1

Optimization of Spectrum Management for

Cognitive Radios
Ayaz Masood Kamran
1
, Muhammad Khalid Khan
1
, Muhammad Aamir
2
1
PAF Karachi Institute of Economics and Technology, Karachi, Pakistan
2
Mehran University of Engineering & Technology, Jamshoro, Pakistan
{kami2982@yahoo.com, muaamir5@yahoo.com, khalid.khan@pafkiet.edu.pk}

Abstract: Spectrum management can be optimized by providing intelligent access
to pre-allocated spectrum which may enable utilization of licensed spectrum by
unlicensed clients on the pre-agreed condition of nominal interference. This
research paper presents a strategy study to optimize cognitive radio networks
against the problems of spectrum deficit by accommodating unlicensed clients in
less populated underutilized fragments of the spectrum. The main information
source for a cognitive radio can be formed by means of spectrum sensing which is
significantly important to ensure that the unlicensed clients do not offer
intolerable levels of interference to licensed clients. This paper is focused on
spectrum sensing scenario that takes advantage of sensing diversity which may
become possible due to multiple nodes combining their sensing results. It provides
the capability to cognitive networks to overcome problems related to hidden
primary clients. In this research, an innovative approach has been proposed which
is based on grouping among cooperating secondary clients for deployment of an
agility driven tasking method that will result in reduction of interference to
primary users through their early detection. Moreover, the proposed approach of
optimization is also suitable to determine the effects of varying group size on
agility. This intended optimization is supported by MatLAB as simulation
platform
Keywords: Cognitive radio, Spectrum sensing, Sensing diversity, Group size,
Agility, MatLAB.

1 Introduction
The last few decades have seen a vast increase in wireless standards. While these standards have
mostly resulted in enhanced data rates and spectral efficiency, they have also resulted in an
increase in spectrum allocations. The result is spectrum deficiency. To address this problem,
various techniques including spread spectrum and opportunistic access to communications can be
2
employed [1]. The Federal Communications Commission issued a notice of proposed rule making
[2] for unlicensed access to television (TV) bands. It also proposed the definition of a metric on
the basis of aggregate interference temperature to ensure right of spectral access to primary users.
This accelerated further research on concepts such as software defined radio (SDR) [3] and
cognitive radio (CR) [4], which laid the foundation of intelligent spectrum access for unlicensed
users.
The SDR promised to provide enhanced flexibility through its ability to modify its operating
parameters in the context of its environment. CR [5] is a radio that fulfils the cognitive cycle of
observe, orient, plan, decide and act phases. For the observe phase, it considers various stimuli
including spectrum, network, policy, user requirements, geographic-location, etc [5]. Accordingly,
spectrum sensing is central to the awareness concept of CR. The software radio has the potential to
serve as a platform for CR [5].
Spectrum sensing serves as the basic stimulus analysis tool that facilitates creation of awareness
in CR. The objective is to find holes in the spectrum so that they can be used by secondary users
till regain by the primary users. Spectrum sensing techniques include matched filtering, cyclic-
stationary feature detection and energy detection [6]. Matched filtering requires demodulation of
primary user signal for effective detection. This implies that a separate receiver would be required
for each class of primary users [6]. Cyclic-stationary feature detection taps the cyclic
characteristics of a modulated signal to identify its presence in a low signal-to-noise ratio regime.
It, however, involves increased complexity for effective performance [6]. Both these techniques
require a-priori knowledge about the primary transmitter to operate efficiently. Energy detection
utilizes implementation of a spectrum analyzer by averaging Fast Fourier Transform bins to
measure power spectral density in sub-bands. This technique is not dependent on a comprehensive
a-priori knowledge base about the primary transmitter and can adapt to a large variety of primary
transmitter classes [6].
Agility [7] (the ability of a cognitive radio to sense vacant spectrum and quickly shift to it) is an
important characteristic that is used to evaluate spectrum sensing techniques. It also indicates the
time taken by a CR to vacate spectrum after it detects a primary user. Agility is measured in terms
of time and its threshold is closely related to the maximum time for which a primary user (PU) can
tolerate interference from a secondary user (SU) [7], [8].
Cooperative spectrum sensing (CSS) [9]-[11] is an environment where two or more spectrum
sensing nodes, that form part of a CR network, combine their spectrum sensing capabilities leading
to centralized [12] or decentralized [13] decision fusion. CSS allows individual nodes to gain a
more global degree of awareness about spectrum occupancy [14]. It also has the inherent
advantages of increased levels of agility as well as greater accuracy due to the ability to detect a
primary user (PU) that is hidden to a sub-set of sensing nodes due to channel behavior [15]. CSS
has to be considered in the context of increased communication overhead [6]. If the inherent
advantages of CSS are more important as compared to the cost of overhead then it is a viable
trade-off.
Recently a large scale measurement campaign (for evaluation of CSS) was conducted [16]
which analyzed various methods of correlation between two sensing nodes. However, this work
[16] was not extended to development of CSS (grouping, hierarchical or other) algorithms.
Grouping of CSS nodes has the potential to reduce the workload on individual nodes. Various
techniques (inspired by multiple access algorithms) were suggested in [17] for handling of CSS
nodes. These techniques incorporated the creation of sub-sets or groups from within the CSS
nodes in a network. However, the criteria for creating groups were not defined.
The proposed technique defines effective criteria for creation of groups. This technique is then
extended to the tasking of CSS nodes through the creation of a hierarchical structure. By
effectively employing grouping criteria, we are able to create a hierarchy that provides advantages
in terms of agility, accuracy and information gain.
2 Theoretical Background
In this section, cross-correlation between individually sensed spectra is being discussed. The
grouping possibilities for CSS nodes are also explained.
3
2.1 Measured Cross-Correlation
In [16], the statistics gathered in various joint measurements of the spectrum were analyzed.
Two units were used; one of them was mobile (on a trolley) while the other was a static unit. Each
unit consisted of a spectrum analyzer with rugged laptops. The units were verbally time
synchronized to analyze the correlation between the measurements made by each unit.
The cross-correlation between two processes U and V is given by:

=
=
L
t
t t
V U
L
V U R
1
.
1
) , (
(1)
Where U and V are the two vectors of length L. The problem with this evaluation methodology
is its dependence on the duty cycle [16]. A higher duty cycle can result in high cross-correlation
whereas the absolute power may not be that correlated. This also highlights the need to have
proper time-synchronization.
Covariance and normalized covariance were also candidates for evaluation of individually
sensed spectra [16]. Covariance addresses the problem of absolute power correlation using
subtraction by mean. However, covariance and normalized covariance are sub-optimal due to the
possible approximation of desired signals as noise [16].
Another candidate was weighted cross-correlation during the measurement campaign [16].
Weighted cross-correlation helps in highlighting the difference between received powers from a
particular PU measured by the two spectrum analyzer units [16].
Besides the above mentioned techniques, binary correlation was also considered in [16]. Binary
correlation analyzes the hypothesis testing (explained in Section III) results obtained by energy
detection methods adopted by each CSS node. The results of the hypothesis testing are fused in the
form of binary vectors. This technique is considered as a near-optimal solution for information
exchange [16].
2.2 Grouping of CSS Nodes
Grouping of CSS nodes was suggested in [17] to facilitate the employment of CSS techniques
(which include Time-Division, Frequency-division (FD), Staggered Frequency-division (SFD),
frequency hopping, irregular sub-band and staggered irregular sub-band). Each of these techniques
has its own advantages in particular scenarios. For example, if a large band is to be sensed then FD
or SFD can be used to cover that band [17]. By division of the large band into smaller sub-bands,
an increase in agility is possible [17]. This requires the definition of suitable criteria for
establishment of groups.


3 Ground Work For Proposed Cooperative
Spectrum Sensing Technique
In this section, few pertinent conclusions are presented that were appeared as the initial ground
work for development of the proposed CSS technique.
3.1 Use of Correlation to Determine Information Gain
CSS nodes can provide information gain to each other due to the diversity in individually sensed
spectra. This can be considered analogous to two people having finite fields of view as shown in
Fig. 1. The potential information gain that they can provide to each other is proportional to the
diversity in their views. Correlation between the individually sensed spectra determines the extent
of information gain that would be possible. An increase in the number of CSS nodes may enhance
the agility and accuracy of the fused decision. However, the information gain is not linearly
proportional to the number of sensing nodes because the varying cross-correlation between sensing
4
pairs tends to increase or decrease the cumulative information gain.

Fig. 1. A correlation analogy to determine information gain.
3.2 Simulation versus Empirical Measurements
In order to analyze and develop techniques for CSS implementation in a networked
environment, it becomes necessary to compare the advantages of using simulations or empirical
measurements. Empirical measurements yield the most realistic results; however, certain practical
aspects/constraints related to an effective measurement campaign have to be considered while
making a choice to use measurements or simulated data.
These practical aspects/constraints include: (a) The minimum equipment composition of each
CSS test node should include a spectrum analyzer, a rugged laptop and a data communication
terminal with time synchronization applications. The test node should be self-contained for power
requirements and mobility. (b) The number of such nodes should be sufficiently large to establish
reasonable statistics about the environment. Two or three such nodes would be too less. (c) Each
of the spectrum analyzers should have the ability to be time synchronized with its peers and should
have equivalent configuration.
Due to non-availability of devices that comprehensively meet these requirements, simulation is
the best available option for development and testing of CSS management algorithms and
techniques.
3.3 Spectral Environment Simulation
PUs and SUs are considered to be randomly placed in a square area. An example of such a
deployment is shown in Fig. 2.
Each PU is randomly allocated a distinct frequency from the usable frequencies. This allows for
extraction of maximum diversity. The scenario is an emulation of simple trunked wireless systems.
5

Fig. 2. Random placement of PUs and SUs in a 30x30 km2 area.


The distance from an SU to each of the PU is calculated using Euclidean [18]:
2 2
) ( ) (
m n m n nm
y y x x d + =
(2)
Where, it is assumed that the total number of SU is N and the total number of PU is M. The x-
and y-coordinates of SU and PU are
) , (
n n
y x
and
) , (
m m
y x
respectively, where m=1,2M and
n=1,2,N. This results in the creation of a distance matrix D.
(
(
(
(
(
(

=
NM N N N N
M
M
M
d d d d d
d d d d d
d d d d d
d d d d d
D
.....
. . . . .
......
......
......
4 3 2 1
3 34 33 32 31
2 24 23 22 21
1 14 13 12 11
(3)
In order to cater for the effects of distance and frequency on propagation, the standard free-
space path loss [19] is considered as follows:
f
r t t
nm r
L d
G G P
d P
2 2
2
) 4 (
) (
t

=
(4)
where P
r
(d
nm
) is the received power, P
t
is the transmitted power, G
t
is the transmitter antenna gain
(for the PU), G
r
is the receiver antenna gain (for the SU), is the wavelength obtained from the
allocated frequency to the concerned PU, d
nm
is the relevant distance from the D matrix and L
f
is
the system loss factor. The antenna gains [19] are computed from:
2
4

t
e
r t
A
G G = =
(5)
where A
e
is the effective antenna aperture. Suitable assumptions related to P
t
, A
e
and L
f
can be
taken while creating the simulation. A random additive white Gaussian noise (AWGN) figure
nm
q is added to the resultant power for each link. Thus , a received power matrix P is obtained
such that:
nm nm r nm
d P P q + = ) (
(6)
6
(
(
(
(
(
(

=
NM N N N N
M
M
M
P P P P P
P P P P P
P P P P P
P P P P P
P
.....
. . . . .
......
......
......
4 3 2 1
3 34 33 32 31
2 24 23 22 21
1 14 13 12 11
(7)
The matrix P is then split into vectors corresponding to each of the SU. This is simply done by
extracting the rows. Thus, the vectors P
n
is obtained for each SU.
| |
nM n n n n
P P P P P .. ..........
3 2 1
=
(8)
The vectors are sorted according to the frequency of the relevant PU and one-to-one
correspondence is established between this vector and the vector of usable frequencies. Thus the
size of the vector P
n
is changed to the size of the usable frequencies vector with interpolation of
AWGN figures at the empty elements. It is further resulted in a simulated spectrum for each SU as
shown in Fig. 3.

Fig. 3. Simulated spectrum for an SU with 50 Pus
4 Proposed Cooperative Spectrum Sensing
Technique
In this section, a novel technique for CSS is being proposed and employed it on the spectrum
simulated in Section 3.
4.1 Individual Sensing Phase
In the first phase of the proposed technique, each CSS node senses the entire spectrum of
concern using energy detection. For this purpose, a hypothesis testing is performed considering a
threshold that is higher than the maximum AWGN level. In the event of a detected PU at a
particular frequency resulted in:
' 1 ' :
1
>
nm
P H
(9)
In the event of no PU at a particular frequency, it is resulted in:
' 0 ' :
0
<
nm
P H
(10)
7
A sparse vector SU
n
is thus created for each SU. This sparse vector [20] is compressed using a
zero-counting algorithm that counts the zeros after each one. This has the advantage of reduction
of data size that has to be shared.
4.2 Distance based Correlation Phase
The Euclidean distance between each SUs is computed. Euclidean distance matrix is created as:

(11)

DistSUmMatrix is a square form matrix. It is used for carrying out clustering analysis. This
analysis can be depicted in the dendrogram shown in Fig. 4.

Fig. 4. Grouping dendrogram created on the basis of distance between sensed spectra by 30 CSS
nodes.
4.3 Grouping Phase
The CDFC establishes groups of CSS nodes on the basis of the pair-wise values determined in
the form of DistSUmMatrix. For this purpose, the built-in clustering function of MATLAB version
7.3.0 was used. First, the DistSUmMatrix was identified as a square form matrix to MATLAB.
Then the single linkage function was used as per the Johnson algorithm. The cluster function of
MATLAB was then used to perform clustering with maxclust parameter set to 30% of the total
number of SUs. This phase resulted in identification of the point at which the hierarchical tree is to
be cut [21]. As a result of this phase, the distinct groups of SU are created as shown in Fig. 5.

8

Fig. 5. Establishment of grouping among SU. The red numerals indicate the group to which each
SU belongs.
4 Sensing Error
In order to establish the accuracy of the proposed CSS methodology, the percentage of missed
PUs analyzed and compared with the peer based sensing. A very low percentage of missed PUs
was observed (Fig.6) and this percentage was comparable to the sensing error in peer based
sensing.

Fig.6. Percentage of missed PUs in the proposed methodology
4.5 Tasking of Grouped CSS Nodes
The actual objective of grouping is to enable the employment of a tasking mechanism within
groups. The grouping mechanism, itself, has ensured that maximum information gain is obtained
through sharing of aggregated sensed spectra between groups.
Therefore, the following tasking mechanism for grouped CSS nodes was proposed:
(a) A group decision fusion centre (GDFC) is defined for each group.
(b) The GDFC is informed about the members of its group by the CDFC.
(c) Each GDFC divides the entire spectrum to be sensed into equal parts according to the number
of group members thus creating equal sub-bands within the group.
9
(d) These non-overlapping sub-bands are sensed by the tasked SU according to the instructions of
the GDFC.
(e) Each GDFC fuses the results obtained from each member SU and passes the aggregated result
to the CDFC.
(f) The CDFC fuses the results from each GDFC to obtain a global picture of the sensed
spectrum.
(g) Fusion of results is optimized by reduction of vector size through creation of sparse vectors
[20].
The tasking mechanism can be understood from the hierarchy shown in Fig. 7.


Fig.7. Hierarchy for tasking and decision fusion of CSS nodes
5 Conclusion
The results obtained using the proposed innovative technique were compared with peer-based
cooperation technique, in which all sensing nodes acted as peers and then combined their results in
a centralized fashion. The proposed approach of optimization is based on grouping among
cooperating secondary clients for deployment of an agility driven tasking method resulted in
reduction of interference to primary users through their early detection. The problem of spectrum
deficit by accommodating unlicensed clients in less populated under-utilized fragments of the
spectrum is addressed as a result of this optimization. A quantum jump in terms of agility without
incurring any cost in terms of accuracy is also achieved. The proposed spectrum management
scheme is also suitable to determine the effects of varying group size on agility

6 Reference
[1] A. Ghasemi and E.S. Sousa, "Collaborative spectrum sensing for opportunistic access in
fading environments," IEEE International Symposium on New Frontiers in Dynamic Spectrum
Access Networks, pp.131-136, November 2005.
[2] Federal Communications Commission, Notice of Proposed Rulemaking, in the Matter of
Unlicensed Operation in the TV Broadcast Bands (ET Docket No. 04-186) and Additional
Spectrum for Unlicensed Devices Below 900 MHz and in the 3 GHz Band (ET Docket No.
02- 380), FCC 04-113, May 2004.
[3] SDR Forum; What is SDR; available at:
http://www.sdrforum.org/pages/aboutSdrTech/whatIsSdr.asp
[4] SDR Forum; SDR Related Technologies; available at:
http://www.sdrforum.org/pages/documentLibrary/documents/SDRF-06-R-0011-V1_0_0.pdf
[5] J. Mitola, Cognitive Radio: An Integrated Agent Architecture for Software Defined Radio,
PhD dissertation, Royal Institute of Technology (KTH) Stockholm, Sweden, 2000.
[6] D. Cabric, S. M. Mishra, and R. W. Brodersen, Implementation issues in spectrum sensing
for cognitive radios, Asilomar Conf. on Signals, Systems and Computers, Pacific Grove, CA,
USA, pp. 772776, November, 2004,.
10
[7] G. Ganesan and Y. Li, "Agility improvement through cooperative diversity in cognitive
radio," IEEE Global Telecommunications Conference, 2005.
[8] A. Ghasemi and E.S. Sousa, "Interference Aggregation in Spectrum-Sensing Cognitive
Wireless Networks," IEEE Journal of Selected Topics in Signal Processing, vol.2, no.1,
pp.41-56, February 2008.
[9] J. Unnikrishnan and V. V. Veeravalli, "Cooperative Sensing for Primary Detection in
Cognitive Radio," IEEE Journal of Selected Topics in Signal Processing, vol.2, no.1, pp.18-
27, February 2008.
[10] G. Ganesan and Y. Li, "Cooperative Spectrum Sensing in Cognitive Radio, Part I: Two User
Networks," IEEE Transactions on Wireless Communications, vol.6, no.6, pp.2204-2213, June
2007.
[11] G. Ganesan and Y. Li, "Cooperative Spectrum Sensing in Cognitive Radio, Part II: Multiuser
Networks," IEEE Transactions on Wireless Communications, vol.6, no.6, pp.2214-2222, June
2007.
[12] J. Unnikrishnan and V. V. Veeravalli, "Cooperative Spectrum Sensing and Detection for
Cognitive Radio," IEEE Global Telecommunications Conference pp.2972-2976, 26-30
November 2007.
[13] J. F. Chamberland and V. V. Veeravalli, Decentralized detection in sensor networks, IEEE
Transactions on Signal Processing, vol. 51, no. 2, pp. 407-416, 2003.
[14] L. Chen, J. Wang and S. Li, "An Adaptive Cooperative Spectrum Sensing Scheme Based on
the Optimal Data Fusion Rule," 4th International Symposium on Wireless Communication
Systems, 2007. ISWCS 2007, pp.582-586, 17-19 October 2007.
[15] R. Chen, J. Park and K. Bian, "Robust Distributed Spectrum Sensing in Cognitive Radio
Networks," The 27th Conference on Computer Communications. IEEE , pp.1876-1884, 13-18
April 2008.
[16] W. Matthias, R. Janne, G. Martin, M. Petri, "Evaluation of Cooperative Spectrum Sensing
Based on Large Scale Measurements," IEEE Symposium on New Frontiers in Dynamic
Spectrum Access Networks pp.1-12, 14-17 October. 2008.
[17] C. Lee and W. Wolf, "Multiple Access-Inspired Cooperative Spectrum Sensing for Cognitive
Radio," IEEE Military Communications Conference, pp.1-6, 29-31 October. 2007.
[18] A.R. Webb, Statistical Pattern Recognition, John Wiley & Sons, Ltd, 2
nd
Edition, 2002.
[19] T. Rappaport, Wireless Communications: Principles and Practice; Pearson Education, Inc.,
2
nd
Edition, Englewood Cliffs, NJ: Prentice-Hall, 2003.
[20] W. F. Tinney, V. Brandwajn and S. M. Chan, "Sparse Vector Methods," IEEE Transactions
on Power Apparatus and Systems, vol. 104, no.2, pp.295-301, February. 1985.
[21] A.K. Jain and R.C. Dubes, Algorithms for Clustering Data, Prentice Hall, Upper Saddle
River, N.J., 1988.