Applied Soft Computing Journal 80 (2019) 154–166

Contents lists available at ScienceDirect

Applied Soft Computing Journal

journal homepage: www.elsevier.com/locate/asoc

Energy-Aware and Density-Based Clustering and Relaying Protocol

(EA-DB-CRP) for gathering data in wireless sensor networks
∗
Khalid A. Darabkh a , , Saja M. Odetallah a , Zouhair Al-qudah b , Ala’ F. Khalifeh c ,
Mohammad M. Shurman d
a
Department of Computer Engineering, The University of Jordan, Amman, 11942, Jordan
b
Department of Communications Engineering, Al-Hussein Bin Talal University, Ma’an 20, Jordan
c
Electrical Engineering and Information Technology, German Jordanian University, Madaba 11180, Jordan
d
Department of Network Engineering and Security, Jordan University of Science and Technology, Irbid 22110, Jordan

highlights

• New energy-aware density-based clustering and routing protocol is proposed for WSN.
• Closed-form expressions are experimentally proposed to describe how to optimally partition the network field.
• New cluster-head and relay-node selections algorithms are proposed.
• Closed-form expressions are derived to deactivate sensor nodes in dense clusters.
• Efficient cluster merge algorithm is proposed to maintain balancing the clusters.

article info a b s t r a c t

Article history: In this paper, we propose a novel energy-aware and density-based clustering and routing protocol
Received 14 September 2018 (EA-DB-CRP) for gathering data in wireless sensor networks which basically aims at distributing the
Received in revised form 5 March 2019 load among available sensor nodes which in turn balances the energy consumption in the network
Accepted 12 March 2019
and consequently elongates the network lifetime. More precisely, we introduce a network model that
Available online 15 March 2019
ends up of having empirical expressions that describe how to partition the network field efficiently
Keywords: into equal-size layers and sub-layers. In each sub-layer, the role of cluster head is pivoted among
Wireless sensor networks all cluster individuals, in a round robin fashion, that are sorted in a list in a descending order based
Dense clusters on an extremely effective cluster head weight. Additionally, there are a minimum number of cluster
Sensor deactivation members maintained to guarantee the feasibility of clusters being formed and this is through proposing
Cluster head weight
a cluster merge algorithm. Not only to this extent, but rather, we maintain the consideration of network
Relay node weight
density over created clusters to balance them and subsequently prolong the network lifetime. Lastly,
an effective relaying algorithm is proposed in which cluster heads get aware of those sensor nodes
located in a layer ahead toward the base station along with their relay-node weights whereas each
cluster head picks the relay node that has the highest weight. Our proposed protocol is evaluated
through conducting various MATLAB simulations. Strikingly, results demonstrated that our proposed
protocol has a momentous change against other related works regarding network lifetime and energy
use.
© 2019 Elsevier B.V. All rights reserved.

1. Introduction the infrastructure-less wireless networks to keep pace with this

huge usage of wireless mobiles and allow users to communicate
The computer networking field developed rapidly in the last easily [3–5]. The immense advancement in wireless network and
decade where it is noticeable that there are rapid growth in microelectronic structures drove the analysts to develop multi-
the number of mobile devices and their users’ dependency on functional devices that are described by having low processing
power, memory, cost, and capabilities. They are described by
the internet which make these devices a vital part of everyday’s
wireless sensor nodes where associating them wirelessly yields
life [1,2]. This requires a great attention to the development of
what is so called wireless sensor network (WSN) [6–9]. These
sensors are conveyed over a geological zone to detect information
∗ Corresponding author. and send it wirelessly to a control cell, called base station to
E-mail address: [email protected] (K.A. Darabkh). be then analyzed and subsequently solve many important design

https://doi.org/10.1016/j.asoc.2019.03.025
1568-4946/© 2019 Elsevier B.V. All rights reserved.
 K.A. Darabkh, S.M. Odetallah, Z. Al-qudah et al. / Applied Soft Computing Journal 80 (2019) 154–166 155

issues related to the many fields out of which health care, military • Proposing a very effective data traffic relaying algorithm in
surveillance, and environment [10–13]. Of interest is that these which a relay node weight is considered for every sensor
sensor nodes have a few constraints and above all power, mem- node to help the cluster heads, located in the adjacent
ory stockpiling, and transmission capacity [14–17]. Along these farther layer (with reference to the base station), in selecting
lines, building up a protocol for sending detected information to relay nodes that have the highest weights and this definitely
the base station which keeps sensor nodes alive as far as might changes every round.
be feasible is a crucial research issue [18,19]. • Conducting several sets of experiments to assess our pro-
Generally and according to the WSN structure, routing pro- posed protocol performance in the perspectives of network
tocols can be classified into flat-based, hierarchical (or cluster)- lifetime and energy utilization without neglecting the com-
based, and location-based protocols [20–22]. Due to the nature parisons with closely related approaches.
of nodes and their robustness, cluster-based routing protocols are
the most common technique used in WSNs [23]. In such proto- It merits saying that prior proposed algorithms, which are sum-
cols, sensor nodes are grouped into clusters where each cluster marized in the next section, demonstrated to affect the network
consists of normal nodes and a cluster head [24]. Usually, the lifespan significantly whereas the simulation results indicated
cluster head is chosen according to some specific criteria which that our proposed protocol has an exceptional change against
mostly take into consideration energy, memory, and bandwidth other firmly related works. Matter of fact, we introduced some
of sensor nodes [23]. It is basically in charge of collecting data precursory results of this work in [28].
from its members, aggregating it, and ultimately relaying it to The rest of this paper is organized as follows: Section 2 sum-
the base station [25]. Cluster heads, nearest to the base station, marizes related works. Section 3 details the proposed protocol.
tend to exhaust their power quicker than others because of their Section 4 illustrates the simulation experiments and discusses
extra task of relaying data traffic toward the base station [26]. the simulation results of our proposed protocol along with those
Intriguingly, if the energy consumption is even among all sensors obtained from relevant works. Section 5 concludes our work.
that are uniformly distributed, then the energy hole is averted
and the network lifetime is improved. Therefore, proposing pro- 2. Related works
tocols that target at maximizing the network lifetime is extremely
challenging [23–27]. Many new cluster-based algorithms were proposed for the
reason of managing the energy consumption and accordingly
1.1. Our methodology and contributions extending the network lifetime [29–44]. In this section, we sum-
marize some of the previous works and especially those quite
To this end, we propose, in this paper, a new Energy-aware related ones. Intriguingly, the father of cluster-based protocols
and Density-based Clustering and Routing Protocol (EA-DB-CRP) is the Low Energy Adaptive Clustering Hierarchy (LEACH) which
for gathering data in wireless sensor networks which primarily was proposed in [36]. In LEACH protocol, the network lifetime
aims at reducing the setup overhead, distributing the power is measured in something called rounds where each round is
evenly among sensor nodes, as well as reducing long-distance composed of setup and steady-state phases. In the setup phase,
communications. The contributions of this manuscript can be cluster heads are elected while in the steady-state phase, normal
summarized as follows bearing in mind that a graphical abstract nodes send their data to their corresponding cluster heads which
of this work is shown in Fig. 1: in turn aggregate and accordingly transmit such data to the base
station. Rather charmingly, LEACH is a well-known distributed
• Proposing a generic network model, in which empirical ex- protocol where every node, in the setup phase of every round,
pressions for the division of this model are proposed, which generates a random number between 0 and 1 and accordingly
aims at ensuring the use of a free space channel propaga- compares it with a formulated probability of that node to be a
tion model rather than two-ray channel propagation model, cluster head. Therefore, the higher probability obtained by that
thereby maximizing the network lifetime. node, the higher opportunity of it to serve as a cluster head.
• Proposing a yet efficient and energy-aware cluster head Tellingly, the exclusion of node’s residual energy and location
selection algorithm among all clusters established. In par- in the aforementioned probability makes LEACH protocol de-
ticular, each node in any cluster is assigned, by its detached ficient. Centralized LEACH-C (LEACH-C) is a revised version of
cluster head, a cluster head weight that is parameterized by LEACH [37]. In this protocol, the base station is in charge of the
the remaining energy of this node, its location among all operations required in the setup phase of every round. Moreover,
other nodes in the sub-layer, as well as its distance to the the optimal numbers of cluster heads along with their best lo-
base station. Accordingly, a detached cluster head maintains cations are found through running simulated annealing or taboo
a list of weights for all its members which in turn sorts and search algorithms.
announces them to its members which result in having all The Threshold-based LEACH (T-LEACH) protocol was proposed
nodes informed by which nodes will serve as cluster heads in [38]. This protocol mainly focuses on delaying the execu-
in the next rounds until the current batch ends. tion of the network setup phase until a cluster head’s power
• Proposing a reliable cluster head replacement algorithm in falls below a pre-defined energy threshold. In different words,
which cluster heads, that are selected to serve in the subse- T-LEACH protocol reduces the overall energy consumption in the
quent rounds as mentioned in former bullet, will continue network through reducing the number of times the cluster head
serving as long as their energy levels do not dip below either is changed (i.e., reducing re-clustering operations) whereas the
a critical threshold or re-clustering threshold. cluster head keeps functioning as long as its power beyond a
• Maintaining a sufficient number of nodes for each cluster pre-defined threshold. The authors in [39] proposed a modified
through incorporating the use of network density where not version of T-LEACH called Modified Threshold-based Cluster Head
only each cluster density should not go beyond the network Replacement (MT-CHR) protocol in which a new probability for
density through deactivating so closer sensor nodes, but a node to serve as a cluster head in any round is proposed
also the number of cluster members should not dip below which basically agrees well with the LEACH assumptions. Fur-
a certain number through the involvement of an efficient thermore, a new energy threshold is proposed which guarantees
merge algorithm. not only avoiding any data loss, but also postponing the first
156 K.A. Darabkh, S.M. Odetallah, Z. Al-qudah et al. / Applied Soft Computing Journal 80 (2019) 154–166

Fig. 1. The graphical abstract of this work.

node death. Authors in [40] proposed a Centralized Density- and cluster report themselves as cluster heads. In addition, COCA
Threshold-based Cluster Head Replacement (C-DTB-CHR) proto- incorporates a multi-hop directing mechanism. Specifically, each
col as a centralized version of MT-CHR protocol. They also tried cluster head picks arbitrarily k heads from the adjoining layers,
to achieve a load balancing through maintaining the employment nearer to the base station, where the sensor that has the most
of almost the same number of nodes for each cluster formed. elevated power among those heads will go about as a hand-off
The authors proposed further C-DTB-CHR with Adaptive Data (relay) node for that cluster head in that round.
Distribution (C-DTB-CHR-ADD) protocol that basically extends The last firmly relevant work is the Energy-Aware Clustering
C-DTB-CHR protocol in a way that avoids the communications and Routing Protocol (EA-CRP) which was discussed in [44]. This
over long distances through the incorporation of an efficient protocol seeks for mitigating the energy consumption among
multi-hop algorithm. sensors. In EA-CRP, the authors concentrated on a rectangular
Power Efficient Gathering in Sensor Information Systems area of the network and subsequently proposed tight closed-form
(PEGASIS) was suggested in [41]. In this protocol, the nodes formulas for the network and cluster divisions. This rectangular
are formed into a chain where each sensor node transmits and zone is partitioned into layers where their widths get shorter
receives data from the closest neighbor. Nodes send data to toward the base station. In this way, the long-distance communi-
each other in a way it reaches the base station via one leader cations are avoided especially for those clusters closer to the base
head, which is chosen randomly from the chain. In PEGASIS, the station as there are extra burden on them as of relaying data.
amount of energy consumed per round is reduced due to the Each layer is further composed of a certain number of clusters
equal distribution of power consumption among all nodes. (sub-layers) where clusters within the same layer have the same
As a matter of fact, there are three quite related works to size. Moreover, the authors employed double heads (i.e., cluster
ours [42–44]. The first one is the Unequal Cluster-based Routing and leader heads) for each cluster. In particular, inside any layer,
Protocol (UCR) which was presented in [42]. In this convention, sensors transmit their detected information to their leader head
the nodes are created into unequal-estimate clusters. Literally, utilizing intra-cluster correspondence. The leader head at that
clusters, closer the base station, have smaller sizes than clus- point totals this information, packs it into a solitary packet, and
ters more distant from the base station. On the other hand, thus transfers it to the cluster head situated in a similar sub-
the authors proposed another hybrid separation to keep away layer which thus transmits it to the nearest leader head in the
from long-distance transferring between cluster heads and base adjacent layer. The cluster and leader heads are selected based
station. In particular, if the separation between a cluster head and on an election process tightened by formulas adopted in every
the base station is higher than a pre-characterized edge, at that round. Interested readers may refer to page 6 of [44].
point, it transfers the information to its adjoining cluster heads
that are situated ahead toward the base station. Else, it sends the 3. The proposed protocol
information specifically to the base station.
The second pertinent convention is the Constructing Optimal To this extent, we develop a novel protocol, namely, EA-DB-
Clustering Architecture (COCA) which was proposed in [43]. The CRP that aims at preserving the energy consumption among all
primary idea of this convention is to boost the network lifetime variable nodes. In the next subsection, protocol’s assumptions are
by constraining the energy consumption among all sensors. It introduced. Section 3.2 presents necessary details of the energy
seeks for apportioning the network into fixed-size square units and channel propagation models adopted accompanied with an
and expanding the quantity of clusters as moving toward the extensive description of the operation of the proposed protocol
units closer the base station since clusters nearer to the base through Sections 3.3–3.9.
station exhaust more power for transferring information than
more distant ones. Along these lines, the power devoured by 3.1. Assumptions
sensors is adjusted. The decision of a cluster head relies upon the
remaining power of a sensor. As it were, every sensor broadcasts The assumptions considered in our proposed protocol are
a control message which mainly contains its ID and remaining summarized as follows which are the same as those considered
power whereas sensors with the most noteworthy power in the in [42–44]:
 K.A. Darabkh, S.M. Odetallah, Z. Al-qudah et al. / Applied Soft Computing Journal 80 (2019) 154–166 157

1. Links are omnidirectional. merge mechanism is proposed that will be talked about later
2. The WSN comprises of a base station and numerous sen- in this paper. The network model of the proposed protocol is
sors. descriptively summarized in Fig. 2.
3. Nodes are arbitrarily appropriated over a rectangular re- The network lifetime of EA-DB-CRP is parceled into batches
gion. (i.e., multiple of rounds). A round can be additionally isolated into
4. All sensors conveyed in the territory are homogeneous and stages: the setup stage in which the cluster heads and hand-off
have similar handling abilities. (relay) nodes are picked and the steady-state stage in which the
5. There is a base station, which is situated outside the con- member nodes send their information to their cluster heads. The
veyed territory of the sensors, and has limitless power. cluster heads in turn aggregate received data and send it to either
6. The sensors and base station have stationary areas. the downstream relay node (the one closer to the base station) or
7. Every sensor has a remarkable identifier ID relegated to it directly to the base station. Impressively, in EA-DB-CRP, the setup
just before deployment. phase, which is considered basically as an overhead as too many
8. Batteries of sensors are non-battery-powered. As it were, control messages are generated, is required only once per batch
sensors cannot be stimulated after organization (or deploy- not per round. In other words, there might be no setup overhead
ment). for many rounds.
9. Nodes dependably have information to send amid their
designated time division multiple access slots.
3.4. Divisions of sensor network model
10. As widely known [1,38–46], computation power is rel-
atively low differentiated to the communication power.
Therefore, it is overlooked in our simulations. In order to achieve the goal of this article, we have conducted a
significant number of simulations for proposing all division’s for-
3.2. Radio energy and chanel propagation models mulas discussed below. Furthermore, these experimental closed-
form formulas can be applied to any sensor network area. There-
There are numerous energy consumption models proposed fore, when assuming that the height and width of the sensor
in the literatures. In our simulations, we have used the ra- network are H and W, respectively, then, the number of layers
dio energy and channel propagation models that are utilized in created (NL ) can be defined as:
[42–46]. In the radio energy model, the energy is expended to ⌊√ ⌋
W ×H ×N
run the electronic transmit circuit and power enhancer at the NL = + 1, (4)
transmitter side and further to run the radio electronics at the 2 × (H + W )
receiver side. where, N represents the number of nodes in the sensor network.
Transmitting q-bit message over a distance (d) requires deplet- Furthermore, the number of sub-layers (clusters) formed at the
ing energy by the radio energy model as [42–46]: first layer, which is the closest layer to the base station, can be
ETX (q, d) symbolized by NCL and accordingly expressed by:
1

Eelec × q + εfree_space_amp × q × d2 d < dcrossov er

{
N εfs M
=
Eelec × q + εMultipath_fading_amp × q × d4 d ≥ dcrossov er
(1) NCL = √ × × 2 , (5)
1
2π ε mpf dtoBS
The channel propagation models used depend on the thresh- where, M denotes for the size of the sensor network (i.e., W × H),
old distance, dcrossov er . In other words, if the distance between
and d2toBS represents the average square distance from all nodes to
transmitter and receiver, annotated as d, is less thandcrossov er ,
the base station. Therefore, the number of sub-layers (clusters) in
then the model used is the free space channel model. Otherwise,
the subsequent layers NCL can be defined as:
the multipath fading channel propagation model is used. Conse- i

quently, εmultipath_fading_amp and εfree_space_amp are the amplification

NCL = NCL − 1, 2 ≤ i ≤ NL . (6)
parameters required when using multipath fading and free space i i−1

channel propagation models, respectively. It is worth mentioning that these divisions are performed
On the other hand, receiving a q-bit message requires energy by the base station which consequently informs about them all
consumption by the receiver electronics as [42–46]: member nodes bearing in mind that all nodes should initially
ERX (q) = ERX-elec (q) = Eelec × q (2) send their locations to the base station. In specific, the base sta-
tion sends an informative control message to be then received by
where Eelec is the energy consumed by the transmitter and the all nodes in the vicinity. This message is at the form of list which
receiver to run the electronic circuits. Moreover, dcrossov er is found has the following attributes: (a) Layer ID, (b) Cluster ID, (c) Node
as follows [42–46]: ID, (d) Node’s location, and (e) Activation flag. In fact, the node’s
√
εfree_space_amp location and activation flag are necessary for the criteria of select-
dcrossov er = (3) ing the cluster head or relay nodes and activating/deactivating
εMultipath_fading_amp
some senor nodes, respectively, which both will be discussed
3.3. Network topology model shortly. Interestingly, in EA-DB-CRP, the cluster head selection
is based on a round robin fashion and this is to distribute the
In our proposed protocol, we center on rectangular-shaped energy consumption among all member nodes as it is known that
networks. Of interest to mention that this rectangular-shaped cluster heads deplete more energy than member nodes, thereby
zone is partitioned into layers that are characterized of having the achieving load balancing. In point of fact, the node that appears
same length (or size). Each layer is involved a particular number first in the list for any cluster will serve as the cluster head in
of sub-layers, called clusters. Clusters inside a layer have a similar that cluster for the current round and the one appears next will
size. Then again, clusters in different layers have diverse sizes to serve as a cluster head in that cluster for the next round and so
dodge the ‘‘power gap’’ issue. At first, the clusters in each layer on. On the other hand, the base station sorts the nodes for each
have a similar size, however subsequent to working for specific cluster in the list based on a certain weight formula, which will
rounds, the span of these clusters may vary, thus, an efficient be discussed in the cluster head selection subsection.
158 K.A. Darabkh, S.M. Odetallah, Z. Al-qudah et al. / Applied Soft Computing Journal 80 (2019) 154–166

Fig. 2. The network model of the proposed protocol.

3.5. Merging clusters algorithm that neighbor cluster head beaning in mind that this message
contains further all information collected by that neighbor cluster
For the sake of reducing the communication cost and conse- head (i.e., layer ID, cluster ID, node ID, node’s energy, node’s
quently conserving the energy efficiently, we employ a cluster location, and activation flag). Consequently, the detached cluster
merging technique which basically merges the clusters that have head updates its list and starts reselecting cluster heads based
member nodes underneath a certain merge threshold. Actually, on a certain weight formula, which will be discussed in the next
cluster merging is performed by the base station at the initial subsection.
stage of the network creation and later (i.e., after the creation of
the network) by detached cluster heads, those become incapable 3.6. Cluster heads selection algorithm
of serving any further due to their energy shortages, specifically
at the beginning of the setup phase of a new batch. In order to guarantee the even distribution of the energy
In specific, after portioning the network and before informing consumption amongst all nodes in the sensor network, the base
the member nodes, the base station examines the number of station, at the beginning of setup phase of the first batch, or
member nodes for each cluster. If it is less than four, which is the detached cluster heads select the proper cluster head for each
proposed merge threshold, then it will merge that cluster with sub-layer. Interestingly, this is done by assigning weights for all
its neighbor cluster in the same layer. Afterwards, it updates all nodes inside each sub-layer as follows: Firstly, the base station, at
member nodes of all necessary information, mentioned earlier. It the start of the first round in the first batch, or detached cluster
is interesting to mention that at the initial network operation, the heads, at the subsequent ( rounds, ) compute the average energy for
base station is in charge of maintaining this merge threshold all all nodes in a cluster i EAv g (Ci ) as follows:
over the network but after the network gets functioning, main-
∑NMNCi
taining this threshold becomes the duty of detached cluster heads En
at the beginning of a setup phase of a new batch. In other words, EAv g (Ci ) = n=1
, (7)
NMNC
in the setup phase, every member node is responsible of sending i

its energy level to its detached cluster head as a response to its where, NMNC represents the number of member nodes in a sub-
i
disjoin message, that will be discussed further shortly. Therefore, layer i, and En is the residual energy of a node n.
the detached cluster head is kept posted by those dead member Secondly, the base station, initially, or detached cluster heads,
nodes (if any) which as a result facilitates its task of merging later, compute the average distance
( between
) a node n and other
clusters. In fact, the detached cluster head will also update the member nodes in a cluster i DAv g (nCi ) as follows:
list received originally from the base station through inserting
a new field called ‘‘node’s energy’’ that is also required for the
∑NMNCi
j=1,j̸ =n DnC →jC
criteria of reselecting cluster heads and relay nodes, that will both DAv g (nCi ) = i i
, (8)
be discussed shortly. NMNC − 1
i
It cannot be left to be mentioned that any detached cluster Lastly, the cluster head weight of a node n in a cluster i(WCH
head, which decides that a merger is required for its cluster,
)
(nCi ) is formulated as follows:
will send a MEGRE_MSG to only neighboring clusters. Any cluster
head of adjacent clusters, which also decides to merge its cluster, En Lw Lw
WCH (nCi ) = + + , (9)
will immediately respond by sending a TO_DO_MEGRE_MSG to EAv g (Ci ) DAv g (nCi ) DnC →BS
i
 K.A. Darabkh, S.M. Odetallah, Z. Al-qudah et al. / Applied Soft Computing Journal 80 (2019) 154–166 159

where, DnC →BS is the distance between a node n in a cluster i and the prior batch. If so, it begins a new batch and immediately starts
i
base station, Lw represents the layer width. As shown above, the sending a disjoin query message all over the network, announcing
WCH (nCi ) is parameterized by the remaining energy of the node, a new cluster heads reselection phase.
its location among all other nodes in the sub-layer, as well as its Upon receiving a disjoin query message, every member node
distance to base station. n, of every cluster, sends its residual energy to its detached cluster
According to their weights, all nodes in the same cluster will head utilizing a broadcast range BR(nLi ) which is defined as:
be sorted in a descending manner in a list, which is maintained 
( )2

definitely by their detached cluster head, which is sent either  LH
initially from the base station to all nodes or latterly from a
√ 2
BR(nLi ) = Lw + , (11)
NCL
detached cluster head to its cluster members. As a result, the i

cluster head role will be rotated throughout the nodes in a round where Lw , LH , and NCL refer to the layer width, layer height, and
robin fashion. It is good to concentrate on the point that the i
the number of clusters in the layer Li , respectively.
detached cluster head sends its updated list in a broadcast range It deserves mentioning that as soon as any detached cluster
equals the diagonal of two layers. The main reasoning of choosing
head sends a disjoin query message, which should reach the
this range is that this list is of important use in the relay node
whole network, all cluster heads in the network become detached
selection algorithm that will be discussed in a little while. In
cluster heads. In other words, to guarantee the time synchroniza-
fact, the aforementioned list is required to largely minimize the
tion among all sensor nodes, re-clustering operation occurs for all
control overhead. Therefore, in this list, all member nodes will be
network clusters.
sorted in a descending order based on their weights, mentioned
earlier, and a cluster head rotation will be applied. It is worth
3.8. Relay nodes selection algorithm
saying that this list will be so useful for also the relay node
selection process that will be discussed shortly.
This subsection explains how the data is relayed by cluster
It is good to mention that as soon as a detached cluster head
clusters until reaching the base station. Therefore, the relay nodes
sends a disjoin message and its member nodes send their energy
are basically the cluster heads in the downstream layer (i.e., layer
levels in return, it amends its list through adding a flag called
closer to the base station). It is mentioned formerly that each
‘‘cluster head possibility’’, which accepts either 0 or 1 indicating
node in the network has an updated list, attributed by layer ID,
‘‘not being possible’’ and ‘‘being possible’’ cluster heads, respec-
cluster ID, node ID, node’s energy, node’s location, and activation
tively. In particular, the detached cluster head will check whether
flag of all member nodes in the same cluster. It is good to
the energy level of every member node dips below the critical
keep in mind that there is a cluster head rotation among cluster
threshold. If so, then it will include ‘‘0’’ in its just mentioned flag.
members which is based on round robin fashion. Furthermore,
Otherwise, ‘‘1’’ will be inserted.
any detached cluster head will send to its members an updated
In light of the above and soon as the detached clusters send
list every new batch (i.e., every setup of the first round in the new
their updated and sorted lists, nodes will serve as cluster heads
batch) where it reaches not only its cluster members, but also
in a round robin fashion excluding those ‘‘difficult’’ nodes where
those nodes in the adjacent layers. Utilizing the layer ID, every
their ‘‘cluster head possibility’’ flags are all zeros. One condition
node can determine what its relay node candidates are as relay
should be satisfied further that the activation flags of nodes, that
nodes should be in the downstream layer (i.e., the one that is
are possible to serve cluster heads, are set (i.e., in active state) in
closer to the base station). In a nutshell, any node, when its turn
that batch.
definitely comes to serve as a cluster head in a round r, knows
well from the received lists which are the cluster heads in a layer
3.7. Cluster head replacement algorithm
ahead that are serving as cluster heads in that round (i.e., r). As
every node has to serve as a cluster head in a certain round and
In EA-DB-CRP, we make sure that the cluster head energy is
it may receive more than a list from its neighboring sub-layers,
beyond a critical threshold to guarantee that the cluster head
more than a relay node candidate exists. Therefore, every node
works properly where it can receive, aggregate and transmit data.
considers a weight for every relay node candidate as follows:
This critical threshold is proposed as follows:
Lw Lw
ECriticalThreshold = EReceiv ingCost + ETransmittingCost + EaggregationCost (10) WR (mj+1 ) = 0.5 × Em j+1 + 0.2 × + 0.3 × , (12)
Dn j →m j+1 Dmj+1 →BS
At the start of every round, every cluster head checks its
energy level. If it is below than that threshold, then the cluster where WR (mj+1 )refers to the weight of a node m located in the
head cannot do its work properly and will send an announcement layer ahead (j+1). Em j+1 represents the energy of a node m located
message (i.e., disjoin query message) to begin a new batch and in the layer ahead (j+1). Dn j →m j+1 denotes for the Euclidean
enter a cluster heads reselection phase. It is good to mention that distance between the node n in a layer j (which, in other words,
we assume that all sensors are timely synchronized. Thus, this the one that puts or assigns the weights for the nodes in the layer
announcement message should reach all nodes in the network ahead j+1) and the node m in the layer j+1. Dmj+1 →BS refers to
and a complete re-clustering operation is going to occur. the Euclidean distance between the relay node candidate (node
It is worth to take into consideration that the energy of nodes m at layer j+1) and BS. Due to the use of Euclidean distances,
decreases as the number of rounds increases. According to this, the existence of node’s location in the list, mentioned earlier,
we employ a re-clustering threshold that depends primarily on becomes urgent and undeniable. Finally, Lw denotes for the layer
the initial energy of the node at the start of a batch. Choosing width, as previously used.
the best re-clustering threshold is so challenging matter but after It is noteworthy to mention that we consider factors of 0.5, 0.3,
running multiple simulations in the consideration of different and 0.2 for the candidate node’s energy, distance from current
thresholds, the one that maximizes the network lifetime (or cluster head to relay node candidate, and distance from relay
ensures the energy distribution between nodes) is found to be node candidate to base station, respectively. Indeed, these factors
50% of the node’s energy in the prior batch, thereby prolonging have been chosen after simulating different factors where these
the early deaths of cluster heads. Consequently, each cluster head have proven to be the ones that maximize the performance of the
checks whether its energy level dips below 50% of its energy in sensor network employed. Actually, if the cluster head is located
160 K.A. Darabkh, S.M. Odetallah, Z. Al-qudah et al. / Applied Soft Computing Journal 80 (2019) 154–166

Fig. 3. The main operation of the proposed protocol.

in the first layer, which is the nearest layer to the base station, • We calculate the cluster activity factor (CAF ji ) of a cluster j
then the cluster head will send its data straightforwardly to the in a layer i as follows:
base station. In other words, there is no need for any relay node NMNC
,
j
weight assignment. CAFji = (14)
After determining the weights of all its relay node candidates, NMNL
i
the corresponding node picks the node that has the highest where NMNC and NMNL represent the number of member
j i
weight to be its relay node in the downstream layer in that nodes in a cluster j and its layer i, respectively.
corresponding round (i.e., the round in which that corresponding • We only deactivate (turn off) some sensor nodes in dense
node serves as a cluster head). It is interesting to state that relay clusters where their CAF s are greater than NAF. To(calculate
)
nodes may differ in various batches which basically prevents
the number of active member nodes in a cluster j NAMNC ,
any heavy burden on such relay nodes, thereby distributing the j
we propose the following:
energy consumption and achieving efficient load balancing. ( )
NMNC × NAF
.
j
NAMNC = Floor (15)
3.9. Sensor activation/deactivation algorithm j CAFj

Because of the irregular nodes’ organization over a network re- Accordingly, the number of inactive member nodes in a cluster
j is found as follows:
gion, there may be clusters with low number of sensors. This issue
is unraveled by utilizing the consolidating (merging) mechanism, NIAMNC = NMNC − NAMNC . (16)
j j j
talked about prior. An issue may emerge; imagine a scenario in
which few clusters are thick (i.e., has a colossal number of nodes) Excitingly, the base station selects sensor nodes, which are
required to be deactivated, based on their energy levels and
and others are definitely not. To manage this issue, an interesting
distances to the nearest neighbors. As the sensor nodes initially
sensor activation /deactivation mechanism is proposed in this
have the same energy, the base station selects those ones that are
manuscript, which is illustrated as follows:
the closest to their neighbors. Afterwards, the detached cluster
heads should take this role (activation/deactivation role) and
• We calculate the normalized activity factor (NAF i ) of any
consider, based on aforementioned equations, deactivating nodes
cluster in a layer i as follows:
that have the lowest energy levels and closest to their neighbors.
1 Actually, this criterion helps much in delaying the early node
NAFi = , (13) deaths and ultimately prolonging the network lifetime. That is
NCL
i
why the activation field is necessary to be incorporated in the
where, NCL represents the number of clusters in layer i. list that is maintained by every member node.
i
 K.A. Darabkh, S.M. Odetallah, Z. Al-qudah et al. / Applied Soft Computing Journal 80 (2019) 154–166 161

Fig. 4. Detached cluster head processing block.

There is no doubt to be intrigued to know that this algorithm Table 1

is performed by detached cluster heads just before having the Simulation parameters.
cluster head weights of member nodes. In other words, the cluster Parameter Value
head selection algorithm is applied just for those nodes that are Network field 60 × 120 m
chosen, by detached cluster heads, to be activated in a corre- Base station location X = 30, Y = −90
Number of sensor nodes 100
sponding batch. The whole operation of our proposed protocol
Network density 0.014
is described in Fig. 3. Initial energy 0.5 J
εf 10 PJ/bit/m2
4. Simulation results εm 0.0013 PJ/bit/m4
Eelec 50 nJ/bit
4.1. Simulation environment EDA 5 nJ/bit
dcrossov er 87 m
Data packet size 6400 bit
The majority of the simulations have been completed utilizing Control packet size 200 bit
MATLAB. In fact, Table 1 lists the simulation parameters that are
utilized in our analyzes where they are expected as default values
unless otherwise mentioned in the discourse.
162 K.A. Darabkh, S.M. Odetallah, Z. Al-qudah et al. / Applied Soft Computing Journal 80 (2019) 154–166

Fig. 5. Number of alive nodes in opposition to rounds at a network size of 60 × 120 m.

4.2. Performance metrics all results get worse as the network functions longer. This is
quite expected since in every round, the sensor nodes drain their
The performance of our proposed protocol is evaluated with energy to send their data.
reference to the following metrics: We can further notice from Fig. 5 that our proposed protocol
achieves the best performance with regard to the number of
• Network Lifetime, where two definitions have been used alive nodes where it remains active for 800 rounds, while COCA,
which include First Node to Die (FND) and Last Node to Die UCR and EA-CRP remain active for 254, 283, 432 rounds, respec-
(LND) which are defined as follows: tively. This is due the improvements incorporated in the proposed
protocol. Firstly, EA-DB-CRP is a batch-based protocol which mit-
1. FND, which represents the time elapsed from the
igates the setup overhead and increases the energy efficiency.
initial deployment of the nodes in the entire network
Secondly, dividing the network area into equal-size layers and
until the first node dies, which is measured in rounds.
each layer into equal-size clusters (sub-layers), merging clusters
2. LND, which represents the time elapsed from the ini-
with other nearby clusters (when there is a tiny number of cluster
tial deployment of nodes in entire the network until
members), as well as applying the density technique to balance
the last node dies.
the energy consumption among cluster members all help much in
• Network Utilization, which is defined as the ratio of the prolonging the network lifetime. Thirdly, employing an efficient
energy depleted for transferring data messages to the overall mechanism for selecting the cluster heads which ends up in
network energy consumption. sorting members of a cluster in a list based on an effective cluster
head weight formula that is parameterized by node’s residual
4.3. Simulation experiments and results energy, node’s average distance from the other member nodes,
and node’s distance from the base station bearing in mind the
There are various simulation sets of experiments conducted cluster head role will be assigned to all members in the list in
to evaluate the performance of our proposed protocol. Firstly, we a descending order and round robin fashion. Finally, the use of
examine the number of the alive nodes versus the number of multi-hop routing, in which every cluster member node, when
rounds. Secondly, we study the impact of varying the network being served as a cluster head in certain round, determines the
size on the number of rounds until the last node dies. Thirdly, we relay node weights for those cluster heads in the adjacent closer
examine the network utilization against network size. Finally, we layer (with reference to the base station) in its corresponding
inquire the number of rounds until the last node to die versus round whereas it selects the one that has the highest relay node
initial energy. All results of the proposed protocol are compared weight, guarantees the use of free space channel propagation
to aforementioned related protocols, namely, UCR, COCA, and model which as a result contributes efficiently in extending the
EA-CRP bearing in mind that comprehensive discussion regarding network lifetime.
all noticed observations are provided. In what follows, these sets EA-CRP shows a superior improvement in the performance
of experiments are extensively illustrated. (especially in LND) over both UCR and COCA but worse than
EA-DB-CRP, which is our proposed protocol. In brief, EA-CRP
4.3.1. Experiments set 1: Number of alive nodes against round num- employs algorithms that not only divide the network into unequal
ber layer sizes, but also shorten the widths of the layers nearby the
As shown in Fig. 5, we investigate the number of alive nodes base station which in turn lessen the long-distance communica-
confronting the number of rounds considering our proposed pro- tions and consequently preserve the nodes’ energy. Furthermore,
tocol (EA-DB-CRP) and other related protocols (i.e., UCR, COCA, EA-CRP employs cluster and leader heads for each cluster and
and EA-CRP). In this set of experiments, we distribute 100 sensor uses a shorter broadcast range than that in UCR and COCA. Ac-
nodes randomly over a 60 × 120 m area. Nevertheless, through- tually, the leader head is located at the middle of a cluster and
out the results, we can be observed a common trend, which is, responsible of aggregating the traffic received from its cluster
 K.A. Darabkh, S.M. Odetallah, Z. Al-qudah et al. / Applied Soft Computing Journal 80 (2019) 154–166 163

Fig. 6. Number of alive node opposed to rounds at a network size of 60 × 180 m.

members while the cluster head is located at the border of a clus- other one is the link cost. As far as the first one is concerned,
ter toward the base station and basically in charge of receiving transmitting data relies on the residual energy of relay node.
the compressed traffic from its corresponding leader head. The In other words, if the energy of the relay node dips below the
residual energy is further considered when electing the cluster energy of the transmitting node, then the transmitting node will
and header heads. Additionally, it operates a multi-hop algorithm send its data straightforwardly to the base station. Genuinely, this
in which the cluster heads deliver their traffic to their adjacent consumes more energy contrasted to sending data by a relying
leader heads until reaching the base station. The major drawback node.
of this protocol is the setup overhead introduced in every round. Fig. 6 depicts the number of alive nodes in opposition to
COCA algorithm splits the network field into equally sized rounds considering COCA, UCR, EA-CRP and EA-DB-CRP where
square layers by taking the ratio of the length of the network the network area selected is 60 × 180 m. The number of nodes
to its width with the involvement of more clusters in the layers used is 152. We used this area dimension since the length is
nearer the base station. In most of our simulations, the network greater three times than the width. Therefore, COCA partitions
area in COCA is partitioned into two layers as the length of the the network into three layers where each of them is of length
network is double the width. Unfortunately, in COCA, the division 60 × 60 m, thus, it performs better than UCR. This concludes
of networks does not take into consideration the avoidance of that as the number of COCA layers increases more and more, the
communication over large distance. In other words, all nodes in broadcast range decreases more and more, thereby performing
a layer have to communicate with each other at every round for better than UCR. The EA-CRP has better performance than both
sending their energy levels where nodes with the highest residual COCA and UCR. Although a bigger network size is considered, our
energy levels announce their cluster head role which makes the proposed protocol still outperforms all of the other protocols.
communication or transmission range longer, thereby consuming
more energy. COCA employs a multi-hop routing technique in 4.3.2. Experiments set 2: Rounds till last node dies opposing network
which the transmitter nodes choose the relay nodes randomly size
from selected multiple cluster heads in a layer ahead. Indeed, this In this set of experiments and as sketched in Fig. 7, we study
simple technique introduces a problem as it may select not only a the impact of varying the network size on the network lifetime
farther relay node which brings on expanding the long-distance when it is expressed by LND performance metric and considering
communications, but also those that might have critical energy UCR, COCA, EA-CRP and our proposed protocol. It can be seen
levels, thereby having possible data loss. from this figure that the network lifetime of all protocols de-
UCR performs better than COCA as it has a pre-determined creases as the network area increases which is quite expected as
and fixed broadcast range (R0 ) used in sending control messages. increasing the network size prompts to increasing the commu-
Actually, the R0 , used in UCR, is less than that used in COCA nication distance between nodes which gives rise to consuming
where is utilized for selecting multiple candidate cluster heads more energy. Among all aforementioned protocols, the proposed
to participate in the cluster head election process. Therefore, as protocol performs the best all over network sizes considered.
R0 increases, more tentative clusters are expected to participate in For instance, in EA-DB-CRP and considering the network size of
the election process. Moreover, each sensor node picks a random 60 × 120 m, the network stays functioning for 802 rounds while
number between 0 and 1 and if that random number is less in COCA, UCR and EA-CRP, it remains active for 254, 330 and 432,
than a pre-defined threshold, then that sensor node will be a respectively. Rather charmingly, Table 2 provides numeric values
candidate cluster head even if its energy is critical. Furthermore, for the LND, performance metric, considering UCR, COCA, EA-CRP
if any candidate cluster head does not win the cluster head and our proposed protocol when employing network sizes of not
election, then it will send a control message to quit the election. only 60 × 120 m, but also 80 × 160 m, 100 × 200 m, and
Likewise, the winner sensor node sends a cluster head message 160 × 320 m.
to announce itself to all member nodes in the cluster. Truth be
told, these control messages put an extra burden on the energy 4.3.3. Experiments set 3: Energy utilization versus network size
of sensor nodes. UCR considers two ways in its multi-hop routing In this set of experiments, we examine the influence of chang-
algorithm. The first one is the node’s residual energy and the ing the network size on the energy utilization considering our
164 K.A. Darabkh, S.M. Odetallah, Z. Al-qudah et al. / Applied Soft Computing Journal 80 (2019) 154–166

Fig. 7. Number of rounds versus network size.

Table 2 5. Conclusions
Rounds till last node dies contrary to network size.
Network size UCR [42] COCA [43] EA-CRP [44] EA-DB-CRP Conserving the residual energy of nodes and increasing the
(Proposed protocol)
network lifetime are the major concerns in designing wireless
60 × 120 m 330 254 432 802 sensor networks protocols. The cluster-based routing protocols
80 × 160 m 252 170 401 783
refer to a widely known category as of being capable of extending
100 × 200 m 249 121 401 701
160 × 320 m 180 64 364 561 the network lifetime in WSNs more than other categories. In
this manuscript, we propose an Energy-aware and Density-based
Clustering and Relaying Protocol (EA-DB-CRP) for gathering data
Table 3 in wireless sensor networks in which new innovative algorithms
Energy utilization in contrast to network size.
have been incorporated for the sake of remarkably lengthening
Network size UCR [42] COCA [43] EA-CRP [44] EA-DB-CRP
the network lifetime. In specific, one algorithm refers to the net-
(Proposed protocol)
work and cluster divisions whereas the network is portioned into
60 × 120 m 42.45% 38.59% 50.25% 98.50%
80 × 160 m 31.73% 26.42% 38.68% 97.71%
similarly estimated layers where each one is further portioned
100 × 200 m 28.87% 19.66% 37.94% 96.32% into equally sized sub-layers (clusters). Moreover, a batch-based
160 × 320 m 21.89% 14.26% 27.87% 95.20 approach is adopted instead of round-based, which is the known
approach in the literature, thereby mitigating the massive num-
ber of control messages generated. Another algorithm belongs
to the cluster head selection policy in which detached cluster
proposed protocol and other closely related protocols (UCR, COCA, heads find a cluster head weight for each of their members based
and EA-CRP). Thereupon, Fig. 8 illustrates the energy utilization on a specific formula, sort them in lists in a descending order
against network sizes of 60 × 120 m, 80 × 160 m, 100 × 200 based on those weights, and accordingly announce them to their
m, and 160 × 320 m for earlier protocols. Obviously, increasing member nodes whereas the cluster head role will be rotated
the network size leads to decreasing the energy utilization. This among them in a round robin fashion bearing in mind that those
is due to deploying more nodes in larger networks. Thus, more members nodes, included in the lists, have to have the flags of
control packets will be sent as over longer distances. Table 3 cluster head possibility and activation set (i.e., ‘‘1’’), resulting in
contains a summary of numeric values of the energy utilization balancing the energy consumption in the network. There is also
for earlier protocols and various network sizes. It is interesting a merge algorithm employed in which clusters are maintained
to mention that the energy utilization of our proposed protocol to be balanced. One more interesting algorithm refers to the
is not sensitive much as others to any increase in the network possibility of deactivating so closer nodes in clusters that are
size which undoubtedly proves the scalability affordable in our highly dense, thereby persevering more energy that helps much
proposed protocol. in prolonging the network lifetime. The last interesting algorithm
refers to a multi-hop relaying technique in which every member
4.3.4. Experiments set 4: Network lifetime versus initial energy node finds relay node weights for those nodes that will serve as
Is the increase in the initial energy directly or inversely pro- cluster heads in a layer head (toward the base station) during
portional to the network lifetime, expressed by LND? Fig. 9 de- its turn of serving as a cluster head and ultimately selects the
picts the network lifetime in contrast to the initial energy when one the has the highest relay node weight to relay its traffic.
it ranges from 0.5 to 3 joule considering UCR, COCA, EA-CRP, Actually, this algorithm maintains not only the communications
and EA-DB-CRP. It is catching the attention that our proposed over shorter distances (i.e., through choosing relay nodes that are
protocol outperforms all others considering all initial energy val- the closest to the corresponding cluster heads and base station),
ues employed. It is entirely fascinating to see the fast increment but also the selection of relay nodes that have the highest energy.
in the number of rounds, achieved in our proposed protocol, Impressively, all of these algorithms have contributed efficiently
compared with other former proposed protocols as the initial in making our proposed protocols outperform its counterparts
energy increases. This is due to the numerous procedures used (i.e., UCR, COCA, and EA-CRP) in the perspectives of network
to conserve the node’s energy as previously justified. lifetime and energy utilization.
 K.A. Darabkh, S.M. Odetallah, Z. Al-qudah et al. / Applied Soft Computing Journal 80 (2019) 154–166 165

Fig. 8. Energy utilization in contrast to network size.

Fig. 9. Network lifetime versus initial energy given a network size of 60 × 120 m and a deployment of 100 nodes.

Acknowledgment [8] Santar Pal Singh, S.C. Sharma, A survey on cluster based routing protocols
in wireless sensor networks, Procedia Comput. Sci. 45 (2015) 687–695.
This work is supported in part by the North Atlantic Treaty [9] Khalid A. Darabkh, Wijdan Y. Albtoush, Iyad F. Jafar, Improved clustering
Organization (NATO) SPS Project No. SPS G4936. algorithms for target tracking in wireless sensor networks, J. Supercomput.
73 (5) (2017) 1952–1977.
[10] A. Abbasi, M. Younis, A survey on clustering algorithms for wireless sensor
References
networks, Comput. Commun. 30 (14) (2007) 2826–2841.
[11] Khalid A. Darabkh, Mohammad Z. El-Yabroudi, A reliable relaying protocol
[1] Khalid A. Darabkh, Mohammad Z. El-Yabroudi, Ali H. El-Mousa, BPA-CRP: A
in wireless sensor networks, in: Proceedings of 2017 IEEE European Con-
balanced power-aware clustering and routing protocol for wireless sensor
ference on Electrical Engineering & Computer Science, Bern, Switzerland,
networks, Ad Hoc Netw. 82 (2019) 155–171.
[2] Kemal Akkaya, Mohamed Younis, A survey on routing protocols for 2017, pp. 56–60.
wireless sensor networks, Ad Hoc Netw. 3 (3) (2005) 325–349. [12] S. Vhatkar, M. Atique, Design issues and challenges in hierarchical routing
[3] Mohammed Hawa, Khalid A. Darabkh, Raed. Al-Zubi, Ghazi Al-Sukkar, A protocols for wireless sensor networks, in: 2014 International Conference
self-learning MAC protocol for energy harvesting and spectrum access in on Computational Science and Computational Intelligence, Las Vegas, NV,
cognitive radio sensor networks, J. Sensors 2016 (2016) 1–18, 9604526. 2014, pp. 90–95.
[4] Khalid A. Darabkh, Ola Alsukour, Novel protocols for improving the [13] Hamzah A. Yaseen, Mohammad Alsalamin, Abdallah Jarwan, Mamoun F.
performance of ODMRP and EODMRP over mobile ad hoc networks, Int. J. Al-Mistarihi, Khalid A. Darabkh, A secure energy-aware adaptive water-
Distrib. Sens. Netw. 2015 (2015) 1–18, 348967. marking system for wireless image sensor networks, in: Proceedings of
[5] A.F. Khalifeh, M. AlQudah, R. Tanash, K.A. Darabkh, A simulation study the 15th IEEE Multi-Conference on Systems, Signals, and Devices, SSD’18,
for UAV- aided wireless sensor network utilizing zigbee protocol, in: Hammamet, Tunisia, 2018.
Proceedings of the 2018 14th International Conference on Wireless and [14] J. Yick, B. Mukherjee, D. Ghosal, Wireless sensor network survey, Comput.
Mobile Computing, Networking and Communications, WiMob, Limassol, Commun. 52 (12) (2008) 2292–2330.
Cyprus, 2018, pp. 181–184.
[15] H. Ammari, The Art of Wireless Sensor Networks, Springer, 2014.
[6] K. Darabkh, S. Ismail, M. Al-Shurman, I. Jafar, E. Alkhader, M. Al-Mistarihi,
Performance evaluation of selective and adaptive heads clustering algo- [16] V. Vermaa, S. Singhb, N. Pathakc, Analysis of scalability for AODV routing
rithms over wireless sensor networks, J. Netw. Comput. Appl. 35 (6) (2012) protocol in wireless sensor networks, Optik 125 (2) (2014) 748–750.
2068–2080. [17] Khalid A. Darabkh, Nareman R. Alsaraireh, A yet efficient target tracking
[7] S. Gowrishankar, T.G. Basavaraju, D.H. Manjaiah, Subir Kumar Sarkar, Issues algorithm in wireless sensor networks, in: Proceedings of the 15th IEEE
in wireless sensor networks, in: Proceedings of the World Congress on Multi-conference on Systems, Signals, and Devices, SSD’18, Hammamet,
Engineering 2008, WCE 2008, vol. I, London, U.K., 2008, pp. 1–12. Tunisia, 2018.
166 K.A. Darabkh, S.M. Odetallah, Z. Al-qudah et al. / Applied Soft Computing Journal 80 (2019) 154–166

[18] T. Gnanasekaran, S.A.J. Francis, Comparative analysis on routing proto- [32] B. Baranidharan, B. Santhi, Ducf: Distributed load balancing unequal
cols in wireless sensor networks, in: 2014 International Conference on clustering in wireless sensor networks using fuzzy approach, Appl. Soft
Computer Communication and Informatics, Coimbatore, 2014, pp. 1–6. Comput. 40 (2016) 495–506.
[19] Khalid A. Darabkh, Laith Al-Jdayeh, A new fixed clustering based algorithm [33] Mamoun F. Al-Mistarihi, Islam M. Tanash, Fedaa S. Yaseen, Khalid A.
for wireless sensor networks, in: Proceedings of the 14th International Darabkh, Protecting source location privacy in a clustered wireless sensor
Wireless Communications and Mobile Computing Conference, IWCMC networks against local eavesdroppers, Mob. Netw. Appl. (2019) in press,
2018, Limassol, Cyprus, 2018. http://dx.doi.org/10.1007/s11036-018-1189-6.
[20] P. Parwekar, Comparative study of hierarchical based routing protocols: [34] Raed T. Al-Zubi, Noor Abedsalam, Ahmad Atieh, Khalid A. Darabkh, LBCH:
Wireless sensor networks, in: S.C. Satapathy, P.S. Avadhani, S.K. Udgata, S. Load balancing cluster head protocol for wireless sensor networks, Int. J.
Lakshminarayana (Eds.), ICT and Critical Infrastructure: Proceedings of the Informatica 29 (4) (2018) 633–650.
48th Annual Convention of Computer Society of India- Vol I, in: Series of [35] Raed T. Al-Zubi, Noor Abedsalam, Ahmad Atieh, Khalid A. Darabkh,
Advances in Intelligent Systems and Computing, vol. 282, Springer, 2014, Lifetime-improvement routing protocol for wireless sensor networks, in:
pp. 277–285. Proceedings of the 15th IEEE Multi-conference on Systems, Signals, and
[21] K. Tran, S. Hyun-Oh, A data aggregation based efficient clustering scheme Devices, SSD’18, Hammamet, Tunisia, 2018.
in underwater wireless sensor networks, in: Y.-S. Jeong, Y.-H. Park, C.-H. [36] W. Heinzelman, A. Chandrakasan, H. Balakrishnan, Energy-efficient com-
Hsu, J. Park (Eds.), Ubiquitous Information Technologies and Applications, munication protocol for wireless microsensor networks, in: Proceedings of
in: Lecture Notes in Electrical Engineering, vol. 282, Springer, 2014, pp. the International Conference on system sciences, Hawaii, USA, 2000, pp.
541–548. 1567–1576.
[22] Khalid A. Darabkh, Wala’a S. Al-Rawashdeh, Raed T. Al-Zubi, .Sharhabeel H. [37] Wendi B. Heinzelman, Application-Specific Protocol Architectures for Wire-
Alnabelsi, A new cluster head replacement protocol for wireless sensor less Networks (Ph.D. dissertation), Massachusetts Institute of Technology,
networks, in: Proceedings of 2017 IEEE European Conference on Electrical USA, June 2000.
Engineering & Computer Science, Bern, Switzerland, 2017. [38] Jiman Hong, Joongjin Kook, Sangjun Lee, Dongseop Kwon, Sangho Yi,
[23] G. Raghunandan, B. Lakshmi, Comparative analysis of routing techniques T-LEACH: The method of threshold-based cluster head replacement for
for wireless sensor networks, in: Proceedings of the National Conference wireless sensor networks, Inf. Syst. Front. 11 (2009) 513–521.
on Innovations in Emerging Technology, NCOIET2011, Erode, Tamilnadu, [39] Khalid A. Darabkh, Wala’a S. Al-Rawashdeh, Mohammed Hawa, Ramzi
Febreuary 2011, pp. 17–22. Saifan, MT-CHR: A modified threshold-based cluster head replacement
[24] A. Papadopoulos, N. Navarra, J. McCann, C. Pinotti, VIBE: An energy efficient protocol for wireless sensor networks, Comput. Electr. Eng. 72 (2018)
routing protocol for dense and mobile sensor networks, J. Netw. Comput. 926–938.
Appl. 35 (4) (2012) 1177–1190. [40] Khalid A. Darabkh, Wala’a S. Al-Rawashdeh, Raed T. Al-Zubi, Sharhabeel H.
[25] Khalid A. Darabkh, Wala’a S. Al-Rawashdeh, Mohammed Hawa, Ramzi Alnabelsi, C-DTB-CHR: Centralized density- and threshold-based cluster
Saifan, Ala’ F. Khalifeh, A novel clustering protocol for wireless sensor head replacement protocols for wireless sensor networks, J. Supercomput.
networks, in: Proceedings of 2017 International Conference on Wire- 73 (12) (2017) 5332–5353.
less Communications, Signal Processing and Networking, WiSPNET 2017, [41] S. Lindsey, C.S. Raghavendra, PEGASIS: Power efficient gathering in sensor
Chennai, India, March 2017. information systems, in: Proceedings of the IEEE Aerospace Conference,
[26] Z. Yulan, J. Chunfeng, Research about improvement of LEACH protocol, in: Big Sky, Montana, 2002, pp. 1125–1130.
Proceedings of the 2nd International Conference on Information Science [42] G. Chen, C. Li, M. Ye, J. Wu, An unequal cluster-based routing protocol in
and Engineering, ICISE, China, 2010, pp. 2281–2284. wireless sensor networks, Wirel. Netw. 15 (2) (2009) 193–207.
[27] S. Zin, N. Anuar, L. Kiah, A. Pathan, Routing protocol design for secure [43] H. Li, Y. Liu, W. Chen, W. Jia, B. Li, J. Xiong, COCA: Constructing optimal
WSN: Review and open research issues, J. Netw. Comput. Appl. 41 (2014) clustering architecture to maximize sensor network lifetime, Comput.
517–530. Commun. 36 (3) (2013) 256–268.
[28] Khalid A. Darabkh, Saja M. Odetallah, Zouhair Al-qudah, Ala’ F. Khalifeh, [44] Khalid A. Darabkh, Noor J. Al-Maaitah, Iyad F. Jafar, Ala’ F. Khalifeh, EA-CRP:
A new density-based relaying protocol for wireless sensor networks, in: A novel energy-aware clustering and routing protocol in wireless sensor
Proceedings of the 14th International Wireless Communications and Mobile networks, Elsevier Comput. Electr. Eng. 72 (2018) 702–718.
Computing Conference, IWCMC 2018, Limassol, Cyprus, 2018. [45] Khalid A. Darabkh, Jumana N. Zomot, An improved cluster head selection
[29] S.A. Sert, H. Bagci, A. Yazici, Mofca: Multi-objective fuzzy clustering algorithm for wireless sensor networks, in: Proceedings of the 14th In-
algorithm for wireless sensor networks, Appl. Soft Comput. 30 (2015) ternational Wireless Communications and Mobile Computing Conference,
151–165. IWCMC 2018, Limassol, Cyprus, 2018.
[30] R. Logambigai, A. Kannan, Fuzzy logic based unequal clustering for wireless [46] Khalid A. Darabkh, Riham Z. Muqat, An efficient protocol for minimizing
sensor networks, Wirel. Netw. 22 (3) (2016) 945–957. long-distance communications over wireless sensor networks, in: Proceed-
[31] D. Agrawal, S. Pandey, Fuca: Fuzzy-based unequal clustering algorithm to ings of the 15th IEEE Multi-conference on Systems, Signals, and Devices,
prolong the lifetime of wireless sensor networks, Int. J. Commun. Syst. 31 SSD’18, Hammamet, Tunisia, 2018.
(2) (2018) e3448.