Figure 1. Clustering Protocols Classification
Wireless Sensor Networks (WSNs) have tight energy constraints than MANET because of limited energy capacity. So the design principles of conventional networks are not appropriate for WSNs. It needs innovative perspective to arrive at solutions for its own kind of problems. Hierarchical architecture based on clustering suits well due to the uniqueness of WSNs. Even though clustering involves some overhead, its effect on the behavior of WSN is tremendous. In this paper, a new Hybrid Layering with Selective Cluster Rotation (HUCSR) protocol for the energy efficient operation of WSN has been designed. In this protocol, the cluster head rotation is done selectively. Instead of rotating it in all clusters of the WSN scenario, rotation is done based on whether it has been actively participated in data handling. Simulation results show a slight increase in the lifetime of the WSN.
Wireless Sensor Network (WSN) is an ad hoc network of sensor nodes that can monitor the surroundings for an event. These sensor nodes are tiny, and energized by limited power batteries. Due to variety of sensors like temperature, proximity, pressure, light, etc., WSN can support diverse applications. Its data traffic is completely different from MANET and other conventional networks (Sharma, 2015). The data traffic pattern could be like periodic, query based, or event based one. The following are the major constraints of WSN: limited energy capacity, limited hardware resources, randomly and densely deployed, dynamic and unreliable environment, no human intervention, and limited radiating power ( Ferng et al., 2010).
Assume a Telos mote (market name for sensor node is 'mote'). It can also be termed as 'smart dust'. It uses 2 AA batteries for power supply. One AA alkaline battery provides 2160 mWh. So for 2 batteries, total mWh is 2 x 2160 = 4320 mWh. Telos mote's total power needed when it is active is 41 mW (Jan et al., 2017; Muhammad et al., 2017).
From the above example the reason of energy constraints of WSN could be understood (Yue et al., 2013). This is the logic behind the massive (in thousands) deployment of sensor nodes in a target region.
It is the time duration from the start to when the First Node Dies (FND). Due to limited energy and its intrinsic characteristics, this is very short for WSN than MANET. It is heavily influenced by the residual energy available in the sensor nodes across the network (Huang et al., 2012; Maheswari, 2018). To improve the network lifetime, the nodes must handle the traffic load evenly. Mainly, the hot spot nodes must handle balanced load among them (Mahmoud et al., 2017). This impacts heavily the choice of the WSN network architecture, and it would be the clustered architecture. Other non-hierarchical architectures cannot support the various initiatives for the cause of lifetime improvement of WSN.
Sensor nodes are divided into different virtual groups according to set of rules such as cluster size, distance to sink, residual energy etc. A cluster is a hierarchical structure. It contains two types of nodes, namely CH (Cluster Head) and CM (Cluster Member) (Lin et al., 2012). Clustering is the only architecture that enables WSN to perform in-network processing to avoid redundant information getting passed through the network. As we avoided unwanted redundant communications, it means that we are saving some crucial residual energy of sensor nodes. There are two major categories in clustering algorithms for WSN ( Ray & De, 2016). Firstly, static clustering, and secondly the Dynamic clustering. In the first one cluster size is constant throughout the protocol execution.
The roles of the CHs are: Collecting the data from the cluster members, doing data aggregation on the collected data, providing intra-cluster communication schedule, computing path to reach sink node, forwarding the collected data to sink, and managing the cluster members (Thanigaivelu & Murugan, 2012).
The above said classification of clustering is based on the frequency of clustering. In static clustering, clusters are formed once, and CHs not changed, once elected. So CHs are overloaded, and they drain their energy early (Zhang et al., 2014). But clustering overhead is reduced. In dynamic clustering, numbers of clusters are changed periodically, and clusters are reformed. Here, new CHs are elected whenever the clusters are changed to increase the network lifetime, but it increases clustering overhead marginally (Thanigaivelu & Murugan, 2012). Figure 1 shows clustering protocols classification.
Figure 1. Clustering Protocols Classification
It provides better adaptability. Clustering can be done in distributed or centralized manner. In distributed manner, the algorithm does clustering without any contribution from sink node. In centralized manner, sink node has the sole responsibility for clustering. Hybrid clustering uses both centralized and distributed techniques. It is not as the combination of both static and dynamic clustering (Babaie & Pirahesh, 2012; Chen et al., 2015).
It is very important to avoid overloading of cluster heads. New CH can be elected by message exchanges with members or by competition. Cluster heads are elected based on individual or combination of the following factors: distance to sink node, residual energy available, and intra-cluster communication cost (Thanigaivelu & Murugan, 2012). Even though rotation of cluster heads are not free (some clustering overhead involved), it must be done selectively for the improvement of network lifetime of WSN (Kosar et al., 2011).
Some of the clustering protocols are,
LEACH - Equal cluster size; Dynamic; Single hop inter-cluster communication.
UHEED, EADUC, ECDC, UCR - Unequal; Dynamic; Multi hop.
HUCSR - Unequal; Hybrid; Multi hop.
In the past, several clustering protocols have been developed to address the energy constraints of WSN. Here, few of the major protocols are discussed to understand the cognizance of various energy conserving methods adopted. The first one of such protocol is LEACH (Lowenergy Adaptive Clustering Hierarchy) protocol. It uses two phases for its working. In the first one it setup the topology dynamically, and in the second one it transmits data to sink node. This protocol employs single hop communication of cluster heads CHs to the sink. Due to this direct communication with sink node, CHs drain their battery power quickly ( Kumar et al., 2011; Liu et al., 2012). Unequal Cluster-based Routing (UCR) protocol deals with the hotspot problem WSN. It uses uneven cluster sizes in which, faraway clusters have more cluster members than nearby clusters of sink node. In the UHEED (Unequal Hybrid Energy- Efficient Distributed) protocol, the CH selection is based on the distance to the sink node. It shows slight improvement in lifetime of the WSN compared to its preceding protocol HEED (Bhakara et al., 2012; Ever et al., 2012).
Another protocol that uses uneven clustering is EADUC (Energy Aware Distributed Unequal Clustering). It is suitable for both homogeneous and heterogeneous networks. Here, the CHs are selected based on the residual energy available on the nodes. It tries to mitigate the energy hole problem of WSN (Liu et al., 2010). ECDC (Energy and Coverage-aware Distributed Clustering) protocol concentrates on both coverage and energy. It tries to reduce the energy consumption by providing better coverage among all the nodes in the network (Uppu et al., 2008; Yun & Xia, 2010).
The working of HUCSR protocol is shown in Figure 2 as a flowchart. This protocol has the following major phases: cluster head election, cluster formation, forward path computation, data transmission with in-network data aggregation, and finally the selective cluster head rotation.
Figure 2. Flow of the HUCSR Protocol
Algorithm
Begin
if cluster state n[i] = “formed”;
send Head message to neighboring nodes;
store head message received from others;
wait for random time;
if neighbors joined cluster
become “CH”
broadcast cluster advt message;
else
become “CM”
join cluster with nearest “CH”
if n[i] = “CH”
perform neighbor CH discovery;
compute the forwarding path;
else n[i] = “CM”
perform neighbor discovery;
compute nexthop table;
if handled data transmission
set data transmissiont bit;
compute residual energy;
stop;
else already n[i] = “CH”
if data transmission bit set?
compute the residual energy and distance to sink;
Initiate cluster rotation;
exchange forward path table with new CH;
compute the residual energy;
stop;
End
The simulation has been done using 100 nodes with both uniform and random distribution of nodes in the sensor region. Life time is given as number of setup rounds at which the last node of the sensor network dies. For random distribution of sensor nodes, the curves are little bit steep indicating that the chunks of nodes are dying simultaneously. The graph shows that the performance of the HUCSR protocol is better than other protocols for both distribution of the sensor nodes. The outcome of the simulation has been shown in Figures 3 and 4.
Figure 3. Lifetime Improvement of Uniform Distribution
Figure 4. Lifetime Improvement of Random Distribution
The HUCSR protocol proposed in this paper performs well for smaller network size. To support thousands of sensor nodes, various scalability issues must be solved. The overlapping of clusters necessitates more investigation. That is, a single CM can be a part of multiple clusters. The cluster rotation process is being hampered by these nodes. As a result, additional attention is required in this area in order to optimise cluster formation. The HUCSR protocol is predicted to produce even better outcomes with thousands of nodes in the simulation scenario once cluster optimization is more successful. As cluster head rotation is not performed on all clusters, the clustering overhead of this protocol is decreased, helping it perform better than its predecessor.