1. Introduction
Wireless sensor networks (WSNs) consist of sensor nodes with sensing and communication capabilities that have gained enormous attention for their usage in many applications, such as internet of things (IoT) [
1,
2]. Most wireless sensors are powered with batteries as limited energy sources. In addition, non-rechargeable energy sources strongly require minimizing energy consumption of the nodes and then maximizing network lifetime. Therefore, a lot of research has focused on energy conservation of sensor nodes in long-run operations of WSNs.
As a somewhat effective technique, clustering is an efficient way to save energy consumption of the sensor nodes. In the clustering process, sensor nodes are grouped into clusters. A node is selected as a cluster head, a leader in a cluster, and the remaining nodes within the cluster are considered cluster members. Sensor nodes sense the physical parameters related to their environment and send the information to their corresponding cluster heads. Cluster heads then aggregate the data and send it to a remote base station (BS) or sink using single-hop or multi-hop that depends on the distance of the BS.
Several cluster-based routing techniques can be found in [
3,
4] where periodic cluster head election and re-clustering is performed. In most of these techniques, cluster heads are elected either independently or based on competition of different weight functions. Similar cluster formations are followed by a joining message, either to a nearby cluster head or to a cluster head with a higher residual energy [
5,
6,
7,
8,
9,
10,
11,
12,
13,
14,
15,
16,
17,
18,
19,
20,
21,
22,
23,
24,
25,
26,
27]. This can balance the energy consumption of the nodes but does not ensure the maximization of the steady state (the time the first node dies in the network) and the network lifetime as well. Because only the residual energy of a node is one of the key criteria to cluster head selection, not considering the average distance from member nodes does not guarantee that the energy consumption of the nodes is saved during intra-cluster communication. Moreover, periodic cluster-head election and re-clustering processes of the protocols require the broadcasting of control messages in each round, which is the energy waste of the power-limited sensor nodes. One solution is to piggyback the weight value of the nodes with the local data sent to a cluster head to hand over the role [
26,
27]. Thus, the number of control messages is minimized regarding the cluster-head elections throughout the network lifetime. But, consideration of energy saving in intra-cluster communication of the protocols does not exist yet.
In this paper, we propose an energy-centric cluster-based routing protocol called ECCR (energy-centric cluster-based routing) for WSNs that addresses the aforementioned issues. Firstly, we propose predefined static clusters that reduce the control message overhead of the cluster formation issue. Secondly, we introduce a caretaker for cluster-head election technique inspired by L. Malathi et al. [
26], where the ranks’ information of the nodes are piggybacked along with the local data. A former cluster head from the previous round is responsible to hand over the role to a prospective cluster head in the current round. This reduces the control message overhead regarding the cluster head elections throughout the network lifetime. For data gathering and forwarding, the cluster heads are elected and selected based on the higher rank among the nodes. The rank of a node is defined by the factors which have a major influence on the energy consumption of the node. Experiments are performed on the proposed ECCR to compare with existing protocols such as EADUC, HUCL and IEADUC in [
19,
26,
27], respectively.
The rest of this paper is organized as follows. We review related works in
Section 2.
Section 3 presents the system model. The proposed protocol is described in
Section 4. The results and comparison with existing protocols are given in
Section 5, followed by the conclusion in
Section 6.
2. Literature Review
Many routing protocols have been proposed for WSNs in the last decade [
3]. It has been proven that distributed clustering with data-fusion technique is more energy-efficient when compared to other routing protocols [
4]. Previous research works have been undertaken in the area of cluster-based routing algorithms. These clustering algorithms are commonly based on rotating the role of cluster heads and re-clustering in every round to prolong the network lifetime. The low energy adaptive clustering hierarchy (LEACH) protocol [
5] is a pioneer work available in this category. The cluster head election process of LEACH is periodic and for every round, new cluster heads are elected. Although the LEACH protocol distributes the energy consumption among the nodes equally, it leads to additional routing overhead, resulting in excessive use of limited energy of a cluster head. Besides, the protocol assumes single-hop communication between the nodes and BS. It constructs the network by making it less scalable and is unsuitable for a large scale WSN.
LEACH gives birth to many protocols that have been developed in the last few years that improved over the network lifetime. The stable election protocol (SEP) has been proposed in reference [
6] for heterogeneous WSN. It suggests that a percentage of nodes are equipped with more energy called advanced nodes compared to rest of the nodes. The advanced nodes create heterogeneity in terms of node’s initial energy. The cluster head election probabilities are weighted by the initial energy of a node relative to that of other nodes. This prolongs the steady state and the network lifetime as well. A new evolutionary-based routing protocol (ERP) has been proposed to extend the network lifetime [
7]. It adopts a fitness function which incorporates two clustering aspects: intra-cluster distance in cluster formation and inter-cluster distance in clusters separation. ERP achieves better performance over SEP. S. Kumar et al. proposed an enhanced threshold-sensitive SEP (ETSSEP) [
8]. It is based on dynamically changing the cluster-head election probability. ETSSEP selects cluster heads on the basis of residual energy level of nodes and minimum number of clusters per round. The election process also prolongs the network lifetime, which is longer than SEP. Besides, a multi-hop-based LEACH (M-LEACH) [
9] and a distance-threshold-based cluster-head election on LEACH (LEACH-DT) [
10] have been proposed. These two protocols achieved an enhanced network lifetime over LEACH.
A hybrid energy-efficient distributed clustering (HEED) has been proposed by O. Younis and S. Fahmy [
11]. It selects cluster heads according to the residual energy of nodes and intra-cluster distance as the primary and secondary criteria. Although it is successful in prolonging the network lifetime, it introduces an extra communication overhead to compute the communication cost with its neighbor nodes by exchanging a large number of control messages. Meanwhile, HEED may not be effective in load balancing as the nodes nearby to the BS die quickly. Many other clustering algorithms have also been proposed in the literature [
12,
13,
14,
15] which are similar to HEED and introduce high control messages overhead during cluster head selection and cluster formation. A distributed energy-efficient clustering (DEEC) protocol has been proposed by L. Qing et al. [
16]. The authors proposed a cluster-head election method based on the ratio of residual energy of a node and the average energy of the network. Although, in all of these clustering and routing methods, a periodic rotation of cluster-head election based on different weight functions introduce the balanced energy consumption of nodes, none of these algorithms can take into account the “
energy hole” problem in many-to-BS data-gathering.
Several methods have been proposed in literature to resolve this “
energy hole” problem and thereby maximize the network lifetime. S. Soro and W. B. Heinzelman proposed an unequal clustering algorithm which was the first initiative regarding this issue [
17]. It adopts unequal cirques called clusters where the clusters in the different cirques have different sizes. Some high-energy nodes can be deployed to take on and balance the energy consumption of these cluster heads. This can effectively prolong the network lifetime, but the position of cluster heads must be known previously. A further solution step has been taken through an energy-efficient unequal clustering (EEUC) mechanism [
18]. It adopts a cluster-head election algorithm, where cluster heads are elected on the basis of residual energy of the nodes. A node is elected as a tentative cluster head with a probability
T (a waiting time to broadcast an announcement message as a cluster head). Cluster heads use uneven competition ranges to form clusters of uneven sizes. The clusters nearby the BS have smaller sizes than the faraway clusters from the BS. The closer cluster heads consume less energy and preserve an amount of energy for the inter-cluster communication resulting in the balanced energy consumption of the cluster heads. Due to the quality of generated cluster heads being biased by the
T, some nodes can be isolated in some cases of
T.
An energy-aware distributed unequal clustering (EADUC) algorithm has been proposed by J. Yu et al. that addresses the “
node isolation” problem [
19]. Here, a cluster head is elected based on the ratio of residual energy of a node and the average residual energy of its neighbor nodes. If a node does not receive any head message from the tentative and neighbor cluster head(s), the node elects itself as a cluster head and broadcasts the head message. The nodes will join with the nearby cluster head to form a cluster. A relay node is selected during inter-cluster communication on the basis of minimum distance towards the BS. This achieves an enhanced network lifetime by contrast with the previous protocols, but a relay node can be selected repeatedly due to the bias of the relay function. As a result, the node dies quickly.
To resolve the issue of losing the nodes, an energy-aware routing algorithm (EADC) has been proposed in [
20]. The proposed cluster-head election is similar to EADUC, but the relay function differs; instead, it uses a node’s residual energy and the number of member nodes as the key criteria to be a relay node. However, in the clusters’ formation of these algorithms [
19,
20], each node chooses the nearby cluster head in order to join without considering the residual energy of the cluster head. One of common problems in these techniques does not assure the relaying load of the cluster heads, which are balanced with respect to their residual energy. In other words, all the cluster heads are not participating in relaying the data for other cluster heads, so it results in an imbalanced energy consumption of the cluster heads, thus limiting the network lifetime. An energy-efficient multi-level and distance-aware clustering (EEMDC) has been proposed by A. Mehmood et al. [
21] in the same regard. It divides the network into three logical levels that are based on the hop-count according to the BS. The levels are namely first-level clusters, second-level clusters, and third-level clusters that include nodes having a hop-count value of 1 to 2, 3 to 5, and 6 to more, respectively. Initially, the BS broadcasts a cluster-initiative packet of hop-count value over the network. A node receives the message and determines which cluster-level it belongs to. The nodes in different levels are determined by the distance of the BS. The number of nodes in a cluster and the number of clusters in different levels are varied due to the position of the nodes with respect to the BS. The proposed cluster head election is similar to LEACH’s. Unlike the first round as LEACH, the competition of a cluster head election is based on the residual energy and hop-count value of nodes from the next round. If there are multiple nodes with the same residual energy in a cluster, a node with the lower hope-count value is selected as a cluster head. The data-routing of EEMDC maintains a hierarchy from cluster heads to the BS according to the levels. This policy ensures a minimum distance route towards the BS. Meanwhile, the energy consumption of nodes is balanced and distributed over the network; thus, the network lifetime is enhanced over the protocols.
An energy and coverage-aware distributed clustering (ECDC) has been proposed in [
22]. ECDC integrates the network coverage issue with the energy saving of nodes. The point and area coverages are studied along with the distributed routing. In both cases, the nodes are elected as cluster heads based on the residual energy and coverage importance metrics, respectively. The information of the metrics is shared among the neighbors within a range. The cluster head election process is similar to LEACH’s, where the minimum coverage importance and higher residual-energy-obtained nodes are elected as cluster heads. In ECDC, a data-routing algorithm is adopted where a relay node is selected based on the relay metric of the cluster heads towards the BS. The protocol achieves a lower energy consumption of nodes and better in-coverage performance compared to other protocols.
An energy-aware routing algorithm (ERA) has been proposed by T. Amgoth and O. K. Jana [
23] that addresses the problem of EADUC and EADC. In ERA, nodes consider a cluster head with higher residual energy to join in order to form a cluster. Unlike the relay functions of the route selection policies of the protocols, a directed virtual backbone (DVB) of the cluster heads has been adopted. The sink initiates broadcasting a route request message to the cluster heads. Cluster heads receive the message and increment its level to one higher than the sink (i.e., the level of the sink is assumed to be at zero,
L(sink) = 0, so that
L(
u) =
L(sink) + 1). Recursively, all the cluster heads broadcast the message to complete the process of forming a DVB. A cluster head selects the relay nodes based on the ratios of the average residual energy of the cluster heads in different levels and distributes all the aggregated data packets towards the sink sequentially. A decentralized energy-efficient hierarchical cluster-based routing algorithm (DHCRA) has been proposed by M. Sabet and H. R. Naji [
24]. The key approach of the protocol scheme is that cluster heads are selected at the tree edges based on effective local information. The cluster heads selection is based on nodes’ residual energy and distance from the BS along with the constructed routing tree. This reduces a number of control messages and consequently saves energy consumption of the nodes. Hence, the network lifetime enhances, but the steady state of the network is not convincing.
A distributed and adaptive routing protocol in cluster-based (DARC) routing has been proposed in [
25]. The key approach of DARC is to adaptively adjust the routing mode of cluster heads to balance the energy consumption during inter-cluster communication. Even though the periodic clustering can distribute the energy consumption of cluster heads among all nodes, the imbalance in energy consumption exists due to the random location of nodes with respect to the distance between the BS. To address this problem, a relay cluster head is selected among the cluster heads on the basis of relay mode (i.e., CH-L and CH-H denote a cluster head with low and high energy) that is defined by the residual energy of the cluster heads. A CH-H selects the BS as its next hop and acts as a relay node for a CH-L. This process balances the energy consumption among the cluster heads by distributing the relay tasks to the CHs-H, which results in an improved network lifetime.
Unlike the typical cluster head election processes of the protocols, hybrid unequal clustering (HUCL) protocol suggests cluster head-hand-over and piggybacking techniques [
26]. In the head-hand-over technique, a former cluster head delivers the role of cluster head to an elected node. In the piggybacking technique, the value of weight function is sent along with the local data to an associated cluster head. This reduces the control message overhead during cluster-setup phases. Hence, the steady state and the network lifetime are prolonged compared to other protocols. However, the number of neighbor nodes is not considered in defining competition radius along with the distance from the BS, although it uses the number of neighbor nodes while computing a waiting time to become a cluster head. To resolve this problem, an improvement of EADUC called IEADUC has been proposed in [
27]. It combines the cluster head-hand-over technique of HUCL with the same cluster-head election technique of EADUC. Meanwhile, it modifies the relay function of a relay-node selection towards the BS. Unlike the EADUC, a relay node is selected on the basis of relay values defined by three factors; the residual energy of a cluster head, number of member nodes associated with the node, and the costs of data aggregation and transmission to a next hop. This achieves an enhanced network lifetime by contrast with the EADUC and HUCL. IEADUC can prolong the steady state and the network lifetime over the protocols.
In all the algorithms, the cluster-head elections are based on different weight functions, where the factors are mostly residual energy of a node, number of neighbor nodes, and distance from the BS. But, in these weight functions, the intra-cluster communication distance has also not taken into consideration, whereas it has a significant impact on the overall energy consumption of the nodes. The cluster-heads election are biased by the time
T and competition range of broadcasting a head message. Therefore, a cluster might have a less residual-energy-obtaining node as a cluster head compared to some of the member nodes, i.e., in [
19,
20,
21,
22,
24,
25,
26,
27]. On average, a high number of control messages is required to complete a single data transmission to the BS. Although some methods of the protocols are very effective to balance the energy consumption among the nodes, the control messages overhead use an excessive amount of energy of the nodes, which limits the network lifetime as a whole.
Table 1 summarizes the comparison between the protocols on the basis of clustering, scalability, control messages overhead in clustering, and energy efficiency that results in steady state and network lifetime.