Review Article

Clustering Analysis in Wireless Sensor Networks: The Ambit of

Performance Metrics and Schemes Taxonomy

Asim Zeb,1 A. K. M. Muzahidul Islam,1 Mahdi Zareei,1 Ishtiak Al Mamoon,2

Nafees Mansoor,3 Sabariah Baharun,1 Yoshiaki Katayama,4 and Shozo Komaki1
1
Malaysia-Japan International Institute of Technology (MJIIT), Universiti Teknologi Malaysia (UTM), Jalan Semarak,
54100 Kuala Lumpur, Malaysia
2
ECE Department, Presidency University, Dhaka 1209, Bangladesh
3
Department of Computer Science and Engineering, University of Liberal Arts Bangladesh (ULAB), Dhaka 1209, Bangladesh
4
Nagoya Institute of Technology, Gokiso-cho, Showa-ku, Nagoya, Aichi 466-8555, Japan

Correspondence should be addressed to A. K. M. Muzahidul Islam; [email protected]

Received 20 January 2016; Revised 8 May 2016; Accepted 10 May 2016

Academic Editor: Xiangjie Kong

Copyright © 2016 Asim Zeb et al. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Research on wireless sensor network (WSN) has increased tremendously throughout the years. In WSN, sensor nodes are
deployed to operate autonomously in remote environments. Depending on the network orientation, WSN can be of two types: flat
network and hierarchical or cluster-based network. Various advantages of cluster-based WSN are energy efficiency, better network
communication, efficient topology management, minimized delay, and so forth. Consequently, clustering has become a key research
area in WSN. Different approaches for WSN, using cluster concepts, have been proposed. The objective of this paper is to review
and analyze the latest prominent cluster-based WSN algorithms using various measurement parameters. In this paper, unique
performance metrics are designed which efficiently evaluate prominent clustering schemes. Moreover, we also develop taxonomy
for the classification of the clustering schemes. Based on performance metrics, quantitative and qualitative analyses are performed to
compare the advantages and disadvantages of the algorithms. Finally, we also put forward open research issues in the development
of low cost, scalable, robust clustering schemes.

1. Introduction homes, manufacturing environments, body area networks,

and underwater sensor networks [3–8].
Wireless sensor networks (WSNs) are highly resource con- In traditional flat WSN, node’s status and functionality
strained with limited power, bandwidth, processing capa- are identical, and node acts as a data generator and router.
bilities, storage, and computational capabilities. Therefore, Flat network is not efficient in energy conserving as com-
sensor nodes are mostly inoperable and irreplaceable when pared to cluster-based WSN wherein network size is larger.
failure occurs due to energy depletion. Increasing net- Transmission happens in traditional flat network in the form
work sustainability and lifetime are the key issues for the of flooding [9]. In flooding, message is sent from the source
contemporary studies in sensor domain. Normally, energy node to destination where the entire network is used for
depletion is highly dominated by radio transmission. The a single operation. However, such technique causes data
energy depletion of radio communication is directly related redundancy. To overcome the problem of data redundancy in
to any transmission in the network. Clustering technique flat network, directed diffusion technique has been suggested
reduces the number of radio transmissions and increases in [10, 11]. Direct Diffusion continuously monitors redundant
sensor network lifetime [1, 2]. Thus, clustering technique can, data. However, this technique is not efficient where data
efficiently, increase lifetime of various sensor applications, input stream is large like environment monitoring. Another
such as robot control, environmental control, offices, smart technique has been proposed to avoid redundancy in flat
2 International Journal of Distributed Sensor Networks

network is rumor. However, this technique also gives poor significance of clustering in WSN and the accessibility of
results when the number of events gets larger [12]. Flat thorough works in this field, a holistic overview is vital at
sensor network scheme is mainly proactive which yields this stage. The objective of this paper is to critically review
unsatisfactory performance for highly dense network [13–15]. and analyze the prominent state of the art of cluster-based
Large sensor network generates more data that highly affects WSN algorithms. First, various cluster-based schemes are
flat network performance. Communication overhead of flat distributed into goal specific categories. Then, the advantages
network routing protocols is 𝑂(𝑛2 ), where 𝑛 represents the and disadvantages of the cluster-based schemes are high-
total number of nodes in the network [3]. This result implies lighted. Moreover, comparative study of various approaches
that such algorithm increases routing overheads. Therefore, through qualitative and quantitative analysis is presented,
flat network performs efficiently in a situation where network which assists in developing the efficient cluster-based proto-
size is small. cols. Finally, open research issues are discussed which can be
To achieve the small sensor network features in a large the future line of directions.
sensor network, various solutions have been proposed to The rest of the paper is presented as follows. Section 2
break a sensor network into smaller groups. Clustering is provides the existing review of articles on cluster-based WSN.
the one that demonstrates scalable results. The basic idea Overview of cluster-based WSN is discussed in Section 3.
behind clustering is to group down the network into small Section 4 provides an overview of clustering schemes, where
networks. Clustering provides logical organization of small various cluster performance metrics are presented. Promi-
units and hence is easy to manage. The structured networks nent cluster-based WSN protocols are discussed and analyzed
have many advantages as compared to flat network, such (qualitatively and quantitatively) in Section 5. Section 6
as data aggregation, reducing communication overhead, discusses the open research areas in cluster-based WSN.
ease of managing, minimizing overall power consumption, Conclusions of this paper are drawn in Section 7.
energy efficiency, and prolonging sensor network life time.
Moreover, clustering results in efficient dynamic routing from
sensor to sensor or to specific nodes (sink nodes) [5, 16, 17]. 2. Earlier Review of Articles in
Few challenges of WSN are as follows: Cluster-Based WSNs
(i) Overall network performance is affected by the lim-
Cluster-based WSNs are a highly dynamic research area. The
ited power in sensor nodes.
published articles in this domain are extremely diverse in
(ii) New deploy nodes execute cluster formation scheme terms of their approaches and implementations. However,
while new redeployed nodes execute cluster main- there exist a few published survey papers, such as [18, 19],
tenance. Such schemes are very critical for large which provide a diverse comparison approach. Article [18]
network. has become obsolete and does not show the latest dominant
(iii) In multihop networks, every node acts as a data study in the area, while article [19] is the most recent review
originator and data router. Therefore, node(s) mal- work among the survey articles that have been considered
function(s) causes topological manipulation which in this paper. A brief survey on cluster-based algorithms is
requires network reorganization and packet rerout- presented in [19], where different taxonomy for measuring
ing. cluster-based architectures is presented with their advan-
(iv) Sensor nodes have low transmission bandwidth and tages and disadvantages. However, article [19] explores the
low processing capabilities. limited performance metrics. It overlooks some imperative
performance metrics in exploring clustered protocols, such
(v) Data transmission difficulties are unpredictable in as communication cost. Moreover, quantitative analysis is not
WSN; thus, a proper fault tolerance or reliability presented in [19].
schemes are essential. Article [20] presents an extensive review on clustering
(vi) Data collection is performed by cluster head(s). This protocols for WSN. This article mentions taxonomy of clus-
may result in redundant data which misuses sensor ter structures and then shortens several cluster procedures
power. on convergence time based protocols. It compares these
(vii) Hidden terminal problems may cause collision which clustering methods based on cluster overlapping, location
results in energy wastage. awareness, cluster stability, and convergence rate.
A review paper presented in [21] discusses clustering
(viii) Network performance degrades when the maximum
protocols in WSN, which are categorized based on cluster
extent of nodes in a network is undefined.
formation and cluster-head (CH) selection. The authors
This paper addresses a few prominent issues in cluster perfor- discuss various significant design issues, and present various
mance metrics, which are absent in the recent review articles, performance problems linked with the clustering algorithms.
such as domino effects, computational round, communi- Another study, conducted in [22] on CH election policies,
cation complexity, cluster maintenance, ID-based heuristic, presents various types of taxonomy, such as deterministic,
degree-based heuristic, collaborative cover heuristic, and adaptive and joint metric schemes. The CH election cost is
weight based heuristic clustering. Moreover, this work is the compared with cluster development and CHs distribution.
comprehensive review of every aspect regarding the cluster- Moreover, necessity of scalable, energy efficient and stable
based algorithms and schemes for WSN. Because of the clustering schemes is placed forward for WSN.
International Journal of Distributed Sensor Networks 3

Here, in [23], the authors state the overall cluster forma-

tion techniques in WSN. They review a simple organization
of the only three considerations during cluster formation,
namely, centralized or distributed CH formation, single-hop
or multihop intra- and interclusters communication. They
also highlight some issues in WSN and introduce few routing
methods.
Review paper in [24] discusses three noticeable benefits of
clustering approaches, such as high scalability, less outflows,
and easy preservation. The authors present an organization
of WSN cluster structures and eight cluster characteristics.
They have studied some energy efficient cluster-based WSN
protocols, namely, HCC, LEACH, PEGASIS, HEED, TEEN,
APTEEN, ACE, EECS, EEUC, PEACH, FLOC, LID, DCA,
3HBA, and CDS. In [24], each algorithm is measured based
on clustering head election characteristic and clustering Cluster head
characteristics. A survey on cluster-based WSN protocols Gateway
discusses the important issues of five famous cluster-based Member node
WSN algorithms, namely, TL-LEACH, EECS, EEUC, HEED,
Figure 1: Cluster-based network.
and LEACH in [25]. The authors compare all these cluster-
based protocols based on various metrics, such as uniformity
of CH distribution, residual energy, distance of hop, cluster
size, and delay and cluster formation techniques. Another flat sensor network and to make the network operations
study is conducted in [26] which compares various cluster- more efficient. In clustering, the network is organized into
based algorithms. They present some basic ideas associated logical groups which depend on network characteristics and
with the clustering procedure. They examine LEACH-based applications’ requirements. Cluster-based WSN has various
protocols with active and reactive protocols in WSN and advantages as compared to flat WSN, such as energy effi-
compare the major performance metrics of these algorithms. ciency and prolonging network lifetime [17]. Cluster-based
Design issues and comparative study of cluster-based WSNs are defined as a hierarchal organization of sensor
WSN algorithms to improve the network lifetime are studied network. Numerous researches emphasize the operative and
in [27]. Authors in [20] highlight various challenging ele- proficient clustering structures for WSN. Typically, in a
ments that affect the design of clustering protocols. Moreover, cluster-based architecture, network is divided into virtual
several effective classical cluster-based schemes in WSNs with groups as shown in Figure 1.
comparative study are conversed in the article. Figure 1 describes cluster-based network where the net-
It is observed that some measurements are missing in work is logically divided into clusters represented by dotted
the abovementioned recent survey articles, namely, domino lines. There may exist three types of nodes in the network,
effects, computation round, communication complexity, namely, cluster head (CH), gateway (GW), and member node
cluster maintenance, and cluster-head election which are (MN). In a cluster, cluster-head is the local coordinator that
based on ID-based heuristic, degree-based heuristic, collab- aggregates and forwards data to base station. Meanwhile,
orative cover heuristic, or weight based heuristic. Without member nodes (MNs) are the leaf nodes that send data
considering these measurement parameters, exploring and to cluster-head. Nodes, which lay between two clusters, are
comparing WSN clustering schemes may not provide efficient known as gateway nodes, where the gateways connect two
outcomes. These performance metrics also have significant or more cluster heads. The advantage of the gateway node is
impact on sensor’s energy depletion. Thus, this paper com- to form a predefined multihop intercluster communication
pares various cluster-based algorithms, and in each category route, known as backbone of the network. Each CH retains
quantitative and qualitative analyses are done to determine neighboring GWs’ information in its routing table where
the efficiency of the cluster-based algorithms. In order to the routing table helps the CHs to make routing decision
attain holistic knowledge of this domain and development of promptly. Backbone makes the data communication more
the algorithms, issues presented in this paper are vital in the efficient. The various features of cluster-based WSN are
development of cluster-based WSNs algorithm. summarized as follows.

3. Overview of Cluster-Based WSN 3.1.1. Features of Cluster-Based WSN

3.1. What Is Cluster-Based Network? The performance of Data Fusion. During data fusion, CHs gather data from differ-
the flat network may degrade once the size of the network ent nodes and send them to the base station (BS). Data fusion
increases. This is because of the fact that increasing the also eliminates redundant data on CH level, which eliminates
network size and control overhead in wireless sensor network extra burden on sensor nodes during communication. Hence,
also leads to the relevant increases. Clustering is one of data fusion enhances the overall network lifetime [28] and
the widely investigated solutions to scale down the large reserves the entire network energy [29].
4 International Journal of Distributed Sensor Networks

Data Load Management. Cluster provides efficient data load in harsh atmosphere, cluster-based protocols are suitable.
management and uniform network lifetime. CHs, closer to Therefore, WSNs must have the ability to reconfigure them-
the BS, experience extra data loads in order to relay data selves without human intervention, particularly in harsh
from upper layers. To overcome this problem, CHs, which environments and inaccessible locations. In order to secure
are closer to the BS, keep lesser member nodes to diminish aggregated data, fault tolerance technique needs to be consid-
load. Thus, the entire nodes have equal energy depletion, and ered during protocol design stage. Cluster maintenance and
network lifetime becomes uniform [30]. CH backup are more feasible techniques to secure the entire
network reconstruction when CH is malfunctioning [34, 35].
Efficient Energy Saving. In flat networks, data is transmitted
through flooding, whereas in cluster-based network data is Data Communication Assurance. CH sends the aggregated
aggregated on CH level and sends it to the BS via multihop data to the base station through single-hop or multihop
routing. Multihop routing in cluster-based network helps in routing. In mobile network, the probability of data loss
decreasing the number of transmission paths, which saves takes keen interest in recent researches because of its high
energy exponentially [31]. chance of occurrence. To handle such problems in mobile
node, the node sends a joint request to its CH before
the actual data communication takes place. If the sender
Relay Node. Network is partitioned or disconnected when receives the acknowledgement message, then it initiates data
nodes fail to communicate. The relay node is used to reestab- transmission; otherwise the sender node considers that it is
lish the path and join the partitions. Relay node can be static no more a part of the network and that it needs to rejoin
or mobile. However, initial task of a static node is to find the network. When the node is rejoined, the network then
out the disjoint portion and then to deploy relay node there. initiates data sending to parent node. Thus, connectivity
However, mobile relay node is a special type of node which assurance between member nodes and their CH is a crucial
places itself in disjoint portion [32]. task for successful data delivery [36].
Robustness. Once the cluster-based WSN is formed, the Deadlock Prevention. In multihop communication, data is
second important step is cluster maintenance. Cluster main- transmitted to the base station using intermediate nodes.
tenance is useful to maintain the network integrity. It handles In this criterion, different nodes relay the data to the base
different scenarios, such as changing network size, move- station. Thus, the node closer to the sink node is overbur-
ments in nodes, and unexpected operational flaw. Cluster- dened with more information as compared to far nodes.
ing algorithms merely require managing these variations Therefore, nodes, closer to the BS, deplete energy quicker, and
within each cluster. Therefore, cluster maintenance makes the deadlock occurs near the BS. This may cause the partition of
network highly robust and more convenient in topological the entire network into groups. Consequently, the far nodes
manipulations. may not be able to approach BS because of the limited range.
Meanwhile, other nodes still have energy. To handle such
Collision Avoidance. In sensor network, when single channel
problems, load balanced clusters were investigated, where a
is considered it is shared among sensor nodes. Thus, the
cluster, nearer to base station, retains smaller number of MNs
performance of the network decreases when many nodes
than a cluster far from the base station. Therefore, nearer CH
send data concurrently which causes collision. This can be
maintains enough energy for intercluster communication.
efficiently solved in cluster-based WSN, where CH assigns
Consequently, deadlock prevention can be efficiently handled
unique time slot to every member node via scheduling [33].
by using unequal size cluster [37, 38].
Latency Reduction. Latency refers to the total time that a
message requires to travel from source to destination node. Network Lifetime. Increasing of network lifetime is an impor-
Cluster-based WSN enhances the delivery performance of tant consideration as nodes retain limited power, bandwidth,
the packet by maintaining routing table at the CH level and processing capabilities. Typically, it is a highly crucial
to make efficient routing decision. Moreover, cluster-based task to optimize a few problems in WSNs, such as intracluster
networks grounded on connected dominating set (CDS) communication cost, redundant data gathering, and uniform
form a predefined communication pathway called backbone cluster loads. Such factors are taken into consideration during
tree, which enables quick and efficient multihop routing [30]. CH election, which extends the lifetime of the network.
Moreover, higher energy route is prioritized for data trans-
Secure Data Communication. As data aggregation is per- mission, where such criterion adopts uniformed energy
formed by CH, malicious nodes may attack to alter or hack depletion in the network and enhances network lifetime [39,
data. In cluster-based WSN, strong authentication schemes 40].
are developed to avoid malicious nodes joining the network.
These schemes improve data integrity and confidentiality Efficient Quality of Services. The functionalities and network
[28]. applications of WSN prompt the prerequisite of quality of
service (QoS). Typically, effective QoS parameters are end-
Fault Tolerance. Sensor nodes may be affected by hardware to-end delay, reliability, throughput, jitter, and bandwidth. It
failure, delay, interference, energy exhaustion, and so forth. is difficult to satisfy all the necessities of QoS parameters in
Due to such constraints, where the nodes are not replaceable cluster-based protocols. Trade-off is required to consider one
International Journal of Distributed Sensor Networks 5

or more QoS parameters based on application requirements. weight based heuristic. In ID-based heuristic, node
The state-of-the-art cluster-based protocols emphasize the ID is taken into consideration for CH election, like
energy efficiency, rather than QoS. QoS issues are considered a smallest ID node becomes CH. In degree-based
for real time application domain, such as healthcare applica- heuristic, quantity of neighbors is considered for CH
tion, battlefield applications, and event observing [41, 42]. election, while collaborative cover based heuristic
considers average hop distance between two commu-
nicating nodes. It is indicated in [46] that the degree-
4. Performance Metrics of Cluster-Based WSN
based heuristic is better than ID-based heuristic in
In this section, a set of performance metrics are enumerated recognizing smaller size CDSs. However, collabora-
which can be used to categorize and differentiate cluster- tive cover heuristic [45] is better than degree-based
based WSN algorithms. One of the benefits of clustering is heuristic in recognizing smaller size CDSs. Moreover,
to make network scalable in situation when sensor nodes' in weight based clustering, various parameters are
number is huge. Nevertheless, there are downsides of using a considered to elect CH, such as remaining energy,
cluster-based network, such as higher cost overhead during communication cost, and distance. In weight based
network construction as compared to flat sensor network. criterion, a node is elected as a CH based on energy
Cost of clustering is an important parameter to authenticate cost.
the effectiveness of the scheme. Moreover, it also refers to
the improvement of network structure in terms of network Cluster Complexity. Cluster complexity defines the trans-
scalability. Cost of the clustering schemes in this paper is mission complexity of the network. There are two types of
evaluated qualitatively and quantitatively. The effectiveness of cluster complexity, namely, computational complexity and
each algorithm as well as their shortcomings are determined. communication complexity.
In this part, various performance metrics of cluster-based
(i) Computational Round/Time Complexity. Computa-
WSNs are discussed. Based on these parameters, the cost of
tional round specifies the total number of rounds
clustering is evaluated more efficiently. Figure 2 describes
in which cluster formation is accomplished. Com-
various performance metrics of cluster-based WSN and each
putational round is a significant metric in cluster
performance metric is discussed afterwards.
formation for static and mobile sensor network. It
Cluster Formation. Cluster formation is the setup phase of indicates an unbound time complexity in mobile
building cluster-based architecture from flat sensor network. sensor nodes. Hence, the more round results more
Cluster formation is divided into two categories, namely, data communication which decreases the efficiency of
network model and cluster-head election. clustering algorithms [47].
(ii) Communication Complexity/Message Complexity.
(i) Network Model. Network model represents the char- Message complexity is categorized into three
acteristics of a network. Two basic components of types that are data aggregation, broadcasting, and
network model are described below. multicasting. Converge-casting is an example of
data aggregation that is performed at CH level and
(a) Node Type. A node can be of two types, either initiated from bottom to top manner towards the
mobile node or stationary node. In the former base station (BS). In broadcasting, messages are
way, CHs, MNs, or GWs or all three can be disseminated from top (base station) and go down
mobile. Therefore, mobile node (CH or MN) in the entire network [48], while in multicasting,
changes its position dynamically in terms of messages are disseminated from one node to set of
other nodes. A challenging problem in such nodes. Moreover, communication complexity is also
scenario is to retain cluster for long time and to dependent on the number of edges.
overcome problems associated with packet loss.
(iii) Control Message. During network formation and
On the other hand, in stationary nodes, CHs,
maintenance, nodes exchange control information,
MNs, and GWs are the static nodes that do not
which is unlike data message. Control message is
change their positions in terms of other nodes
directly proportional to energy depletion of a node.
[20].
Moreover, the control information results in more
(b) Network Type. In WSN, cluster formation is energy depletion and vice versa. All the studied
either distributed or centralized. In centralized algorithms in this paper are evaluated via three scales:
technique, a base station or CH needs universal low, medium, and high [49, 50].
information about the sensor network. In the
distributed technique, a node becomes either
CH or member node without the entire network Cluster Communication. Cluster communication is a data
information. sending mechanism from MNs to CH and from CH to
base station. There are two types of data communication
(ii) Cluster-Head Election. CH election can be of different mechanism and those are intracluster and intercluster.
types: ID-based heuristic [43], degree-based heuris- (i) Intracluster Communication. In cluster-based WSN,
tic [44], coverage based heuristic [45], and greater intracluster communication is diversified into two
6 International Journal of Distributed Sensor Networks

Static
Node type
Mobile
Network model
Centralized
Network type
Distributed
Cluster formation

ID based heuristic

Degree-based heuristic
CH election
consideration
Celebrative cover based heuristic

Weight based heuristic

Performance metrics of
cluster-based WSNs

Computational complexity

Cluster complexity Communication complexity

Control message

Single-hop
Intracluster
Multihop
Cluster communication
Single-hop
Intercluster
Multihop

Cluster maintenance
Cluster management
Domino effects

Figure 2: Performance metrics of cluster-based WSN.

approaches, such as single-hop intracluster com- Cluster Management. Cluster management deals with the
munication manner and multiple-hop intracluster topological manipulation in the cluster-based WSN. It is
communication manner. In the case of single-hop categorized into two types: cluster maintenance and domino
intracluster, all MNs in the cluster send data to effects.
the corresponding CH straightly, while in multihop
intracluster data moves through intermediate MNs (i) Cluster Maintenance. Cluster-based network forma-
in order to convey the message to the correspond- tion deals with the clusters formation, where cluster
ing CH. Single-hop intracluster performs efficiently maintenance handles the topological changes when
comparatively multihop intracluster communication clusters are formed. Cluster topology manipulates
in terms of energy conservation [51, 52]. new neighboring node discovery or the existing
node leaving the cluster-based network. Thus, cluster
(ii) Intercluster Communication. In cluster-based WSN, maintenance deals with updating the cluster structure
intercluster communication is also diversified into according to the change network topology. If cluster-
two classes which are single-hop intercluster commu- ing scheme is not scalable enough to facilitate cluster
nication manner and multiple-hop intercluster com- maintenance, then it results in domino effects. Thus,
munication manner. In the case of intercluster single- the whole network needs to be rebuilt from scratch
hop, all CHs communicate with the BS directly. In [54, 55].
contrast, data is relayed through intermediate nodes (ii) Domino Effects. There are some situations where
towards base station in intercluster multiple-hop. cluster-based network is rebuilt from scratch due to
To increase scalability of sensor network, multihop damage or movement of sensor node. Such situation
intercluster communication performs efficiently as occurs when cluster has no maintenance mechanism.
compared to single-hop intercluster routing [53]. In other words, domino effect results in reclustering
International Journal of Distributed Sensor Networks 7

the entire network when the existing nodes want to body area networks, different types of nodes are used in order
leave or new nodes want to join the network while to gather different types of data which is also heterogeneous
maintenance mechanism is absent in the network network.
[56].
Another type of heterogeneous network consists of nodes
which are of different resources in terms of energy and
5. Taxonomy of Cluster-Based WSN Schemes processing power. Mostly, CHs are preassigned and have extra
capabilities in the network [64, 65].
In this section, cluster-based algorithms are divided into Based on the mentioned classifications, in each category,
seven groups as described in Table 1. Different clustering the pros and cons of each algorithm are compared quali-
algorithms are studied under one of the seven scenarios. tatively and quantitatively. The summary of each group is
In each group, algorithms are compared qualitatively and described in Table 1. Figure 3 shows different algorithms
quantitatively to identify different algorithms’ characteristics. which are discussed in the next section.
Qualitative and quantitative analysis is performed which
is based on the performance metrics as discussed in the Network Type. In WSN, cluster formation is either
previous section. It is worth mentioning that generally a distributed or centralized. In centralized technique, a
cluster algorithm retains more than one single objective base station or CH needs universal information about
which is also discussed in this section. the sensor network. In the distributed technique, a
node becomes either CH or member node without the
Connected Dominated Set Based Cluster. Considering wireless entire network information.
sensor network as graph 𝐺, a vertex (node) subset 𝑆 of 𝐺 is a
dominating set (DS) if each vertex in 𝐺 either belongs to 𝑆 or Cluster-Head Election. CH election can be of
is adjacent to at least one vertex in 𝑆. Connected dominating different types: ID-based heuristic [43], degree-
set (CDS) forms the backbone tree of the network. Each CH based heuristic [44], coverage based heuristic
in the backbone has either direct connection with other CH [45], and greater weight based heuristic. In ID-
or indirect connection with other CH through other nodes. based heuristic, node ID is taken into con-
Backbone tree is a predefined communication path to send sideration for CH election, like a smallest ID
data to base station. The advantage of the backbone is to speed node becomes CH. In degree-based heuristic,
up the routing decision from source node to destination node quantity of neighbors is considered for CH elec-
[57, 58]. tion, while collaborative cover based heuristic
considers average hop distance between two
Mobility Aware Cluster. Mobility aware cluster comprises communicating nodes. It is indicated in [46]
mobile sensor nodes. Mobile sensor nodes move with respect that the degree-based heuristic is better than
to place and thus cause reclustering. To reduce reclustering ID-based heuristic in recognizing smaller size
issue, identical speed of nodes is grouped to form clusters. CDSs. However, collaborative cover heuristic
Thus, it results in prolonging cluster lifetime which also saves [45] is better than degree-based heuristic in
the energy during reclustering [35, 59]. recognizing smaller size CDSs. Moreover, in
weight based clustering, various parameters are
Energy Efficient Cluster. Sensor nodes mostly have limited considered to elect CH, such as remaining
and irreplaceable supply of energy. The nodes lifetime can be energy, communication cost, and distance. In
saved using energy efficient clustering techniques [60]. weight based criterion, a node is elected as a CH
based on energy cost.
Load Balancing. Load balancing puts limitation on nodes’
density to specific area during cluster formation phase. Thus,
such consideration forms sensor network in a way that the 5.1. CDS-Based Cluster. DS-Based Cluster considers two
data load is distributed equally among all sensor nodes. types of nodes which are CH and MN. CH gathers data from
Consequently, lifetime of the network is uniform [61, 62]. MNs and forwards it to other nodes via routing table. Each
CH acquires routing table and retains information of all its
Dynamic Cluster. Cluster maintenance is a very important neighboring nodes. CHs are a dominating set which is a
step to maintain all clusters of the network whenever any subset of the set consisting of all the nodes in the network
topological changes occur. Cluster maintenance makes the [6]. Considering wireless sensor network in graph 𝐺, a vertex
network dynamic when a new node joins or the existing node (node) subset 𝑆 of 𝐺 is a DS if each vertex in 𝐺 either belongs
leaves the cluster. The dynamic feature of the network, effi- to 𝑆 or is adjacent to at least one vertex in S.
ciently, handles the topological manipulation of the network When the dominating nodes are directly connected or
without affecting the whole network [2, 63]. indirectly connected through intermediate nodes, then they
form connected dominating set. The CDS forms virtual
Homogenous and Heterogeneous Cluster. In homogeneous backbone in the network which provides efficient energy
network, all nodes/data are of the same type. Therefore, utilization of sensor nodes and efficient data communication
CHs are selected in a random way while in heterogeneous in the network. Let an undirected graph represent 𝐺 = (𝑉, 𝐸),
networks node(s)/data are of different types. Moreover, in where 𝑉 is a set of nodes and 𝐸 is a set of links between nodes.
8 International Journal of Distributed Sensor Networks

Table 1: Brief overview of seven cluster schemes.

Number Cluster schemes Objective
All dominating nodes are joined together and form a subgraph to enhance
5.1 CDS-based cluster
communication routing and energy efficiency.
Mobile nodes of cluster are constructed and maintained with relative speed to
5.2 Mobility aware cluster
tighten nodes within same cluster.
Sensor nodes are irreplaceable and of limited energy. Suitable cluster leads to
5.3 Energy efficient cluster
extending life of network.
Spreading the workload of the network equally. In such scenario, CHs apply
5.4 Load balancing cluster
restriction on child nodes in order to accept a certain limit of child nodes.
When a new node is joining and existing node is leaving in existing network,
5.5 Dynamic cluster
dynamic cluster handles such situation.
Homogenous and In homogenous network, node/data is of same type, while in heterogeneous
5.6
heterogeneous cluster network node/data is of different type.

Classification of cluster-based WSNs schemes

CDS-based Mobility aware Energy efficient Homogenous and

Load balancing
cluster cluster cluster Dynamic cluster heterogeneous
cluster
cluster

DACDS MBC HEED AMC CNet(G) TL-LEACH

MCDS-CCH LECH-M DWEHC LBGC CBNet(G) UCS

DBCDS VGDRA PACDS LBCD-GA PPIA CDR

GBCHS NNDBC

Figure 3: Illustrating different algorithms highlighted in specific group.

𝑉󸀠 is a subset of 𝑉and 𝑉󸀠 forms CDS of graph 𝐺, if each vertex Root node (level 0) Level 1
𝑢 is a member of 𝑉, such that (𝑢, V) is a member of 𝐸 and Link: physically Backbone
subgraph is induced by 𝑉󸀠 . An efficient CDS should be more existing but
robust and of smaller size. logically not
considered

5.1.1. DACDS. New Distributed Algorithm connected dom-

inating set is proposed in [44] to build DS. Initially, in [44]
it uses the distributed leader election algorithm [66] to build Link: logically
a rooted spanning tree. Later on, every node finds its level as considered
follows. Firstly, a root node announces its lowest level 0. Other
nodes, upon receiving the level message from their parent
node, increase the level by 1. Secondly, a labeling strategy is Gateway
used to distribute the nodes in the tree to be either gray or Member node
black according to their level (the organize pair of id and Figure 4: Illustrating DACDS protocol.
level). The labeling mechanism starts from the root node and
finishes at the leaf node. The marking process is complete
when it reaches the leaf nodes. An MIS can be constructed
in the following way. receive a dominator message for the first time mark them-
Figure 4 shows the node which is in the lowest rank level selves as gray nodes and declare themselves as dominatee
0. It becomes the root node and marks itself as a black node. nodes (i.e., MNs). Then, the dominatee node broadcasts a
Black nodes become dominator nodes (i.e., CH) and message to its neighboring nodes about its dominatee status.
send dominator message to their neighbors. The nodes which When a node receives the dominatee messages from all of its
International Journal of Distributed Sensor Networks 9

Link: logically is in the range of sink node. Once the ready message is
considered Link: physically received, the node replies to the sink by sending its location
existing but information. After gathering information of all nodes, the
C logically not
considered sink executes a centralized algorithm to determine the CDS
B D among the nodes and adjust the right location of dominated
nodes in the network. Consequently, the dominated nodes
are moved and adjust their location accordingly, which is also
in the communication range of dominating nodes. Thus, the
high coverage network is also achieved. Advantages of the
proposed algorithm are improved network coverage rate and
efficient data routing, and a single node is used to collect
information of the specific location in the network.
A
5.1.4. GBCHS. Grid based cluster-head selection (GBCHS)
is presented in [69] to elect the CH among new deployed
nodes while clusters are formed based on data transmission.
Dominator The proposed algorithm works in centralized manner and
Member node
Connector
needs global information of the network about location. Each
node sends its location information to the sink node. After
Figure 5: Describing MCDS using celebrative cover heuristic. gathering information of the location, sink node finds the
center point of each grid and sends this information to new
deployed nodes. Once nodes receive the message from the
neighbors with lower ranks, it marks itself black and declares sink node, they determine their distance from center point.
itself as a dominator by broadcasting a dominator message. A node whose distance is very close to the center point
In the last phase, black nodes form connected dominating declares itself as a CH. This process is carried out in each grid.
set (CDS). Minimum connected dominating set (MCDS) of Advantages of such algorithms include less overhead to elect
DACDS is 8opt + 1, where opt represents size of black nodes. CH and energy efficiency as nodes distance is average.

5.1.2. MCDS-CCH. Minimum connected dominating set 5.1.5. Qualitative Analysis of CDS-Based Cluster. Table 2
(MCDS) using a collaborative cover heuristic (CCH) for describes all the three algorithms’ features. A CDS improves
WSN is proposed in [45]. The CCH is based on size data routing in WSNs in terms of smaller end-to-end delay,
optimization of MIS and CDS. Proposed algorithm in [45] less energy usage, and so forth. An efficient CDS is to be
optimizes independent nodes (i.e., CHs) based on the inspi- smaller, more robust, and of low stretch (backbone routing
ration of coverage instead of ID- or degree-based heuristic. should not be longer than the shortest path). A minimum
After electing independent nodes, the neighboring nodes are connected dominating set (MCDS) has many advantages
connected. The duplicated vertices are recursively removed in terms of data traffic, less communication overheads,
via Steiner tree. The algorithm is self-organized when node less energy consumption, efficient use of bandwidth, and
physically moves. It has been concluded that the size of increasing network lifetime. MCDS is an NP-hard problem
MCDS of CCH algorithm is intuitively efficient as compared in unit disk graph [70, 71]. DACDS is a degree-based
to [67]. heuristic to form CDSs, and MCDS-CCH is a collaborative
Figure 5 defines MCDS via celebrative cover heuristic. coverage heuristic to form CDSs. In terms of minimum
The nodes A, B, C, and D are an independent set of nodes connected dominating set (MCDS), collaborative coverage
and are connected with each other through a connector. The heuristics MCDS-CCH is better than degree-based heuristics
connection is formed, based on independent nodes coverage. DACDS, whereas degree-based heuristics DACDS is better
The nodes which are within the coverage of the independent than ID-based heuristics CDS-ER [45]. Computation round
node become the member nodes of their independent node. of MCDS-CCH is higher than DACDS, whereas DACDS
computational round is higher than CDS-ER. In terms of
5.1.3. DBCDS. New node deployment based on connected communication complexity, CDS-ER is higher than DACDS,
dominating set (DBCDS) is proposed in [68] which forms whereas DACDS is higher than MCDS-CCH.
connected dominating set (CDS), when nodes are deployed. Some of the disadvantages of DBCDS algorithm are as
The DBCDS provides better solution to nodes deployment follows. Firstly, the algorithm is operated in a centralized
problem in wireless sensor networks (WSNs). New nodes manner which requires global optimization location infor-
are randomly deployed in 3D monitoring space, where mation of the nodes to form CDS. Thus, the network may
the disconnected nodes move towards the sink node until suffer from large number of messages when the number of
connection is achieved. The sink node broadcast the ready nodes in the network is exceeding. Secondly, the network
message in order to collect the information of all nodes to requires an efficient collision avoidance scheme in order
form the network. A node keeps moving towards the sink to successfully send the messages. Thirdly, communication
node until it receives the ready message to confirm that it overheads highly affect network performance in terms of
10 International Journal of Distributed Sensor Networks

Table 2: Quantitative analysis of connected dominating set based cluster.

5.1.1 5.1.2 5.1.3 5.1.4

Cluster detail DACDS MCDS-CCH DBCDS GBCHS
CH election Degree based Collaborative Cover based Degree based Degree based
Intracluster routing Single-hop Single-hop Single-hop Single-hop
Intercluster routing Multihop Multihop Multihop Single-hop
Control message High High High Medium
Domino effect Yes Yes Yes Yes
Computation round 𝑂(𝑛) 𝑂(𝑛) 𝑂(𝑃𝑡 ∗ 𝑛) and 𝑂(𝑃𝑡 ∗ (𝑛 − 1)) 𝑂(𝑛 log 𝑛)
Communication complexity 𝑂(𝑛 log 𝑛) 𝑂(𝑛) 𝑂(log 𝑛) 𝑂(𝑞)
Stationary or mobile Static Static Mobile Mobile
Maintenance No No No No
Basic control Distributed Distributed Centralized Centralized

node energy and delay. Fourthly, nodes leaving happening in receives at time slot 𝑡2 . The distance 𝑑𝑖𝑗 (𝑡) = Radio velocity ∗
the network may cause rebuilding the network from scratch. |𝑡2 − 𝑡1 |. When a message is received, it is easy to determine
However, in GBCHS the network lacks backbone formation the mobility factor of 𝑁 nodes.
which ultimately limits the network coverage.
5.2.2. MBC. Mobility based cluster (MBC) protocol is sug-
5.2. Mobility Aware Cluster. Mobility aware clustering defines gested in [73] which is suitable when nodes are mobile. MBC
the mobility variations of sensor nodes. It describes topology considers two efficient parameters for CH election which
manipulation and route annulment at the time when the are remaining energy and the expected connection time. For
sensor node moves. In mobility aware clustering, the key longer network lifetime, CH election considers a node having
consideration for cluster formation is to group the nodes, more energy as compared to other nodes. Furthermore, to
based on identical motion. Thus, the links in inter- and overcome the issues of data loss, the expected connection
intraclustering can get more intensely connected, which time between CH and MNs is considered. MBC algorithm
decreases the reclustering and packet loss issues. considers two important phases for the network, namely,
setup phase and steady state phase. In the network setup
5.2.1. LEACH-ME. Leach Mobile Enhancement Protocol phase, nodes decide their status as either CH or MN based
(LEACH-ME) is a cluster-based protocol which is suitable for on the below formula:
mobile nodes in WSNs [72]. The design objective is to choose 𝑝 𝐸
a node with low mobility as a CH for longer cluster life. In 𝑇 (𝑛)new = ( 𝑛 current
1 − 𝑝 × [𝑟 mod (1/𝑝)] 𝐸max
order to quantify the low mobility factor, two parameters (2)
are considered, namely, remoteness and time. Remoteness is V − V𝑛 current
measured by data communication link changes rate. Thus, ⋅ max ) , ∀𝑛 ∈ 𝐺,
Vmax
uniform speed clusters are developed where CH has the
least speed in the group. The advantage of such cluster where 𝑝 is the expected percentage of CHs (e.g., 𝑝 =
development is to maintain least maintenance in the network. 0.05), 𝑟 describes current trip, 𝐸𝑛 current describes current
Consequently, CHs are retained for longer time. Remoteness energy, V𝑛 current describes current speed, 𝐸max describes
is measured, based on the mentioned equation. initial energy, and Vmax describes maximum speed of the
Let 𝑛𝑖 (𝑡), 𝑖 = 0, 1, 2, 3, . . . , 𝑁−1, where 𝑁 describes nodes node. When CH is selected, it informs the neighboring
density, be the location vector of sensor node 𝑖 at time 𝑡. nodes via sending a message. After receiving the message, the
And 𝑑𝑖𝑗 (𝑡) = |𝑛𝑗 (𝑡) − 𝑛𝑖 (𝑡)| is the distance between sensor neighbor nodes send join appeal to the CH. When the join
node 𝑖 and node 𝑗 at time 𝑡. So, the remoteness from sensor appeal is received, CH makes reservation of a timeslot for the
node 𝑖 to sensor node 𝑗 at time 𝑡 is 𝑅𝑖𝑗 (𝑡) = 𝐹(𝑑𝑖𝑗 (𝑡)), where newly MN.
𝐹 describes remoteness function. Distance from node 𝑖 to The algorithm considers two phases, namely, steady state
node 𝑗 is described by remoteness. Thus, nodes remoteness phase and setup phase, in order to reduce the number of
regarding time is measured as lost messages. In steady state phase, member node (MN)
checks its link status before sending data to CH. When
1 𝑁−1 󵄨󵄨 󵄨
∑ 󵄨󵄨𝑑 (𝑡)󵄨󵄨󵄨 ,
an acknowledgement message is sent to MN from CH, it
𝑀 (𝑡) = (1)
𝑁 − 1 𝑗=0 󵄨 𝑖𝑗 󵄨 describes that MN is still a child of CH. Consequently, MN
initiates data sending to its parent node (CH). Moreover, to
where 𝑑𝑖𝑗 describes the total distance from 𝑖th node to 𝑗th maintain reliability of path between MN and CH, both nodes
node and 𝑁 describes the total number of nodes in the maintain cluster time information. When the time hits on the
network. The data transmission starts at active time slot 𝑡1 and CH level, then a message is broadcast to cluster members.
International Journal of Distributed Sensor Networks 11

of mobile sink and guaranteed data delivery, this work

Base station
is, further, enhanced in [75] where appropriate forwarder
nodes are selected along mobile sink’s trajectory. The role
Cluster 2
of cell-header is progressively delegated to other nodes that
Cluster 1
help to achieve balanced energy consumption and improve
Cluster 3
network’s lifetime. Advantages of these algorithms are mini-

ge
Link: physically Link: logically

Ran
existing but considered mum communication overheads and improved data delivery
logically not 7
performance.
considered 10
Cluster 5
Cluster 4
5.2.4. Qualitative Analysis of Mobility Aware Cluster. Table 3
describes all three algorithms features. The intracluster com-
munication of all the three algorithms is considered as single-
Range

hop while intercluster communication is multihop. During

cluster formation phase, nodes send a node joining request
message. The nodes are considered as mobile nodes; hence,
Cluster head
New node network topology changes rapidly. All the three algorithms
Member node also invoke reclustering of the network; since there are no
cluster maintenance considerations.
Figure 6: Illustrating MBC algorithm. To address the mobility issues, LEACH-ME considers low
mobility node to be CH. The LEACH-ME requires 𝑂(𝑛2 )
Old TDMA rounds for cluster formation and reconfiguration and thus
scheduling Node 3 Node 4 Node 5 Node 7 Node 6 Node 8 provides stable clustering as compared to LEACH-M [76].
Each mobile node sends 𝑂(|𝑞|) messages to its neighbor to
New TDMA determine the relative speed, where 𝑞 represents one-hop
scheduling Node 3 Node 4 Node 5 Node 10 Node 6 Node 8
direct neighbor. Each mobile node aggregates node mobility
information, and a node having less mobility declares itself as
Figure 7: Illustrating TDMA of cluster 1. a cluster head.
In MBC, reclustering is lesser than CBR and Leach-M
because of its core features. In case of vast mobile node
As described in Figure 6, node number 7 withdraws its old environment, the MBC protocol decreases the packet loss
cluster (cluster 1) and rejoins a new cluster (cluster 4). On the by 25% comparatively to CBR [77] and 50% comparatively
other hand, node number 10 withdraws its old cluster (cluster to LEACH-M [76]. However, it considers centralized cluster
4) and rejoins a new cluster (cluster 1). Therefore, cluster 1 and formation, and, therefore, nodes suffer from extra communi-
cluster 5 of new TDMA scheduling are adjusted on CHs level. cation overheads.
In Figure 7, when node 7 leaves the network, then its time slot
is also removed. Hence, the time slot of node 7 is unoccupied
as node leaves the network. When cluster-head receives the 5.3. Energy Efficient Cluster. One of the key challenges in
joining message from node 10, then the unoccupied time slot WSNs is increasing lifetime of the network, because sensor
is reserved for node 10. The time slotting is altered when a nodes are mostly constrained by limited energy and irreplace-
node joining and node leaving happens. able power supply. Thus, energy of the sensor nodes needs
In MBC, every MN is assigned a timeslot for data trans- to be used wisely. Energy efficient clustering is one of the
mission. MN sends its data within its timeslot boundaries and techniques that can extend sensor network lifetime. Cluster-
broadcasts a new joint message to a new cluster at the time it based WSNs consist of CH and MNs, where CH handles
drops the connection with the respective CH. extra work as compared to the MNs, such as data aggregation
(from MNs) and data routing (from lower level). Thus, CHs
have excess of chances to “die” soon. Enhancing the network
5.2.3. VGDRA. Virtual Grid Based Dynamic Routes Adjust- lifetime is a critical issue in wireless sensor network.
ment (VGDRA) scheme presented in [74] aims to maintain
nearly optimal routes to the latest location of a mobile
sink at the expense of least communication overheads. It 5.3.1. HEED. Hybrid Energy Efficient Distributed Clustering
partitions the sensor field into square number of equal sized (HEED) is energy efficient clustering [78]. HEED considers
cells where each cell is administered by a locally elected two important parameters for CH election, and these are
cell-header. To cope with dynamic network topology caused node’s residual energy as a primary parameter and node’s
by sink’s mobility, it defines a set of communication rules degree or intracluster communication cost as a secondary
that govern the routes readjustment process. Following those parameter. Comparing HEED with LEACH, HEED performs
communication rules, only a subset of the elected cell- better in terms of enhancing network life time. LEACH elects
headers takes part in routes readjustment process, thereby CH randomly which makes few nodes die fast. In HEED, each
offering nearly optimal data delivery routes while incurring MN is associated with a cluster. In HEED, the total clustering
least communication overhead. To cope with speed variation cost is minimized as the CHs are distributed properly across
12 International Journal of Distributed Sensor Networks

Table 3: Quantitative analysis of mobility aware cluster.

5.2.1 5.2.2 5.2.3

Cluster detail Leach ME MBC VGDRA
CH election Weight based Weight based Weight based
Intracluster routing Single-hop Single-hop Single-hop
Intercluster routing Multihop Multihop Multihop
Control message Medium Medium High
Domino effect Yes Yes Yes
Computation round 𝑂(𝑛2 ) 𝑂(𝑛 log 𝑛) 𝑂(𝑛)
󵄨 󵄨
Communication complexity 𝑂(󵄨󵄨󵄨𝑞󵄨󵄨󵄨) 𝑂(log 𝑛) 𝑂(log 𝑛)
Stationary or mobile Mobile Mobile Mobile
Maintenance No No No
Basic control Distributed Centralized Distributed

the network. Each node finds its own probability to become

Base station
CH, based on the below mentioned equation:
𝐸residual
CHprob = 𝐶prob . (3)
𝐸max
Link: physically
𝐶prob describes the number of network nodes to become CHs existing but Link: logically
logically not considered
(e.g., 5%), 𝐸residual describes the present node energy, and considered
𝐸max describes maximum energy. CHprob is not allowed to be
lesser than a certain limit of the threshold. Therefore, every
node executes certain iteration until it finds the CH. A node
would be a tentative CH, if its probability is less than 1. A node
becomes MN when it finds other less cost CH. On the other
hand, a node acts as a permanent CH when its probability
reaches 1.
HEED considers multihop inter- and intracluster com- Cluster head
munication. The algorithm considerations are as follows: Member node
firstly, a Distribute clustering scheme where CH election
considers two significant parameters as mentioned earlier; Figure 8: Describes HEED protocol.
secondly, the clustering process terminating within a specific
cycle limit; thirdly, reducing control overhead; fourthly, the
intercluster communication is multihop between CHs as weight becomes the CH. All first level members have direct
described in Figure 8. communication with CH and the other nodes with lower
energy access indirectly to the CH through other MNs. Each
5.3.2. DWEHC. Distributed Weight Based Energy Efficient node finds the relay node to approach CH at lesser cost.
Hierarchical Clustering (DWEHC) is introduced in [79] Based on node’s distance to its neighboring node, it decides
which is more efficient and goal-oriented as compared to to remain in the first level or move to the second level. This
HEED. DWEHC is a distributed multihop intraclustering process is repeated until all nodes settle on energy efficient
topology as described in Figure 9. Every sensor node calcu- intraclustering. Every CH allocates a range in which all MNs
lates its weight by the below equation and either declares itself must exist within that range. Thus, it limits the maximum
as a CH or locates itself as a MN: number of levels up to a certain range.

(𝑅 − 𝑑) ] 𝐸residual (𝑠)
𝑊weight (𝑠) = [ ∑ × , (4) 5.3.3. PACDS. On calculating power-aware connected dom-
6𝑅 𝐸initial (𝑠) inating sets are proposed in [80] to discuss efficient routing
[𝑢∈𝑁𝛼,𝑐 (𝑠) ]
in cluster-based WSN. In the proposed scheme, connected
where 𝑅 is cluster range and 𝑑 is distance from node 𝑠 to node dominating set is formed, based on the sensor node degree
𝑢 where 𝑢 is the neighboring node of node 𝑠 and 𝐸residual (𝑠) and the level of energy of every host. The aim is to offer
and 𝐸initial (𝑠) are the residual and initial energy in 𝑠, which is CH election scheme so that overall energy depletion is well-
considered uniform for all the nodes. adjusted in the network, and a relatively small connected
DWEHC takes two parameters into consideration for dominating set is produced at the same time. DS nodes
CH election: preserved energy and the neighbor proximity. retain some remarkable batteries as compared to non-DS
In neighboring node, the node that retains the highest nodes because nodes of DS execute some extra tasks, like
International Journal of Distributed Sensor Networks 13

get neighbor information that requires 𝑂(𝑛 ∗ 𝑟) messages

Base station
where 𝑟 denotes the number of rotations. On the other
hand, DWEHC requires remaining energy and neighbor
proximity information requires 𝑂(𝑛 + 𝑐) rounds where 𝑐
MN upper
bound limit in denotes proximity information. DWEHS suffers from huge
all clusters Link: logically number of control message. In the first phase, it gathers
considered information from each node energy and then decides a higher
Range Range energy node to be in DS. In the second phase, it tries to reduce
the size of a DS to avoid unnecessary energy consumption.
Link: physically Thus, it requires 𝑂(𝑛2 ) rounds to form the network, which is
existing but higher than 𝑂(𝑛) and 𝑂(𝑛 + 𝑐) rounds.
logically not
considered
5.4. Load Balancing Cluster. The basic idea of load balancing
cluster is to achieve uniform lifetime of the network. Load
Cluster head balancing cluster algorithms handle optimum quantity of
Member node nodes in cluster. If the numbers of nodes is too large in
a cluster, CH is overloaded as it performs multiple tasks.
Figure 9: Illustrates DWEHC protocol.
On the other hand, cluster formation with small number
of nodes causes the high number of clusters. Therefore, the
network increases the hierarchical length and larger end-to-
data gathering, routing information, and data routing. Thus, end delay. Load balancing limits the number of nodes in a
it is compulsory to reduce energy utilization of DS. One cluster. Whenever cluster size exceeds its predefined limit,
technique is to reduce DS density, and, thus, unnecessary load balancing procedure accommodates extra node in other
mobile sensor nodes are removed from DS. This technique clusters or forms a new cluster.
secures their energy depletion for operating as cluster-heads.
In [43], certain extra procedures are recommended to remove 5.4.1. AMC. An adaptive multihop cluster-based scheme
needless dominating nodes. is proposed in [81] which is a load balancing clustering.
In [80], some procedures, based on energy level, are also Initial construction of cluster formation is not described
suggested to define whether a node can be in DS or not. in AMC. However, for cluster maintenance, every node
The mobile sensor node is excluded from the dominating broadcasts its information, including its ID, status, and CID
set when the node and its members are adjacent to other (CH/member/gateway) to other nodes in the cluster. A CH
DS. The suggested scheme in [80] is more energy efficient can get information of each neighboring cluster node. Each
as compared to other DS-based clustering procedures as it cluster can handle up to certain limit of MNs via upper and
attempts to remove mobile nodes for DS that has lesser lower bounds (𝑈 and 𝐿).
residual energy. A flat architecture is depicted in Figure 10(a). In AMC, cluster merging occurs when one MN |C𝑖 |
Figure 10(b) is a cluster-based architecture. Figure 10(b) distance is less than 𝐿 from cluster C𝑗 . Cluster C𝑖 searches for
shows the CHs, namely, A, B, C, and D, which retain less its neighboring cluster C𝑗 to assure |C𝑖 +C𝑗 | ≤ 𝑈 and increase
energy, and its MNs are also within the range of higher energy. the total number. In another situation when |C𝑖 | + |C𝑗 | > 𝑈
Thus, the CHs along with their associated child nodes become for the entire neighboring clusters, it minimizes the value via
as a member node of higher energy cluster-head, and its child searching C𝑗 . After joining two clusters, the CH keeps more
nodes are also moved to other cluster-heads. Figure 10(c) member nodes.
describes the graph after excluding the less energy CHs. Figure 11 describes AMC protocol. There are three clusters
in the network: cluster A, cluster B, and cluster C. Each
5.3.4. Qualitative Analysis of Energy Efficient Cluster. Table 4 cluster must maintain a specific number of member nodes to
describes all three algorithms’ features. The three algorithms maintain proper load balancing. If the member nodes of the
are weight based cluster-head election. Intracluster is single- cluster are less than lower bound, then the cluster merges with
hop, and intercluster is multihop, and which are considera- the neighboring cluster. In Figure 11, cluster C considers that
tions. HEED and DWEHC are considered stationary nodes a mobile node j moves to cluster B. After moving to cluster
while PACDS are considered mobile nodes. If a few nodes B, the cluster size becomes lesser than lower bound. Hence,
are faulty in HEED or DWEHS, this may cause severe to merge cluster C with either cluster A or cluster B, CH of
problem. Such situation makes the network into a disjoint cluster C gets information about the neighboring cluster size.
portion where one portion may become out of range from As the size of cluster A is lesser than the size of cluster B,
another portion. Therefore, stationary nodes may suffer therefore, cluster C merges with cluster A. In situation when
more as compared to mobile node. Since no algorithm has the member nodes increase from upper bound, cluster splits
cluster maintenance, therefore, network is restructured from into two clusters.
scratch whenever a node dies in any of the algorithms. In
HEED, tentative CHs leave some unconnected nodes, when 5.4.2. LBGC. Load Balancing Group Clustering (LBGC) is
permanent CH gets elected. In HEED algorithm, all nodes proposed in [82]. LBGC mainly focuses on multihop routing
14 International Journal of Distributed Sensor Networks

Cluster

D
A

C
Cluster head Cluster head
Member node Member node Member node
(a) (b) (c)

Figure 10: Describes PACDS algorithms.

Cluster A Cluster B Cluster A Cluster B

b b l
Link: physically k l a k
a
existing but d
d n c n
logically not c
m m
considered
e p e p

o o
Link: logically g g
considered f f
j
j
Cluster h
h i Cluster C
i Cluster C

Cluster head Cluster head

Member node Member node
(a) (b)

Cluster A Cluster B Cluster A Cluster B

b l b l
a k a k

c d n d n
c
m m
e p e p

o o
g g
f f
j j
h h
i Cluster C i

Cluster head Cluster head

Member node Member node
(c) (d)

Figure 11: Describing AMC algorithm.

International Journal of Distributed Sensor Networks 15

Table 4: Quantitative analysis of energy efficient cluster.

5.3.1 5.3.2 5.3.3

Cluster detail HEED DWEHC PACDS
CH election Weight based Weight based Weight based
Intracluster routing Single-hop Single-hop Single-hop
Intercluster routing Multihop Multihop Multihop
Control message High High High
Domino effect Yes Yes Yes
Computation round 𝑂(𝑛 ∗ 𝑟) 𝑂(𝑛 + 𝑐) 𝑂(𝑛2 )
Communication complexity — 𝑂(𝑛) 𝑂(𝑛 + 𝑚)
Stationary or mobile Stationary Stationary Mobile
Maintenance No No No
Basic control Distributed Distributed Distributed

Table 5: Quantitative analysis of load balancing cluster.

5.4.1 5.4.2 5.4.3

Cluster detail AMC LBGC LBCD-GA
CH election Weight based Weight based Weight based
Intracluster routing Single-hop Single-hop Single-hop
Intercluster routing Multihop Multihop Multihop
Control message High High High
Domino effect No Yes Yes
Computation round — Two rounds 𝑂(𝑛 log 𝑛)
Communication complexity — — —
Stationary or mobile Mobile Stationary Stationary
Maintenance Yes No No
Basic control Distributed Distributed Distributed

mechanism to reduce energy consumption. But this may is mainly population based search algorithm. It considers the
cause problem of “hot-spots” in sensor networks. For exam- number of dominators in dominating set. GAs are heuristic
ple, the energy of the CH, which is near to the BS, depletes search based algorithms on evolutionary thoughts of natural
faster due to higher number of transaction as compared to assortment and genetics [84]. Figure 12 illustrates the load
those nodes which are far from the BS. To overcome this LBCD-GA algorithm. Figure 12(a) illustrates a flat network.
problem, they use unequal-grouping method to attain load In Figure 12(b), there are two dominator nodes: dominator
balance among the entire CHs. The distance among the group 4 and dominator 5. If each node sends the same amount of
head (GH) and BS serves as grouping condition to make the data at fixed rate, then dominator 4 exhausts energy quicker
groups which are near to the BS and smaller than the far ones than dominator 7. Therefore, in Figure 12(c), the number of
possible. dominators is increased. Thus, in terms of load balancing,
LBGC protocol, periodically, elects CHs, and it imple- Figure 12(c) gives results better than Figure 12(b). However,
ments dynamic route calculation according to the state of still dominatee balance is not uniform. Thus, in Figure 12(d),
energy allocating of sensor network. LBGC reelects the CH dominatee, that is, node 4, connects with dominator, that
by considering its transmission costs and extra energy and is, node 6. Subsequently, in terms of enhancing network life
rearranges cluster. Load of the nodes is distributed, based on time, Figure 12(d) provides results better than Figures 12(b)
their remaining energy among all the nodes. It proves better and 12(c).
load balance and increases life for the sensor network. If each
dominator in CDS is not equal, then the heavy dominator 5.4.4. Qualitative Analysis of Load Balancing Cluster. Table 5
nodes decrease their energy very rapidly. summarizes three load balancing cluster-based schemes as
discussed before. AMC adopts multihop communication in
5.4.3. LBCD-GA. Load Balance Connected Dominating Set both intra- and intercluster communication. AMC has no
Genetic Algorithm (LBCDS-GA) is proposed in [83], which domino effect of reclustering and retains good cluster stability
assures that each dominator (i.e., CH) is balanced in terms whereas both the LBGC and LBCD-GA have no maintenance
of work load in MCDS and allocates dominatee (i.e., MN) mechanism and result in domino effects. Making a MCDS
node to every dominator (i.e. CH). To overcome “hot-spots” for a WSN is an NP-hard problem even in the unit disk
problem, the author proposes Genetic Algorithm (GA). GA graph [83] model. Many research works have been devoted
16 International Journal of Distributed Sensor Networks

8 8 8

7 7 7
6 6 6
1 1 1
5 5
5 4 4
4
3 3 3

2 2 2

Dominators Dominators Dominators

Dominatees Dominatees Dominatees Dominatees
(a) (b) (c) (d)

Figure 12: Illustrating LBCD-GA algorithm.

to achieving a better performance ratio. For CDS, algorithms Root node Cluster
are categorized into two stages: one-stage and two-stage Link: physically Backbone
algorithms. The objective of one-stage algorithms [85] is to existing but
build a CDS directly, while two-stage algorithms build a CDS logically not
considered
in two phases. First phase builds minimum DS, while second
phase uses Steiner Tree method to build a CDS [86]. In [85],
the author proposes two centralized greedy algorithms. The
first algorithm is a one-stage strategy with an approximation
ratio of 2𝐻(Δ) + 2, where Δ is the maximum node degree in a Link: logically
network and 𝐻 is a harmonic function. The second strategy is considered
a two-stage strategy with an approximation ratio of 2𝐻(Δ)+2
proposed in [85].
Cluster head
Gateway
5.5. Dynamic Cluster. The sensor network has sufficient Member node
scalability to permit new nodes to join, or the existing nodes
to leave the network without interrupting network data flow. Figure 13: Describing CNet(G) topology.
However, it has been explored that flat sensor network is
not scalable as cluster-based WSN [13, 14, 87]. In cluster-
based network, cluster formation and cluster maintenance three types of nodes in CNet(G), namely, MN, CH, and
methods are used to maintain the structure of the network. GW. MNs are single-hop intraconnected with CH. CHs are
Cluster formation deals with building of cluster structure. connected through gateway (GW). Thus, CHs and GWs form
Cluster maintenance deals with updating the cluster network a predefined routing path which is called backbone. Network
when a new node wants to join and the existing node leaves maintenance is performed through two operations, namely,
the network. Dynamic clustering ensures network robustness node-move-in and node-move-out. Node-move-in describes
wherein new node joining and fault tolerance techniques are a single node joining while node-move-out describes a single
addressed in the network environment. The objective of the node leaving the CNet(G). When a new node needs to join the
new nodes deployment in the network can be divided into network, then it is deployed to the network field. The node is
two types, such as relay node and new nodes. Relay node joins joined and gets its status, based on in-range node status. The
the disjoint paths of the network and new nodes increase the node that is within the range of the new node is described by
coverage area of the network. The cluster becomes unstable the neighbor node.
whenever cluster maintenance is absent in the algorithm The new node searches the neighbor nodes based on
[2]. Thus, it results in loss of routing information at the priority in order to join as described in Figure 14. Firstly,
time when changes occur in the network. Therefore, overall new node searches a CH in neighbor, if CH does not exist
network performance degrades. To overcome this problem, then new searches a GW node. When the CH and GW do
it is suggested to build efficient cluster maintenance scheme. not exist in neighbor, then new searches a MN. Otherwise,
Consequently, it avoids loss of routing information and when neighboring node of new node is gateway node, then
improves overall performance of routing protocol [88]. new node joins the network as CH. If the neighboring node
is member node, then neighbor node changes its own status
5.5.1. CNet(G). Construction and Maintenance of Novel from member node to gateway node and new joins as CH.
Cluster-based Wireless Sensor Network (CNet(G)) is pro- As described earlier, node-move-out is a single node
posed in [89, 90] as described in Figure 13. There are leaving algorithm and sends “leaving message” to child nodes
International Journal of Distributed Sensor Networks 17

Link: physically
Start
existing but
Backbone logically not
New node sends considered
“AddMe” message

No CH exists in the Yes

neighbor

GW exists in
No the neighbor Yes New node becomes MN
Link:
logically
New node becomes CH Stop considered
MN becomes GW

Cluster head Gateway

MN = member node Cluster forwarding node Forwarding node
CH = cluster head
GW = gateway node Figure 16: Describing network topology of CBNet(G).
Figure 14: Describing new node joining in CNet(G).

routing protocol. CBNet(G) considers multihop intra- and

intercluster communication as described in Figure 16. Two
maintenance algorithms are also developed for CBNet(G),
DCWSN Root node
namely, node-move-in and node-move-out. Node-move-in is
a single new node joining algorithm while node-move-out is
a single existing node leaving algorithm. Both algorithms are
described below.
When a new node is deployed in the network, then the
Leaving node node calls node-move-in to join the network. New node
determines its status according to the in-range node of the
network. When there is a CH(s) in the range of the new
Subtree node, then the node joins as an ordinary node. Whereas
when the new node finds any cluster forwarding node in
its neighbor, then the new node becomes as an ordinary
node and chooses one of the cluster forwarding nodes as its
parent node, which is having the least distance to its CH.
When there is a forwarding node within the range of the new
node, the new node becomes ordinary node and also elects a
Figure 15: Describing the leaving node procedure.
forwarding node which is having the least distance to its CH
as its parent. Else, if the new node has an ordinary node in
its range, which is also closer to CH 𝑘 (𝑘 ≥ 2), the new node
in order to leave the network. When the child nodes receive elects a parent node which has the minimum distance to its
the message, then they withdraw the network and rejoin CH. Thus, the new node becomes an ordinary node, and the
from the scratch as described in Figure 15. In Figure 15, the selected node changes its status to the forwarding node. Else,
decedent nodes of the leaving node withdraw the network if the new node finds gateway nodes in its neighbor, the new
and rejoin via calling node-move-in algorithm. one becomes a CH. Else, ordinary node is within the range of
The merits of structured network provide key features the new node whose distance to their CHs is 𝑘. The new selects
such as minimizing communication overheads, choosing an ordinary node as its parent and becomes a CH. The chosen
data aggregation points, increasing the probability of aggre- node changes its status from the ordinary node to gateway
gating redundant data, and minimizing the overall power node and all other nodes, which are fallen on the root from
consumption [91]. the gateway node to its CH, also change their status to cluster
forwarding node.
5.5.2. CBNet(G). Novel Cluster-Based Architecture and a
Routing Protocol for Dynamic Ad Hoc Radio Networks 5.5.3. NNDBC. Node nonuniform deployment based on
CBNet(G) are proposed in [92] which is mainly designed to clustering (NNDBC) algorithm for UWSNs is introduced
support a timely and energy efficient, loop-free, on demand in [93] which addresses nodes joining and nodes leaving
18 International Journal of Distributed Sensor Networks

Table 6: Quantitative analysis of cluster maintenance.

5.5.1 5.5.2 5.5.3

Cluster detail CNet(G) CBNet(G) NNDBC
CH election Degree based Degree based Degree based
Intracluster routing Single-hop Multihop Single-hop
Intercluster routing Multihop Multihop Multihop
Control message Medium High High
Domino effect No No No
Computation round 𝑂(log 𝑞) 𝑂(𝑞 + 𝑘) —
Communication complexity 𝑂(2𝑝 − 1) 𝑂(𝑟 + 𝑙) —
Stationary or mobile Stationary Stationary Mobile
Maintenance Yes Yes Yes
Basic control Distributed Distributed Distributed

procedures. The coverage targets of NNDBC algorithm are and ordinary nodes. Similarly, CBNet(G) addresses single
the isolated events whose distribution is usually nonuniform node joining and single node leaving the network. The new
in the mentioned space. NNDBC aims to optimize network node waits to listen from all the nodes before joining and
connectivity rate and improve network lifetime for nonuni- then chooses a suitable node to establish the join connection.
form deployed networks. The clustering formation is based The time complexity and message complexity of CBNet(G)
upon communication range which is used to determine the are higher than CNet(G) as described in Table 6.
network connectivity. The nodes in the cluster consist of in- The disadvantages of the NNDBC algorithm are as
cluster nodes and cluster-head node. The in-cluster nodes follows: high communication overheads occur in the network
are connected to their own cluster-head node, whereas the as the algorithm requires running on centralized manner
cluster-head nodes are connected to the closest node (say [19]. Secondly, the algorithm has lack of minimizing number
gateway node) of the other cluster. of messages to form the network. Thirdly, the descendant
After the network formation, steady state phase is exe- nodes of leaving node need to rejoin the network from scratch
cuted. In this phase, when a node 𝑖 energy is lesser than which degrades network performance.
the threshold energy then the node is substituted by another
node. Thus, node cluster head 𝑐(𝑖) of 𝑖 broadcast help-need
message to other clusters. A cluster head that has received the 5.6. Homogeneous and Heterogeneous Cluster. In homoge-
message checks a node whose contribution degree is smaller neous network, all nodes/data are of the same type; thus, CHs
than node 𝑖. The cluster head sends help-gives message when are selected in random way, while in heterogeneous networks
the node exits. Thus, the node 𝑖 moves to the new cluster node(s)/data are of different types. In case of WBANs,
head upon receiving the help-gives message. Advantages of different types of nodes are required to monitor different
the algorithm are improved network connectivity and lower health parameters of the human beings. Thus, WBANs is a
aggregate contribution degree to substitute the dying node. type of heterogeneous network [4].

5.5.4. Qualitative Analysis Cluster Maintenance. Table 6 5.6.1. TL-LEACH. TL-LEACH is introduced in [94] which
summarizes the features and objectives of three main- maintains two-level hierarchy of CH: primary CH (CH𝑖 )
tenance schemes. CNet(G) is a single-hop intracluster and secondary CH (CH𝑖𝑗 ) as presented in Figure 17. MNs
while CBNet(G) is multihop intracluster communication. In send the data to secondary CH; secondary CH sends the
Table 6, 𝑞 represents neighboring nodes of the new joining aggregated data to primary CH, and finally primary CH sends
node; 𝑝 represents the number of CH in CNet(G); 𝑘 is the the aggregated data to the base station. In TL-LEACH, energy
maximum radius of cluster; 𝑟 is the number of intracluster is equally assigned to all the nodes. The algorithm consists
nodes, and 𝑙 is the size of intercluster. CNet(G) is a single node of four basic phases: in the first phase, nodes send request
joining and single node leaving algorithm. When a single to form the network and it is called “advertisement phase.”
node is deployed to join the network, it gets the neighbor Second phase is cluster setup phase; third phase is schedule
nodes’ information. The computational round of CNet(G) is creation, and the last phase is data transmission. During the
𝑂(log 𝑞) which is lower than CBNet(G). However, in node second phase, each node is primary CH or secondary CH
leaving procedure, the decedent nodes of the leaving node or ordinary node. If a node becomes primary CH, then it
rejoin the network. It happens that the size of the subtree of informs the neighboring nodes about its status. Thereafter,
leaving node is bigger than the other subtree. Therefore, the the secondary CH decides which primary CH it belongs to,
complexity of joining subtree may be very near to the network and it informs its primary CH by sending a message. In the
formation. same way, each MN decides which secondary CH the node
CBNet(G) consists of five types of nodes: CH nodes, belongs to, and it informs its secondary CH by sending a
forwarding nodes, cluster forwarding nodes, gateway nodes, message.
International Journal of Distributed Sensor Networks 19

Base station
Link: logically
considered
Link: physically
existing but
Link: physically logically not
considered
existing but
logically not Link: logically
considered considered

Layer 1 cluster head

Layer 2 cluster head
Primary cluster head Base station
Secondary cluster head
Figure 18: Describing UCS network.
Figure 17: Describing TL-LEACH topology.

layer are identical. However, the two layers of cluster are

To avoid collision, Carrier Sense Multiple Access (CSMA) different which make uniform lifetime of the network.
is used. In the third phase, each primary CH forms a TDMA
schedule and assigns it to secondary CH in its group. In the 5.6.3. Critical Data Routing in WBSNs (CDR). The algorithm
same way, every secondary CH creates scheduling and assigns in [96] describes dominating set based new nodes joining.
it to their MNs. In the last phase, data transmission happens The data communication is divided into three tiers: Intra,
with respect to TDMA scheduling. Figure 17 describes the Inter, and Extra [4]. In the first tier, that is, intra, a new joining
network in three levels. In the first level, CHs are in direct node forwards the sensed data to the local coordinator which
communication with the base station. In the second level, acts as a cluster head [4]. The three main issues with intra
secondary CHs send the aggregated date to primary CHs. In are as follows: firstly, heterogeneous nature of the sense
the third level, MNs gather the data from the environments data which requires different quality of service parameters.
and send them to secondary CHs. Secondly, temperature rise of the implanted sensor nodes
which may be harmful for human tissue. Thirdly, high and
dynamic path loss due to the postural movements [96–98].
5.6.2. UCS. Unequal Cluster Size (UCS) is suggested in [95]
In [97], the authors have classified the sensed data by
to balance energy of the network and to increase lifetime of
new node into reliability sensitive data and normal data,
the network. UCS is heterogeneous cluster model for WSN.
while in [98] the data is being classified into critical data and
It is considered that the base station is laying in the middle of
normal data. However, in [96] the data is categorized into
the network. All CHs are preordered positions and arranged
three classes, that is, reliability sensitive data, delay sensitive
in circle around the base station. In UCS, it is considered as
data, and normal data. The aforementioned three schemes are
a sensing field in circle. Each circle is, further, divided into
modular based where different modules are used to perform
layers. As shown in Figure 18, it is considered that the shape
the various tasks. The routing decisions are based on the
and size of the entire cluster within layer are identical.
nature of data, temperature rise, and path loss. All these
However, the shape and size of clusters in two layers are
schemes performs better as compared to other state-of-the-
different. The CH, which is near the base station, consumes
art schemes due to multiobjective selection criteria. However,
more energy due to high burden of relaying packets coming
in node joining, messages are transmitted among new node
from upper CH. Therefore, the CH nodes, which are near to
and existing node. These messages use node’s energy in data
BS, keep lesser number of MNs as compared to CHs far away
transmission. The algorithm has lack of a mechanism to
from the base station which makes uniform lifetime of the
minimize the messages.
time.
The energy usage of a CH depends on inter- and
intracluster communication. However, the energy usage on 5.6.4. Qualitative Analysis of Homogenous and Heterogeneous
intercluster data communication is a property of expecting Cluster. In a homogeneous network, CH is highly-loaded
load from other CHs. A CH near the base station suffers from by member nodes and other CH(s) in sending information
more energy depletion due to relaying other CH data. It is towards the remote base station. If a CH has 𝑛 child node(s)
concluded that cluster size optimization based on distance then it might have 𝑂(𝑛 + Δ) data overheads where 𝑛 is the
with the BS maximizes the life time of the network. For number of its child nodes, and Δ is the additional data load of
example, in Figure 18, there are two layers of CH: inner CH relaying information. Therefore, the chances of CH(s) dying
and outer CH. The shape and size of the entire cluster within earlier are more than the other nodes. It is indigent that all
20 International Journal of Distributed Sensor Networks

Table 7: Quantitative analysis of homogeneous and heterogeneous network.

5.6.1 5.6.2 5.6.3

Cluster detail TL UCS CDR
CH election Degree based Weight based Weight based
Intracluster routing Single-hop Multihop Single-hop
Intercluster routing Multihop Multihop Multihop
Control message High High Medium
Domino effect Yes No No
Computation round — — —
Communication complexity — — —
Stationary or mobile Stationary Mobile Stationary
Maintenance No Yes No
Basic control Distributed Distributed Distributed

the nodes use their battery at the same period and very less work needs to be done to improve CCH to reduce
energy is left behind. routing complexity.
TL-LEACH is a homogeneous cluster-based scheme
(ii) Nodes mobility may also introduce cluster reclus-
which forms primary and secondary CH. It rotates the roles
tering and reconfigurability. Auxiliary cluster head
of both primary and secondary CH to maintain efficient load
might be elected from the cluster members, which can
distribution.
act as a cluster head. This technique may make the
On the other hand, UCS is a heterogeneous cluster-
cluster more stable.
based scheme. CHs are designated super nodes where the
overall network lifetime is determined by the life time of (iii) During cluster formation, control messages transmit
CH nodes [95]. CHs have enough energy to meet 𝑂(𝑛 + Δ) between nodes which consume the node’s energy.
data overheads where 𝑛 is the child nodes, and Δ is the Control messages might be decreased when the com-
additional data load of relaying information. In UCS, the plexity of the cluster formation depends upon the
number of MNs in each CH is not constant. CH changes neighbor nodes instead of entire network.
its number of nodes based on the estimated communication
load. Therefore, UCS claims that it maintains more identical
energy dissipation between the CHs. Thus, energy dissipation Cluster Maintenance
for each CH is similar which increases the total network
(i) Network maintenance is an inseparable part for sen-
lifetime.
sor network architecture. At the network reconfigura-
In UCS, CH nodes are super nodes which are deployed
tion phases, nodes need to discover the neighboring
in preassigned location that reduces universality. Moreover,
nodes, where the neighbor discovery process requires
two-hop intercluster transmission decreases the network
synchronization among the nodes. Therefore, an effi-
communication range. Furthermore, CH node serves as the
cient synchronization scheme is a challenging issue to
fusion point as well as the command center of its cluster. As
discover the neighbor nodes.
a result, when a CH node fails, all the sensor nodes in that
cluster have to be reassigned to other neighboring clusters (ii) The state-of-the-art of cluster-based WSNs focuse on
(Table 7). single node deployment, where multinodes deploy-
ment is still missing in dynamic clustering. If the
channel is to be considered as single, then data and
6. Open Research Areas in beacon collisions may happen that interrupt multi
Cluster-Based WSNs node joining. Randomize algorithms might be one of
the solutions to overcome collision and to successfully
Recent proposed cluster-based schemes have a lot of critical join multi nodes.
concerns for the concrete deployment of wireless sensor
network. Nevertheless, it is observed that some issues are still (iii) The state of the art of cluster-based WSNs addresses
pertaining in the latest research. The issues, open for research only single node leaving the network. However, mul-
in cluster-based WSNs, are as follows. tiple nodes may leave the network. Leaving nodes
might be parent nodes, and leaf nodes depend on
Cluster Formation the minimum energy level of sensor meets. If leaving
nodes are parent node, then centralizing algorithm,
(i) An efficient minimum connected dominating set using DFS algorithm, might be one of the solutions
(MCDS) exhibits less routing complexity due to its to handle multiple nodes leaving. If leaving nodes are
shortest path. Collaborative cover heuristic CCH the leaf nodes, then multiple node leaving might be
provides efficient results in terms of MCDS. More possible without informing centralize node.
International Journal of Distributed Sensor Networks 21

(iv) New nodes need to discover all their neighbor nodes an important parameter that highly affects energy, where high
to join the network. However, gathering the entire control overheads result in more energy depletion.
neighbors’ information requires high number of com- We, thoroughly, discussed various goal specific algo-
putation rounds. The number of computation rounds rithms to identify the impact of clustering and dynamic
might be reduced when new nodes gather informa- challenges. This paper helps to examine the vital challenges
tion about a specific node in their neighborhood to during algorithm development, like cluster stability and data
join the network. Therefore, algorithms need to be loss issues, cluster-head election where energy consumption
developed to optimize computation rounds during is the consideration, uniform load balancing, efficient net-
neighbor nodes discovery. work maintenance, and role of heterogeneous network to
increase network lifetime. Each goal specific cluster scheme
(v) The key challenges in mobile aware nodes are to select
has its own scenario and objective. In the general con-
the accurate CH to retain network for long time.
sideration, for any goal specific scenario, it is important
Dynamic strategies need to be developed to exchange
to consider control overhead and network maintenance.
reliable information among the neighboring CH elec-
With this survey, readers can design efficient goal specific
tion.
algorithms, bearing in mind other important considerations
too. We have, similarly, highlighted the open issues for future
Others research in cluster-based WSN.

(i) With the increasing sensor nodes, the probability Competing Interests
of the redundant information also increases. Up
to some extent, redundant information is worthy The authors declare no conflict of interests.
for data reliability. However, redundant data wastes
sensor energy. Thus, a trade-off is required between Acknowledgments
redundancy minimization and data reliability.
This work is partially supported by grants GUP Tier 1, 2014-
(ii) More research is recommended to address QoS issues
2015, with Vote no. 05H61, GUP Tier 1 with Vote no. 11H39,
in cluster-based WSNs. This is mainly considered
2015–2017, and Malaysia-Japan and International Institute of
in all real time applications domain. Emergency
Technology and (MJIIT) of Universiti Teknologi Malaysia
monitoring in medical care is a real time application
(UTM) Research Grant with Vote no. 4J044, Ministry of
where data is required without any delay due to path
Higher Education (MoHE), 2012–2017.
loss or any other factors.

References
7. Conclusions
[1] M. A. Razzaque, C. Bleakley, and S. Dobson, “Compression in
For the best use of cluster-based WSN, a robust cluster-based wireless sensor networks: a survey and comparative evaluation,”
WSN algorithm is an absolute requirement. In this paper, a ACM Transactions on Sensor Networks, vol. 10, no. 1, article 5,
complete study on the latest cluster-based WSN with their 2013.
qualities and impediments is presented. Acknowledging the [2] Z. Han, J. Wu, J. Zhang, L. Liu, and K. Tian, “A general
restrictions of non-cluster-based architecture, cluster-based self-organized tree-based energy-balance routing protocol for
architecture is a better answer for WSN. Cluster divides the wireless sensor network,” IEEE Transactions on Nuclear Science,
large distributed networks into groups to reduce overhead. vol. 61, no. 2, pp. 732–740, 2014.
Thus, it makes it possible to retain the features of small [3] E. M. Belding-Royer, “Hierarchical routing in ad hoc mobile
network in dense network. networks,” Wireless Communications and Mobile Computing,
Intra- and intercluster routing, control message, domino vol. 2, no. 5, pp. 515–532, 2002.
effects, computation round, communication complexity, [4] J. I. Bangash, A. H. Abdullah, M. H. Anisi, and A. W. Khan, “A
node type, basic control, cluster maintenance, cluster-head survey of routing protocols in wireless body sensor networks,”
Sensors, vol. 14, no. 1, pp. 1322–1357, 2014.
election, ID-based heuristic, Degree-based heuristic and
collaborative cover heuristic, and weight based heuristic are [5] D. Karaboga, S. Okdem, and C. Ozturk, “Cluster based wireless
sensor network routing using artificial bee colony algorithm,”
the primary performance metrics that have been recognized
Wireless Networks, vol. 18, no. 7, pp. 847–860, 2012.
to assess the cluster-based WSN.
[6] P. Jiang, Y. Xu, and F. Wu, “Node self-deployment algorithm
Communication cost is directly proportional to sensor based on an uneven cluster with radius adjusting for underwater
energy. Therefore, more communication cost results in more sensor networks,” Sensors, vol. 16, no. 1, p. 98, 2016.
energy depletion. An efficient maintenance mechanism over- [7] P. Jiang, J. Liu, B. Ruan, L. Jiang, and F. Wu, “A new node
comes the problem of reclustering. Single-hop intercluster deployment and location dispatch algorithm for underwater
communication gives good results as compared to multihop sensor networks,” Sensors, vol. 16, no. 1, p. 82, 2016.
intercluster communication in terms of energy depletion, [8] E. Valero, A. Sivanathan, F. Bosché, and M. Abdel-Wahab,
while multihop intercluster communication provides good “Musculoskeletal disorders in construction: a review and a
results as compared to single-hop intercluster communica- novel system for activity tracking with body area network,”
tion in terms of network scalability. Control overhead is also Applied Ergonomics, vol. 54, pp. 120–130, 2016.
22 International Journal of Distributed Sensor Networks

[9] J. Zymunt and J. Y. H. Haas, “Gossip-based ad hoc routing,” Wireless Communications and Networking Conference (WCNC
in Proceedings of the 21st Annual Joint Conference of the IEEE ’09), Budapest, Hungary, April 2009.
Computer and Communications Societies (INFOCOM ’02), vol. [25] O. Boyinbode, H. Le, and M. Takizawa, “A survey on clustering
3, pp. 1707–1716, IEEE, 2002. algorithms for wireless sensor networks,” International Journal
[10] C. Intanagonwiwat, R. Govindan, D. Estrin, J. Heidemann, and of Space-Based and Situated Computing, vol. 1, no. 2/3, pp. 130–
F. Silva, “Directed diffusion for wireless sensor networking,” 136, 2011.
IEEE/ACM Transactions on Networking, vol. 11, no. 1, pp. 2–16, [26] L. M. Arboleda C and N. Nasser, “Comparison of clustering
2003. algorithms and protocols for wireless sensor networks,” in Pro-
[11] Y. Pang, Y. Guo, X. Xue, and C. F. Martin, “A POMDP based ceedings of the Canadian Conference on Electrical and Computer
routing model to enhance directed diffusion in wireless sensor Engineering (CCECE ’06), pp. 1787–1792, Ottawa, Canada, May
networks,” in Proceedings of the International Conference on 2006.
Control Engineering and Communication Technology (ICCECT [27] M. Haneef and D. Zhongliang, “Design challenges and compar-
’13), pp. 180–183, IEEE Computer Society, 2013. ative analysis of cluster based routing protocols used in wireless
[12] D. Braginsky and D. Estrin, “Rumor routing algorthim for sensor networks for improving network life time,” Advances in
sensor networks,” in Proceedings of the 1st ACM International Information Sciences and Service Sciences, vol. 4, no. 1, pp. 450–
Workshop on Wireless Sensor Networks and Applications (WSNA 459, 2012.
’02), pp. 22–31, ACM, Atlanta, Ga, USA, September 2002. [28] S. Ozdemir and Y. Xiao, “Secure data aggregation in wireless
[13] P. Gupta and P. R. Kumar, “The capacity of wireless networks,” sensor networks: a comprehensive overview,” Computer Net-
IEEE Transactions on Information Theory, vol. 46, no. 2, pp. 388– works, vol. 53, no. 12, pp. 2022–2037, 2009.
404, 2000. [29] R. Rajagopalan and P. K. Varshney, “Data aggregation tech-
[14] K. Xu, X. Hong, and M. Gerla, “An ad hoc network with mobile niques in sensor networks: a survey,” IEEE Communications
backbones,” in Proceedings of the International Conference on Surveys and Tutorials, vol. 8, no. 4, pp. 48–63, 2006.
Communications (ICC ’02), pp. 3138–3143, May 2002. [30] J. He, S. Ji, Y. Pan, and Y. Li, “Greedy construction of load-
[15] A. Sinha and D. K. Lobiyal, “Performance evaluation of data balanced virtual backbones in wireless sensor networks,” Wire-
aggregation for cluster-based wireless sensor network,” Human- less Communications and Mobile Computing, vol. 14, no. 7, pp.
Centric Computing and Information Sciences, vol. 3, article 13, 673–688, 2014.
pp. 1–17, 2013. [31] K. Maraiya, K. Kant, and N. Gupta, “Wireless sensor network:
a review on data aggregation,” International Journal of Scientific
[16] C. E. Perkins, E. M. Royer, S. R. Das, and M. K. Marina,
& Engineering Research, vol. 2, no. 4, pp. 1–7, 2011.
“Performance comparison of two on-demand routing protocols
for ad hoc networks,” IEEE Personal Communications, vol. 8, no. [32] S. Lee, M. Younis, and M. Lee, “Connectivity restoration
1, pp. 16–28, 2001. in a partitioned wireless sensor network with assured fault
tolerance,” Ad Hoc Networks, vol. 24, pp. 1–19, 2015.
[17] N. Vlajic and D. Xia, “Wireless sensor networks: to cluster or not
to cluster?” in Proceedings of the International Symposium on a [33] S. H. Lee, S. Lee, H. Song, and H. S. Lee, “Gradual cluster head
World of Wireless, Mobile and Multimedia Networks (WoWMoM election for high network connectivity in large-scale sensor
’06), pp. 258–266, IEEE Computer Society, Buffalo-Niagara networks,” in Proceedings of the 13th International Conference on
Falls, NY, USA, June 2006. Advanced Communication Technology: Smart Service Innovation
through Mobile Interactivity (ICACT ’11), pp. 168–172, February
[18] J. Y. Yu and P. H. J. Chong, “A survey of clustering schemes 2011.
for mobile ad hoc networks,” IEEE Communications Surveys &
[34] M. Yu, H. Mokhtar, and M. Merabti, “A survey of network man-
Tutorials, vol. 7, no. 1, pp. 32–48, 2005.
agement architecture in wireless sensor network,” in Proceedings
[19] X. Liu, “A survey on clustering routing protocols in wireless of the 6th Annual PostGraduate Symposium on The Convergence
sensor networks,” Sensors, vol. 12, no. 8, pp. 11113–11153, 2012. of Telecommunications, Networking and Broadcasting, 2006.
[20] A. A. Abbasi and M. Younis, “A survey on clustering algorithms [35] M. Younis, I. F. Senturk, K. Akkaya, S. Lee, and F. Senel,
for wireless sensor networks,” Computer Communications, vol. “Topology management techniques for tolerating node failures
30, no. 14-15, pp. 2826–2841, 2007. in wireless sensor networks: a survey,” Computer Networks, vol.
[21] P. Kumarawadu, D. J. Dechene, M. Luccini, and A. Sauer, 58, no. 1, pp. 254–283, 2014.
“Algorithms for node clustering in wireless sensor networks: a [36] N. M. Freris, H. Kowshik, and P. R. Kumar, “Fundamentals
survey,” in Proceedings of the 4th International Conference on of large sensor networks: connectivity, capacity, clocks, and
Information and Automation for Sustainability (ICIAFS ’08), pp. computation,” Proceedings of the IEEE, vol. 98, no. 11, pp. 1828–
295–300, December 2008. 1846, 2010.
[22] B. P. Deosarkar, N. S. Yadav, and R. P. Yadav, “Clusterhead [37] A.-F. Liu, P.-H. Zhang, and Z.-G. Chen, “Theoretical analysis
selection in clustering algorithms for wireless sensor networks: of the lifetime and energy hole in cluster based wireless sensor
a survey,” in Proceedings of the International Conference on networks,” Journal of Parallel and Distributed Computing, vol.
Computing, Communication and Networking (ICCCN ’08), pp. 71, no. 10, pp. 1327–1355, 2011.
1–8, IEEE, St. Thomas, Virgin Islands, USA, December 2008. [38] G. Chen, C. Li, M. Ye, and J. Wu, “An unequal cluster-
[23] C. Wei, J. Yang, Y. Gao, and Z. Zhang, “Cluster-based routing based routing protocol in wireless sensor networks,” Wireless
protocols in wireless sensor networks: a survey,” in Proceedings Networks, vol. 15, no. 2, pp. 193–207, 2009.
of the International Conference on Computer Science and Net- [39] A. A. Aziz, Y. A. Sekercioglu, P. Fitzpatrick, and M. Ivanovich, “A
work Technology (ICCSNT ’11), pp. 1659–1663, IEEE, Harbin, survey on distributed topology control techniques for extending
China, December 2011. the lifetime of battery powered wireless sensor networks,” IEEE
[24] C. Jiang, D. Yuan, and Y. Zhao, “Towards clustering algorithms Communications Surveys & Tutorials, vol. 15, no. 1, pp. 121–144,
in wireless sensor networks-a survey,” in Proceedings of the IEEE 2013.
International Journal of Distributed Sensor Networks 23

[40] Y. Liao, H. Qi, and W. Li, “Load-balanced clustering algorithm [55] R. C. Carrano, D. Passos, L. C. S. Magalhaes, and C. V. N. Albu-
with distributed self-organization for wireless sensor networks,” querque, “Survey and taxonomy of duty cycling mechanisms in
IEEE Sensors Journal, vol. 13, no. 5, pp. 1498–1506, 2013. wireless sensor networks,” IEEE Communications Surveys and
[41] D. Sahin, V. C. Gungor, T. Kocak, and G. Tuna, “Quality-of- Tutorials, vol. 16, no. 1, pp. 181–194, 2014.
service differentiation in single-path and multi-path routing for [56] C.-C. Chiang, H.-K. Wu, W. Liu, and M. Gerla, “Routing
wireless sensor network-based smart grid applications,” Ad Hoc in clustered multihop, mobile wireless networks with fading
Networks, vol. 22, pp. 43–60, 2014. channel,” in Proceedings of the IEEE Singapore International
[42] D. I. Tapia, R. S. Alonso, Ó. Garcı́a, F. de la Prieta, and B. Conference on Networks (SICON ’97), 1997.
Pérez-Lancho, “Cloud-IO: cloud computing platform for the [57] O. Dagdeviren, K. Erciyes, and S. Tse, “Semi-asynchronous and
fast deployment of services over wireless sensor networks,” distributed weighted connected dominating set algorithms for
in 7th International Conference on Knowledge Management wireless sensor networks,” Computer Standards and Interfaces,
in Organizations: Service and Cloud Computing, vol. 172 of vol. 42, pp. 143–156, 2015.
Advances in Intelligent Systems and Computing, pp. 493–504, [58] S. Tahouri, R. E. Atani, A. H. Karbasi, and Y. Deldjoo, “Applica-
Springer, Berlin, Germany, 2013. tion of connected dominating sets in wildfire detection based on
[43] J. Wu and H. Li, “On calculating connected dominating set for wireless sensor networks,” International Journal of Information
efficient routing in ad hoc wireless networks,” in Proceedings Technology, Communications and Convergence, vol. 3, no. 2, pp.
of the 3rd International Workshop on Discrete Algorithms and 139–160, 2015.
Methods for Mobile Computing and Communications, pp. 7–14, [59] A. Ahmad, S. Jabbar, A. Paul, and S. Rho, “Mobility aware
ACM, Seattle, Wash, USA, August 1999. energy efficient congestion control in mobile wireless sensor
[44] P.-J. Wan, K. M. Alzoubi, and O. Frieder, “Distributed construc- network,” International Journal of Distributed Sensor Networks,
tion of connected dominating set in wireless ad hoc networks,” vol. 2014, Article ID 530416, 13 pages, 2014.
in Proceedings of the 35th Annual Hawaii International Confer- [60] D. C. Hoang, P. Yadav, R. Kumar, and S. K. Panda, “Real-time
ence on System Sciences (HICSS ’02), pp. 1597–1604, June 2002. implementation of a harmony search algorithm-based clus-
[45] R. Misra and C. Mandal, “Minimum connected dominating tering protocol for energy-efficient wireless sensor networks,”
set using a collaborative cover heuristic for ad hoc sensor net- IEEE Transactions on Industrial Informatics, vol. 10, no. 1, pp.
works,” IEEE Transactions on Parallel and Distributed Systems, 774–783, 2014.
vol. 21, no. 3, pp. 292–302, 2010. [61] P. Shrivastava and S. B. Pokle, “Energy efficient scheduling
[46] M. Cardei, X. Cheng, X. Cheng, and D.-Z. Du, “Connected strategy for data collection in wireless sensor networks,” in Pro-
domination in multihop ad hoc wireless networks,” in Proceed- ceedings of the International Conference on Electronic Systems,
ings of the 6th Joint Conference on Information Sciences (JCIS Signal Processing, and Computing Technologies (ICESC ’14), pp.
’02), pp. 251–255, March 2002. 170–173, January 2014.
[47] B. Zhang, R. Simon, and H. Aydin, “Harvesting-aware energy [62] N. Kaur Kapoor, S. Majumdar, and B. Nandy, “Techniques
management for time-critical wireless sensor networks with for allocation of sensors in shared wireless sensor networks,”
joint voltage and modulation scaling,” IEEE Transactions on Journal of Networks, vol. 10, no. 1, pp. 15–28, 2015.
Industrial Informatics, vol. 9, no. 1, pp. 514–526, 2013. [63] S. Deng, J. Li, and L. Shen, “Mobility-based clustering protocol
[48] X.-Y. Li, Y. Wang, and Y. Wang, “Complexity of data collection, for wireless sensor networks with mobile nodes,” IET Wireless
aggregation, and selection for wireless sensor networks,” IEEE Sensor Systems, vol. 1, no. 1, pp. 39–47, 2011.
Transactions on Computers, vol. 60, no. 3, pp. 386–399, 2011. [64] V. Singhal and S. Suri, “A comparative study of hierarchical
[49] P. Huang, L. Xiao, S. Soltani, M. W. Mutka, and N. Xi, “The routing protocols in wireless sensor networks,” in Proceedings
evolution of MAC protocols in wireless sensor networks: a of the 2nd IEEE International Conference on Computing for
survey,” IEEE Communications Surveys & Tutorials, vol. 15, no. Sustainable Global Development (INDIACom ’15), pp. 1018–1023,
1, pp. 101–120, 2013. New Delhi, India, March 2015.
[50] L. A. Villas, A. Boukerche, H. A. B. F. De Oliveira, R. B. De [65] D. Kumar, “Performance analysis of energy efficient clustering
Araujo, and A. A. F. Loureiro, “A spatial correlation aware protocols for maximising lifetime of wireless sensor networks,”
algorithm to perform efficient data collection in wireless sensor IET Wireless Sensor Systems, vol. 4, no. 1, pp. 9–16, 2014.
networks,” Ad Hoc Networks, vol. 12, no. 1, pp. 69–85, 2014. [66] S. Vasudevan, J. Kurose, and D. Towsley, “Design and analysis
[51] L. Wang, J. Yang, Y. Lin, and W. Lin, “Keeping desired QoS by of a leader election algorithm for mobile ad hoc networks,”
a Partial coverage algorithm for cluster-based wireless sensor in Proceedings of the 12th IEEE International Conference on
networks,” Journal of Networks, vol. 9, no. 12, pp. 3221–3229, Network Protocols (ICNP ’04), pp. 350–360, IEEE, Berlin,
2014. Germany, October 2004.
[52] P. T. A. Quang and D.-S. Kim, “Throughput-aware routing for [67] S. Balaji and N. Revathi, “An efficient heuristic for the min-
industrial sensor networks: application to ISA100.11a,” IEEE imum connected dominating set problem on ad hoc wireless
Transactions on Industrial Informatics, vol. 10, no. 1, pp. 351–363, networks,” Proceedings of WASET, vol. 68, pp. 1045–1051, 2012.
2014. [68] P. Jiang, J. Liu, F. Wu et al., “Node deployment algorithm for
[53] S. K. Gupta, N. Jain, and P. Sinha, “Energy efficient clustering underwater sensor networks based on connected dominating
protocol for minimizing cluster size and inter cluster communi- set,” Sensors, vol. 16, no. 3, p. 388, 2016.
cation in heterogeneous wireless sensor network,” International [69] K. Haseeb, K. A. Bakar, A. H. Abdullah, and A. Ahmed,
Journal of Advanced Research in Computer and Communication “Grid based cluster head selection mechanism for wireless
Engineering, vol. 2, no. 8, 2013. sensor network,” Telkomnika (Telecommunication Computing
[54] S. Sharma and S. K. Jena, “Cluster based multipath routing pro- Electronics and Control), vol. 13, no. 1, pp. 269–276, 2015.
tocol for wireless sensor networks,” ACM SIGCOMM Computer [70] R. Asgarnezhad and J. A. Torkestani, “A survey on backbone
Communication Review, vol. 45, no. 2, pp. 14–20, 2015. formation algorithms for Wireless Sensor Networks: (a new
24 International Journal of Distributed Sensor Networks

classification),” in Proceedings of the Australasian Telecommuni- [85] S. Guha and S. Khuller, “Approximation algorithms for con-
cation Networks And Applications Conference (ATNAC ’11), pp. nected dominating sets,” Algorithmica, vol. 20, no. 4, pp. 374–
1–4, Melbourne, Australia, November 2011. 387, 1998.
[71] Z. Liu, B. Wang, and L. Guo, “A survey on connected dominat- [86] D. Du and X. Hu, Steiner Tree Problems in Computer Communi-
ing set construction algorithm for wireless sensor networks,” cation Networks, World Scientific, 2008.
Information Technology Journal, vol. 9, no. 6, pp. 1081–1092, [87] X. Hong, K. Xu, and M. Gerla, “Scalable routing protocols for
2010. mobile ad hoc networks,” IEEE Network, vol. 16, no. 4, pp. 11–21,
[72] G. S. Kumar, M. V. Vinu Paul, G. Athithan, and K. P. Jacob, 2002.
“Routing protocol enhancement for handling node mobility in [88] N. S. Yadav, B. P. Deosarkar, and Y. Yadav, “A low control
wireless sensor networks,” in Proceedings of the IEEE Region 10 overhead cluster maintenance scheme for mobile ad hoc NET-
Conference (TENCON ’08), Hyderabad, India, November 2008. works (MANETs),” International Journal of Recent Trends in
[73] S. Deng, J. Li, and L. Shen, “Mobility-based clustering protocol Engineering, vol. 1, no. 1, pp. 100–104, 2010.
for wireless sensor networks with mobile nodes,” IEEE Xplore: [89] J. Uchida, A. K. M. M. Islam, Y. Katayama, W. Chen, and K.
IET Wireless Sensor Systems, vol. 1, no. 1, pp. 39–47, 2011. Wada, “Construction and maintenance of a novel cluster-based
[74] A. W. Khan, A. H. Abdullah, M. A. Razzaque, and J. I. Bangash, architecture for ad hoc sensor networks,” Journal of Ad Hoc and
“VGDRA: a virtual grid-based dynamic routes adjustment Sensor Wireless Networks, vol. 6, no. 1-2, pp. 1–31, 2008.
scheme for mobile sink-based wireless sensor networks,” IEEE [90] A. K. M. M. Islam, K. Wada, J. Uchida, and W. Chen, “A better
Sensors Journal, vol. 15, no. 1, pp. 526–534, 2015. dynamic cluster-based structure of wireless sensor network for
[75] A. W. Khan, A. H. Abdullah, M. Abdur Razzaque, J. I. Bangash, efficient routing,” International Journal of Innovative Comput-
and A. Altameem, “VGDD: a virtual grid based data dissemi- ing, Information and Control, vol. 8, no. 10, pp. 6747–6760, 2012.
nation scheme for wireless sensor networks with mobile sink,” [91] S. Basagni, M. Mastrogiovanni, and C. Petrioli, “A performance
International Journal of Distributed Sensor Networks, vol. 2015, comparison of protocols for clustering and backbone formation
Article ID 890348, 17 pages, 2015. in large scale ad hoc networks,” in Proceedings of the IEEE
[76] D.-S. Kim and Y.-J. Chung, “Self-organization routing protocol International Conference on Mobile Ad-Hoc and Sensor Systems,
supporting mobile nodes for wireless sensor network,” in Pro- pp. 70–79, October 2004.
ceedings of the 1st International Multi-Symposiums on Computer [92] A. M. Islam, Y. Katayama, W. Chen, and W. Wada, “A novel
and Computational Sciences (IMSCCS ’06), vol. 2, pp. 622–626, cluster-based architecture and routing protocols for dynamic
Hangzhou, China, June 2006. ad-hoc radio networks,” Journal of Electrical Engineering, The
[77] Y. Luo, W. Zhang, and Y. Hu, “A new cluster based routing Institution of Engineers, Bangladesh, vol. EE33, no. 1-2, 2006.
protocol for VANET,” in Proceedings of the 2nd International [93] P. Jiang, J. Liu, and F. Wu, “Node non-uniform deployment
Conference on Networks Security, Wireless Communications and based on clustering algorithm for underwater sensor networks,”
Trusted Computing (NSWCTC ’10), pp. 176–180, April 2010. Sensors, vol. 15, no. 12, pp. 29997–30010, 2015.
[78] O. Younis and S. Fahmy, “HEED: a hybrid, energy-efficient, [94] V. Loscri, G. Morabito, and S. Marano, “A two-levels hierarchy
distributed clustering approach for ad hoc sensor networks,” for low-energy adaptive clustering hierarchy (TL-LEACH),” in
IEEE Transactions on Mobile Computing, vol. 3, no. 4, pp. 366– Proceedings of the IEEE 62nd Vehicular Technology Conference
379, 2004. (VTC ’05), pp. 1809–1813, IEEE, 2005.
[79] P. Ding, J. Holliday, and A. Celik, “Distributed energy-efficient [95] S. Soro and W. B. Heinzelman, “Prolonging the lifetime of
hierarchical clustering for wireless sensor networks,” in Proceed- wireless sensor networks via unequal clustering,” in Proceedings
ings of the IEEE International Conference on Distributed Com- of the 19th IEEE International Parallel and Distributed Processing
puting in Sensor Networks, pp. 322–339, Santa Clara University, Symposium (IPDPS ’05), April 2005.
July 2005. [96] J. I. Bangash, A. W. Khan, and A. H. Abdullah, “Data-centric
[80] J. Wu, F. Dai, M. Gao, and I. Stojmenovic, “On calculating routing for intra wireless body sensor networks,” Journal of
power-aware connected dominating sets for efficient routing Medical Systems, vol. 39, no. 9, pp. 1–13, 2015.
in ad hoc wireless networks,” Journal of Communications and [97] J. I. Bangash, A. H. Abdullah, M. Abdur Razzaque, and A. W.
Networks, vol. 4, no. 1, pp. 59–70, 2002. Khan, “Reliability aware routing for intra-wireless body sensor
[81] T. Ohta, S. Inoue, and Y. Kakuda, “An adaptive multihop networks,” International Journal of Distributed Sensor Networks,
clustering scheme for highly mobile ad hoc networks,” in vol. 2014, Article ID 786537, 10 pages, 2014.
Proceedings of the 6th International Symposium on Autonomous [98] J. I. Bangash, A. H. Abdullah, A. W. Khan, M. A. Razzaque,
Decentralized Systems (ISADS ’03), pp. 293–300, April 2003. and R. Yusof, “Critical Data Routing (CDR) for intra wireless
[82] Y. Deng and Y. Hu, “A load balance clustering algorithm body sensor networks,” TELKOMNIKA, vol. 13, no. 1, pp. 181–
for heterogeneous wireless sensor networks,” in Proceedings 192, 2015.
of the International Conference on E-Product E-Service and
E-Entertainment (ICEEE ’10), pp. 1–4, IEEE, Henan, China,
November 2010.
[83] Y.-C. Liang, K.-C. Chen, G. Y. Li, and P. Mähönen, “Cognitive
radio networking and communications: an overview,” IEEE
Transactions on Vehicular Technology, vol. 60, no. 7, pp. 3386–
3407, 2011.
[84] D. E. Goldberg and J. H. Holland, “Genetic algorithms and
machine learning,” Machine Learning, vol. 3, no. 2, pp. 95–99,
1988.