Energy Efficient Algorithm for Wireless Sensor Network using Fuzzy C-Means Clustering

Energy efficiency is a vital issue in wireless sensor networks. In this paper, an energy efficient routing algorithm has been proposed with an aim to enhance lifetime of network. In this paper, Fuzzy C-Means clustering has been used to form optimum number of static clusters. A concept of coherence is used to eliminate redundant data generation and transmission which avoids undue loss of energy. Intra-cluster and inter-cluster gateways are used to avoid nodes from transmitting data through long distances. A new strategy has been proposed to select robust nodes near sink for direct data transmissions. The proposed algorithm is compared with LEACH, MR-LEACH, MH-LEACH and OCM-FCM based upon lifetime, average energy consumption and throughput. From the results, it is confirmed that the performance of the proposed algorithm is much better than other algorithms and is more suitable for implementation in wireless sensor networks. Keywords—WSN; clustering; sleep-awake; virtual grids; multihop; routing


INTRODUCTION
Wireless Sensor Networks (WSNs) consist of thousands of micro-sized low power Sensor Nodes (SNs) randomly deployed in the Sensor Field (SF).These SNs sense local environmental statistics, aggregate and communicate sensed information to sink.For each operation SN consumes its battery power.As SFs are hostile in most of the applications of WSN, it is not possible to replace batteries of SNs [1].In order to enhance lifetime of network, the routing algorithms in WSN mainly focus on energy efficiency.
Clustering is one of the most effective techniques in routing to preserve the energy of the network.The whole network is organized into small groups of SNs called clusters.
In each cluster, one node is elected as Cluster Head (CH) which performs the task of collecting data from all Cluster Member (CMs) nodes, aggregation of data and forwarding it to the sink directly or in multi-hop manner [2], [3].The aggregation may or may not be perfect depending upon the relation between the sensed data.Perfect aggregation means many k-bit data packets are compressed to a single k-bit packet.Perfect-fusion, a simple technique has attracted many researchers' interest [2], [4].LEACH (Low Energy Adaptive Clustering Hierarchy) [2], EEUC (Energy Efficient Unequal Clustering) [3], IB-LEACH (Intra-Balanced LEACH) [5], MR-LEACH (Multi-hop Routing LEACH) [6], MH-LEACH (Multi-Hop LEACH) [7], EMRP (Energy efficient multi-hop routing protocol) [8], ACEEC [9], OCM-FCM (Optimal Clustering Mechanism Fuzzy-C Means) [10], SEECP (Stable Energy Efficient Clustering Protocol) [11] and MLRC (Multi-Level Route-aware Clustering) [12] are some of the popular clustering protocols.In these protocols, all nodes in the network actively sense from the environment and continuously generate data.The nodes which are placed close to each other tend to generate redundant data due to phenomenon of coherence.A major amount of network energy is wasted in transmitting redundant data to sink which directly affects the network lifetime.Protocols like Span [13] and LEACH-SM (LEACH Spare Management) [14] keep a small subset of SNs active in such a way that these nodes cover the whole network while other nodes in the network are kept in sleep mode.In [15] an energy efficient sleep scheduling has been proposed in order to maintain network coverage and connectivity.Consequently, the energy consumed per round is reduced, thus, increasing the WSN lifetime.
Major part of energy is consumed by SNs in transmitting data which is proportional to the distance between sensor nodes raised to power ( 2)  .Therefore, long-distance transmissions should be minimized for optimizing the usage of network energy [16].Multi-hop data transmission between Gate Way nodes (GWs), CHs and sink can be used to reduce energy consumption, but SNs (may be CHs or GWs) near sink are more probable to be selected for transmitting the data to sink, resulting in their early death and thus, effecting the overall performance of the network.
In many clustering algorithms, number of clusters in each round is not fixed and employs poor CH election and cluster formation techniques.As a result, total inter and intra cluster distance becomes large, resulting in high energy consumption of the network.Therefore, soft computing techniques can be beneficially employed in cluster based routing protocols [16], [17].Fuzzy-C means clustering groups the SNs based upon their degree of membership in each cluster.The aim is to minimize the sum of distances between the SNs and the centroid of their cluster.OCM-FCM uses Fuzzy-C means clustering to form more uniform and correlated clusters in order to improve the lifetime of WSN [10].
In this paper, a multi-hop efficient routing algorithm for WSN has been proposed.The algorithm emphasizes to optimize the energy usage in the network in order to enhance the network lifetime.It eliminates redundant data generation in the network by keeping only one node active from the set of nodes that are sending redundant data and also renders full network coverage.Initially, the whole network is divided into optimal number of static clusters using FCM.CH in each www.ijacsa.thesai.orgcluster changes deterministically during the network operation.In order to reduce long distance transmissions and balance load, multi-hop routes are established using gate way nodes.In multi-hop routing, the nodes which lie near the sink are more probable to be selected for establishing direct link to the sink despite of having lower residual energy.Therefore, SNs near the sink will die soon and create energy hole in the network.To avoid it, a strategy is designed which selects the nearest eligible SN for transmitting data to sink, therefore consuming least amount of energy.It also ensures that this SN has sufficient residual energy so that it may not become dead.
The remainder of the paper is organized as follows: Section 2 gives brief review of the related work.Network model and assumption made for the proposed work has been presented in Section 3. Section 4 describes the proposed algorithm for an energy efficient multi-hop adaptive routing using FCM clustering.The results and their comparison with other algorithms have been discussed in Section 5. Finally concluding remarks are given in Section 6.

II. RELATED RESEARCH Many clustering protocols have been explored in literature
for WSN in the last few years to increase network lifetime.LEACH is forerunner decentralized clustering protocol uses random rotation of CH in each round to evenly distribute the load in the network.CHs are responsible to transmit cluster information to the sink [2].CHs far away from the sink consume lot of the energy to transmit data to sink.Randomized selection of CHs in each round adds extra overhead to the network.The algorithm does not assure even distribution of CHs in the network which may result into uneven energy consumption in different parts of the network.Moreover, the clusters formed in each round may or may not be correlated which increases the intra cluster distances.Therefore, LEACH is not suitable for large scale networks.IB-LEACH is an improvement over LEACH [5].CH selection of method is complex and involves lot of computations.Some nodes in the network are overburdened.
MH-LEACH and MR-LEACH are the variants of LEACH, which follow the same CH selection methods as proposed in LEACH [6], [7].These algorithms establish the multi-hop routes to transmit data to the sink and reduce number of long distance transmissions in the network.CH nodes which are near the sink (inner CH) are responsible for transmitting data to the sink and other CHs (outer CH) will transmit their data to the nearby CH.When sink lies outside the sensor field, inner CH will consume higher energy than outer CHs and may die soon.
EEUC and EMRP divide the network into unequal sized clusters [3], [8].The clusters that are formed near the sink are smaller in size as compared to clusters far away from the sink.There is an additional overhead involved for CH selection in each round.Nodes near sink are responsible to transmit data directly to sink even if they have low energy.ACEEC and THCEEC [9] are the centralized routing algorithms which results in better network lifetime as compared to the conventional distributed routing protocols like LEACH.The algorithm divides the whole network in regions which act as static clusters for WSN.The sink finds CH in each region and efficient path to route region data to sink.This centralized scheme is helpful in planning power-aware, well-organized routes in the network.Network status needs to transmit to the sink in each round which increases energy consumption in all SNs and overall performance of network degrades.
Soft computing methods like Fuzzy C-Means clustering (FCM), K-means clustering, Genetic Algorithms (GA) and Particle Swarm Optimization (PSO) are recently used to overcome problems of non-correlated clusters [10], [17].It helps to prolong the life of WSN.However, all these methods run at centralized node i.e. at sink and need physical location of SNs.In OCM-FCM reduces the cluster formation overhead on SNs as sink divides WSN in K optimal number of static clusters.After first round, the current CH selects a highest energy node in the cluster and elects it as CH for the next round.This algorithm uses multi-hop technique by finding Secondary Cluster Heads (SCHs) from already chosen CHs if sink is far away from the sensor field [18], [19].It will result in reduction of long distance transmissions.The drawbacks of the algorithm are: 1) Poor CH selection; 2) CHs are overburdened as compared to other SNs; 3) SCHs consume more energy as compared to primary CHs.In all the algorithms discussed, all nodes are regularly sensing and transmitting data to sink.Thus, large amount of network energy is wasted in redundant data transmissions which need to be reduced.

III. NETWORK MODEL AND ASSUMPTIONS
Total N sensor nodes in WSN are randomly deployed in of X m by Y m rectangular sensor field.The assumptions made for the implementation of the proposed work are as given under: 1) All nodes are stationary and have unique ID.
2) It is assumed that all SNs find their locations either through GPS system or using some localization algorithm.
3) All nodes are static and have same initial energy init E 4) All the active sensor nodes regularly generate information from the environment.
5) All SNs are capable to perform data aggregations in perfect fusion mode.
6) In the cluster formation process, each SN can be a member of only one cluster and can act as CH for the same cluster.
7) The sink has unlimited energy and computational resources.
8) First order radio energy model described in [2] is utilized for sensor nodes during their communications procedure.

IV. PROPOSED ENERGY EFFICIENT ALGORITHM
The paper proposes an energy efficient algorithm for WSN using multi-hop routing.Initially it runs a network dimensioning phase where sink divides the whole network into desired number of static clusters using FCM technique and turns off all SNs producing redundant data.Thereafter, it organizes the network operation into a series of rounds where each round is divided into network setup phase and network communication phase.In network set up phase, new CHs are elected in static clusters if required and energy efficient multihop paths are established from each SN to the sink.The network communication phase deals with intra-cluster and www.ijacsa.thesai.orginter-cluster multi-hop routing and data transmissions.In network communication phase every CH allocates TDMA slots to CMs in each cluster and for inter-cluster communication is achieved through CDMA codes.

A. Network Dimensioning Phase a) Division of SF in granules:
WSN is initially divided uniformly into granules.The size of granule may vary according to degree of coherence in sensed data.In a SF, if information to be sensed is more coherent, then size of granule may be taken large and vice-versa.The entire SF is divided All SNs save this information for further use.

b) Initial Active-Sleep Methodology:
Once SNs find their associated granule ID, each SN will broadcast a message ( , ) S NodeID GranID to all SNs in its radius com R .Where com R is equal to the diagonal of a granule and is given by following equation: Initially, SNs do not need to broadcast their residual energy as all SNs have same init E .Each SN saves NodeID of the received message if it belongs to its granule node and discards otherwise.In this way, all SNs maintains NodeID and resi E of its granule SNs.Once, all SNs save information of their associated GranID , granule SNs.SN with highest node ID in a granule will activate itself and broadcast its location and ID information to the sink.Remaining nodes in the granule shall be in sleep mode.c) Clusters Formation: At the end of active-sleep process, the sink will receive location information of all n active nodes.The sink first calculates their associated GranIDs using (1), then divides whole WSN into K optimal number of clusters using FCM clustering algorithm.The SN is associated with the cluster in which it has the highest membership value based upon its distance from the centroids of the clusters.The sink broadcasts messages   , ClusterID List of associated Gra S nIDs   for each cluster to all SNs.SNs belonging to GranID specified in the message will save this message in its cluster table.

d) Initial Selection of Cluster Heads:
The sink divides the whole network into clusters.Thereafter, the sink evaluates the fitness of all SNs in each cluster on the basis of an objective function given by (3).The SN with highest fitness value will be elected as CH in each cluster.The sink broadcasts this information to each cluster.
where () fi is the fitness of th i member of th c cluster.
( , ) dis i j is the Euclidean distance between th i and th j member of th c cluster.The value of  is taken as 2 for ( , ) dis i j less than threshold distance, otherwise 4. e) Logical Region Division: In many multi-hop routing algorithms proposed by various researchers, SNs nearer to sink carry heavier traffic loads and are responsible for routing data to the sink.Therefore, SNs around sink deplete their energy at faster rate and may create energy hole near the sink.To avoid the problem, the SF under consideration is divided into three logical regions as Low Energy Area (LEA), Medium Energy Area (MEA) and High Energy Area (HEA) depending upon the energy required by SNs in these regions to transmit data directly to the sink.The sink assigns region IDs to all SNs.The logical region division has been shown in Fig. 1.The SNs from the region specified by the sink will only establish direct communication to it.Initially, LEA becomes region of interest.b) Multi-hop route establishment: CH in a cluster receives information from its CMs, aggregates the received data with its own sensed data and forwards it to sink either in single hop or in multi-hop.CMs are accountable to sense environmental data and transmit sensed information to CH.Therefore, CH in a cluster does more work than other nodes and consumes much higher energy.To reduce its work, the task of inter cluster communication is shifted to another node known as Intra Cluster GW (IntraCGW).The cluster CH will choose IntraCGW in the direction of logical region announced by sink.If the selected IntraCGW is in the announced logical region, link is established directly to sink depending upon a probability function described in (4), otherwise inter cluster path is devised.Then, IntraCGW finds a nearest inter cluster SN in a cluster, currently not connected in order to avoid loops in communication path.The selected inter cluster node called "Inter Cluster GW (InterGW)" forms the link between the two clusters.The multi-hop topology formed by the proposed algorithm is shown in Fig. 2.
where P is the desired percentage of direct links to the sink, r is the current round, R avg E is the average energy of CMs of the cluster in announced region E avg is the average energy of the active nodes in the network.Each CH in c th cluster in announced region generates a random number between 0 and 1.The CH will establish a link to the sink through selected IntraGW, if generated random number is less than threshold value of the cluster as calculated in (4).The proposed algorithm to establish multi-hop routes is given in Fig. 3.

c) Region Announcement by Sink:
In order to attain nearly balanced energy depletion and to increase the reliability of WSN, it is desirable that all nodes remain alive for larger time period.SNs in LEA region consume lowest energy in direct data transmissions than MEA and HEA region nodes.Therefore, nodes from LEA region are desirable for establishing direct link to sink, but if these nodes continue to transmit information directly to sink, they will consume energy continuously and ultimately become dead.To avoid such situation, a shifting strategy between three logical regions has been proposed in this paper.A threshold parameter R th E is taken to decide this shifting strategy.

Initially,
R th E is set to 30% of E init energy.Depending upon its value, the sink announces the region of interest in each round.

If residual energy of all SNs in the SF goes below
E is decreased to 10% of E init and announce LEA as region of interest.
Algorithm: Network Setup Phase

C. Network Communication Phase
In communication phase, there are two types of communication takes place in the network: intra-cluster www.ijacsa.thesai.orgcommunication and inter cluster communication.For intra cluster communication CH node assigns TDMA schedule to its entire CMs.Each CM in a cluster transmits its sensed data to CH in its allotted slot.With TDMA scheme lot of energy of CMs is saved as CM goes in sleep mode when it is not transmitting any data.CH of each cluster receives data packets from CMs and aggregated the received information with its own sensed data and forwards it to IntraGW.It will then aggregate the received data packet with its own and forward it to InterGW using CDMA codes.In each round, active SN those go below R th E transmit its energy status to the sink.Depending upon information received, it announces the region of interest.

V. RESULTS AND DISCUSSIONS
The proposed methodology has been implemented with MATLAB and executed on 100 randomly deployed wireless sensor networks in order to evaluate the performance parameters: network lifetime, energy consumption and throughput.Network lifetime is considered as the number of rounds until the complete network is dead.Energy consumption in each SN of the network is measured as the sum of energy consumed in receiving, aggregating and forwarding the data to other intermediate nodes or to the sink.Throughput is defined as sum of unique data packets generated in each round.For example, four data packets are generated from an active granule by four SNs in that granule.We consider it as one because other three carries same data as that of first packet.For simplicity, it is assumed that there is no data collision and packet loss in the wireless channel.Likewise, control packets have also been ignored for the evaluation of performance parameters.
In this simulation setup, one hundred static sensor nodes are randomly distributed in a SF of area 200m X 200m and considered as one WSN deployment.The sink is placed at location (100,100) and is immobile.Initial battery power for all SN is taken as 0.5J.Further, to ensure the generalisation of results, the simulation experiments were conducted for one hundred WSN deployments.Each simulation is run until all sensor nodes in the network become dead.The energy consumed in the network is calculated for transmitting, receiving and aggregating data packets.The energy consumed in aggregating and transmitting 1 bit data over distance d is calculated using (5).
Energy consumed in receiving 1 bit data is given as under: ( , ) *( ) Where is the length of data packet taken as 4000 bits.To decide optimal number of clusters K opt for the proposed algorithm, the algorithm is executed on all 100 simulations for varying K, i.e. number of clusters, from 1 to 20.The average energy consumption per round for each value of K is calculated.The minimum average energy consumption per round is found at K =10.Hence, the value of K opt is taken as10.
In order to generalize the results, all the five algorithms (LEACH, MR-LEACH, MH-LEACH, OCM-FCM and EEA-FCM) were implemented on the same hundred deployments and averaged results obtained from them have been used to present the comparative account of the algorithms.The results are shown in Fig. 4, 5, 6, 7 and 8.     Therefore, network load is uniformly distributed amongst all SNs in EEA-FCM and consume almost equal amount of energy which leads to more stable and reliable network operation.
A comparative graph of throughput for all five algorithms is presented in Fig. 8. OCM-FCM and MR-LEACH acquire less amount data from the SF as their nodes have lesser lifetime as compared to EEA-FCM.In EEA-FCM, SNs are not generating redundant data from SF and remain alive for longer period.Therefore, throughput of the network is improved.Total throughput in case of EAA-FCM is 7.6% higher than MR-LEACH and 28% more than OCM-FCM.Thence, the proposed algorithm shows a significant improvement in throughput.
Node density of WSN may greatly affect SF coverage, network connectivity and network life time.Therefore, it is an important factor that needs to be considered while evaluating the performance of WSN.To verify the effect of node density on the performance of the network, additional simulation experiments are conducted by varying the number of nodes, N, in the WSN deployments as N=100, 400 and 1000.Please note that N=400 in SF of 200m X 200m is equivalent 100 nodes in an area of 100m X100m.
The percentage of dead nodes for varying number of rounds has been shown in Fig. 4 (N=100) and Fig. 10 (N=400) and Fig. 11 (N=1000).Comparison for all three algorithm for N=100 has already been discussed earlier.Taking the discussion further, it is seen that when network node density increased to 400, within first 1100 rounds, all nodes for LEACH, MH-LEACH and OCM-FCM become dead, but all the four hundred nodes for EEA-FCM are still alive.When 15% nodes of EEA-FCM are dead, 90% nodes in MR-LEACH has already become dead.Further, the slops of five curves in Fig. 6 have been observed.It is clearly seen that the slop of curve for EEA-FCM is much less than other two curves.This indicates that the rate of nodes becoming dead is much slower in EEA-FCM.Lifetime of WSN for the proposed algorithm is almost double than LEACH, MH-LEACH and OCM-FCM algorithms.Further, when number of nodes in WSN increased from 400 to 1000, similar behaviour of algorithms has been observed.There is no improvement in network lifetime of LEACH, MH-LEACH, MR-LEACH and OCM-FCM.Network lifetime of EEA-FCM is far improved over other two algorithms.It is approximately 4.5 times that of LEACH, MH-LEACH and OCM-FCM.The network throughput obtained in LEACH, MR-LEACH, MH-LEACH, OCM-FCM and EEA-FCM with varying N is shown in Fig. 9 and Table I.It is observed, for N=100, OCM-FCM, MR-LEACH and EEA-FCM generate 96641, 114973 and 123577 data packets in its lifetime, respectively.There is a 63% improvement in throughput of EEA-FCM over OCM-FCM for N=400.By varying nodes from 400 to 1000, throughput obtained in EEA-FCM is almost three times as compared to OCM-FCM.The throughput of EEA-FCM is improved in many folds compare to LEACH and OCM-FCM with the increase in node density.The value of rounds for First Node Dead (FND), Half Node Dead (HND) and Last Node Dead (LND) for all algorithms under examination at N=100, 400 and 1000 are shown in Fig. 10 and 11.For N=100, OCM-FCM improves the round number for FND by 4% over MH-LEACH.EEA-FCM improves FND of MH-LEACH further by another 11%.The round number for HND is 694 th , 1171 th , 984 th , 977 th and 1220 th round in LEACH, MR-LEACH, MH-LEACH, OCM-FCM and EEA-FCM respectively.MR-LEACH improves the round number for HND for OCM-FCM by 19%.EEA-FCM improves it further by 5%.The LND round number is almost same for LEACH and OCM-FCM.The network lifetime for EEA-FCM is improved 58% over LEACH.It is also observed that by changing N from 100 to 400, FND values of all three algorithms are improved.EEA-FCM improves HND value of OCM-FCM by 63%.Network lifetime for EEA-FCM is almost double as compared to the other two algorithms.For N=1000, there is not much improvement in LEACH and OCM-FCM.The round number for FND, HND and LND in EEA-FCM is delayed by 182, 1396 and 2063 rounds respectively as compared to their values at N=400.Therefore, network operation got improved in EEA-FCM on increasing the node density.
into g n granules where x n granules are along x-axis and y n granules along y-axis.The size of granule is w x by w y subject to constraint that w x and w y are integer multiple of width and height of SF respectively.Any th i SN with co-ordinates   , ii xy finds its associated granule ID using formula given as under:

Fig. 1 .Fig. 2 .
Fig. 1.Logical Region Division of SF in LEA, MEA and HEA (this figure is created for this research work).

E
is energy dissipated per bit to run transmitter or receiver circuit, taken as 50nJ/bit.DA E is energy consumed in aggregating one bit data and set to 5nJ/bit/signal.fs E and mp E are energy consumed in free space model and energy used in multipath fading model taken as 10 pJ/bit/m 2 and 0.0013 pJ/bit/m 4 respectively.The threshold distance o d is equal to square root of ratio fs mp

Fig. 8 .
Fig. 8. Percentage of dead nodes verses rounds at number of nodes is 400.

Fig. 9 .
Fig. 9. Percentage of dead nodes verses rounds at number of nodes is 1000.
What is the highest residual energy of the granule?" to all CMs and activate active sleep process.3.Each active SN in the cluster finds other SN in its granule with highest

TABLE I .
LIFETIME AND THROUGHPUT FOR VARYING NODE DENSITY N=100, 400 AND 1000