International Conference on Robotics, Automation, Control and Embedded Systems – RACE 2015

18-20 February 2015, Hindustan University, Chennai, India.

Localization with Beacon Based Support

Vector Machine in Wireless Sensor
Networks
Z. Mary Livinsa Dr. S. Jayashri
Director, Adhiparasakthi Engineering College,
Research scholar, Sathyabama University,
Melmaruvathur, Tamil Nadu, India.
Chennai, India
[email protected]
[email protected]

Abstract: Recent developments in radio technology Basically a WSNs consists of two types of
and processing systems, Wireless Sensor Networks nodes has shown in fig1.The foremost node is an
(WSNs) are tremendously being used to perform anchor node (beacon node) or seed node or
an assortment of tasks from their atmosphere. locator. These nodes know their own position
Localization plays the most important task in
information either through the GPS or manual
WSNs. Accuracy is the one of the major problems
facing localization. In this paper, we propose an location which provide the location information
improved localization algorithm based on the to another node. The second node is the ordinary
learning concept of support vector machine (SVM). node or unknown node which position is
In SVM classification the finite size of grid cells unknown. The unknown node position is
offer the localization accuracy. The localization determined using the anchor nodes. Since anchor
error using the proposed algorithm is calculated nodes are very more expensive than the ordinary
and compared with basic SVM and fuzzy logic. node, a localization algorithm with less number
Simulation result demonstrates that the improved of anchor nodes are favored. Some Metrics
support vector machine can effectively reduce the
associated with localization are energy
localization error and thus achieve the objective of
better accuracy. efficiency, accuracy and security [2].

Keywords: Wireless sensor networks, support vector

machine, localization, fuzzy logic

I.
INTRODUCTION
II.
Wireless Sensor Networks (WSNs) are
Fig.1 Basic types of Node
composed of a large number of sensor nodes that
consists of four components: sensing element,
The organization of this paper consists of
processing element, transmission element and
the following section: Section II classification of
power element. The potential applications of
localization, Section III proposed method,
Wireless Sensor Networks are environmental
Section IV performance analysis, Section V
monitoring, target tracking, surveillance,
Conclusion.
military, security, etc. In Wireless Sensor
III. CLASSIFICATION
Networks, the fundamental problem is
localization i.e, determining the location of
In WSNs research on localization can be
nodes. Global Positioning System (GPS) is the
classified two broad categories: Centralized
most obvious solution of this problem. But the
localization and Distributed localization.
huge, expensive cost and power consumption of
2.1. Centralized localization
GPS, different algorithms are developed for the
localization. While designing the localization In this localization, all nodes collect the data
algorithms the key aspects or issues such as: in the network is sent to the central processing
limited resource, the number and density of node unit (central base station). The central processing
network topology, device capability, localization unit collects the distance measurements of all
accuracy, kind of signals and mobility of node inter-node and compute the localizations back to
(static node or mobile node) must be considered the respective nodes of the entire network. The
[1].

ISBN: 978−81−925974−3−0

Authorized licensed use limited to: Jaypee Insituite of Information Technology-Noida Sec 128 (L3). Downloaded on March 12,2024 at 08:51:53 UTC from IEEE Xplore. Restrictions apply.
 International Conference on Robotics, Automation, Control and Embedded Systems – RACE 2015
18-20 February 2015, Hindustan University, Chennai, India.

advantage of this localization is it eliminates the 2.2.5. Interferometric Algorithms: In this

computation and communication cost [3]. algorithm the necessary ranging information is
obtained by emitting interfering radio waves
2.1.1. Multi Dimensional Scaling (MDS): The from two different locations at slightly different
one of the best schemes of centralized frequencies [5].
localization is Multi Dimensional Scaling
(MDS). In MDS a set of node whose position is 2.2.6. Error Propagation aware localization: In
unknown and the measured distance between WSNs the entire nodes communicate with each
each pair of nodes are given. MDS display the other. During communication error propagation
structure of distance as a geometrical picture. It can be occurred due to unavoidable environment
can be used to localize sensor nodes in a such as noise and channel fading. An algorithm
network. Generally, MDS is applied to each called Error Propagation Aware Algorithm
node to estimate a local map. Local maps are (EWA) has proposed to reduce this error
then combined together to form a global map. propagation [6].
Absolute position can be computed if enough
anchor nodes available. Then transforming the IV. PROPOSED METHOD
global map in an absolute map. MDS can Our main goal is finding the accurate
achieve very accurate location when comparing location of the unknown node with the help of
to other schemes [4]. minimum number of anchor nodes. In an existing
2.2. Distributed Localization system fuzzy logic-based approach for mobile
node localization is challenging environments. It
In this localization, the entire computation task is more expensive because it needs more
relies on the nodes themselves (self localization) training, not more accuracy [7]. For that, one
and nodes in the network communicate with each machine learning technique is used for finding
other to estimate their position. Distributed the accurate location called Localization based
localization can be categorized into many on Support Vector Machine (LSVM). We
classes. proposed Improved LSVM-a novel solution that
localizes the network using the learning concept
2.2.1. Beacon-based distributed algorithms: This
of Support Vector Machines (SVMs). This is
algorithm starts with some beacon node and more applicable for larger scale networks
sensor node to obtain the distance measurement because it is based on connectivity like DV-Hop
in the network and then using this measurement [8] rather than the direct signal strength Indicator
to determine their position. Beacon based (RSSI) [9].
algorithm can be categorized in Diffusion,
Bounding Box and Gradient. 3.1 Support vector machine :
Support Vector Machine (SVM) is a
2.2.2. Relaxation- Based distributed algorithms:
machine learning method used for classification
This algorithm uses a grain algorithm (spring
and regression [10]. This classification launches
model, cooperative ranging approach) to roughly
a hyper plane or set of hyper planes in a high-
localize the nodes in the network. In this
dimensional space. It has a noble functional
algorithm, each node adjusts its position to
margin and lower generalization error. Basically
approximate the optimal solution.
an SVM classification method consists of two
2.2.3. Coordinate system, stitching based modules. The first one is a Kernel function
distributed algorithms:In this algorithm the which is demarcated based on the beacons hop
network is divided into small subgroups. Each counts. The second method sets of support
subgroup estimates the distance and creates an vectors obtained from the training phase, which
optimized local map. Then all the local maps is the set of beacons. In SVM classification the
combined together and form a global map. Some finite size of the grid-cells provide the
approaches of this algorithm are: Cluster based localization accuracy. Finer grid cells with a
approach, Construction of Global Coordinate higher value of classes will lead to a smaller
System in a network of Static Computational
localization error and give the best localization
Nodes from Inter node distance.
accuracy at a fixed computation cost [11].
2.2.4. Hybrid localization algorithms: Hybrid Improved support vector machine is designed by
algorithm uses two different techniques such as: further dividing and classifying the finite grid
Multi Dimensional Scaling (MDS) and cells and calculated localization error. The main
Proximity based Map (PDM) or MDS and Ad- advantage of SVM is that it does not require any
hoc Positioning System (APS). These techniques extra devices and require only the hop-count
are used to minimize the communication and
computation cost.

ISBN: 978−81−925974−3−0

Authorized licensed use limited to: Jaypee Insituite of Information Technology-Noida Sec 128 (L3). Downloaded on March 12,2024 at 08:51:53 UTC from IEEE Xplore. Restrictions apply.
 International Conference on Robotics, Automation, Control and Embedded Systems – RACE 2015
18-20 February 2015, Hindustan University, Chennai, India.

information. So it circumvents high so the number of samples is calculated by M-

expenditures. 1=35 (samples) with the total geographic area is
denoted by D, where, D=200 in X and Y
3.2 The simple procedure for SVM dimension. Suppose a node is a nearby cx3 line
A large number of sensor nodes and (i.e., right side) then the position is calculated by
least number of anchor nodes are deployed (i+1/2) D/M. Similarly a node is on left side,
randomly in two dimensional geographical areas then the position is calculated by (I-1/2) D/M.
with a communication range, which should be The maximum error will be calculated using
less than zero. At the initial step, each anchor D/M√4. In Improved LSVM, when M increases
node sends a message to every other anchor node in the set of classes, then the localization error
across the complete network. Each anchor node decreases.
receives the same message by another anchor
node and defines its hop count distance. Then, V. PERFORMANCE ANALYSIS
each anchor node newscast another information In this analysis, we have deployed 100
message to the base station, which contains the sensor nodes randomly in 200m X 200m square
position of the source node and the hop count area. The localization error (m) based on anchor
distance. Therefore, the base station recognizes population (%) for the proposed improved
the position of every anchor node and hop count LSVM is calculated and then compared with
distance between every two anchor node [12]. fuzzy localization is shown in fig 3. This
Our assumption is more flexible because each performance shows that while the anchor
node can communicate with a beacon node by a percentage increases, the location error are
multihop path. Next, the base station starts the decreasing. It shows that when compared to
SVM training phase procedure on two classes. fuzzy localization our proposed method is having
less error. Similarly, improved LSVM with
3.3 Algorithms and protocol for SVM LSVM is shown in fig 4.
A Radix-2 decision algorithm is
organized to classify the x- dimension classes,
demonstrated in Fig 2. The classes are delegated
to the tree nodes and we traverse the tree in the
order of left child →parent → right child .So
the result of the order list is cx1→ cx2→ … →
cxM-1.LSVM organizes the cx-classes in a
hierarchy tree. Each tree node is a cx class and
the two outgoing links represent the outcomes
(0: “not belong”, 1: “belong”) is shown in figure
of classification in this class.

Fig. 3 Localization Error (m) Vs. Anchor (%)

Fig 2. Radix2 decision algorithm for m=4

This is the algorithm for finding the

unknown nodes X- dimension location. Similarly Fig. 4 Localization Error (m) Vs. Anchor (%)
for Y dimension we can calculate the Y position
In result, we showed that by having the
of the unknown node. For example, consider
minimum number of anchor i.e., even 2% of
m=4 then we know M=2m. Therefore, M=24=16,
anchor also can achieve the less localization

ISBN: 978−81−925974−3−0

Authorized licensed use limited to: Jaypee Insituite of Information Technology-Noida Sec 128 (L3). Downloaded on March 12,2024 at 08:51:53 UTC from IEEE Xplore. Restrictions apply.
 International Conference on Robotics, Automation, Control and Embedded Systems – RACE 2015
18-20 February 2015, Hindustan University, Chennai, India.

error of 1.56m. Also for 4%, 6%, 8%, 10% of REFERENCES

anchor population, we can also easily find the [1] Lina M.Pestanna Leao de Brito,Laura M.Rodriguez
location of the remaining nodes with less Peralta: “An Analysis of Localization Problems and
localization error in improved LSVM.We Solutions in Wireless Sensor Networks”,April 2008,vol
v1,ISSN:1645-9911.
calculated the density of the anchor node and the
[2] Srinivasan and Jie Wu Florida Atlantic University,Boca
density of the node. The performance graph is Raton,FL,USA: “ A survey on secure localization in
plotted and shown in Fig 5, and 6. wireless sensor network”-2010
[3] Amitangshu Pal:”Localization algorithms in wireless
sensor networks:Current Approaches and Future
Challenges”-:ISSN 1943-3581-Network Protocols and
Algorithms.www.macrothink.org/npa, 2010.
[4] Zhang Shaoping,Li Guohui,Wei and Yang Bing: “A
Novel Iterative Multilateral Localization Algorithm for
Wireless Sensor Networks”-Journal of Networks-2010.
[5] Rui Huang, Gergely V. Zaruba, and Manfred Huber,
“Complexity and Error Propagation of Localization
Using Interferometric Ranging”, in Proceedings of
IEEE International Conference on Communications
ICC 2007, pp. 3063-3069, Glasgow, Scotland, June
2007.
[6] N. A. Alsindi, K. Pahlavan, and B. Alavi, “An Error
Fig 5. Anchor density Vs Localization error (m) Propagation Aware Algorithm for Precise Cooperative
Indoor Localization”, in Proceedings of IEEE Military
Communications Conference MILCOM 2006, pp. 1-7,
Washington, DC, USA, October 2006.
[7] Harsha Chenji, and Radu Stoleru, “Toward Accurate
Mobile Sensor Network Localization in Noisy
Environments” IEEE transactions on mobile
computing, vol. 12, no. 6, june 2013.
[8] Livinsa, Z. Mary, and S. Jayashri. "LS and MMSE
based Localization Algorithm for WSNs amid
obstacles." Indian Journal of Computer Science and
Engineering, E-ISSN 0976 – 5166,2014.
[9] Livinsa, Z. Mary, and S. Jayashri. Performance analysis
of diverse environment based on RSSI localization
algorithms in wsns." IEEE Conference on Information
& Communication Technologies (ICT),. ISBN-978-1-
4673-5758-6, 2013.
Fig.6 Node Density Vs Localization error(m) [10] Samira Afzal: “ Localization Algorithm for Large Scale
Mobile Wireless Sensor Networks”, Journal of Basic
and Applied Scientific Research, 2 (8)7589-7596, ISSN
VI. CONCLUSION 2090-4304, 2012.
In this paper, we proposed an improved [11] D.A. Tran and T. Nguyen, “Localization in Wireless
localization algorithm based on support vector Sensor Networks Based on Support Vector Machines,”
machine (SVM).The localization error is IEEE Trans.Parallel Distributed Syst., vol. 19, no. 7,
calculated using the proposed algorithm and pp. 981-994, July 2008.
compared with basic SVM and fuzzy logic. In [12] Reza Samadian and Seyed Majid Noorhosseini
“Probabilistic Support Vector Machine Localization in
the simulation result, we exposed that by having Wireless Sensor Networks”, ETRI Journal, Volume 33,
the minimum number of anchor node also can Number 6, December 2011.
get an accurate location and achieve the less
localization error.

ISBN: 978−81−925974−3−0

Authorized licensed use limited to: Jaypee Insituite of Information Technology-Noida Sec 128 (L3). Downloaded on March 12,2024 at 08:51:53 UTC from IEEE Xplore. Restrictions apply.