computer
Communications, vol. 30, pp. 277-2785, June 2007.
[10] R. Ramanathan, R. Rosales-Hain, Topology control of
multihop wireless networks using transmit power
adjustment, in:Proceedings of the 19th INFOCOM,
Tel Aviv, Israel, March 2000, pp. 0–13.
[11] Barrenechea, G.; Beferull-Lozano, B. & Vetterli,
M.(200). “Lattice Sensor Networks:Capacity Limits,
Optimal Routing and Robustness to Failures,”
Proceedings of thethird international symposium on
Information processing in sensor networks, pp. 186–
195,Berkeley, California, April 200.
of Computer Science & Engineering at Kamla Nehru Institute of
Technology, Sultanpur (Uttar Pradesh) as Lecturer in 2000 and
became Assistant Professor in 2007. His teaching and research
interests include Computer Networks, Wireless Networks and
Mobile Ad-Hoc Networks. He is pursuing PhD program in the
department of Computer Science & Engineering, MNNIT,
Allahabad (Uttar Pradesh) in the area of Mobile Ad-hoc Networks.
Ajeet Kumar Post graduated from Madan
Mohan Malaviya Engineering college,
Gorakhpur (U.P.) in Master of Computer
Application 2003. I pursuing our Master degree
(M.Tech.) in Software Engineering from
SRMS CET Bareilly affiliated to Uttar Pradesh
Technical University, Lucknow. I have joined
the department of Master of Computer
Application in SRMS CET Bareilly (Uttar
Pradesh) as Lecturer in 2003. His teaching and
research interests include Object Oriented System, Software
Engineering, Mobile Computing, Wireless Networks and Mobile
Ad-Hoc Networks.
Rati Agarwal received her B.Tech degree
from Uttar Pradesh Technical University
Lucknow. She is pursuing M.Tech from
Karnataka University. Presently, she is
working as a Lecturer (CS/IT Deptt) in
SRMSWCET, Bareilly. She has written a
book on “Soft Computing”. She has attended
a faculty development program based on
“Research methodologies” and has also
undergone a training in Infosys Campus
Connect Program held at S.R.M.S.C.E.T. Bareilly.

Authors Profile
Dr. Neeraj Tyagi has completed B.E. degree
in Computer Science & Engineering from
Motilal Nehru National Institute of
Technology, Allahabad (Uttar Pradesh) in
1987, his M.E. degree in Computer Science &
Engineering from MNNIT, Allahabad (Uttar
Pradesh) in 1997 and PhD degree in
Computer Science & Engineering from
MNNIT, Allahabad (Uttar Pradesh) in 2008.His teaching and
research interests including Computer Networks, Mobile Ad-Hoc
Networks, wireless Networks and Operating Systems. He joined
the department of Computer Science & Engineering in 1989 at
MNNIT, Allahabad and he has Also worked in Warman
International- Australia, G.E (Capital) - U.S.A, Electronic Data
Systems- U.S.A during 1999-2001. Presently he is working as
Associate Professor in the department of Computer Science &
Engineering, MNNIT, Allahabad(Uttar Pradesh), India.

Awadhesh Kumar graduated from Govind

Ballabh Pant Engineering college, Pauri
(Garhwal) in Computer Science &
Engineering in 1999. He obtained his Master
degree (M.Tech.) in Computer Science from
Uttar Pradesh Technical University,
Lucknow in 2006. He joined the department
 (IJCNS) International Journal of Computer and Network Security, 177
Vol. 2, No. 5, May 2010

Centralized Framework for Real Time Traffic in

Vehicular Ad hoc Networks
Nazish Khan1, Dr. Anjali Mahajan2
1
IV Semester M.E. Wireless Communication and Computing,
Computer Science and Engineering Department,
G. H. Raisoni College of Engineering, Nagpur, India
nazish07@rediffmail.com
2
Professor, Department of Information Technology,
G. H. Raisoni College of Engineering, Nagpur, India
armahajan@rediffmail.com

Abstract: Vehicular networks are formed by vehicles
communicating in an ad hoc manner or through cellular and
IEEE 802.11 base stations. Routing in vehicular networks is
challenging due to highly dynamic network topologies. Previous
research done has proposed the idea of Road-Based using
Vehicular Traffic information (RBVT) routing. In RBVT, routes
are sequences of road intersections, and geographical
forwarding is used for sending packets between the
intersections.
This paper proposes centralized framework for real time traffic
in Vehicular Ad hoc NETworks (VANETs). Already available
protocols are completely distributed protocol that unicasts
connectivity packets in the network to find the roadbased
topology graph. This graph represents a real-time view of the
roads with enough vehicular traffic to ensure connectivity
between intersections and is disseminated to all nodes. Proposed
framework uses a similar concept, but the road connectivity
information is gathered and sent separately to a server via
cellular links. The server centrally computes the connectivity
graph and disseminates it to the nodes in the network.
The proposed framework is compared against existing routing
protocols. Simulation results for city scenarios show that the
proposed framework performs better than the existing protocols,
with reduction in end-to-end delay and increase in delivery ratio.
needed for most of the applications to come to reality. The
development of new routing protocols is important because
not only will they enable the addition of new dimensions to
the vehicle through various applications in areas such as
vehicular safety, but they will also transform it from being a
mere transportation means to a smart vehicle. The
requirements of routing protocol in VANET are
significantly different than what they are in wired network
or Mobile Ad hoc NETwork (MANET). Some of the major
differences in VANET are (i) Mobility: due to which the
topology is highly dynamic, (ii) Road layout: a node moves
on the road, which constraint the packet to follow the road
path, and (iii) Obstacles: due to which the communication
on close-by parallel streets is restricted.
In past, the research has been focused on finding variants to
MANET protocols and changing some of their parameters
(to overcome issue of highly dynamic topologies) or
developing reactive/geographical VANET protocols. Some
of these protocols [11][12] do use the right approach of
using road intersections as part of their routes but fail to use
the real time traffic information or propose an expensive
solution of deploying infrastructure such as deploying
sensors to get the real-time traffic information.

Keywords: VANETs, Dissemination, Centralized Information 2. Related Work

Packet, Wi-Fi Link, Cellular Communication.
1. Introduction
In recent years there has been a growing interest in field of
vehicular networks, primarily due to the advancement in
technology and need for safety and entertainment
applications. The current generation of vehicles have
embedded computers, GPS receivers, and number of
sensors, whereas the next generation of these vehicles will
potentially have short-range wireless network interfaces and
wireless access to the Internet. A vehicular network in
general is a network of vehicles that can communicate with
each other; the network could be formed by use of
infrastructure such as cellular or road side fixed equipment
or on-the-fly in an ad hoc manner. The vehicular network
that is formed in ad hoc manner without need of
infrastructure is called a Vehicular Ad-hoc NETwork
(VANET).
One of the core issues of vehicular networks is designing
efficient and scalable routing protocols. Such protocols are
The research group in the UbiNetS lab at NJIT proposed
Road-Based using Vehicular Traffic (RBVT) routing. As the
name suggests, RBVT leverages real-time vehicular traffic
information to create road-based routes. The created routes
are successions of the road intersections that have, with high
probability, network connectivity between them. A packet
between consecutive intersections is forwarded using
Geographical routing; this approach decouples routes from
individual node movements. There are two main advantages
in RBVT: (a) adaptability to topological changes by
incorporating real-time vehicular traffic information and (b)
stability in route by use of road-based routes and
geographical forwarding.
Route Construction: Routing is a process of moving
packets through the network. It consists of two separate, but
related, tasks: (a) Defining paths for the transmission of
packets through the network and (b) Forwarding packets
based upon the defined paths. The tasks are achieved with
the help of routing table. RBVT being proactive protocols
178
the routes tables are updated for every known change of
topology. Usually the protocols [8, 9, 10] that have been
derived from MANET use node centric approach, these
protocols do not work well due to the dynamic nature of the
network. The other protocols [11, 13, 14, 15] use the road-
based approach, these protocols use the right approach but
most of them [11, 15] do not factor in the current road
traffic and rather try to use the shortest distance to reach
from source to destination, which currently may not have
any traffic. The protocols fail to consider traffic flow; some
of the protocols [14, 12, 16, 17] try to work around this
problem by using historic data. Due to the events such as
road constructions or traffic redirections which are not rare,
the historical data does not remain an accurate indicator of
the current road traffic flow.
The RBVT protocols use road intersection as their building
block. When a source wants to transfer data to the
destination, instead of tying particular nodes to the route
towards the destination it uses the road intersections. As
long as the road segments remain connected it does not
matter if the connectivity between the individual nodes on a
road segment exists. The fault in binding a particular node
to a route is due to the node movement, it may not remain in
the sight of the route.
Forwarding: A node periodically broadcasts hello
messages; the receiving node updates its neighbor table
using the information provided in hello message. If a hello
message is not received from a particular neighbor within a
time interval the entry is purged; non-receipt of hello
messages is a strong indication that the other node is
travelled out of communication range. When the source
node wants to transmit data to destination node, it queries
the routing table for the available routes to the destination; if
there are multiple routes from the source to destination, the
route with minimum number of intersections is used. This
desired route is stored in the packet for reference of the
intermediate node. The protocol uses loose source routing to
forward data packets in order to improve the routing
performance. The intermediate nodes are free to change the
path in the packet if they have latest connectivity
information. The node that stores or updates the route
information in the packet also writes the timestamp of the
connectivity graph that was used for determining this route.
The intermediate node compares its connectivity graph
timestamp to decide if a better routing decision can be made.
Due to the high volatility there could be route breaks, in
such cases the intermediate node tries forwarding the data
packet using geographical forwarding method until it
reaches a node that has fresher information and there exists
a route from that node to the destination; this newly found
route is updated in the packet and forwarded further towards
the next intersection. In between intersections the packet is
forwarded using geographical forwarding; this is
advantageous as the node always selects the farthest possible
neighbor node to forward the packet.
In this class of routing there are particularly two advantages:
(a) it is adaptive to network topology changes by using real-
time vehicular traffic information and (b) the stability of
route through road-based routes and use of geographical
forwarding.
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
2.1 RBVT-P
RBVT-P is a proactive routing protocol for VANET; it
periodically discovers and disseminates the real-time road
connectivity graph. The source node uses the connected
shortest available path to the destination. The path consists
of intersections that are found connected in the real-time
view. RBVT-P uses a location service, to find the
destination position. Due to the dynamic nature of the
network, RBVT-P does not bind the forwarding of packets
to a particular node like AODV (which are node centric).
Topology Discovery: The topology is discovered by use of
Connectivity Packet (CP). The CP is a packet that traverses
the connected road segments and stores the visited
intersections in it. This information is later used to create
the connectivity graph which is then disseminated in the
network to other nodes. The CPs are generated periodically
by a number of randomly selected nodes in the network. The
CP packet traverses the road intersection by use of algorithm
similar to DFS graph traversal, as the packet moves from
one intersection to other intersection the road segment is
marked as connected otherwise as un-connected. The CPs
are periodically generated by a CP generator node, they are
unicasted to discover the network topology.
Topology Dissemination and Route Computation: The
node that finally receives the CP on the generator segment
extracts the information in the CP to generate a
Connectivity Graph Update (CGU) packet and disseminates
the CGU to other nodes in the connected network. Upon
receiving the latest CGU, each node updates its routing table
and re-computes the shortest paths to all other connected
intersections. One of the fields of each entry in CGU is
timestamp, with the use of this field the node determines
whether the information it already has or the information
the CGU contains is the latest one. The timestamp of each
entry is retained in the routing table to identify and remove
the stale entries at later point of time.
When the source has to transmit information to the
destination it determines and uses the shortest path through
the connected intersections. The path is made of sequence of
intersections and is stored in the header of the data packet.
The packet is forwarded geographically between the
consecutive intersections, defined in the path. The protocol
uses loose source routing so the intermediate node is free to
update the path if it has more recent information about the
topology.
Route Maintenance: The RBVT-P generates CP frequently
(which in turn generates CGU) that keeps the nodes updated
with the topological changes. The CPs are generated from
different section of the network so that the information of
particular section is gathered by the nodes in that section.
The node in an isolated section cannot know the
connectivity information about other section till the
disconnection between these isolated networks is bridged.
The frequency of generating CP would depend on the
network size and density of nodes in the section of the
network. Higher the density, lower number of packets are
required as the probability of finding the replacement node
increases. In a CP generation interval, multiple CP are
generated to avoid the problem of loosing connectivity
information in a CP round when CP packet is lost. The CP
 (IJCNS) International Journal of Computer and Network Security, 179
passing protocol is not a reliable protocol but a best effort
delivery so there are chances of them getting lost.
3. Proposed Work
Importance of this proposed solution come from the use of
real-time traffic information and the way route between the
source and destination is created. Unlike the MANET
protocols, the proposed framework use road intersections as
a part of route. Road intersections are stationary; having
them in route is more advantageous than having a moving
vehicle. The difference with respect to other VANET
protocol is that the proposed solution use real-time traffic
information rather than the historic data. The use of road
intersection as the guiding point helps when a moving
vehicle goes out of range from the guided point. This
vehicle (that moved out of range) can be replaced with other
vehicle near the road intersection to forward the packet
towards the destination. The need for such mechanism
arises due to the highly dynamic changes in the topology.
The proposed framework being proactive is expected to be
better in terms of end to end delay i.e. lowering the delay of
end to end communication.
3.1 Design
Centralized framework for real time traffic in Vehicular Ad
hoc NETworks uses the concept of proactive protocol. It
uses a concept called as Centralized Information Packet
(CIP) that is similar to CP used in RBVT-P. The car node
on the road considers itself the most forward if it does not
receive any packet for a timeout interval; similarly it
considers itself most backward if it does not find any node to
which it can send the CIP. The nodes backward/forward
direction is irrespective of the driving direction; each road
in the map is assigned a direction that is stored along with
digital map in the node. The communication between the
node and the server takes place using the Cellular
technology whereas the communication between the nodes is
through Wi-Fi wireless technology.
Topology Discovery: The most forward node on the road
sends the CIP to the node behind it. The intermediate node
continues forwarding this packet to the node behind it,
eventually reaching the most backward node (no other node
behind it on that road which this backward node can
communicate). This most backward positioned node on the
road sends the gathered connectivity information through its
GPRS interface to the server. The connectivity information
entry is of the form <FirstIntersectionID,
LastIntersectionID, ExpirationTime>; where
FirstIntersectionID indicates the intersection id from where
the connectivity begins, LastIntersectionID indicates the end
intersection till the connectivity on the road segments exists
and the ExpirationTime indicates the time till the road
segments are valid. The expiration time is calculated by
addition of Maximum connectivity valid period (protocol
configurable value) and current time. The connectivity
information is gathered for the segments through which a
CIP traveled. The CIP may also implicitly indicate the
nonconnectivity of two segments the one before the
Vol. 2, No. 5, May 2010
FirstIntersectionID and the one after LastIntersectionID to
the block of connected segments.
The server receives one CIP from each set of connected
nodes; there can be multiple CIPs for the same road
depending on the connectivity pattern on the road. For e.g.
in Figure 1 on road segment S1 of road R3, the five nodes
create an isolated network from the nodes on segment S2 on
the same road, in these cases there are two different CIPs
that are sent to the server from these two sections on the
same road. In the figure R1, R2 and R3 indicate roads and
S1, S2 and S3 designate the segments on the roads. The pair
<Road><Segment> uniquely identifies the segments. The
car nodes on the roads are named N1, N2, N3 and N11.
Figure 1. Most backward cars on the each road send CIP to
server. After computing the real-time graph it is sent to the
node which sent this CIP.
The car N1 being the most forward node on road R1 starts a
CIP, every node that has the responsibility to send this
packet tries to send it the most farthest possible node so that
the discovery of information is faster and this also makes the
protocol more efficient.
In the example N1 skips node N2 and sends the packet
directly to N3 as N3 is in the wireless range of N1. The
packet is similarly forwarded to the most backward node (in
this case N11), which then sends the packet to the server for
re-computing the real-time graph. I1 to I8 indicate the
intersections or end points of the roads.
Topology Dissemination and Route Computation: On
receipt of each CIP the server updates the connectivity
graph. In comparison with RBVT-P, this same task is done
by the node that finally received the CP from the segment
where it was initiated (this indicates that an iteration of
connected graph is complete). Once the server rebuilds the
connectivity graph, it transmits this graph back to the node
that sent the CIP that triggered the graph computation. The
node then disseminates this information by broadcasting it
to the neighbors. This continues till the most forward node
receives the information. Similar to RBVT-P each node then
runs Dijkstra’s algorithm on the newly received connectivity
graph to find the shortest path to all the connected segments
in the map. Further on, when a source node wants to
transmit data, it stores the connected available shortest path
in the packet header. These routes, represented as sequences
of intersections, are stored in the data packet headers and
180
used by intermediate nodes to geographically forward
packets between intersections.
To give an example of how the above process contribute to
the view of a node. In figure 1, say node N11 is the last node
to send the CIP out of the five CIP s sent to the server. The
time difference of the CIP sent should not be more than α,
where α is the period between the round of CIP generation.
On receipt of CIP from N11, the server recomposes the
graph and sends it to node N11. The node views the
received graph as shown in figure 2, the dotted lines
indicate no connectivity between the intersections (or on
that segment) and the bold lines imply the connected
component of the graph from the intersection I5 at which
node N11 is present. The road segment S2 of road R2 is not
reachable as there is no node in the connected graph that
could communicate with the cluster of nodes that are present
towards intersection I1. Similarly as there is no node
towards the intersection I6, that intersection remains
unconnected from intersection I5.
Figure 2. View of most backward node on road R1 after
receiving the updated connectivity graph from the server.
From figure 1 it is seen on road R3, the segments S1 (I5 to
I8) and segment S2 (I2 to I5) generated two different CIP s,
which mean that nodes on those segments are not able to
communicate due to the distance between them, but as seen
in figure 5.2 the graph is created as showing them
connected; the server generated the graph with global view.
The nodes at the intersection I5 on road R1 can bridge the
gap to forward the traffic between the disconnected
segments S1 and S2 on road R3.
Route Maintenance: In framework, the forward node on
the road frequently generates the CIP, which eventually
gathers the connectivity information of the road segments
that are connected and sends it to the server. The server
sends the computed graph to the node that recently reported
the connectivity (CIP). This node disseminates the
information to its neighbor and those neighbors send it to
their neighbors and so on.
4. Simulation Setup
The evaluation of proposed framework is done using
Network Simulator NS-2 [7] simulation. The movement of
the vehicles is generated using open-source microscopic,
space-continuous and time discrete vehicular traffic
generator package called as SUMO [18]. SUMO uses a
collision-free car-following model. The map is inputted into
(IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010
SUMO, the information such as road speed limits, traffic
lights and number of lanes is also inputted. The output file
from SUMO is converted into the required node movement
format of NS-2 simulator.
For the wireless configuration, at the physical layer, the
shadowing propagation model is used to characterize
physical propagation. The communication range of 400m
with 80% probability of success for transmissions is set. The
obstacle model simulates buildings in a city environment;
the contour of each street can either be a building wall (of
various materials) or an empty area. Thus for each street
border, the signal attenuation value is set to a randomly
selected between 0dB and 16dB.
5. Metrics
The evaluation performance of these routing is done using
different Constant Bit Rate (CBR) data rates, different
network densities and different numbers of concurrent
flows. Following are the various metrics to evaluate the
performance:
Average delay: This metric defines the average delay
occurred for the transmitted data packets that are
successfully delivered. The average delay characterizes the
latency introduced by each routing approach. For a proactive
protocol this is the primary metric, unlike reactive protocols
proactive protocols maintain routes between source and
destination which leads to reduced delays.
Average delivery ratio: This metric defines the number of
data packets successfully delivered at destinations per
number of data packets sent by sources (duplicate packets
generated by loss of acknowledgments at the MAC layer are
excluded). The average delivery ratio gives the measure of
the routing protocol to transfer end-to-end data successfully.
Average number of hops: This metric defines the average
number of nodes that are part of successful packet delivery
from source to the destination. Historically, the average
number of hops was a measure of path quality. This metric
is used to study if there is a correlation between the number
of hops and average delivery ratio and average delay,
respectively.
Overhead: This metric is defined as the number of extra
packets per number of unique data packets received at
destinations. The number of extra packets consists of
routing packets (i.e. routing overhead) and duplicate data
packets. Therefore, the overhead measures the total
overhead per successfully delivered packet.
6. Results
The results are shown between RBVT-P and the proposed
framework, the simulation uses 250 nodes.
Average delay: Figure 3 shows the end to end delay for
RBVT-P and proposed framework.
The proposed framework posses’ lower delay than RBVT-P
for two reasons (a) the traffic in the network gets split due to
the addition of cellular link (b) the server can quickly gather
and disseminate the CGU to the network. Due to the
introduction of server with cellular link there is reduced
traffic at the Wi-Fi link also as the node has more accurate
view of the network proposed framework has reduced delays
in data delivery.
 (IJCNS) International Journal of Computer and Network Security, 181
References
Figure 3. Average Delay with density of 250 nodes.
Average delivery ratio: Figure 4 shows the average
delivery ratio for RBVT-P and proposed framework. The
RBVT-P demonstrates lower delivery ratio performance
than the proposed framework as in RBVT-P along with the
data packets, the connectivity packets (that gather the real-
time connectivity graph of the network) and the CGU
packets (that disseminate the collected connectivity graph
view) participate in network traffic at Wi-Fi interface. The
proposed framework shows 10% gain in delivery ratio this is
due to the fact that the control plane traffic of the network is
transmitted via cellular link and Wi-Fi link is used for data
plane.
Figure 4. Average Delivery Ratio with density of 250 nodes.
7. Conclusion
This paper presented the framework for real time traffic
environments that take advantage of the road topology to
improve the performance of routing in VANET. Simulation
results show that the proposed framework outperform
existing approaches in terms of average delay, average
delivery ratio and low overhead. Because the RBVT
protocols forward data along the streets, not across the
streets, and take into account the real-time traffic on the
roads, they perform well in realistic vehicular environments
in which buildings and other road characteristics such as
dead end streets, traffic lights are present. As cellular link
would more commonly be available, we used a server with
cellular communication to show how the performance can
increase if the control plane (i.e., routing traffic) is routed
across the cellular link whereas data is transmitted over the
Wi-Fi link in ad hoc manner.
Vol. 2, No. 5, May 2010
[1] Bernd Roessler, Christian Bruns. Replication of Local
Road Networks for Regional Information
Dissemination in VANETS. WORKSHOP ON
POSITIONING, NAVIGATION AND
COMMUNICATION (WPNC’06).
[2] Mitko Shopov, Evtim Peytchev, Generation and
dissemination of traffic information in Vehicular Ad-
hoc
Networks, Computer Science’2005, Technical
University of Sofia, Bulgaria.
[3] Emanuel Fonseca, Andreas Festag, Roberto Baldessari,
Rui Aguiar, Support of Anonymity in VANETs–
Putting Pseudonymity into Practice, IEEE Wireless
Communications and Networking Conference
(WCNC),Hong Kong, March 2007.
[4] Tarik Taleb, Ehssan Sakhaee, Abbas Jamalipour,
Kazuo Hashimoto, Nei Kato, and Yoshiaki Nemoto,
“A Stable Routing Protocol to Support ITS Services
in VANET Networks”IEEE TRANSACTIONS ON
VEHICULAR TECHNOLOGY, VOL. 56, NO. 6,
NOVEMBER 2007
[5] Maria Kihl, Mihail L. Sichitiu, and Harshvardhan P.
Joshi “Design and Evaluation of two Geocast
Protocols for Vehicular Ad-hoc Networks”
JOURNAL OF INTERNET ENGINEERING, VOL.
2, NO. 1, JUNE 2008
[6] Ioannis Broustis and Michalis Faloutsos “Routing in
Vehicular Networks: Feasibility, Modeling, and
Security” Hindawi Publishing Corporation
International Journal of Vehicular Technology
Volume 2008, Article ID 267513
[7] The network simulator ns-2
http://www.isi.edu/nsnam/ns2/
[8] Perkins, C. and Royer, E. (1999). Ad hoc On Demand
Distance Vector (AODV) Routing. In Proceedings
2nd Workshop on Mobile Computing Systems and
Applications. New Orlean s, LA, USA: IEEE,
February 1999, pp. 90–100.
[9] Clausen, T. and Jacquet, P. (2003). Optimized Link
State Routing Protocol (OLSR). RFC 3626, IETF
Network Working Group.
[10] Perkins, C., and Bhagwat, P. (1994). Highly dynamic
Destination-Sequenced Distance-Vector routing
(DSDV) for mobile computers. SIGCOMM Computer
Communication.
[11] Lochert, C., Hartenstein, H., Tian, J., Füßler, H.,
Hermann, D., and Mauve M. (2003). A routing
strategy for vehicular ad hoc networks in city
environments. In Proceedings of IEEE Intelligent
Vehicles Symposium, pages 156--161.
[12] Jerbi, M., Meraihi, R., Senouci, S., and Ghamri-
Doudane Y. (2006). Gytar: improved greedy traffic
aware routing protocol for vehicular ad hoc networks
in city environments. In Proceedings of the 3rd ACM
International Workshop on Vehicular Ad hoc
Networks (VANET), Los Angeles, CA, USA, pp. 88–
89.
[13] Naumov, V., and Gross, T., (2007). Connectivity-
Aware Routing (CAR) in vehicular ad hoc networks.
182 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

In Proceedings IEEE International Conference on

Computer Communications, Anchorage, AK, USA.
[14] Li, T., Hazra, S., and Seah, W. (2005). A position-
based routing protocol for metropolitan bus networks.
In Proceedings IEEE 61st Vehicular Technology
Conference VTC-Spring, Stockholm, Sweden.
[15] Tian, J., Han, L., Rothermel, K., and Cseh, C. (2003).
Spatially aware packet routing for mobile ad hoc
inter-vehicle radio networks. In Proceedings IEEE
Intelligent Transportation Systems, Shanghai, China.
[16] Zhao, J., and Cao, G. (2006). Vadd: Vehicle-assisted
data delivery in vehicular ad hoc networks. In
Proceedings IEEE 25th International Conference on
Computer Communications, Barcelona, Spain.
[17] Wu, H., Fujimoto, R., Guensler, R., and Hunter M.
(2004). MDDV: A Mobility-Centric Data
Dissemination Algorithm for Vehicular Networks. In
Proceedings of the 1st ACM International Workshop
on Vehicular Ad hoc Networks (VANET).
Philadelphia, PA, USA: ACM.
[18] “Sumo - simulation of urban mobility,”
http://sumo.sourceforge.net/

Authors Profile

Dr. Anjali R Mahajan , currently

working as Professor in Department
of Information Technology , G H
Raisoni College of Engineering ,
Nagpur, completed her B.E. in
Computer Science and Engineering
from Government College of
Engineering, Amravati in the year
1994, M.Tech from Sant Gadge
Baba Amravati University in the year 2002 and Ph.D. from
Sant Gadge Baba Amravati University in 2010. Dr.
Mahajan has to her credit nine publications in International
Journals and 48 publications at the International
conferences and 5 publications in National Conferences.
She is a member of IEEE, CSI, LM of ISTE.
Her areas of interest are Compiler optimization, Artificial
Intelligence, Evolutionary Computing and Image Processing

Nazish Khan received the B.E. degree

in Computer Technology from G. H.
Raisoni College of Engineering in
2002. She is pursuing Master of
Engineering degree [M.E.] in Wireless
Communication and Computing in
Computer Science and Engineering
Department from G. H. Raisoni
College of Engineering, Nagpur, India.
She is working as lecturer in Computer Science and
Engineering Department in engineering college from last
six years. Her areas of interest are computer networks,
Mobile Ad hoc Networks and Vehicular Ad hoc networks.
 (IJCNS) International Journal of Computer and Network Security, 183
Vol. 2, No. 5, May 2010

Energy-efficient Algorithm for a novel clustering

approach in w/s networks
S. Sarah 1, P. Kumaran2 and Dr. A. Arul Lawrence Selvakumar 3

1
M. E Final Year, Dr. Paul’s Engineering College, Pondicherry,
Anna University, Chennai, Tamilnadu, India
sarahnixon@rediffmail.com
2
Assistant Professor, Department of Computer Science and Engineering,
Dr. Paul’s Engineering College, Pondicherry,
Anna University, Chennai, Tamilnadu, India
kumaran.periyasamy@yahoo.co.in
3
Professor & Head, Department of Computer Science and Information Technology,
Kuppam Engineering College, Kuppam, Andhra Pradesh, India
aarul72@hotmail.com

During data collection, two mechanisms are used to

Abstract: Energy efficiency operations are essential in
extending Wireless Sensor Networks lifetime. Among the energy reduce energy consumption: message aggregation and
saving- based solutions, clustering sensor nodes is an interesting filtering of redundant data. These mechanisms generally use
technique for achieving these goals. . In this work, we propose a clustering methods in order to coordinate aggregation and
new energy-efficient approach for clustering nodes in adhoc filtering.
sensor networks. The main challenge in wireless sensor network
deployment pertains to optimizing energy consumption when Hierarchical (clustering) techniques can aid in reducing
collecting data from sensor nodes. Compared to other methods useful energy consumption. Clustering is particularly useful
(CPLEX-based method, distributed method, simulated for applications that require scalability to hundreds or
annealing-based method), the results show that our approach thousands of nodes. Scalability in this context implies the
returns high-quality solutions in terms of cluster cost and need for load balancing and efficient resource utilization.
execution time. As a result, this approach is suitable for
handling network extensibility in a satisfactory manner.
Applications requiring efficient data aggregation (e.g.,
Keywords: wireless sensor networks, clustering, energy- computing the maximum detected radiation around an
efficient, distributed. object) are natural candidates for clustering. Routing
protocols can also employ clustering. Clustering was
1. Introduction proposed as a useful tool for efficiently pinpointing object
locations.
Sensor networks have recently emerged as an important
computing platform. Sensor nodes are typically less mobile
Clustering can be extremely effective in one-to-
and more densely deployed than mobile ad-hoc networks
many, many - to- one, one-to-any, or one-to-all (broadcast)
(MANETs). Sensor nodes must be left unattended e.g., in
communication. For example, in many-to-one
hostile environments, which makes it difficult or impossible
communication, clustering can support data fusion and
to re-charge or replace their batteries (solar energy is not
reduce communication interference. The essential operation
always an option). This necessitates devising novel energy-
in sensor node clustering is to select a set of cluster heads
efficient solutions to some of the conventional wireless
among the nodes in the network, and cluster the rest of the
networking problems, such as medium access control,
nodes with these heads. Cluster heads are responsible for
routing, self-organization, bandwidth sharing, and security.
coordination among the nodes within their clusters (intra-
Exploiting the tradeoffs among energy, accuracy, and
cluster coordination), and communication with each other
latency, and using hierarchical (tiered) architectures are
and/or with external observers on behalf of their clusters
important techniques for prolonging network lifetime
(inter-cluster communication).
This necessity implies the emergence of new kinds of
Energy efficiency operations are essential in
networks, which are typically composed of low-capacity
extending Wireless Sensor Networks lifetime. Among the
devices. Such devices, called sensors, make it possible to
energy saving- based solutions, clustering sensor nodes is an
capture and measure specific elements from the physical
interesting alternative that features a reduction in energy
world (e.g., temperature, pressure, humidity). Moreover,
consumption through: (i) aggregating data; (ii) controlling
they run on small batteries with low energetic capacities.
transmission power levels (iii) balancing load; (iv) putting
Consequently, their power consumption must be optimized
redundant sens or nodes to sleep.
in order to ensure increased lifetime for those devices.
184 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

Since sensor nodes carry limited and non-replaceable power

sources, the efficiency of energy use determines the lifetime Centralized clustering is modeled as hyper graph
of the sensors and, consequently, the duration of the sensing partitioning. The novel method proposes the use of a tabu
task and the network lifetime. Here, the network lifetime is search heuristic to solve this problem. The existing
defined as the time elapsed until the first node (or the last centralized clustering methods cannot be used to solve this
node) in the network depletes its energy (dies). Energy issue due to the fact that our approach to model the problem
saving in sensor networks is an open challenging issue, assumes that the numbers of clusters and cluster heads are
where a substantial amount of works has been done. Among unknown before clusters are created, which constitutes
these works, clustering-based approaches are showing the another major original facet of this paper.
most exiting result through their ability to reduce energy
consumption by multiple ways. In fact, hierarchical- 2. Data Collection Mechanism
clustering reduces: (i) the amount of query packets via inter-
Generally, sensor networks contain a large quantity of nodes
cluster query dissemination; (ii) the amount of data packets
that collect measurements before sending them to the
by aggregating collected data. Further, hierarchical-
applications. If all nodes forwarded their measurements, the
clustering needs a minimal number of active nodes to cover
volume of data received by the applications would increase
the target area. This is achieved by putting redundant sensor
exponentially, rendering data processing a tedious task. A
nodes in sleep mode. At this point, hierarchical-clustering is
sensor system should thus contain mechanisms that allow
particularly useful for applications that require scalability to
the applications to express their requirements in terms of the
hundreds or thousands of nodes. However, hierarchical-
required quality of data. Data aggregation and data filtering
clustering scheme deals with some constraints. The most
are two methods that reduce the quantity of data received by
relevant concerns the Cluster-Head (CH) reconfiguration
applications. The aim of those two methods is not only to
process. Actually, to alleviate the large amount of energy
minimize the energy consumption by decreasing the number
consumption required by a CH, frequent reconfigurations of
of messages exchanged in the network but also to provide
clusters are needed.
the applications with the needed data without needlessly
overloading them with exorbitant quantities of messages.
They model this problem as a k-means clustering problem,
which is defined as follows : let P be a set of n data points in
The aggregation data mechanism allows for the gathering
d-dimensional space Rd and an integer k, and the problem
of several measures into one record whose size is less than
consists of determining a set of k points in Rd, called
the extent of the initial records. However, the result
centers, to minimize the mean squared distance from each
semantics must not contradict the initial record semantics.
data point to its nearest center. In centralized proposed
Moreover, it must not lose the meanings of the initial
version of Low Energy Adaptive Clustering Hierarchy
records. The data filtering mechanism makes it possible to
(LEACH), their data collection protocol, in order to produce
ignore measurements considered redundant or those
better clusters by dispersing cluster head nodes throughout
irrelevant to the application needs. A sensor system provides
the network. In this protocol, each node sends information
the applications with the means to express the criteria used
regarding its current location and energy level to the sink
to determine measurement relevancy, e.g., an application
node, which computes the node’s mean energy level, and
could be concerned with temperatures, which are 1) lower
nodes, whose energy level is inferior to this average, cannot
than a given value and 2) recorded within a delimited zone.
become cluster heads for the current round. Considering the
The sensor system filters the network messages and
remaining nodes as possible cluster heads, the sink node
forwards only those that respect the filter conditions.
finds clusters using the simulated annealing algorithm in
order to find optimal clusters.
Applications that use sensor networks are generally
concerned with the node measurements within a certain
This algorithm attempts to minimize the amount of energy
period of time. Hence, the most important key indicators in
required for non cluster head nodes to transmit their data to
sensor networks are the quality of the measurements and the
the cluster head, by minimizing the sum of squared
network lifetime. An application designed to record the
distances between all non cluster head nodes and the closest
mean temperature in zones where the sensors are deployed
cluster head. The energy map, the component that holds
could be associated with a set of requirements in terms of
information concerning the remaining energy available in
measured frequencies (e.g., the sensor system must record a
all network areas, can be used to prolong the network’s
measurement every 15 minutes), in terms of a measurement
lifetime.
discrepancy thresholds (e.g., the sensor system must ignore
This paper proposes a new centralized clustering data whose result is less than 10 percent of the previous
mechanism equipped with energy maps and constrained by value), and in terms of the sensor lifetime (e.g.,
Quality-of-Service (QoS) requirements. Such a clustering measurements must be provided for one year).
mechanism is used to collect data in sensor networks. The
first original aspect of this investigation consists of adding In [3], we propose a novel data collection approach for
these constraints to the clustering mechanism that helps the sensor networks that use energy maps and QoS requirements
data collection algorithm in order to reduce energy to reduce power consumption while increasing network
consumption and provide applications with the information coverage. The mechanism comprises two phases: during the
required without burdening them with unnecessary data. first phase, the applications specify their QoS requirements
 (IJCNS) International Journal of Computer and Network Security, 185
regarding the data required by the applications. They send
their requests to a particular node S, called the collector
node, which receives the application query and obtains
results from other nodes before returning them to the
applications. The collector node builds the clusters,
optimally using the QoS requirements and the energy map
information. During the second phase, the cluster heads
must provide the collector node with combined
measurements for each period. The cluster head is in charge
of various activities: coordinating the data collection within
its cluster, filtering redundant measurements, computing
aggregate functions, and sending results to a node collector.
3. Problem Formulation
The considered network contains a set V of m
stationary
nodes whose localizations are known. The communication
model can be described as multihop, which means that
certain nodes cannot send measurements directly to the
collector node: they must rely on their neighbors’ service.
An application can specify the following QoS requirements:
1. Data collection frequency, fq. The network provides
results to the application every time the duration fq
expires.
2. A measurement uncertainty threshold, mut. If the
difference between two simultaneous measurements
from two different nodes in the same zone (fourth
requirement) is inferior to mut, then one of them is
considered redundant.
3. A query duration, T . The network required for the
query run a total time whose value is equal to T .
4. A zone size step. The step value determines the zone
length. Within a single zone, measurements are
considered redundant. If an application requires
more precision, it could decrease the step value or
even ignore the transfer of such value.
The goal of the clustering algorithm is to
1) Split the network nodes into a set of clusters.
2) application requirements,
3) reduce energy consumption, and
4) Prolong the network lifetime.
Clusters are built according to the following criteria:
. Maximize network coverage using the energy map;
. Gather nodes likely to hold redundant
measurements;
. Gather nodes located within the same zone delimited
by the application.
4. A Tabu Search Approach
In order to facilitate the usage of tabu search for CBP, a new
graph called Grow is defined. It is capable of determining
feasible clusters. A feasible cluster consists of a set of nodes
that fulfill the cluster building constraints. Nodes that
Vol. 2, No. 5, May 2010
satisfy Constraint, i.e., ensure zone cover-age, are called
active nodes. The vertices of Gr represent the network
nodes. An edge is defined in graph Gr between nodes i and j
if they satisfy Constraints. Consequently, it is clear that a
clique in Gr embodies a feasible cluster. A clique consists of
a set of nodes that are adjacent to one another.
Five steps should be conducted in order to adapt tabu
search heuristics to solve a particular problem:
1. design an algorithm that returns an initial solution,
2. define moves that determine the neighbourhood N
of a solution s,
3. determine the content and size of tabu lists,
4. define the aspiration criteria,
5. design intensification and diversification
mechanisms.
The algorithm ends when one of the following three
conditions occurs:
1. All possible moves are prohibited by the tabu lists;
2. The maximal number of iterations allowed has been
reached;
3. The maximal number of iterations, where the best
solution is not enhanced successively, has been
reached.
Analysis of the Impact of Tabu Search Parameters
The size of the tabu list has a direct impact on the
quality of the solution. Hence, it is important to analyze its
impact, in order to adjust its value accordingly. Results
reported in [5] show that determining the tabu list size
dynamically is more efficient than fixing its value during
the iterations. This experiment involves a sensor network
composed of 100 nodes. A square topology is used, i.e.,
nodes arise on the summits of squares that cover the entire
network area. To facilitate the analysis, it is assumed that all
nodes are active.
Figure 1. Impact of max no of iterations on the solution
cost.
An experiment was conducted in order to quantify the
impact of the parameter “maximum number of allowed
iterations” on the quality of the solution found and
186 (IJCNS) International Journal of Computer and Network Security,
Vol. 2, No. 5, May 2010

determine the limits of the solution enhancement. A sensor References

network that comprises 1,000 nodes localized in a square
topology is considered. The maximum number of iterations
is modified and the other parameters are set except for the
“maximum number of iterations where the best solution is
not enhanced” parameter, which takes a value that allows
the algorithm to reach the maximum number of iterations.
Fig. 1 shows the impact of this parameter on the percentage
of the improvement of the initial solution cost.
Results show that by increasing the maximum number of
iterations, the solution costs is enhanced; hence, higher
quality solutions are obtained. This occurs until this
parameter reaches a value where solutions have almost
identical cost in spite of the increasing number of iterations.
Indeed, between 0 and 5,000 iterations, the solution quality
is enhanced by a factor of 2.3. However, between 5,000 and
10,000 iterations, such enhancement is evident only 1.35
times. This is mainly due to the influence of the other
parameters and the initial solution’s algorithm, which yields
a satisfactory solution.
Finally, results show that a tabu search-based resolution
method provides quality solutions in terms of cluster cost
and execution time. Furthermore, it behaves well with
network extensibility. Nevertheless, compared to a
distributed approach, this centralized approach suffers from
a major drawback linked to the additional costs generated by
communicating the network node information and the time
required to solve an optimization problem.
[1] P.K. Agarwal and C.M. Procopiuc, “Exact and
Approximation Algorithms for Clustering,”
Algorithmica, vol. 33, no. 2, pp. 201- 226, June 2002.
[2] P. Basu and J. Redi, “Effect of Overhearing
Transmissions on Energy Efficiency in Dense Sensor
Networks,” Proc. Third Int’l Symp. Information
Processing in Sensor Networks (IPSN ’04), pp. 196-
204, Apr. 2004.
[3] A. El Rhazi and S. Pierre, “A Data Collection
Algorithm Using Energy Maps in Sensor Networks,”
Proc. Third IEEE Int’l Conf. Wireless and Mobile
Computing, Networking, and Comm. (WiMob ’07),
2007.
[4] O. Moussaoui, A. Ksentini, M. Naimi, and M. Gueroui,
“A Novel Clustering Algorithm for Efficient Energy
Saving in Wireless Sensor Networks,” Proc. Seventh
Int’l Symp. Computer Networks (ISCN ’06), pp. 66-72,
2006.
[5] T. Kanugo, D.M. Mount, N.S. Netanyahu, C.D. Piatko,
R. Silver- man, and A.Y.Wu, “A Local Search
Approximation Algorithm for k-Means Clustering,”
Proc. 18th Ann. ACM Symp. Computational Geometry
(SoCG ’02), pp. 10-18, 2002.
[6] O. Younis and S. Fahmy, “Distributed Clustering in
Ad-Hoc Sensor Networks: A Hybrid, Energy-Efficient
Approach,” Proc. IEEE INFOCOM, pp. 629-640,
2004.

5. Conclusions

This paper has presented a heuristic approach based on a

energy efficient search to solve clustering problems where
the numbers of clusters and cluster heads are unknown
beforehand. To our knowledge, this is the first time that the
clustering problem is modeled and resolved with these
constraints. The tabu search adaptation consists of defining
three types of moves that allow reassigning nodes to
clusters, selecting cluster heads, and removing existing
clusters. Such moves use the largest size clique in a
feasibility cluster graph, which facilitates the analysis of
several solutions and makes it possible to compare them
using a gain function.
Performance of our central approach with those of a
distributed approach and we conclude that the central
approach is less efficient than the distributed approach in
the cluster building phase. Nevertheless, the central
approach is more efficient in the data collection phase.
Consequently, the central approach is more efficient in a
case where the data collection phase is long. Otherwise, the
distributed approach should be used to run the queries with
a short execution time.