Current issue- International Journal of Advanced Smart Sensor Network Systems (IJASSN)

International Journal of Advanced Smart Sensor Network Systems (IJASSN), Vol 10, No.1/2/3, July 2020
DOI:10.5121/ijassn.2020.10301 1
DEVELOPMENT OF SOM NEURAL NETWORK BASED
ENERGY EFFICIENT CLUSTERING HIERARCHICAL
PROTOCOL FOR WIRELESS SENSOR NETWORK
Md. Tofael Ahmed1
, Bipasa Sharmin Setu1
,
Maqsudur Rahman2
and A. Z. M. Touhidul Islam3
1
Department of Information and Communication Technology,
Comilla University, Bangladesh
2
Department of Computer Science and Engineering,
Port City International University, Bangladesh
3
Department of Electrical and Electronic Engineering,
University of Rajshahi, Bangladesh
ABSTRACT
Cluster-Based Routing Protocols is a renowned scheme to extend the lifetime and energy consumption
simultaneously for the Wireless Sensor Network (WSN). Every sensor node work homogenously or
heterogeneously which is energy constrained when energy and memory capacity is limited. Congregating
information resourcefully in perilous situations in the sensor network for a large-scale area and huge time
is required an effectual protocol. In this paper, we proposed a cluster-based hierarchical routing path
protocol, namely SOM-PEG protocol, which is a modified PEGASIS protocol based on traditional
PEGASIS with the employment of Self Organizing Map (SOM) neural network (NN). The simulation is
performed on MATLAB simulation tool as well as NN GUI. The performance comparison shows that the
proposed protocol provides better network lifetime and ensures less energy consumption compared with
traditional PEGASIS protocol.
KEYWORDS
WSN, PEGASIS, SOM-PEG, NN GUI, SOM, Energy consumption, Lifetime
1. INTRODUCTION
Wireless Sensor Network (WSN)is formed with a huge amount of tiny sensor nodes in a sensor
field, which are non-deterministically installed over the region where human intervention is a
bulky task. At first, wireless sensor networks were developed dedicatedly for the military
applications specially for battlefield surveillance. Nowadays, it enhances its application in
industrial and civilian area where industrial process control and monitoring, machine monitoring,
environment control, and inhabitant monitoring, patient monitoring, smart home and smart traffic
control are performed by the WSN [1]. A typical view of the WSN has shown in Fig.1 using
some internetworking devices. Neural networks are used to solve hardware and software
problems of WSN applications which are also known as Artificial Neural Networks (ANN) [2].
Parallelization, speed, and flexibility are the strength of it. The main focus of these technologies
is solving pattern recognition problems, clustering network, dynamic time series, etc. To
investigate the energy-efficient cluster-based hierarchical WSN protocols a neural network
protocol is used and aimed to enlighten the network lifetime and decrease the energy
consumption.

2
Figure1. A typical view of WSN with central monitoring [7].
To support the data accretion in the network these nodes moves in the small groups called
clusters. Clustering is used to enhance the networks lifetime, which is a primary metric that
evaluates the performance of a sensor network [2]. Figure 2 shown below describes the basic
hierarchy of clustering where a user can receive data from remote sensor through the internet.
Figure 2. Basic Clustering hierarchy.
There are a number of routing protocols used in WSN to make the network user accessible. They
are briefly explained in the following section.
i. LEACH: Low Energy Adaptive Clustering Hierarchy protocol are the special types of WSN
protocol where each sensor node doesn’t send the data to sink or base station to take decision for
the further hop.

3
ii. PEGASIS: Power-efficient gathering in sensor information systems which doesn’t have any
clustering overhead but manages topology dynamically for the cluster. PEGASIS form a chain
among all the sensor nodes to form a intercommunication among the closest neighbor [3].
iii. APTEEN: Adaptive periodic Threshold-Sensitive Energy-Efficient Sensor Network follow
he TDMA technologies so that their signals transmission remains turned off when they arein
inactive mode.
iv. ECHERP: Equalized Cluster Head Election Routing Protocolis follows energy conservation
over balanced clustering [4] which have less network lifetime and higher energy consumption
compared with PEGASIS.
Despite tremendous development, there are still limitations that WSNs suffer. Some challenges
like designing a low power network, data security, and architecture of network have taken the
most attention of researchers in the last years [5]. As it is an one time installation network, predict
the expected lifetime of each sensor nodes before the network installation is very important to
install a network in a remote area.
There is lots of study in several clustering protocols of WSN. The objective of this work is to
propose an algorithm (called SOM-PEG) applied for large number of sensor nodes in a large
distance area, which will provide less energy consumption and improved network lifetime.
This paper is organized as follows. In section I, we have discussed an introduction and objectives
of the work and Section 2 presents literature review. In Section 3, the most important protocol
PEGASIS protocol has been analyzed. Section 4 described SOM neural Networks. Section 5
describes the proposed SOM-PEG protocol and the simulations procedure is presented in Section
6. Section 7 gives simulation results and discussion. Finally, conclusions of the work are
presented in section 8.
2. RELATED WORKS
There are so many researches went through Wireless Sensor Network (WSN). Therefore, the
main goals of all research are to develop WSN routing protocols that extend the network lifetime
and facing the challenging issues of WSN. Yong-Chang et al. [6] proposed a method where a
chain is deducing by adopting a threshold distance in order to decrease the foundation of a long
link. Their model able to prove that the proposed method has better performance than PEGASIS.
Ref. [7] proposed a routing protocol that works hierarchically with stationary wireless sensor
networks and they claimed that their scheme can solved the core problems in PEGASIS which
helps to improve the lifetime of the wireless sensor network. Authors in [8], they proposed a
protocol named EEPB which has improved PEGASIS from two perspectives, firstly, it adopts the
threshold distance to evade PEGASIS in order to forming a long chain. Secondly, the early death
problem of chain nodes can be solved by this protocol.
In [4], the authors have developed a double cluster heads in one chain and to avoid the existing
long chain they prepared a hierarchical structure in the new algorithm. Their simulation result
shows that, this algorithm is able to increase the productivity of energy-using the load balancer,
which help to extend the lifetime of the whole network. A clustering protocol where the Kohonen
SOM concept are used for clustering has proposed in [9].The unpredictable behaviors of network
parameters estimation and corresponding applications presentation was the major focus of their
works. A LEACH protocol is proposed where SOM neural networks select cluster Heads to infer
further decision in [10]. In case of cluster heads estimation, all the SOM inputs are used as
parameters of cluster heads. There are lots of the SOM based clustering protocols was developed

4
for the purpose of lilfe time maximization and less energy consumption of wireless sensor
network [10-14].
3. PEGASIS PROTOCOL
The chain building process of PEGASIS is to lessen the total length during processing which is as
similar as the traveling salesman problem. In PEGASIS each node intercommunicates to close
neighbors to send control to the base station as it is the clustering head of this particular network.
The energy allocation process among the sensor nodes follow the clustering heads estimation
method consistently [15]. Primarily the nodes has place casually in the playfield and that is why
all the nodes are organized as at a random location. Using a greedy algorithm the sensor nodes
proficiently organized the chain starting from some high energy nodes [16].The base station also
able to form this chain and can broadcast it to all the nodes in the network. The PEGASIS
protocol can save the huge amount of energy during the configuration of cluster and sensing data
delivery method for the purpose of network lifetime maximization [15].
Instead of cluster information in LEACH, the PEGASIS protocol constructs a chain efficiently
for the head of the cluster which is responsible to carry data to the base station. The energy
consumption consistently depends on the head nodes in the networks [15]. In that case PEGASIS
protocol performs twice better than LEACH. The hierarchical routing protocol architecture of
PEGASIS works in three steps as shown in the Fig. 3.
Figure 3. WSN with PEGASIS Protocol Architecture.
There formed a PEGASIS chain under each cluster nodes through the greedy approach. Each
sensor nodes collects, process, and defuse data to its neighbor and finally to its Cluster Head
through the chain by token passing approach.

5
Figure 4. Process of leader selection in PEGASIS and Chain Formation.
4. SELF-ORGANIZING MAP (MOP) NEURAL NETWORK
The weight of the winner and its surrounding neurons in the SOM topology is updated regardless
of the input vector [17]. During the weight update, the distant and nearest neurons to the 1-
vicinity of the winning neuron and the winning frequency of each neuron are found and taken
into consideration [17]. The learning performance is estimated using three standard
measurements in new SOM and it also used in input data sets for the further steps. Using the
SOM protocol cluster, the whole sensing area can efficiently covered. The base station transmits
and receives data with the interconnected SOM clustered head nodes.
Initial Network
All nodes
Calculate distance from node n to sink
Calculate Qnfor select leader Qn=En/Dn
Qn> Qn+1?
Select node n as leader
Select another node in ascending order
Compute energy consumption of every node
Life time
close?
Output
No
Yes

6
Figure 5. Flowchart of Cluster-Based SOM Neural Network.
The flowchart of cluster-based SOM neural network represents the overall process of SOM
formation. To calculate the nearest distance, it uses the SOM algorithm in which Euclidean
distance is used to look for activation of nodes.
5. PROPOSED SOM-PEG PROTOCOL
Motivated by original PEGASIS protocol, other hierarchical architecture and cluster-based SOM
protocol of Neural Network, we have proposed a modification to the process of chain formation
and cluster-based network selection. Our main goal is to minimize the energy consumption and
maximize the network alive time. For a wireless sensor network, we assume the following
conditions as constraints.
 Fixed base station with long distance between sensor nodes.
 Heterogeneous sensor nodes with limited-energy.
 The energy consumption depends on the distance.
 No mobility of sensor nodes.
 Nodistributed sensor nodes.
 One time installation.
 Equal competences.
 Power control capabilities.

7
5.1. Algorithm
The Algorithm is mainly complete in two steps which are given below.
(1) Cluster Head (CH) Selection by SOM-NN
i. Each node weights are randomly initialized from a sensor field. Each node has a
topological position (x,y coordinates), weight vector w.
ii. Choose an input vector node and find the node whose weight vector is neighboring to the
selected point node. The distance is calculated by the Euclidean distance.
iii. For i=1 to n
j = 1 to m
D(j) = ∑ 𝑝𝑛
𝑖=1 ∑ (𝑥𝑖 − 𝑤𝑗)2𝑚
𝑗=1 (1)
Where D(j) is the minimum winning unit index,
𝑥𝑖 is the position of each node vector,
𝑤𝑗 is the weight vector of each node
iv. Calculate the Best Matching Unit (BMU) from shortest distance nodes.
v. The neighborhood of BMU is defined as all the nodes lying within its radius of influence.
vi. The weight vector is associated with a neighbor node. Also BMU is updated by means of
the following calculation
vii.
𝑖𝑤( 𝑞) = 𝑖𝑤( 𝑞 − 1) + 𝛼𝑝( 𝑞) − 𝑖𝑤( 𝑞 − 1) (2)
Where p(q) is the input vector chosen,iw(q-1) is the weight vector associated with node i and
iw(q) is the updated value of the weight vector.
viii. Repeat from step ii until the iteration limit has been reached.
(2) PEGASIS Chain Formation in Each Cluster
i. Chain formation phase: Each cluster creates a chain. The first node is selected from the
clustered nodes located farthest from CH.
ii. Leader selection: A node is selected as a leader considering the residual energy.
iii. Transmission phase: Data transmission occurs in.
a) Each node in the chain contracts with two messages; one it receives and another it
transmits.
b) The sensor nodes collect and forward the data until it reaches to the base station.
c) The leader node receives the collected information, it forwards it to the base station.

8
Equation:
Chain leader selection of each cluster is random and determined by
𝑄 𝑛 =
𝐸 𝑛
𝐷 𝑛
(3)
Where En = Residual energy of nth node, Dn = distance between base station and nth node, Qn =
deterministic weight of the nth node.
Figure 6. Flowchart of Proposed SOM-PEG Cluster Formation.

9
And for the leader selection,
CH = 𝑄 𝑛 > 𝑄 𝑛+1 (4)
Where 𝑄 𝑛 is the leader node.
Other nodes are selected in descending order of their weight which is called the Greedy approach.
5.2. Flow Chart
Flow chart of proposed SOM-PEG is given in Figs.6 and 7.
Figure7. Flowchart of PEGASIS Chain Formation of Each Cluster.
6. SIMULATION PROCEDURES
6.1. Environment Setup
A variety of tools has been used for the simulation of WSN routing protocol, such as OPNET,
MATLAB, OMNET++, and NS, etc. In order to fully simulate the protocol here, we use the
popular network simulation platform MATLAB and Neural Network Graphical User Interface
(NN GUI) for the experiment.
In WSN there are a lot of parameters to evaluate a clustering algorithm. The parameters used for
simulation are shown in TABLE I. Extensive simulation is carried out by varying the node
density and initial energy.

10
6.2. Simulation Matrices
6.2.1. Network Lifetime
Network lifetime is usually measured using three matrices [2].
i) FND - First Node Dies means 10% die of total nodes.
ii) HND - Half of the Nodes Die revels 50% die of total nodes.
iii) LND - Last Node Dies represents 90% die of total nodes.
When last nodes are going to die the network, lifetime turns to failure that requires re-
synchronization.
Table 1. Simulation Parameters.
6.2.2. Energy Consumption
Besides network lifetime, another metric to check the efficiency of a routing protocol is its
energy consumption per transmission. The energy consumption of the network totally depends on
the lifetime of the network. The number of dead nodes revels the balance of energy consumption,
as much as less the nodes die means as much as higher efficiency of energy usage. When a
node’s contains less than zero energy, this node’s treat as a dead node [5].
6.2.3. Convergence Indicator
Convergence Indicator (CI) is used to estimate network conjunction. It is assumed that the higher
value of CI is better than the fixed energy consumption of the network.
CI=
LND−HND
HND−FND
(5)
Parameter Name Value
Area 100 X 100
Number of Nodes I. 50
II. 100
III. 200
IV. 500
V. 1000
Initial Energy /Node I. 0.25 J
II. 0.50 J
III. 2 J
Relative Position of BS (50,60)
Number of Dead Nodes in the beginning 0
Energy Required for Transmission 50*10^ (-9) J/b
Energy Required for Receiver 50*10^ (-9) J/b
Amplifying Energy Required by the Transmitter 100*10^ (-12) J/b/m^2
Energy required to run circuitry (both for transmitter
and receiver) 50*10^ (-9); units in Joules/bit
Packet Size 4000 bits

11
7. RESULTS AND DISCUSSIONS
Here we measured the performance of both the PEGASIS and SOM-PEG protocols and tried to
compare them. We put number of nodes 50, 100, 200, 500, and 1000 respectively with initial
energy 2J by MATLAB simulator. The simulation results are presented in Figures 8 to 12.
Figure 8. Number of Nodes vs FND (Rounds).
The number of rounds for first 10 % nodes dies (FND) increses with the increse of total nodes for
the simulation. In every step SOM-PEG gets more rounds to dies its first nodes (Fig. 8) compare
with PEGASIS protocol. From Fig. 9, we see that SOM-PEG perform better results compare with
traditional PEGASIS protocols in every class of simulations with incresing the total number of
nodes in case of half number of nodes (HND) destroyed.
Figure 9. Number of Nodes Vs HND (Rounds).
0
1000
2000
3000
4000
5000
6000
7000
8000
50 100 200 500 1000
FND(Roundes)
Number of Nodes
Number of Nodes Vs. FND (Rounds)
SOM-PEG PEGASIS
0
1000
2000
3000
4000
5000
6000
7000
8000
50 100 200 500 1000
HND(Rounds)
Number of Nodes
Number of Nodes Vs. HND (Rounds)
PEGASIS SOM-PEG

12
Figure 10. Number of Nodes Vs LND (Rounds).
Figure 11. Number of Nodes Vs Convergence Indicator (CI).
0
1000
2000
3000
4000
5000
6000
7000
8000
50 100 200 500 1000
LND(Rounds)
Number of Nodes
Number of Nodes Vs LND (Rounds)
PEGASIS SOM-PEG
0
0.1
0.2
0.3
0.4
0.5
0.6
50 100 200 500 1000
CI
Number of Nodes
Number of Nodes Vs. CI
PEGASIS SOM-PEG

13
Figure 12. Number of Nodes Vs Energy Consumption (J).
Figure 10 shows the best result for the SOM-PEG protocol and PEGASIS protocol in case of last
nodew dies (LND). The total number of roundes for last nodes dies takes more rounds for SOM-
PEG protocol than PEGASIS. SOM-PEG protocols perform extra rounds for last nodes dies
means it can survive more time than PEGASIS protocol.
Generally, highest convergence indicator is the sign of maximum lifetime. In the Fig. 11, the
proposed protocol shows that it perform high convergence of indicator with the increase of total
number of nodes per simulation. The convergence of indicator shows the high performance up to
200 nodes and after that its value gradually decreases. In every case SOM-PEG shows best
results compared with PEGASIS protocol.
In case of energy consumption, the proposed SOM-PEG protocol shows better results in every
step of simulations. The traditional PEGASIS protocol consume high energy with the increase of
total number of nodes and it takes extra high energy if the total number of nodes increase more
than 500, whereas the proposed SOM-PEG protocol shows stable energy consumption with high
number of nodes (Fig. 12).
From all of the above graphical results (Figs. 8 to 12), we can say that the proposed SOM-PEG
protocol provides better outcomes than traditional PEGASIS protocol in every performance
metrics with less energy consumption and improved network lifetime.
8. CONCLUSIONS
The main purpose of designing a routing protocol for wireless sensor networks is to make it as
energy-efficient as possible that will keep the network runs for a longer period. Here, we
proposed a cluster-based hierarchical routing path protocol, called SOM-PEG protocol, which is
a modified PEGASIS protocol based on traditional PEGASIS with the employment of SOM
neural network. The SOM-PEG protocol consider the location of the sensor nodes, residual
energy of each nodes, distance between nodes and helps to ignore the unwanted process of the
data which can save the energy of nodes as well as extend network lifetime. Based on the
comparative performance analysis of the two protocols, we may conclude that the proposed
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
50 100 200 500 1000
EnergyConsumption(J)
Number of Nodes
Number of Nodes Vs. Energy Consumption
(J)
PEGASIS SOM-PEG

14
SOM-PEG protocol is more energy efficient and provides improved network lifetime than the
traditional PEGASIS protocol.
REFERENCES
[1] Deepak Goyal and Malay Ranjan Tripath, “Routing Protocols in Wireless Sensor Networks: A
Survey”, Second International Conference on Advanced Computing & Communication Technologies,
INSPEC Accession Number: 12616438, Publisher IEEE (2012).
[2] Umesh M. Kulkarni, Deepti V Kulkarni, Harish H. Kenchannavar, “Neural network-based energy
conservation for wireless sensor network ”, International Conference On Smart Technologies For
Smart Nation (SmartTechCon), INSPEC Accession Number: 17754733, Publisher: IEEE (2017).|
[3] R. Sittalatchoumy, L. Sivakumar, “Performance Enhancement of Energy Efficient Protocols for
application in WSN”, Journal of Theoretical and Applied Information Technology, vol. 60, no. 1, pp.
112-118 (2014).
[4] Stefanos A. Nikolidakis, Dionisis Kandris, Dimitrios D. Vergados, and Christos Douligeris,
Algorithms volume, “Energy Efficient Routing in Wireless Sensor Networks Through Balanced
Clustering”, vol. 6, issue 1, pp. 29-42 (2013).
[5] Alaa Eisa, Hazem M. El-Bakry, Samir M. Abd Elrazik, Sajjad Q. Hasan, Asaad Q. Hasan, Samir
Zaid, “Challenges in Wireless Sensor Networks”, International Journal of Advanced Research in
Computer Science & Technology, Vol. 4, Issue 4, pp. 22-27 (2016)
[6] Y. Yong-chang, G. Wei, “An improved PEGASIS in Wireless Sensor Network”, Electronic Journal,
(2008).
[7] Hasan Al-Hasan, Mohammad Qatawneh, Azzam Sleit, Wesam Almobaideen, “EAPHRN: Energy
Aware PEGASIS-Based Hierarchal Routing Protocol for Wireless Sensor Networks,” Journal of
American Science, vol. 7, no. 8, 753-758 (2011).
[8] Abdelbari Ben Yagouta, Maher Jabberi, Bechir Ben Gouissem “Impact of Sink Mobility on Quality
of Service Performance and Energy Consumption in Wireless Sensor Network with Cluster-Based
Routing Protocols”, 14th International Conference on Computer Systems and Applications
(AICCSA), pp. 1125-1132, Publisher: IEEE (2017).
[9] P.J. Chandra, M.P. Kumar, and C. Goutam, “Development of Laguerre Neural-Network-Based
Intelligent Sensors for Wireless Sensor Networks”, IEEE Transactions on Instrumentation and
Measurement, vol. 60, no. 3, (2011).
[10] Tao Li1, Feng Ruan1, Zhiyong Fan1, Jin Wang2 and Jeong-Uk Kim3 “An Improved PEGASIS
Routing Protocol Based on Neural Network and Ant Colony Algorithm” International Journal of
Future Generation Communication and Networking, Vol. 8, No. 6, pp. 149-160 (2015)
[11] Gail A. Carpenter and Stephen Grossberg, " A self-organizing neural network for supervised learning,
recognition, and prediction", IEEE Communication Magazine, Volume: 30 , Issue: 9, pp. 38-49
(1992 ).
[12] Mario Cordina and Carl J. Debono “Increasing wireless sensor network lifetime through the
application of SOM neural networks”, 3rd International Symposium on Communications, Control
and Signal Processing, Publisher: IEEE, INSPEC Accession Number: 10044621 (2008).
[13] Nauman Ahad, Junaid Qadir, and Nasir Ahsan "Neural network in the wireless network: Techniques,
applications, and guidelines", Journal of Network and Computer Application, vol. 68, Issue: C, pp. 1-
27 (2016).
[14] Vesanto J, Alhoniemi E. "Clustering of Self Organizing Map", IEEE Transactions on Neural
Networks, Vol. 11, No. 3, pp. 586-600 (2000)
[15] T. Liu and F. Li, “Power-efficient clustering routing protocol based on applications in wireless sensor
network”, Proceedings of 5th international conference on Wireless Communications, Networking and
Mobile Computing, WiCOM, (2009).
[16] S. Lindsey and C.S. Raghavendra, “Pegasis: Power-efficient Gathering in Sensor Information
Systems”, Proc. of IEEE International Conference on Computer Systems, San Francisco, USA
(2002).
[17] H. Shahbazi, M.A. Araghizadeh and M. Dalvi, "Minimum Power Intelligent Routing In Wireless
Sensors Networks Using Self Organizing Neural Networks", IEEE International Symposium on
Telecommunications, pp. 354–358 (2008).

Current issue- International Journal of Advanced Smart Sensor Network Systems (IJASSN)

More Related Content

What's hot (18)

Similar to Current issue- International Journal of Advanced Smart Sensor Network Systems (IJASSN) (20)

Recently uploaded (20)

Current issue- International Journal of Advanced Smart Sensor Network Systems (IJASSN)