A Survey of Energy Conservation Mechanisms for Dynamic Cluster Based Wireless Sensor Networks

WSN (Wireless Sensor Network) is an emerging technology that has unlimited potential for numerous application areas including military, crisis management, environmental, transportation, medical, home/ city automations and smart spaces. But energy constrained nature of WSNs necessitates that their architecture and communicating protocols to be designed in an energy aware manner. Sensor data collection through clustering mechanisms has become a common strategy in WSN. This paper presents a survey report on the major perspectives with which energy conservation mechanisms has been proposed in dynamic cluster based WSNs so far. All the solutions discussed in this paper focus on the cluster based protocols only.We have covered a vast scale of existing energy efficient protocols and have categorized them in six categories. In the beginning of this paper the fundamentals of the energy constraint issues of WSNs have been discussed and an overview of the causes of energy consumptions at all layers of WSN has been given. Later in this paper several previously proposed energy efficient protocols of


BACKGROUND
In WSN there are several reasons that cause necessary and/or unnecessary utilization of energy. Energy consumptions occur at all layers of networking; physical layer, MAC layer, network layer and is also dependent on the application requirement. Many approaches have been adopted by the researchers in the past to conserve the scarce node energy as much as possible such as [1][2]. In this section we give an overview of causes of energy utilization in WSNs.

Energy Consumption at MAC Layer
MAC layer provides the accessibility to communication channels and there are two basic strategies used for this purpose: contention based and schedule (contention free) based [3][4].
Most of the WSN protocols use combination of these two strategies on their MAC layer. For example, in LEACH    We have categorized previously proposed cluster based protocols according to their energy conserving mechanisms. Table 1 gives an overview of the classification of these protocols.

Data Aggregation and Compression Solutions
To get a better coverage and robust fault tolerance in the network usually more than required numbers of nodes are deployed in WSN. Since the cost and sizes of SN are trivial, therefore it is possible to have redundant nodes in the network and these spare nodes also generate redundant data flow in the network. To reduce the redundant data transmissions several data collection methods have been used [10][11]. One of the most commonly used methods is the efficient aggregation of data in the network.   In this section a brief overview of different types of data aggregations is given. Since in many cases data compression has been considered as a part of aggregation process therefore some compression techniques including the conventional compression techniques are also discussed.

Data Aggregation
Data aggregation process is based on the methods of gathering and compressing data at an aggregation point in a network [12][13]. The basic motivation behind any aggregation process is to suppress the redundancy in the collected data and to reduce the overall transmission cost of the network. In a cluster based network it is the duty of the CHs to aggregate the data before forwarding it to the sink [14].
We have classified the data collection methods into different categories as shown in Fig Packet merging is also a very simple method of accumulating data in one place without reducing the data sizes. The packets coming from different sources are merged into one packet at a collection point or an aggregation point, without processing the value of data [15]. In these cases, the packet sizes at the aggregation points are assumed to be big enough to accommodate large amount of data.
Generally, in literature, there have been two perspectives of looking at the data aggregations in a WSN. One is by the means and methods used to route and forward data through the best path to the sink while collecting data on the way. This method is usually called as In-network Another way of looking at the data aggregation process, that has been discussed in literature widely is the methods of suppressing or compressing the data at an aggregation point (for example CHs in cluster based networks) such as aggregation processes described in [12,14]. The basic idea behind these methods is to suppress the redundancy in the collected data to reduce the overall transmission cost of the network. Since spatial and temporal correlations among the sensed data are one of the basic characteristics of WSN, therefore, this unique feature of WSN has been extensively exploited for the purpose of data aggregation and compression [13].

Summary Based Aggregation
Taking mean, median and max/min values of a set of collected data are the simplest form of data compression.This method is also called Summary based aggregation.These methods take minimum amount of bits that is why many of the WSN protocols have acquired it for their data aggregation processes [19][20]. Since in these methods there is a single data value that represent multiple sensors readings, therefore, these methods can give the highest degree of inaccuracy. These data compressing methods cannot give good data representation.

Traditional Data Compression
There have been many traditional compressing schemes, compared and analyzed for different types of sensor networks, such as Huffman coding, Adaptive coding and Delta coding. Although these algorithms are loss less and provide light compressions for linearly varying data values, but they require a heavy code book for the recovery of data at the receiver. These schemes suit well for limited sensor data types and cannot be applied to a WSN consisting of multiple data types which have nonlinear behavior [13].

Centralized Source Coding Based Data Compression
Centralized source coding based aggregation is a method where the compression is done at the aggregation points rather than on the data originating nodes. In a large scale cluster based WSN with spatially correlated nodes, the centralized source coding algorithm, suits more to the data collection strategies and give efficient data aggregation environments However, there are some cluster based schemes that com-press the data at their originating sources instead of their CHs, for example in [13]. In our view, considering the optimal size of packets in a WSN and the general size of a sensed data, the need of compressing data at the originating source is not much fruitful. In order to reduce the redundant data transmissions in the network, either the SN should transmit their data packets otherwise they should be kept in sleep mode for a considerable amount of time.
Some of the centralized source coding based protocol that has been proposed so far is RDAC [21]. In these methods the data are collected at the CHs where aggregation with compression algorithms is performed.
In RDAC [12] where the clusters are formed repeatedly. Almost the same method of forming correlation matrix has been adapted in [21].
In 3D-DCT (Discrete Cosine Transform) technique has been used which exploits the spatial-temporal correlation for the aggregation of data. In this method the data are transferred into uncorrelated frequency domain coefficients at an aggregation point.A 3D-Zigzag sorting algorithm makes sure that the aggregation point transfers frequency coefficients from lower frequencies that contain the main energy of the original data to higher frequencies [22].
The aggregation techniques proposed in RDAC [12,21] and 3D-DCT [22] do not consider some important factors of clustering like the size of clusters or the amount of data collected from a cluster and the packet sizes available at a CH for data aggregations. The ADiDA (Adaptive Differential Data Aggregation) technique [23] is also a centralized source coding based aggregation scheme.
ADiDA considers the aspects like correlation among data and sensing ranges of nodes, but beside these aspects the main parameters that regulate the adaptive nature of ADiDA are the amount of data collected from a cluster and the size of packet available for the aggregation of these data. Both of these parameters can vary at run time therefore the proposed aggregation method is adaptive to the variations of these two parameters whereas it gives a trade-off for the distortion level in the aggregated data.
Based on the proposed method an optimal payload size has also been estimated. ADiDA is discussed in [23] in detail. Since amount of data in a cluster is dependent on the size of cluster in terms of number of nodes, therefore the efficiency of ADiDA is also dependent on the uniformity in the cluster sizes. In the following section we have discussed various uniform clustering mechanisms.

Load Balancing through Uniform Clustering
Usually highly variable as shown in Fig. 2(b).
The major drawback of unequal clustering is that it can create few large sized clusters in the network which require large sized payloads so that any amount of data can be aggregated into it. In the clusters that are spread on large regions of network, there can be several nodes which needto contact distantly located CHs to forward their data. For example, in clusters 1 and 2 of Fig. 2(b).
Therefore, long distance intra-cluster communication can create high energy consumption in the transmission of data in the network. The SN in these clusters placed at far off distances provide less degree of spatial correlations with each other, therefore spatial correlation based data compressions cannot be used easily.
There are some solutions that are proposed to overcome the variations in cluster sizes of the network. These mechanisms can be divided into three categories: centrally controlled cluster formation protocols, grid-cluster based protocols and radio range based clusteringprotocols.

Centrally Controlled Protocols
In centrally controlled cluster formation protocols the clusters are formed dynamically but the process is

Grid-Cluster Based Protocols
In grid-cluster based protocols unlike centrally controlled cluster formation protocols, the CH selection decisions are distributed among the nodes themselves. Therefore, it can work well in a large scale network. The fundamental idea of the grid-cluster based protocols of WSN is to divide the whole network area into equal sized virtual grids where each grid is considered as a cluster with one CH in each cluster [9,26,27]. The role of CH is rotated among different nodes within a cluster. In [28] we have proposed a mechanism named as DUCA

Radio Range Based Protocols
In this approach CHs through controlled range of radios broadcast their advertisements therefore, the nodes beyond the range cannot listen and cannot join the cluster hence in this manner the sizes of clusters are controlled. To minimize the energy consumption in CH rotation, the CH rotation takes place uniformly after an optimal number of rounds. The optimal numbers of rounds are precalculated based on the total energy consumed per round in the cluster and the total initial energy of that cluster.
In short, the proposed method in VCCBC [29]  In this [28]we have proposed a mechanism named DUCA which can give better results in terms of uniformity of clusters and conservation of energy in the network, discussed in [28]. We have further reduced the sizes of clusters by exploiting the redundant and overlapped sensing ranges, which occur due to the random placements of sensor nodes in the network. Table 1 gives a comparison of the uniform clustering protocols discussed in this section.

Reduction in Cluster Setup Overhead
In large scale WSN, the data collection through dynamic clustering was introduced by Heinzelman [24]. In dynamic clustering, network lifetime is divided into data collection rounds, where in each round the job of CH is assigned rotationally to different nodes and subsequent clusters are formed repeatedly in the network. The idea behind this process is to evenly distribute the consumption of energy among all the nodes. To achieve this, a mechanism called re-clustering is done in each round.
Re-clustering is a process where in every data collection round, different set of nodes become CHs and these CHs form clusters with their neighboring nodes. Since CHs consume more energy than the regular nodes, the job of CH is handed-over to different nodes in different rounds. The most common method to avoid re-clustering is to fix the clusters once they are formed and rotate the job of CH among the nodes within a cluster [29]. The essential requirement of one-time-clustering approach is that the clusters formed should be of uniform sizes so that the load is evenly balanced among all the clusters. One of the approaches used is the centralized control system for example LEACH-F [24]. In [19] we have proposed a dynamic clustering mechanism called RR (Round Rotation) method which can reduce up to 80% of the overhead energy consumed in cluster setups only. Unlike fixed clusters based approaches, the basic feature of RR method that differentiate it from existing solutions is that it can work for dynamic cluster based protocols. Value to judge different paths and choose best optimized path whose energy consumption is lesser than other routing paths. To construct CH based energy efficient routing tree in the network, the important factor that should be fulfilled is the uniform distribution of CHs in the network. Therefore, location based CH selection and cluster formation is necessary for this approach. Many locations based clustering approach has been used in previous works like LEACH-C [24] and GROUP [31]. In [32] before the formation of clusters, DAT (Data Aggregation Tree) is formed using N t nodes out of total N nodes in the network. These N t nodes collect the data from its neighboring nodes, aggregate the data and then route the data through the DAT towards the sink. To make sure that all other non-tree nodes can find at least one N t node in its close proximity, the nodes of DAT are well spread and uniformly distributed in the entire region.

Cluster Head Selection Criteria
In research literature it is generally found that cluster LGCA [35] uses game theory for the clustering of nodes where the nodes compete selfishly with its neighbor nodes which are in their communication ranges and suitable for the CH selection. In EEMA [36], the CH selection criterion has been enhanced by considering the centrality of the node and the proximity to the other neighboring CHs besides the residual energy of the nodes.

Cluster Formation on the Basis of Distance to Sink
The problem, but majority of them rely on unequal and distance to sink based sizing of clusters, for example [6,37].
For example, EECS [38] Therefore, such division of region produces smaller sized clusters closer to the sink compared to those farther away from the sink. Similarly, in E2GBR protocol [41], instead of fan shaped the authors have divided the region into unequal and non-uniform sized grids, small grids close to sink and larger grids farther from the sink. These grids are assumed as clusters. The sizes of grids are adjusted by varying the transmission range of nodes. Table 2 gives an overview of all the energy conservation protocols.
In all of the protocols mentioned above the basic intention is to reduce the transmission load on the nodes closer to sink and balance the overall consumption of energy in a cluster based WSN. For that, all of them have aimed to fairly distribute the consumption of energy between the inter-cluster and intra-cluster communication of data. Table 3 gives a comparison overview of all the types of energy conservation mechanisms that have been discussed in this paper.

CONCLUSION
In this paper initially we have discussed the reasons and factors that generally affect the energy constraints of   s  m  s  i  n  a  h  c  e  M  t  n  e  i  c  i  f  f  E  y  g  r  e  n  E  s  e  l  p  m  a  x  E  l  o  c  o  t  o  r  P   .  1  g  n  i  t  u  o  r  r  e  t  s  u  l  c  a  r  t  n  I  r  o  r  e  t  n  I  t  n  e  i  c  i  f  f  e  y  g  r  e  n  E  ,  ]  6  2  [  P  R  C  G  d  e  v  o  r  p  m  I  ,  ]