A Review on Energy Efficient Clustering Protocols of Heterogeneous Wireless Sensor Network

— Today is the era of information technology, collect the information and use it for required application with technology support. Sensor nodes operating remotely are the popular approach for today’s researcher for collecting real time information. But, facing difficulty due to constraints of energy resource for long life monitoring. So there is high need to have energy efficient communication scheme for the betterment of sensor network communication. Clustering protocol is the best option in designing routing protocol for Wireless Sensor Network (WSN). Though we have option of heterogeneity, Clustering approach enhances the energy efficiency of WSN by systematically sharing the load and hence prolongs the lifetime of the network. Most of the researchers achieve energy efficient approach in WSN, by adding different level high energy nodes and use clustering approach to prolong the lifetime of WSN. There are lots of efforts put in reality by researchers for the development of energy efficient schemes with WSN. This paper explored the contribution of different clustering scheme reported in published literature in the three sections as Clustering algorithm basic and its different attributes, suggested Cluster head selection criteria, literature survey and the identified gaps found in published material. The main aim of this paper is to present basic considered in designing clustering algorithm and metrics available for validation.


I. INTRODUCTION
The main aim of WSN is to be a witness and become the reporter for the events of the physical world [44]. Over the last two decade WSNs Popularity has been increased tremendously in several applications like scientific, industrial and social domain to explore events [4,45]. All these applications use static placement of sensor nodes. For example in Scientific domain by putting nodes at appropriate places at top of mountain, at Glaciers field, at suitable point in city and at water dam bottom side [8,45], it is possible to predict the accurate forecast [43]. In industrial domain we can use sensor nodes in electrical machine for monitoring the status of armature, and other parts of electrical motor [38,43]. Sensor nodes can be used, in power plant monitoring. In approximately all cases, sensor nodes have to be dependent on their limited powered battery and processing power resource, and after deployment are usually unattended from sensing regions [52]. In social domain WSN can be used for environmental monitoring, some of the times in wildfires monitoring [40,50]. In Military applications sensor network is used as in target tracking, battlefield surveillance, and container monitoring are some of the dynamic node applications from WSN [8,15,20,36,60]. Sensing the events and routing the recorded valuable data for longer life is found to be extremely important and challenging.
The above said applications demands energy efficient schemes for WSN [23], so that available energy of WSN is use systematically and network can serve for longer life. Solution for above point is, data traffic at run time must be evenly distributed over available sensor node. Network reliability must be maintain during event monitoring. For the entire above requirement, clustering is the best solution [3,5,44]. In clustering heterogeneous WSN (HWSN) sensing work is almost handle by small energy capable node and data from all sensor is collected and forwarded towards sink is done by high energy nodes, other than their regular work (named as cluster head in clustered scenario) [49,52]. With clustering approach in real time scenario different parameters are explored like stability period, throughput, propagation delay and reliable data delivery. Hence lifetime of the network get prolong. Currently about 50 % of the researcher are interested in energy efficient method to prolong the life time of wireless sensor network and 50 % of those are working on different schemes like aggregation, security and different sink arrangement etc. [47]. In latest reported public material shows that clustering is an efficient scheme for hierarchical network routing, balancing the load of traffic, enhancing the stability period and hence prolonging the network lifetime [1,16,60,47] . Clustering is a scheme through which sensor network is divided into groups with central controller for each usually called as Cluster head (CH) with responsibility as to be update and live all the time and maintain intra-connectivity as well as inter connectivity with sink or base station [42,46] . Clustering reduces the overall energy consumption between networks with the expense of small communication effort.
There are basically two approach of clustering as probabilistic, non-probabilistic. Non probabilistic approach further support for bio-inspired schemes of clustering. Clustering schemes can also be grouped as centralized and distributed approach [28,45,62]. This paper is an attempt from our side to highlight the recent development in clustering HWSN based on different criteria as algorithmic strategy, metrics considered, software used other than merits, demerits and number of energy level of nodes. The rest of the paper is organized as basics of HWSN and objectives of HWSN, clustering basics and objectives of clustering, cluster head selection criteria, performance measures of clustering WSN, Data transfer approach in clustered WSN, observations from our side in the form of findings from reviewed literature and summary followed by literature survey and finally there is conclusion.

II. CLUSTERING HETEROGENOUS WIRELESS SENSOR NETWORK
We are presented some basics for clustering Heterogenous WSN, brief about heterogeneous WSN and then clustering are explained with this section.

A. Heterogeneous WSN (HWSN)
WSN's are of two kinds homogeneous and heterogeneous [22]. In homogeneous WSN, all nodes have same capabilities as processing, energy level (in the form of battery) battery source and link bandwidth etc. But, in reality once sensor network start for event monitoring and updating it no longer being a homogeneous network. As energy level available with sensor nodes is not same. Hence it becomes heterogeneous WSN. Hence, Heterogeneous WSN is the suitable sensor network for real time event monitoring [46]. As the WSN which consist different capabilities like processing, energy level and link bandwidth etc. But, node with different energy level is the best option for heterogeneous WSN. As different capabilities like processing and link bandwidth demand energy backup [1,23,29,42,48,53]. Objectives of adding heterogeneous nodes in WSN are listed as follows.
1) Increase the data transmission speed: Due to high capability of processing, energy level and link bandwidth improves data manipulation speed and with high energy level, data can be transmitted at long distance than simple sensor node or normal node. With high link bandwidth no waiting period or data with high rate transfer is achieved [1,23].
2) Enhance the lifetime and network throughput: As high energy node available which compensate for high processing power and high link bandwidth reduce the burden of packet transfer. With high energy node stability of the network get enhance, on the cost of high initial energy in the network [1,23].
3) Enhance the reliability of data communication: As with high energy level, high link bandwidth and with high processing capability makes manipulation more precise and with less effort data can be transferred to base station or sink [1,23].

4) Enhance coverage within network:
With high energy node, connectivity in the network get elongated. It also support for high scalability in the network [1].
B. Clustering basics "To group the same capability or same functionality related physical elements", and assign or arrange for the specific task, for the required outcome is termed as cluster [2,28]. Some of the main objectives of clustering scheme are listed as follows.
1) Systematic Load sharing among the nodes is possible: As there is CH selection or election, then TDMA frame from the CH towards sensor nodes. Every node send collected information in allotted time slot [1,2,45 ].
2) Node failure can be identified: With acknowledgment received from cluster member CM to CH after advertisement. TDMA frame is formed only with replied nodes not with all the available. At the start of every cluster cycle this process is repeated [1,2].
3) Reduce the intra message in the network: Without clustering every node have to participate in completion of data transfer to sink. On the contrary with single hop or multi-hop communication need to adapt for data transfer. Also chain based approach is possible for without clustered environment for data transmission but, if any intermediate node of chain failure results in loss of valuable data [1,2].
C. Clustering process WSN can be reactive or proactive, but the clustering process having some stages [12,13,18]. These stages are common in all kind of clustering scheme, as selection of cluster head, cluster setup phase and steady state phase. First there is the cluster head (CH), which is one of the node from the available network nodes. Which can be fixed or selected based on some strategy, which is totally controlled by algorithm working over the network. Selected cluster head intimated by sink or by the sensor nodes working in the network. Once CH is selected or elected then it intimates to all other nodes can be called as cluster members, which is in coverage of communication radius. Intimation is all the way through advertising its identifiers like its id, energy level available, its closeness with base station or all other cluster member and cluster member reply to CH by updating appropriate information in routing table such as receiving signal strength, hop count, relative distance with CH etc. This is cluster setup phase. With the reception of reply, CH design time schedule as TDMA frame (Time division multiple access frame) as per the number of cluster member and send it over coverage span. CSMA/CA is the collision avoidance scheme most probably used. Here after all the members send their respective collected data to CH only in pre-decided time slot. The time span after issuing TDMA frame to all node reply (all member reply) on their respective time slot, is called cluster cycle. This completes the steady state phase of cluster process [1,2].
There are some of the cluster head selection criteria's, which are explored as follows.

D. CH selection criteria
Clustered network mainly have three entities as cluster head (CH), sensor node (SN) and base station (BS). Main control entity is CH, for selecting CH we listed different attributes from the reviewed literature is as follows.

1) Initial energy:
In this scheme the nodes which have higher energy (battery source) is selected as CH. For example advance nodes or super nodes can be CH [1,20,23].
3) Random or probabilistic: Any node can be CH in this criteria without considering its capability, position and its relative distance etc. Such as random number generated by sensor nodes and whose number is less as compared to other is selected as CH [6,18,38,49,50].

4) Distance from sink:
The node whose distance from the all other cluster member is small with respect to sink node can be selected as CH. Some of the time other criteria is also added for CH selection as node having greater connectivity with all others and distance is also less compared with others is selected as CH [10,14,15,20].

5) Energy consumption rate:
The node whose energy depletion rate is lower compared with other nodes from the coverage zone is selected as CH. Some of the time other parameter is also added with energy required for current round by the selected CH [16,21,22,41].

7)
Node degree: Node with strongest connectivity with all the cluster member is selected as CH. Node who receive maximum level Receive Signal Strength Indicator (RSSI) , so that packet can be received from cluster member is CH for the round [11].

8) Hybrid approach:
Cluster head selection criteria in which more than one parameter considered for CH selection. Means node with higher energy available and small distance with sink, initial energy and remaining node energy with average energy of the network. There can be criteria as epoch period with remaining energy, Node with low hop count towards sink and better connectivity selected as CH [19,27,29,[30][31][32][33]36,37,43,44,[46][47][48]52].
Clustering performance can be evaluated with reference to some of the listed performance measures of clustered WSN

E. Performance measures of Clustering WSN
Mostly considered performance parameter by the researchers is as follows.

1) Stability period:
The time interval between start of network (very first message triggered in the network) to very first node death is termed as stability period. Because it is the interval over which all the nodes from the network are working against energy exertion without sacrifice [1,6,7,15,18,19,22,24,25,27,29,32,34,47,56,64].

F. Data transfer approach in clustered WSN
Data collection from WSN is possible through three different ways, with reference to base station. In heterogeneous network there can be possibility of mix types of nodes instead of same sensing parameter. By the help of query, time and threshold base approach it is possible to collect the information for required sensing event.

1) Query based:
In query based approach the query of administrator is placed from BS over the network towards CH. CH in response forwards the said query for searching information over a network. The sensor node send the collected information to CH as a response. Only matched data will be send by all respective CH to BS. But, while collected data is being send by sensor nodes at the same time there energy level information is also conveyed to respective CH [12,13,57].
2) Time threshold: In this time frame sent by selected cluster head to member and collect the data from those entire member. Transfer the aggregated data after collection followed by CH selection. In this scheme all CH are tightly synchronized with each other [13,16,34].

3) Threshold based:
In case of threshold value based scheme information collected by sensor nodes is monitored an update till the threshold is reached. Once threshold is achieved then the collected data will be sent to CH. Same platform followed by CH, it collect the data from all the cluster member till the threshold of its own is reached from aggregated value before sending towards BS [12,18].

III. LITERATURE SURVEY
According to research effort taken by researcher for the development of best possible routing protocol through clustering scheme based on probability strategy. But, they have their target application or network architecture. On the other side there can be different options possible, one of the option other than probability strategy is combining different parameter for clustering or through non-probabilistic approach reducing energy consumption in the WSN. Which indirectly enhance the lifetime of network.

A. Clustering schemes of Homogeneous WSN
Towards this work many researcher contribute and propose energy efficient scheme for homogeneous and for heterogeneous WSN [23]. Low Energy Adaptive Clustering Hierarchy protocol (LEACH) proposed in [6,7,8] is the protocol more suitable in the environment over which role of CH is rotated after each cluster cycle. So, that load balancing and hence energy efficiency in WSN can be achieved. CH selection is based on random number, in this every sensor node generates random number and if the generated number is less than the threshold given in [6] is elected as CH for current round. CH send advertise message to sensor nodes of his coverage. Cluster member reply to advertised message by sending acknowledgement. After receiving acknowledgement CH prepare a schedule of TDMA frame and publish over the network (Cluster section of his control) every node send the collected data within their allotted time slot. Time span between transfer of TDMA frame from CH to again data collected frame to CH and aggregated data to base station is termed as Cluster cycle or cluster round. LEACH protocol is suitable for homogeneous WSN not for heterogeneous WSN. As LEACH select the CH based on random number without considering the available energy or relative distance etc. hence selected CH is less faithful than expected. To resolve the issue of LEACH, A hybrid energy efficient distributed (HEED) clustering approach for ad hoc sensor networks (HEED) protocol from [10] which undertake residual energy of the node for the selection of CH and relative distance with respect to BS. This protocol us slight faithful CH than LEACH, but having less emphasis on node connectivity and available energy within the network or energy consumption rate of selected node. PEGASIS protocol proposed in [9] is one of the different direction for energy efficient WSN, in the form of chain based data communication to BS. With this study many of the researchers are trying to encompass different parameter for the best choice of CH selection and with multihop scheme [51].

B. Clustering Schemes of Heterogeneous WSN
As in real time WSN, no network found to be homogeneous for long time. Heterogeneous WSN (HWSN) is the popular network architecture for real time sensing environment. Researchers of Heterogeneous WSN protocol are motivated from the basics of homogeneous WSN protocol like [6,7,9,10], so that energy consumption per cluster cycle is to be reduced as much possible. To enhance the energy efficiency and indirectly enhance the stability period of WSN, hence prolong the lifetime of WSN. By rotating the role of cluster head and threshold will be set for the selection of CH, on the same base as LEACH. Very first heterogeneous WSN protocol proposed in [17] is with two type nodes as normal and advanced node. A CH selection criterion for this protocol is based on remaining energy with nodes and not on the network energy. It means selection of CH not taken into account global information of the network. This protocol introduces high energy node, hence stability of the network get enhanced and it prolong the lifetime of WSN. Demerits of this scheme is identified as there is always high energy node get selected as CH, hence there is always punishment to advance node. Researchers contribute towards this approach for better stability and better remaining energy and hence to prolong the lifetime of the network in the form of [18,24,25]. Enhanced Distributed Energy Efficient Clustering (EDEEC) protocol mentioned in [26] is three level node heterogeneous WSN, in this added node termed as super node. Main aim of this scheme is to increase the heterogeneity and improve stability period, hence lifetime of the network by using the base of [24] or indirectly with the base of [6]. CH selection follows the base of probability with the ratio of remaining energy to average energy of the network. Controlling the epoch period also contribute for enhancing stability period and lifetime of the network in [29]. Some of the other detailed contribution from the researchers for different schemes is organized as follows in Table I. Presents comparison details of contribution for different algorithms proposed by the researchers. Table I   In this protocol penalty is always for advanced nodes, in the condition when their energy deplete and in the range of the normal nodes.

MATLAB
Summary: DEEC is a distributed energy efficient clustering algorithm for heterogeneous WSN. In this cluster heads are selected based on probability using the ratio between residual energy of each node and the average energy of the network. So the node with higher residual energy has the more chance to become CH. With the use of varying epoch strategy, distribute the load over the network. Hence prolong the lifetime specially the stability period as compared to SEP and LEACH. 12 [25] To balance the energy consumption inside the network by selecting CH over all the network balance selection of CH. Summary: Developed Distributed Energy-Efficient Clustering (DDEEC) algorithm select the CH based on residual energy. DDEEC adopts the base of DEEC, to select the cluster heads. After CH selection procedure and round completion, when advanced node become normal node at one point selection of advance and normal node is of equal possibility and hence life time and stability is enhance. 13 [35] To increase the stability of the network by enhancing energy efficiency Probabilistic Approach Remaining energy, throughput and no.of alive nodes/ round High assurance for node having higher energy become CH Need to calculate average energy of the network for CH selection, which consume more energy

>3 MATLAB
Summary: This threshold distributed energy efficient clustering (TDEEC) algorithm modifies the scheme of cluster head selection as with normal procedure of LEACH. TDEEC also use the basis as DEEC, ratio of residual energy to average energy and with the value of optimum no.of cluster heads required and compromise between this is selected for further work. Hence prolong the lifetime of network. 14 [61] To Enhance stability by reducing control messages Non-probabilistic approach Instability period, packet delivery/ round, network lifetime and stability Better and reliable communication, more stable network Less importance given to control messages, increase the load over a network, which required high processing energy

Summary:
In this algorithm BS is responsible for CH selection (based on residual energy, average energy and relative distance between nodes and BS).In this nodes contributed for their respective region information communication, This protocol is validate only with selected protocol. 15 [62] To select best location of CH for improving energy efficiency

Random approach No.of clusters and energy consumption/ round
It select best location of CH, non-uniform cluster formation It adapts centralized mechanism for management and network initialization messages have less importance

-
Summary: This scheme select best possible CH such that energy consumption get reduced. CH to BS communication is direct. In case of multiple cluster there is formation of shortest path towards sink and joined to form chain, with the help of single cluster head in the center of cluster algorithm. This scheme consider the boundary condition and density of nodes available in the cluster zone. 16 [63] To enhance the lifetime of sensor network by reducing energy consumption Non-probabilistic, Hybrid approach Lifetime (energy consumption and time) It gives best cluster, reduce intra cluster energy Intra-cluster messages are high 1

NS-2
Summary: Proposed protocol minimize the possibility first dead nodes and the energy consumption, hence extend the network lifetime. In this, each node in order to calculate the node weight sends a message to BS. For CH selection the proposed protocol uses a hybrid scheme of residual energy, distance between nodes and BS, Weightnode. It means cluster head selected must dissipate minimum energy with mechanism of cluster radius, which reduce network load. The node which has high Pch value than neighbour's get elected as CH. CH send only meaningful information from the collected information. 17 [64] To design Energy efficient scheme and take full advantage of heterogeneity IV. FINDINGS FROM REVIEWED LITERATURE As selecting the CH at center offers best connectivity in the network, but this scheme have less assurance that CH is always at the center of cluster. Sending routing schedule over a network with available residual energy and decision based on the same is taken for CH selection. But, this is time consuming and energy consuming process. Such messaging also increases internal messages over a network. A euclidean distance criterion is less faithful than, selecting CH based on energy available. Every time for new round initiation messages transacted over the network are proportional to number of parameter and number of calculations with number of nodes available in the network. Most of the literature assumes that internal messages processing energy is negligible. Assuming CH statically and centrally controlled approach of clustering offers less reliable cluster scheme. Most of the researcher used random deployment for sensor node placement; this is one of the causes for less stable network than uniformly deployed network.
V. SUMMARY OF REVIEWED LITERATURE It is identified that, precise the selection of CH better the clustering connectivity or node degree and hence better the performance measures of clustered WSN. Hence at the end of this study, we found that SEP and its variants consider the residual energy as main parameter for probability ratio calculation. Which contribute all the way for better stability, hence enhances the throughput of the SEP and its variants. On the other side DEEC and its variants are totally targeting for heterogeneity and average energy available within the network, results in life time enhancement. There are some variants of LEACH and HEED offers better result compared to the original LEACH and HEED protocol in terms of adding node distance and heterogeneity of nodes. Other important parameter from the present literature is node degree, available energy on routing path, spanning tree basis and some of the Bio-inspired base like Bee colony and mechanical base of firefly-simulated annealing also add weightage to select best possible routing mechanism with clustering. PEGASIS basis of chain based data transaction is also used in intra-cluster communication and single hop or multihop communication can be used for CH to BS communication. According to some of the authors selected CH must be at center gives best node Connectivity, and better coverage results in best clustering. Hence results in more stability and prolonged lifetime of WSN. Researcher also target for distance between different CH positions, which also contribute in lifetime enhancement of the network. So, there is high need to design a clustering scheme which having the provision of identifying best possible cluster by keeping internal messages as low as possible other than valuable data message. If there is any death of node then there must be provision to trigger the cluster head and base station. Other option for the energy efficient clustering scheme is to check node available in cluster must have minimum energy for communication.

VI. CONCLUSION
With the complete survey of reported literature here it is identified that most of the researchers still follow a random deployment strategy with modified probability weight function and with the threshold comparison decision is taken for cluster head. So there is high need to have intelligent scheme based on automata or neural network such that nodes clustering and head selection is executed based on real time or run time value.
Hence there is high need to design and develop clustering algorithm of Non-probabilistic approach type for balancing the clusters, and must result optimal number of cluster in the network such that network survival time get improved. Hence, indirectly stability of the network get enhances. If life of the network is extending, then it indirectly contributes for packet delivery ratio to BS and hence throughput enhances.

ACKNOWLEDGMENT
The author would like to thanks Prof. Dr. G.K.Kharate and Dr. V.H. Patil from Matoshri College of Engineering and Research Centre, for their continuous support for our research work and valuable suggestion and constructive comments to improve the quality of research work.