You are on page 1of 6

Proceedings of 2015 IEEE 12th International Conference on Networking, Sensing and Control

Howard Civil Service International House, Taipei, Taiwan, April 9-11, 2015

P-phase Picker Using Virtual Cloud-Based Wireless


Sensor Networks
Maki Matandiko Rutakemwa

IvenJose

PhD Student Computer Science PHDCS 1145008


Christ University Bangalore
Bangalore,Karnataka,India
makinsoft@gmail.com,
makimatandiko.rutakemwa@christuniversity.in

Faculty of Engineering
Christ University Bangalore
Bangalore,Karnataka,India
iven@christuniversity.in

Abstract-Wireless

Sensor

Networks,

is often done by graphical scrutiny of veteran seismologists.


Thus, when the bulk and frequency of data capture is huge,
this procedure is exceedingly laborious, timewasting, slow,
inefficient, very long, arduous and subject to discrepancy
across different examiners.
Recently, automated P-phase picking algorithms have been
developed but unfortunately,they require powerful nodes with
weighty computation,stomge and power resources.
Hence, the primary objective of this work is to create a
framework that allows sensor network infrastructures to
take full advantage of cloud computing approach so that
sensing frames can be viewed as an extensible resource
environments that can be dynamically re-purposed and
re-programmed. The main task in apprehending this
journey is the design of a novel wireless sensor network
architecture that supports the dynamic computational
requirements of volcano monitoring.
The second key contribution of the present paper is the
implementation of a new automated signal processing
algorithm for P-phase picking which is going to be
suitable for a cloud-based environment.
The succeeding sections are organized as follows:
Section 2 reviews related work. Section 3 states the
problem. Section 4 presents the virtual sensor cloud
infrastructure and the signal processing algorithm at
sensor cloud,Section 5 presents the results and Section 6
concludes the paper.

mainly

regarded as numerous resource-limited nodes linked


via low bandwidth, have been intensively deployed
for active volcano monitoring during the few past
years. This paper studies the problem of primary
waves received by seismic wireless sensors suffering
from limited bandwidth, processing capacity, battery
life and memory. To address these challenges, a new
P-phase

picking

virtualized

using

approach
cloud

where

computing

sensors

are

architecture

followed by a novel in-network signal processing


algorithm, is proposed. The two principal merits of
this paper are the clear demonstration that the Cloud
Computing model is a good fit with the dynamic
computational requirements of volcano monitoring
and the novel signal processing algorithm for accurate
P-phases picking. The proposed new model has been
evaluated

on

Mount

Eucalyptus/Open

Stack

Nyiragongo

with

using

Orchestra-Juju

for

Private Sensor Cloud then to some famous public


clouds

such

as

Amozon

SensorCloud and Pachube.

EC2,

ThingSpeak,

The testing has been

successful at 75%. The recommendation for future


work would be to improve the effectiveness of virtual
sensors

by

applying

optimization

techniques

and

other methods.
Keywords:

Virtual

Wireless

Sensor Networks; P-

phase Picking; Ubiquitous Computing; Cloud Computing

II. RELATED WORK


I.

INTRODUCTION

The concept of Virtual Sensor has been investigated in a


few research works [4], [2] in the context of WSNs [5].
In [4], authors define Virtual Sensors as soft ware sensors
that provide indirect measurements of abstract conditions
by combining sensed data from a group of heterogeneous
physical sensors. This approach does not consider fine
gmined, signal-processing-oriented virtual sensors and is
mainly based on query-driven virtual sensors (based on a
pull model) instead of data-driven virtual sensors (based
on a push model) that would have higher performance in
Earthquake timing determination.

The current volcano monitoring systems employ


broadband seismometers which, although can yield high
fidelity seismic monitoring signals but they remain
expensive, power hungry, bulky, and difficult to install.
These limitations have largely prevented them from wide
deployment, providing limited coverage and coarse
grain monitoring. This major drawback restricts scientists'
ability to monitor the volcano changing aspects and foresee
eruptions.
Moreover, in current volcano observatories, P-phase picking

978-1-4799-8069-7/15/$3\.00 20151EEE

334

have satisfactory precision and maximize the accuracy of


earthquake hypocenter estimation.

In [2] a framework for building virtual sensors and


actuators in wireless sensors and actuators networks is
presented. In particular, authors propose virtual nodes as
a programming abstraction simplifying the development
of decentralized WSN applications. This technique can
be perceived as the reading of a single virtual sensor
which is not enough in the case of volcano monitoring
problems.
The concept of Virtual Sensor Networks (VSNs) has also
been described in mUltiple works.
In [9], authors present VIP Bridge that aims at
connecting heterogeneous sensor networks with IP based
wired/wireless networks, and integrate these sensor
networks into one VSN. This allows for discovering and
querying of sensor nodes located in different and
heterogeneous sensor networks.

IV. VIRTUAL SENSOR CLOUD A ND SIGNAL PROCESSING


ALGORITHM A T SENSOR CLOUD

Physical sensors map an observed physical quantity,


such as temperature,acceleration, or sound, onto a data
value and produce an output. The output is generated
when inputs change, as the result of an event, or in
response to a (timed) request. Physical sensors are
transducers converting values from one form to another
using physical processes. Signal processing algorithms
convert values using digital processes.

S. Kabadayi et. al [4] present another approach to the


definition and implementation of VSNs that provides
protocol support for the fonnation,usage,adaptation and
maintenance of subsets of sensors collaborating on
specific tasks. VSNs can also enable applications that
involve dynamically varying subsets of sensor nodes
collaborating tightly to achieve the desired outcomes,
while relying on the remaining nodes to achieve
connectivity and overcome the deployment and resource
constraints [10].

Figure i: Physical Sensors VS. Virtual Sensors

This observed similarity is the motivation behind the


virtual sensor abstraction. Every processing task can be
represented as a virtual sensor. Therefore, if we consider
a complete VSN system, we can model its data
processing part as a multi-level hierarchy of virtual
sensors as shown in Fig. 2. Moreover, virtual sensors
may be implemented directly in a programming
language, or as networks of already existing virtual
sensors.

The seismology community previously developed


several algorithms for P-phase picking. They are
typically based on the identification of changes in signal
characteristics such as energy, frequency and
characteristics of autoregressive models [11]. These
methods difficult to adapt to different regions or
temporally changing environments.
Another important category of picking algorithms is
based on autoregressive (AR) models [6]. These methods
pick the time instance to maximize the dissimilarity of
two AR models for signals before and after the picked
time instance. AR-based algorithms need few user
settings and are the most accurate and robust P-phase
picking algorithms to date [3]. However, since both AR
models must be constructed for each time instance, they
incur high computational complexity and memory usage.
In the context of our paper, such proposals can be
customized for enabling the integration of Cloud
computing to wireless Sensors in order to have networks
of Virtual Wireless Sensors that can be created for
monitoring not only a given volcano but also an even
large number of volcanoes.
III.

Computational

Fig. 2. Multi-layer signal processing

Software frameworks are usually introduced to provide


programmers with abstractions to isolate them from low
level implementation details. Virtual sensors provide a
new level of abstraction at the software level by allowing
signal processing tasks to be defined and composed
easily. Furthermore, VS abstractions allow signal
processing tasks to be modified or changed at design or
runtime without affecting the rest of the system. In Fig.2
every component represents a processing task applied to
a stream of data originated from physical sensors and

PROBLEM STATEMENT

The work to be done is focused upon two key objectives.


First, virtualize the physical sensors through cloud
computing model so that the sensors' resources can be
extended. Second, to achieve a strong in network signal
processing in order to ensure that the picked P-phases

335

can be modeled as a virtual sensor. The output of each


virtual sensor is defmed by a set of inputs and its
configuration. More formally, a virtual sensor j, denoted
as VSj is defined as:

,and K is
the of links contWcti no
{VS1' VSm}

inpu ts and

outputs of

As far as in networking signal processing algorithm is


concerned, the P-phase is linearly polarized with respect
to the direction of propagation and the metric used to
measure the degree of linear polarization is helpful to
detect the P-phase arrival [1]. Such a metric known as
the rectilinearity function is defined by Kanasewich [12].
Its equation is:

Equation J: Virtual Sensor Formalization

Where Ej denotes the set of inputs, Fj denotes the set of


outputs, Dj denotes the configuration and Hd denotes the
state of VSj. The configuration of each virtual sensor
defmes the type of its inputs and outputs, the particular
implementation used for a given computational task, and
a set of parameters required for a particular
implementation.
In particular,Dj is defined as:

R = l - (6e-)1
Equation 4: rectilinearity function of Kanasewich

Where ill and il are the largest and the second largest
eigenvalues of the covariance matrix respectively.
If the covariance matrix Equation 5 is diagonalized, an
estimate of the rectilinearity of particle motion trajectory
over the specified time window can be obtained from the
ratio of the principal axis of this matrix [12] i.e. the
rectilinearity estimation can be obtained from the ratio of
the largest and the second largest eigenvalues of the
covariance matrix. The covariance matrix [12] is defined
as:

Equation 2: Definition of the configuration D

Where %n describes the types of inputs, qout for the


outputs, v represents the specific VS implementation,
and w denotes the VS configuration parameters.
This defmition provides high modularity for application
design. In fact, different configurations of the same
virtual sensor can be easily substituted without requiring
changes in the rest of the design. This property therefore
enables a component based approach for application
development in which an application is assembled out of
well-defined components appositely interconnected.
Moreover, it can be used when environmental changes
require a new implementation of a particular signal
processing component for a given application.
Alternative implementations do not need to be loaded
into main memory at all times. They can be stored in
flash memory,or transferred over the air upon request.
VS can be further composed to create higher-level VS.
This allows to define multiple abstraction levels that
capture the successive processing and interpretation of
sensor data and system components that perform data
fusion. High-level VS identify abstractions that are
useful to support code modularity and reusability. In
fact, if an implementation of a VS is replaced with
another one, where one or more VS components are
changed but the interface is the same,there is no need to
change the rest of the system.
More formally, the composition of n Virtual Sensors to
form a higher-level VS can be defined as follows:

cav(X,X)

M = Co-v (!',X)

Co-v(Z,X)

Cav(X,!')
Co-v (!" !')
Co-v (Z,!')

cav(X,Z)
Cav(!' ,Z)
Cav(Z, Z)

Equation 5: Covariance matrix

Where
X: east component wavelet coefficients at scale j;
Y: north component wavelet coefficients at scale j;
Z: vertical component wavelet coefficients at scale j.
While the covariance of X and Y is defined as:

Cov (X,y)

N' L(XCi)
I

[=1

- !-' ..

J (yW

1-'y)

Equation 6: Covariance of X and Y

Where U x and 1-'y are the mean values of X and Y


respectively. The direction of polarization may be
measured by considering the eigenvector of the largest
principal axis.
Hence, the P-phase algorithm can be described in the
following steps:

Equation 3: Definition of the Higher-Level Virtual Sensor


Node

(i) Processing the three component seismic signals using


Discrete Time Wavelet Transform (DTWT)
A three component seismic signals represented by east
(X), north (Y) and vertical (Z) are processed by the
discrete time wavelet transform (DTWT) to calculate the
wavelet coefficients wcI, wc2 and wc3 respectively.

Where

E' El U Ez. ......U Em, F' Fl U F. .... ..U Fm,D " =


W1, D 2 , ... ..Dm},H is t. hB state of {VS1,...... VSm}

336

Each component is decomposed into several scales. So,


the results are the wavelet coefficients xli, x and x} for
east, north and vertical components respectively (j is the
number of scales).

Figure 4 shows the testing result of an earthquake (event


5). It illustrates the composite rectilinearity function Cf
and how it is used to locate the P-arrival time.
Figure 5 shows the tested result of another earthquake
(event 1) and the composite rectilinearity function Cf. It
shows that the peak of this function is used to determine
the P-arrival time. It can be observed from the figure that
the position at this function is a maximum is where P
phase arrives.
In order to test the performance of the algorithm, it can
be observed that the mean magnitude of algorithm error
is low (0.1952 seconds) as compared with STA/LTA
mean magnitude error (0.3982 seconds). This proves that
the performance of the algorithm is efficient.

(ii) Constructing the rectilinearity function


At each scale, a pointwise moving window is
constructed over the three components.
At each window the rectilinearity function is constructed
so that the result is a rectilinearity function (Qi) for each
scale. Then, a composite rectilinearity function (SQ) is
constructed so that the rectilinearity function of each
scale contributes in this function. This composite
function is constructed by (Equation 7).

Equation 7: Rectilinearity function

where i is the scale number.


The location where this function gets its maximum value
is taken as the P-arrival time.

20

40

60

60

100

160

180

.1000

1:------04

-1500

0!;--20==---t.O:----;60!;;--ao;!;-,-!::,OO;:---7.,20-,,':;;.O--,;-!::60,--,ao

1r------r--_-r---,

(iii) Calculating the back azimuth angle


Back-azimuth is the angle measured from north to the
direction from which the energy arrives at a given
station. It is used to determine the longitudinal and
transverse directions for an incoming ray at a prescribed
station. For an incident P wave the backazimuth is
calculated by constructing the rectilinearity function and
finding the eigenvector associated with the maximum
eigenvalue for the detected P-wave. This eigenvector
represents the direction of linear polarization that
specifies the back azimuth angle. This calculation can be
conducted at each wavelet scale as well as on the
original signals. But, since the original three component
signals generally contain noise, the calculation is carried
out at the third and higher scales while the first two
scales that contain high frequency noise are excluded. It
is found that the polarization is conserved along different
scales i.e. the same back azimuth angle is obtained in the
different scales. So, one value is considered in the
processing.
V.

"3
:

20

40

60

80

100

120

140

160

180

20

40

60

60

100

120

140

160

180

20

40

60

80

100

120

140

160

180

20

40

60

80

100

120

140

160

160

20

40

60

80

100

120

140

160

180

20

40

60

60
100
Time (In seconds)

120

140

160

160

<
<
<
<

Figure 3: Three component waveforms of event 21 and


its related rectilinearity functions of six different scales.

CLOUD BASED P-PHASE DETECTION ALGORITHM


RESULTS

The rectilinearity function proved to be an effective


metric in locating the P-arrival time. Figure 3 shows the
rectilinearity functions of six different scales obtained by
testing the algorithm with event 3. It can be observed
that the rectilirJearity function is approximately equal to
1 when the waveform is lirJearly polarized and equal to
zero when there is no linear polarization. This can be
seen clearly in the first four scales.

337

j
j
j
j
j
j

><

lao,

1400
1500

mt

station with epicentral distance of 142.898 km and back


azimuth of 54.03 degrees and the composite function Cf

p,pasea

1300

1'"
100

50

'r

150

200

250

300

350

200

250

300

350

1400
>-

1200

1..

,..
1000
0
1800
N

50

100

150

1800
1400
0

50

100

50

100

150

200

250

300

350

150

200

250

300

350

10

Figure 4: Three component data (east, north and


vertical) of event 5 recorded by JRSN Lwiro station (one
of the northern stations) with epicentral distance of
226.4 km and backazimuth of 106.9 degrees and the
composite rectilinearity function Cf

Figure 6: The three component signals of event 13


zoomed around P-arrival time to compare the analyst
result with the algorithm result.

It can be observed that the mean magnitude of algorithm


error is low (0.1952 seconds) as compared with
STA/LTA mean magnitude error (0.3982 seconds). This
proves that the performance of the algorithm is efficient.

The comparison between the analyst result and the


algorithm result is illustrated in Figure 6. It can be
noticed that the results are consistent with each other.
The three components signals illustrated in Figure 6 are
recorded by a station with noisy local conditions. That's
why; the STA/LTA (short term average/long term
average) method couldn't give a result for P-arrival time
for the data recorded by this station. However, the P
phase detection algorithm gives a good result though of
this noise condition. The algorithm result is compared
with the analyst result as well as with STA/LTA results.
The comparison shows that the three results are
consistent with each other as illustrated in Figure 7.
1'.I>h"scarri\1

1I!11

.L.u" '"

X.13lI
.,2!iI,D
12!II

'fl,q"p1rr'f'
II

!II

III

10

II)

lID

!II

110

,1,,1.1.1.

Figure 7:The result


analyst result and
event 22 recorded
distance of 8.365
degrees.

1, .... ,
'1"1011'1"'1"
1I!11Ji

!II

III

lID

10

110

zl': : : ' .-j


crt .IL.:" :,.,",:.1: 1: I.: :," :.1 l
IIlJi

II

II

!II

!II

10

III

III

.. .

10

Time (in seconds)

II)

II)

!II

" ,
!II

lID

110

of comparing the algorithm with the


STA/LTA result when testing it by
by Kabare station with epicentral
km and backazimuth of 236.22

VI.

CONCLUSION

Virtual sensors allow abstracting even complex systems


that behave like sensors, thus favoring code modularity
and reuse.
We have combine the virtualized wireless coud-based
sensors with an in-situ signal processing algorithm in
order to construct a strong P-phase picking system.
This system is needed for volcanic earthquake

Figure 5: Three component data (east, north and


vertical components) of event 9 recorded by kalehe

338

monitoring.
Within this novel architecture users can share several
types of physical sensors easily and consequently many
new services can be provided via a virtualized structure
which allows allocation of sensor resources to different
users and applications under flexible usage scenarios
within which users can easily collect, access, process,
visualize, archive, share and search large amounts of
sensor data from different applications.

Parolai, Seismic Phase Picking Using a


Kurtosis-Based Criterion in the Stationary
Wavelet Domain, IEEE Transactions on
Geoscience and Remote Sensing 12/2012.

Moreover, an implementation has been achieved using


Arduino-Atmega328 as hardware platform and
Eucalyptus/Open Stack with Orchestra-Juju for Private
Sensor Cloud. Then this private Cloud has been
connected to some famous public clouds such as
Amazon EC2, ThingSpeak, SensorCloud and Pachube.
The testing was successful at 80%.

[7]

R. Sleeman and T. Van Eck. Robust automatic


p-phase picking: an on-line implementation in
the analysis of broadband seismogram
recordings. Physics of the earth and planetary
interiors,113,2009.

[8]

R. Tan, G. Xing, 1. Chen, W. Song, and R.


Huang. Quality-driven volcanic earthquake
detection using wireless sensor networks. In
RTSS,2010.

[9]

G. Werner-Allen, K. Lorincz, J. Johnson, 1.


Lees, and M. Welsh. Fidelity and yield in a
volcano monitoring sensor network. In OSDI,
2012.

[10]

Konstanteli,
K. Cucinotta,
T. ; Psychas,
K. ; Varvarigou, T.A., Elastic Admission
Control for Federated Cloud Services, Dept.
Of Electr. & Comput. Eng., Nat. Tech. Univ.
Of Athens,Athens, Greece,22 October 2014

[11]

M. Withers, R. Aster, C. Young, J. Beiriger,


M. Harris, S. Moore, and J. Trujillo. A
comparison of select trigger algorithms for
automated global seismic phase and event
detection. Bulletin of the Seismological
Society of America,88(1),2011.

The recommendation for future work would be to


improve the effectiveness of virtual sensors by applying
optimization techniques and other methods.
REFERENCES

[1]

[2]

E. Cand'es and M. Wakin. An introduction to


compressive sampling. IEEE Signal Process.
Mag., 25(2),2012.
P. Ciciriello, L. Mottola, and G. P. Picco,
"Building virtual sensors and actuators over
logical neighborhoods," in midsens '06:
Proceedings of the international workshop on
Middleware for sensor networks. New York,
NY,USA: ACM,2013,pp. 19-24.

[3]

E. Endo and T. Murray. Real-time seismic


amplitude measurement (RSAM): a volcano
monitoring and prediction tool. Bulletin of
Volcanology,53(7),2007.

[4]

S. Kabadayi, A. Pridgen,and C. Julien, "Virtual


sensors: Abstracting data from physical
sensors," A World of Wireless, Mobile and
Multimedia
Networks,
International
Symposium on,pp. 587-592,2010.

[5]

F. Lewis, "Wireless sensor networks," in


Smart Environments: Technologies, Protocols,
and Applications. Cook, D.J. and Das, S.K.
(eds). John Wiley / New York, 2014, pp. 1924.

[6]

J.J.

Galiana-Merino,J.L.

[12] E. R. Kanasewich, "Time Sequence Analysis in


Geophysics," University of Alberta Press,Edmonton,
2011.

Rosa-Herranz,S.

339

You might also like