Spatial correlation based clustering algorithm for random and uniform topology in ws ns
1 like231 views
The document presents a distributed clustering algorithm for wireless sensor networks (WSNs), which groups sensor nodes based on spatial and data correlation to enhance energy efficiency. It compares results from the proposed algorithm applied to both uniform and random network topologies using metrics like the average number of clusters formed and energy consumption. The findings indicate that while uniform topology performs better with fewer nodes, random topology is more energy-efficient with larger networks.
![IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
_______________________________________________________________________________________
Volume: 03 Issue: 06 | Jun-2014, Available @ http://www.ijret.org 83
SPATIAL CORRELATION BASED CLUSTERING ALGORITHM FOR
RANDOM AND UNIFORM TOPOLOGY IN WSNs
Bhavana H.T1
, Jayanthi K Murthy2
1
M.Tech Scholar, Dept. of ECE , BMS College of Engineering, Bangalore
2
Associate Professor, Dept. of ECE, BMS College of Engineering, Bangalore
Abstract
In Wireless Sensor Networks (WSN) sensor nodes with similar readings can be grouped such that, it is enough to report a single
reading from the entire group. A representative node is selected from each cluster to do the reporting job. This helps to increase
the battery life of sensor nodes. However, efficiently identifying sensor groups and their representative nodes is a challenging
task. In this paper, a distributed algorithm is proposed to determine a set of representative nodes which exploits the tradeoff
between data quality and energy consumption. In this paper, we group the sensor nodes based on their inherent spatial and data
correlation in WSN. The proposed clustering algorithm is applied for uniform and random topology of sensor network. The results
based on different metrics such as average number of clusters formed, energy consumption and average variation in cluster size
are compared for both topologies.
---------------------------------------------------------------------***--------------------------------------------------------------------
1. INTRODUCTION
A WSN consists of thousands of spatially distributed sensor
nodes. Once deployed, the sensor nodes form a network
through short-range wireless communication. They collect
environmental data and send to the data processing center
(Sink node). Each sensor node collects local data.
Data collection in WSNs consists of two main processes
sampling and wireless communication. As the sensor nodes
are battery operated their energy and communication
bandwidth are limited. Fortunately most of the applications
require less data accuracy. Hence it is unnecessary for all the
sensor nodes to sample and report data. Energy consumption
can be reduced greatly by turning off redundant sensor
nodes and it will be wise choice for long term data
collection.
The region in which all the sensor nodes send the readings
which are similar in nature is known as Spatial Correlation
region and therefore it is enough to send a single report to
represent the correlation region [1]. The readings of a sensor
node may be predicted from that of its nearby sensor nodes
with high assurance.
This paper aims to save energy in continuous data collection
applications. To do this job nodes are partitioned with
similar observations into a cluster. Because of energy
constraint and difficulty involved in discovering the spatial
correlation pattern, spatial clustering in an energy-efficient
and distributed manner is a tough work. A distributed
algorithm is proposed in this paper to form the clusters
based on correlation between sensor nodes. The same
algorithm is applied on two network topologies uniform and
random. The comparison is done between these two
topologies using three different key metrics. A novel
ranking method is used to select efficient representative
nodes to cover entire network.
The rest of the paper is organized as follows: section II
contains the related work about the spatial correlation based
clustering algorithms, the system model is described in
section III, section IV elaborates the algorithm, Results
obtained are discussed in section V and conclusion and
future work are suggested in section VI.
2. RELATED WORK
Clustering is a well known technology which helps in
topology management in WSNs. In [2] [3] and [4] authors
describes taxonomy of clustering and different energy aware
clustering algorithms. These papers give a clear idea about
the cluster properties, cluster head capabilities and
clustering algorithms.
In [5] the author has introduced the spatial correlation
concept in WSNs. It aggregates the data using joint entropy.
A linear model is proposed to capture the spatial correlation.
But in real world most of the systems may not be linear.
To estimate the energy dissipation for sensor networks
author in [6] has proposed simple radio model LEACH.
B. Gedik et al. in [7] have proposed ASAP, a distributed
clustering algorithm which groups the nodes with similar
readings. ASAP can control the number of clusters but it
needs a number of iterations to select efficient cluster heads
to cover all nodes and distribution of cluster heads may not
be so good.
EEDC proposed in [8] considers spatial and temporal
correlation between nodes to group them. The spatial
correlation range (user defined parameter) is used as a
metric to find similarity between nodes. DCglobal proposed
in [9] is a centralized algorithm which selects CH based on
their data coverage range. However, EEDC and DCglobal](https://image.slidesharecdn.com/spatialcorrelationbasedclusteringalgorithmforrandomanduniformtopologyinwsns-140821042422-phpapp01/85/Spatial-correlation-based-clustering-algorithm-for-random-and-uniform-topology-in-ws-ns-1-320.jpg)
![IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
_______________________________________________________________________________________
Volume: 03 Issue: 06 | Jun-2014, Available @ http://www.ijret.org 84
are centralized algorithms hence communication overhead
to select cluster heads and to form the cluster is high.
Chih-Chieh Hung has extended DCglobal to distributed
version known as DClocal it uses Manhattan distance to
measure similarity. The DClocal uses more counters and
hence it requires accurate time synchronization. Zhidan Liu
et al. have proposed a distributed algorithm known as DSCC
in [10]. This algorithm considers both distance and data
similarity to group the sensor nodes. In this paper the
algorithm used till the selection of CH is similar to that of
DSCC.
Topology of nodes play an important role in resource
constrained WSNs [11] [12]. Most of the clustering
algorithms consider the uniform deployment of sensor
networks. But in real time scenarios the sensor nodes will be
random in fashion. We need to design an efficient clustering
algorithm by considering random topology as well.
The efficient clustering algorithms demand less
communication cost, long battery life, low cost and more
data accuracy. The main problems associated with the
existing spatial correlation based clustering algorithms are
1) No strict Requirement on similarity measurement
between nodes. 2) Abundant communication overhead for
spatial clustering. 3) Number of iterations to select enough
cluster heads. 4) Algorithms are limited to uniform
topology.
3. SYSTEM MODEL
In this paper, a WSN consisting of N sensor nodes with the
Sink node located outside is considered. The sensor node i is
denoted as Si and the sensor node set as S = {S1, S2, S3…},
where |S| = N. All sensor nodes and Sink node are stationary
after deployment. The initial energy of all sensor nodes is
same and nodes are homogenous. Once the clustering begins
the nodes can have different residual energy. The
approximate distance between the nodes is estimated
through received signal strength. Finally each sensor node
can adjust the transmission power to change the
communication range with single-hop communication
between an active node and the Sink. After the cluster
formation we assume a simple scheduling property where
only the cluster head will report the reading and all other
cluster members will be in sleep mode.
Fig: 1 System Model
The system model of the proposed algorithm is shown in
figure 1. The model receives two user defined parameters
namely Spatial Correlation Range (RSC) and Error Tolerance
Threshold (ε) to calculate the similarity between nodes.
Each sensor node adjusts the transmission power to
broadcast a “Hello” message (H-msg), which contains its
residual energy and recent reading serial to neighboring
nodes in the range of spatial correlation. Using this
information each node decides its similar nodes. Based on
the local information, energy level and representative
capability each node decides whether to become a cluster
head candidate (CHC). Later, all CHC nodes will compete
to be the final cluster head (CH) by sending the
“Competition” message (C-message) which consists of only
node id to its neighboring nodes. All non-CH nodes will
choose an appropriate cluster to join and hence the
clustering will finish.
4. ALGORITHM TO FIND THE CH
This algorithm ends with the selection of appropriate CH.
The clustering algorithm works according to following
steps.
1) Detection of similar nodes: The Euclidean distance is
used to calculate the geographical distance between the
nodes. We say that Sj is the neighboring nodes of a node Si
only if the distance between two nodes is within a spatial
correlation range RSC.
The Manhattan distance is a metric used to find the
dissimilarity between the nodes using their recent reading
serials. The dissimilarity measurement function fmd (Si, Sj) is
defined as
𝑓𝑚𝑑 𝑆𝑖, 𝑆 𝑗 =
𝑉 𝑘 𝑆 𝑖 −𝑉′
𝑘(𝑆 𝑗 )
𝑞
𝑘=1
𝑞
.................... (1)](https://image.slidesharecdn.com/spatialcorrelationbasedclusteringalgorithmforrandomanduniformtopologyinwsns-140821042422-phpapp01/85/Spatial-correlation-based-clustering-algorithm-for-random-and-uniform-topology-in-ws-ns-2-320.jpg)
![IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
_______________________________________________________________________________________
Volume: 03 Issue: 06 | Jun-2014, Available @ http://www.ijret.org 85
In Equation 1, V (Si) = {V1, V2... Vq} and V’ (Sj) = {V’1,
V’2... V’q} are the sensor reading serials of Si, Sj respectively
and q is the number of readings from each node.
If the dissimilarity measurement between nodes Si and Sj is
less than half of error tolerance threshold i.e., fmd (Si, Sj) ≤
ε/2 and Sj is neighboring node of Si then both nodes are said
to be similar to each other. Greater RSC will lead to fewer
clusters but this leads to larger dissimilarity of cluster. On
the other hand lesser RSC results in less number of clusters
which increases the communication overhead.
2) Computation of Similarity Coverage Rate and
Similarity Difference Rate: The similarity coverage rate Cr
(Si) is the rate of coverage capacity of a node. It is calculated
using equation 2. If the Cr (Si) is high then the capacity of
node to cover its neighboring node is also will be high.
𝐶𝑟 𝑆𝑖 =
𝑆𝑁(𝑆 𝑖)
𝑁𝐵𝑅(𝑆 𝑖)
………………….… (2)
The similarity difference rate Sr (Si) gives the dissimilarity
measurement between the nodes using their Manhattan
distance. It is computed using equation 3. Sr (Si) expresses
the similar degree between Si and its similar nodes. Both Cr
(Si) and Sr (Si) fall in the range of [0, 1].
𝑆𝑟 𝑆𝑖 =
ɛ−
𝑓 𝑚𝑑 (𝑆 𝑖, 𝑆 𝑗 )𝑆 𝑗 ∈𝑆𝑁 (𝑆 𝑖)
𝑚
ɛ
… … … … … … . ..(3)
3) Computation of Representative Capability p (Si) and
Relative Energy Level ER (Si): Equation 4 gives the
expression for representative capability of a node.
𝑝 𝑆𝑖 = 𝛼 ∗ 𝐶𝑟 𝑆𝑖 + 1 − 𝛼 𝑆𝑟(𝑆𝑖)...... (4)
Where α is an adjustable parameter to decide the relative
importance of Cr and Sr.
The relative energy level ER (Si) of a sensor node Si is
calculated using equation 5.
𝐸 𝑅 𝑆𝑖 =
𝐸 𝑆 𝑖 ∗(𝑁−1)
𝐸 𝑆 𝑖 + 𝐸(𝑆 𝑗 )𝑆 𝑗 ∈𝑁𝐵𝑅 (𝑆 𝑖)
… … … … ..(5)
E (Si) is the residual energy of Si. The ordinary sensor node
becomes CHC node only if it has ER (Si) >1 and p (Si) > δ,
where δ is a parameter which controls the number of CHC.
4) Selection of CH: A CHC node and broadcasts
competition message (C-msg) which includes ID and the
size of similar node set. In the perspective of spatial
correlation, node Si defeats node Sj if and only if given
conditions are satisfied. 1) They are similar nodes 2) SN (Si)
contains more nodes than SN (Sj) 3) More than half nodes of
set SN (Sj) are in set SN (Si).
The ranking assignment is performed as per the flowchart
indicated in figure 2. At the expiry of TCHC, each CHC node
confirms its own ranking and algorithm will select the final
CHs in an iterative manner by exploiting these rankings.
This CH algorithm is shown in figure 3.
Fig: 2 Ranking assignment algorithm
Fig: 3 CH Selection Algorithm
The CH selection phase lasts for TCH. In first step loop
checks whether the node is CHC or not, if so then its
ranking is verified. If the ranking is equal to 1, it selects
itself as a final CH and broadcasts a CH advertising message
(A-msg) which only includes its node ID to neighbors in
range of RSC. After receiving an A-msg, any ordinary node
or CHC node whose ranking is larger than 1 will add this
CH to its list and consider itself as being “covered” by a CH
node. At the expiry of Trank every “uncovered” CHC node](https://image.slidesharecdn.com/spatialcorrelationbasedclusteringalgorithmforrandomanduniformtopologyinwsns-140821042422-phpapp01/85/Spatial-correlation-based-clustering-algorithm-for-random-and-uniform-topology-in-ws-ns-3-320.jpg)
![IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
_______________________________________________________________________________________
Volume: 03 Issue: 06 | Jun-2014, Available @ http://www.ijret.org 86
reduces its ranking a half and goes to next iteration. This
loop is executed till a node becomes CH or non-CH.
5) Proposed algorithm for Formation of cluster: The
cluster formation algorithm is shown in figure 4.
Fig: 4 Formation of cluster
In this phase if an ordinary node receives only one A-msg
from any other CHC, it immediately joins that CH. In
second case if any node receives A-msgs from more than
one node it selects the closer node as the CH and joins to
that cluster. In third case if a node does not receive any A-
msg then it finds the nearest CH using Euclidean distance
formula. Finally every ordinary node will join one cluster
and hence the cluster formation process will end.
5. RESULTS OBTAINED
The simulation is carried out in Matlab. The two user
defined thresholds spatial correlation range and error
tolerance threshold are selected as 20 and 0.7 units
respectively. The residual energy of nodes is taken in the
range [4, 5]. The simulation is carried on both uniform and
random distribution of nodes. The two threshold α and δ are
set as 0.5 and 0.4 respectively. For uniform deployment the
distance between immediate neighboring nodes is
considered as 13 units.
Extensive simulations are carried out by varying the sensor
nodes from 16 to100. As the number of nodes varies the
field size is also varied from 52X52 to 130X130
respectively. All results in this paper are average values of
twenty simulations.
The figure 5 shown below is the comparison of number of
clusters formed. This shows that when the sensor nodes are
less in number the number of clusters generated is less. But
as the sensor nodes increase the cluster number rapidly
changes in uniform topology. The graph for random
topology gradually increases as the number of nodes
increases. Less number of clusters facilitates lower power
consumption as in spatial correlation based clustering more
nodes can be turned into sleep mode. For the network with
more number of nodes, random topology is suitable as it
ends with less number of clusters when compared to
uniform topology.
The variation in cluster size is shown in figure 6. The
uniform topology shows less variation in cluster size. Since
the number of clusters formed rapidly increases in uniform
topology it can maintain the same cluster size. The cluster
size variation is more in random topology, however as
explained above more no of nodes can be turned into sleep
mode. Hence there may not be significant effect on energy
efficiency.
Fig: 5 Number of Clusters Formed
Fig: 6 Variation in Cluster Size](https://image.slidesharecdn.com/spatialcorrelationbasedclusteringalgorithmforrandomanduniformtopologyinwsns-140821042422-phpapp01/85/Spatial-correlation-based-clustering-algorithm-for-random-and-uniform-topology-in-ws-ns-4-320.jpg)
![IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
_______________________________________________________________________________________
Volume: 03 Issue: 06 | Jun-2014, Available @ http://www.ijret.org 87
We consider the famous radio model LEACH to estimate
the energy dissipation of sensor nodes.
Fig: 7 Energy Consumption
The energy consumed for CH to transmit data to sink node
is calculated and results are plot in the figure 7. The
parameters assumed for transmission of data from one node
to another are given as Eelec= 50 nJ/bit, Efs=10 pJ/bit/m2
,
Emp=0.0013pJ/bit/m4
. The threshold distance d0 is varied
according to field size.
The above figure clearly indicates that the energy consumed
by uniform topology rapidly increases as the number of
nodes increase. In random topology this curve is gradually
increasing and always it is less than uniform topology.
Hence for network with less number of nodes uniform
topology is suitable. The random topology will be energy
efficient when the sensor nodes are more.
6. CONCLUSIONS AND FUTURE WORK
In the proposed model the cluster formation is done with
reduced communication overhead between the nodes, as the
messages exchanged during clustering is very less. The
computational complexity involved in selecting the efficient
cluster heads is little more. However, the energy required for
communication is more when compared to the computation.
The simulations are carried out for uniform and random
topologies. When the sensor nodes are less in number
uniform topology gives better result in all perspective. When
the number of nodes is more random topology is best suited
since the energy efficiency is more.
As the future work efficient scheduling algorithms can be
implemented in every cluster to make few nodes to work at
a time and hence more data fidelity can be acquired. Along
with this clustering algorithm good routing algorithms can
be integrated to get better results.
REFERENCES
[1]. Arthur Getis, “A History of the Concept of Spatial
Autocorrelation: A Geographer’s Perspective” ISSN, vol. 4
pp.0016-7363, 2007.
[2]. Pedro A. Forero, Student Member, IEEE, Alfonso Cano,
Member, IEEE, and Georgios B. Giannakis, Fellow, IEEE,
“Distributed Clustering Using Wireless Sensor Networks”
IEEE journal of selected topics in signal processing, vol. 5,
no. 4, august 2011.
[3]. Vinay Kumar, Sanjeev Jain and Sudarshan Tiwari,
“Energy Efficient Clustering Algorithms in Wireless Sensor
Networks: A Survey”, IJCSI International Journal of
Computer Science Issues, Vol. 8, Issue 5, No 2, September
2011
[4]. A. Abbasi and M. Younis, “A Survey on Clustering
Algorithms for Wireless Sensor Networks,” Computer
Communications, vol. 30, no. 14-15, pp. 2826–2841, 2007.
[5]. W. Heinzelman, A. Chandrakasan, and H. Balakrishnan,
“An Application Specific Protocol Architecture for Wireless
Microsensor Networks,” Wireless Communications, IEEE
Transactions on, vol. 1, no. 4, pp. 660–670, 2002.
[6]. Sandeep Pattern, Bhaskar Krishnamachari, Ramesh
Govindan, “The Impact of Spatial Correlation on Routing
with Compression in Wireless Sensor Networks”. IPSN, vol.
1, pp. 846. 2004.
[7]. B. Gedik, L. Liu, and P. Yu, “ASAP: An Adaptive
Sampling Approach to Data Collection in Sensor
Networks,” Parallel and Distributed Systems, IEEE
Transactions on, vol. 18, no. 12, pp. 1766–1783, 2007.
[8]. L. Chong, W. Kui, and P. Jian, “An Energy-Efficient
Data Collection Framework for Wireless Sensor Networks
by Exploiting Spatiotemporal Correlation,” Parallel and
Distributed Systems, IEEE Transactions on, vol. 18, no. 7,
pp. 1010–1023, 2007.
[9]. C. C. Hung, W. C. Peng, and W. C. Lee, “Energy-
Aware Set-Covering Approaches for Approximate Data
Collection in Wireless Sensor Net-works,” IEEE
Transactions on Knowledge and Data Engineering, vol. 24,
no. 11, 2011.
[10]. Zhidan Liu, Wei Xing, Bo Zeng, Yongchao Wang,
Dongming Lu, “Distributed spatial Correlation Based
Clustering for WSNs”, IEEE computer society, pp. 56-64,
2013.
[11]. Quazi Mamun, “A Qualitative Comparison of
Different Logical Topologies for Wireless Sensor
Networks”, Sensors 2012, vol.12, pp. 14887-14913, 2012.
[12]. Divya Sharma, Sandeep Verma, Kanika Sharma,
“Network Topologies in Wireless Sensor Networks: A
Review”, IJECT, vol.4, Issue-3, 2013.](https://image.slidesharecdn.com/spatialcorrelationbasedclusteringalgorithmforrandomanduniformtopologyinwsns-140821042422-phpapp01/85/Spatial-correlation-based-clustering-algorithm-for-random-and-uniform-topology-in-ws-ns-5-320.jpg)
Spatial correlation based clustering algorithm for random and uniform topology in ws ns
- 1. IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308 _______________________________________________________________________________________ Volume: 03 Issue: 06 | Jun-2014, Available @ http://www.ijret.org 83 SPATIAL CORRELATION BASED CLUSTERING ALGORITHM FOR RANDOM AND UNIFORM TOPOLOGY IN WSNs Bhavana H.T1 , Jayanthi K Murthy2 1 M.Tech Scholar, Dept. of ECE , BMS College of Engineering, Bangalore 2 Associate Professor, Dept. of ECE, BMS College of Engineering, Bangalore Abstract In Wireless Sensor Networks (WSN) sensor nodes with similar readings can be grouped such that, it is enough to report a single reading from the entire group. A representative node is selected from each cluster to do the reporting job. This helps to increase the battery life of sensor nodes. However, efficiently identifying sensor groups and their representative nodes is a challenging task. In this paper, a distributed algorithm is proposed to determine a set of representative nodes which exploits the tradeoff between data quality and energy consumption. In this paper, we group the sensor nodes based on their inherent spatial and data correlation in WSN. The proposed clustering algorithm is applied for uniform and random topology of sensor network. The results based on different metrics such as average number of clusters formed, energy consumption and average variation in cluster size are compared for both topologies. ---------------------------------------------------------------------***-------------------------------------------------------------------- 1. INTRODUCTION A WSN consists of thousands of spatially distributed sensor nodes. Once deployed, the sensor nodes form a network through short-range wireless communication. They collect environmental data and send to the data processing center (Sink node). Each sensor node collects local data. Data collection in WSNs consists of two main processes sampling and wireless communication. As the sensor nodes are battery operated their energy and communication bandwidth are limited. Fortunately most of the applications require less data accuracy. Hence it is unnecessary for all the sensor nodes to sample and report data. Energy consumption can be reduced greatly by turning off redundant sensor nodes and it will be wise choice for long term data collection. The region in which all the sensor nodes send the readings which are similar in nature is known as Spatial Correlation region and therefore it is enough to send a single report to represent the correlation region [1]. The readings of a sensor node may be predicted from that of its nearby sensor nodes with high assurance. This paper aims to save energy in continuous data collection applications. To do this job nodes are partitioned with similar observations into a cluster. Because of energy constraint and difficulty involved in discovering the spatial correlation pattern, spatial clustering in an energy-efficient and distributed manner is a tough work. A distributed algorithm is proposed in this paper to form the clusters based on correlation between sensor nodes. The same algorithm is applied on two network topologies uniform and random. The comparison is done between these two topologies using three different key metrics. A novel ranking method is used to select efficient representative nodes to cover entire network. The rest of the paper is organized as follows: section II contains the related work about the spatial correlation based clustering algorithms, the system model is described in section III, section IV elaborates the algorithm, Results obtained are discussed in section V and conclusion and future work are suggested in section VI. 2. RELATED WORK Clustering is a well known technology which helps in topology management in WSNs. In [2] [3] and [4] authors describes taxonomy of clustering and different energy aware clustering algorithms. These papers give a clear idea about the cluster properties, cluster head capabilities and clustering algorithms. In [5] the author has introduced the spatial correlation concept in WSNs. It aggregates the data using joint entropy. A linear model is proposed to capture the spatial correlation. But in real world most of the systems may not be linear. To estimate the energy dissipation for sensor networks author in [6] has proposed simple radio model LEACH. B. Gedik et al. in [7] have proposed ASAP, a distributed clustering algorithm which groups the nodes with similar readings. ASAP can control the number of clusters but it needs a number of iterations to select efficient cluster heads to cover all nodes and distribution of cluster heads may not be so good. EEDC proposed in [8] considers spatial and temporal correlation between nodes to group them. The spatial correlation range (user defined parameter) is used as a metric to find similarity between nodes. DCglobal proposed in [9] is a centralized algorithm which selects CH based on their data coverage range. However, EEDC and DCglobal
- 2. IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308 _______________________________________________________________________________________ Volume: 03 Issue: 06 | Jun-2014, Available @ http://www.ijret.org 84 are centralized algorithms hence communication overhead to select cluster heads and to form the cluster is high. Chih-Chieh Hung has extended DCglobal to distributed version known as DClocal it uses Manhattan distance to measure similarity. The DClocal uses more counters and hence it requires accurate time synchronization. Zhidan Liu et al. have proposed a distributed algorithm known as DSCC in [10]. This algorithm considers both distance and data similarity to group the sensor nodes. In this paper the algorithm used till the selection of CH is similar to that of DSCC. Topology of nodes play an important role in resource constrained WSNs [11] [12]. Most of the clustering algorithms consider the uniform deployment of sensor networks. But in real time scenarios the sensor nodes will be random in fashion. We need to design an efficient clustering algorithm by considering random topology as well. The efficient clustering algorithms demand less communication cost, long battery life, low cost and more data accuracy. The main problems associated with the existing spatial correlation based clustering algorithms are 1) No strict Requirement on similarity measurement between nodes. 2) Abundant communication overhead for spatial clustering. 3) Number of iterations to select enough cluster heads. 4) Algorithms are limited to uniform topology. 3. SYSTEM MODEL In this paper, a WSN consisting of N sensor nodes with the Sink node located outside is considered. The sensor node i is denoted as Si and the sensor node set as S = {S1, S2, S3…}, where |S| = N. All sensor nodes and Sink node are stationary after deployment. The initial energy of all sensor nodes is same and nodes are homogenous. Once the clustering begins the nodes can have different residual energy. The approximate distance between the nodes is estimated through received signal strength. Finally each sensor node can adjust the transmission power to change the communication range with single-hop communication between an active node and the Sink. After the cluster formation we assume a simple scheduling property where only the cluster head will report the reading and all other cluster members will be in sleep mode. Fig: 1 System Model The system model of the proposed algorithm is shown in figure 1. The model receives two user defined parameters namely Spatial Correlation Range (RSC) and Error Tolerance Threshold (ε) to calculate the similarity between nodes. Each sensor node adjusts the transmission power to broadcast a “Hello” message (H-msg), which contains its residual energy and recent reading serial to neighboring nodes in the range of spatial correlation. Using this information each node decides its similar nodes. Based on the local information, energy level and representative capability each node decides whether to become a cluster head candidate (CHC). Later, all CHC nodes will compete to be the final cluster head (CH) by sending the “Competition” message (C-message) which consists of only node id to its neighboring nodes. All non-CH nodes will choose an appropriate cluster to join and hence the clustering will finish. 4. ALGORITHM TO FIND THE CH This algorithm ends with the selection of appropriate CH. The clustering algorithm works according to following steps. 1) Detection of similar nodes: The Euclidean distance is used to calculate the geographical distance between the nodes. We say that Sj is the neighboring nodes of a node Si only if the distance between two nodes is within a spatial correlation range RSC. The Manhattan distance is a metric used to find the dissimilarity between the nodes using their recent reading serials. The dissimilarity measurement function fmd (Si, Sj) is defined as 𝑓𝑚𝑑 𝑆𝑖, 𝑆 𝑗 = 𝑉 𝑘 𝑆 𝑖 −𝑉′ 𝑘(𝑆 𝑗 ) 𝑞 𝑘=1 𝑞 .................... (1)
- 3. IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308 _______________________________________________________________________________________ Volume: 03 Issue: 06 | Jun-2014, Available @ http://www.ijret.org 85 In Equation 1, V (Si) = {V1, V2... Vq} and V’ (Sj) = {V’1, V’2... V’q} are the sensor reading serials of Si, Sj respectively and q is the number of readings from each node. If the dissimilarity measurement between nodes Si and Sj is less than half of error tolerance threshold i.e., fmd (Si, Sj) ≤ ε/2 and Sj is neighboring node of Si then both nodes are said to be similar to each other. Greater RSC will lead to fewer clusters but this leads to larger dissimilarity of cluster. On the other hand lesser RSC results in less number of clusters which increases the communication overhead. 2) Computation of Similarity Coverage Rate and Similarity Difference Rate: The similarity coverage rate Cr (Si) is the rate of coverage capacity of a node. It is calculated using equation 2. If the Cr (Si) is high then the capacity of node to cover its neighboring node is also will be high. 𝐶𝑟 𝑆𝑖 = 𝑆𝑁(𝑆 𝑖) 𝑁𝐵𝑅(𝑆 𝑖) ………………….… (2) The similarity difference rate Sr (Si) gives the dissimilarity measurement between the nodes using their Manhattan distance. It is computed using equation 3. Sr (Si) expresses the similar degree between Si and its similar nodes. Both Cr (Si) and Sr (Si) fall in the range of [0, 1]. 𝑆𝑟 𝑆𝑖 = ɛ− 𝑓 𝑚𝑑 (𝑆 𝑖, 𝑆 𝑗 )𝑆 𝑗 ∈𝑆𝑁 (𝑆 𝑖) 𝑚 ɛ … … … … … … . ..(3) 3) Computation of Representative Capability p (Si) and Relative Energy Level ER (Si): Equation 4 gives the expression for representative capability of a node. 𝑝 𝑆𝑖 = 𝛼 ∗ 𝐶𝑟 𝑆𝑖 + 1 − 𝛼 𝑆𝑟(𝑆𝑖)...... (4) Where α is an adjustable parameter to decide the relative importance of Cr and Sr. The relative energy level ER (Si) of a sensor node Si is calculated using equation 5. 𝐸 𝑅 𝑆𝑖 = 𝐸 𝑆 𝑖 ∗(𝑁−1) 𝐸 𝑆 𝑖 + 𝐸(𝑆 𝑗 )𝑆 𝑗 ∈𝑁𝐵𝑅 (𝑆 𝑖) … … … … ..(5) E (Si) is the residual energy of Si. The ordinary sensor node becomes CHC node only if it has ER (Si) >1 and p (Si) > δ, where δ is a parameter which controls the number of CHC. 4) Selection of CH: A CHC node and broadcasts competition message (C-msg) which includes ID and the size of similar node set. In the perspective of spatial correlation, node Si defeats node Sj if and only if given conditions are satisfied. 1) They are similar nodes 2) SN (Si) contains more nodes than SN (Sj) 3) More than half nodes of set SN (Sj) are in set SN (Si). The ranking assignment is performed as per the flowchart indicated in figure 2. At the expiry of TCHC, each CHC node confirms its own ranking and algorithm will select the final CHs in an iterative manner by exploiting these rankings. This CH algorithm is shown in figure 3. Fig: 2 Ranking assignment algorithm Fig: 3 CH Selection Algorithm The CH selection phase lasts for TCH. In first step loop checks whether the node is CHC or not, if so then its ranking is verified. If the ranking is equal to 1, it selects itself as a final CH and broadcasts a CH advertising message (A-msg) which only includes its node ID to neighbors in range of RSC. After receiving an A-msg, any ordinary node or CHC node whose ranking is larger than 1 will add this CH to its list and consider itself as being “covered” by a CH node. At the expiry of Trank every “uncovered” CHC node
- 4. IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308 _______________________________________________________________________________________ Volume: 03 Issue: 06 | Jun-2014, Available @ http://www.ijret.org 86 reduces its ranking a half and goes to next iteration. This loop is executed till a node becomes CH or non-CH. 5) Proposed algorithm for Formation of cluster: The cluster formation algorithm is shown in figure 4. Fig: 4 Formation of cluster In this phase if an ordinary node receives only one A-msg from any other CHC, it immediately joins that CH. In second case if any node receives A-msgs from more than one node it selects the closer node as the CH and joins to that cluster. In third case if a node does not receive any A- msg then it finds the nearest CH using Euclidean distance formula. Finally every ordinary node will join one cluster and hence the cluster formation process will end. 5. RESULTS OBTAINED The simulation is carried out in Matlab. The two user defined thresholds spatial correlation range and error tolerance threshold are selected as 20 and 0.7 units respectively. The residual energy of nodes is taken in the range [4, 5]. The simulation is carried on both uniform and random distribution of nodes. The two threshold α and δ are set as 0.5 and 0.4 respectively. For uniform deployment the distance between immediate neighboring nodes is considered as 13 units. Extensive simulations are carried out by varying the sensor nodes from 16 to100. As the number of nodes varies the field size is also varied from 52X52 to 130X130 respectively. All results in this paper are average values of twenty simulations. The figure 5 shown below is the comparison of number of clusters formed. This shows that when the sensor nodes are less in number the number of clusters generated is less. But as the sensor nodes increase the cluster number rapidly changes in uniform topology. The graph for random topology gradually increases as the number of nodes increases. Less number of clusters facilitates lower power consumption as in spatial correlation based clustering more nodes can be turned into sleep mode. For the network with more number of nodes, random topology is suitable as it ends with less number of clusters when compared to uniform topology. The variation in cluster size is shown in figure 6. The uniform topology shows less variation in cluster size. Since the number of clusters formed rapidly increases in uniform topology it can maintain the same cluster size. The cluster size variation is more in random topology, however as explained above more no of nodes can be turned into sleep mode. Hence there may not be significant effect on energy efficiency. Fig: 5 Number of Clusters Formed Fig: 6 Variation in Cluster Size
- 5. IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308 _______________________________________________________________________________________ Volume: 03 Issue: 06 | Jun-2014, Available @ http://www.ijret.org 87 We consider the famous radio model LEACH to estimate the energy dissipation of sensor nodes. Fig: 7 Energy Consumption The energy consumed for CH to transmit data to sink node is calculated and results are plot in the figure 7. The parameters assumed for transmission of data from one node to another are given as Eelec= 50 nJ/bit, Efs=10 pJ/bit/m2 , Emp=0.0013pJ/bit/m4 . The threshold distance d0 is varied according to field size. The above figure clearly indicates that the energy consumed by uniform topology rapidly increases as the number of nodes increase. In random topology this curve is gradually increasing and always it is less than uniform topology. Hence for network with less number of nodes uniform topology is suitable. The random topology will be energy efficient when the sensor nodes are more. 6. CONCLUSIONS AND FUTURE WORK In the proposed model the cluster formation is done with reduced communication overhead between the nodes, as the messages exchanged during clustering is very less. The computational complexity involved in selecting the efficient cluster heads is little more. However, the energy required for communication is more when compared to the computation. The simulations are carried out for uniform and random topologies. When the sensor nodes are less in number uniform topology gives better result in all perspective. When the number of nodes is more random topology is best suited since the energy efficiency is more. As the future work efficient scheduling algorithms can be implemented in every cluster to make few nodes to work at a time and hence more data fidelity can be acquired. Along with this clustering algorithm good routing algorithms can be integrated to get better results. REFERENCES [1]. Arthur Getis, “A History of the Concept of Spatial Autocorrelation: A Geographer’s Perspective” ISSN, vol. 4 pp.0016-7363, 2007. [2]. Pedro A. Forero, Student Member, IEEE, Alfonso Cano, Member, IEEE, and Georgios B. Giannakis, Fellow, IEEE, “Distributed Clustering Using Wireless Sensor Networks” IEEE journal of selected topics in signal processing, vol. 5, no. 4, august 2011. [3]. Vinay Kumar, Sanjeev Jain and Sudarshan Tiwari, “Energy Efficient Clustering Algorithms in Wireless Sensor Networks: A Survey”, IJCSI International Journal of Computer Science Issues, Vol. 8, Issue 5, No 2, September 2011 [4]. A. Abbasi and M. Younis, “A Survey on Clustering Algorithms for Wireless Sensor Networks,” Computer Communications, vol. 30, no. 14-15, pp. 2826–2841, 2007. [5]. W. Heinzelman, A. Chandrakasan, and H. Balakrishnan, “An Application Specific Protocol Architecture for Wireless Microsensor Networks,” Wireless Communications, IEEE Transactions on, vol. 1, no. 4, pp. 660–670, 2002. [6]. Sandeep Pattern, Bhaskar Krishnamachari, Ramesh Govindan, “The Impact of Spatial Correlation on Routing with Compression in Wireless Sensor Networks”. IPSN, vol. 1, pp. 846. 2004. [7]. B. Gedik, L. Liu, and P. Yu, “ASAP: An Adaptive Sampling Approach to Data Collection in Sensor Networks,” Parallel and Distributed Systems, IEEE Transactions on, vol. 18, no. 12, pp. 1766–1783, 2007. [8]. L. Chong, W. Kui, and P. Jian, “An Energy-Efficient Data Collection Framework for Wireless Sensor Networks by Exploiting Spatiotemporal Correlation,” Parallel and Distributed Systems, IEEE Transactions on, vol. 18, no. 7, pp. 1010–1023, 2007. [9]. C. C. Hung, W. C. Peng, and W. C. Lee, “Energy- Aware Set-Covering Approaches for Approximate Data Collection in Wireless Sensor Net-works,” IEEE Transactions on Knowledge and Data Engineering, vol. 24, no. 11, 2011. [10]. Zhidan Liu, Wei Xing, Bo Zeng, Yongchao Wang, Dongming Lu, “Distributed spatial Correlation Based Clustering for WSNs”, IEEE computer society, pp. 56-64, 2013. [11]. Quazi Mamun, “A Qualitative Comparison of Different Logical Topologies for Wireless Sensor Networks”, Sensors 2012, vol.12, pp. 14887-14913, 2012. [12]. Divya Sharma, Sandeep Verma, Kanika Sharma, “Network Topologies in Wireless Sensor Networks: A Review”, IJECT, vol.4, Issue-3, 2013.