4992 IEEE SENSORS JOURNAL, VOL. 20, NO.

9, MAY 1, 2020

On Area Coverage Reliability of Mobile Wireless

Sensor Networks With Multistate Nodes
Suparna Chakraborty , Neeraj Kumar Goyal , and Sieteng Soh, Member, IEEE

Abstract —Wireless sensor networks (WSNs) are a special

type of infrastructure-less network made up of a large num-
ber of tiny sensor nodes with limited energy, processing,
and communication capabilities. WSNs have applications in
health care, home security, environment monitoring, etc., with
research challenges in energy efficiency, network lifetime,
and network reliability. One of the major research challenges
lies in providing application-specific coverage of the region
of interest and reliable transmission of the gathered data
to the mobile sink in the presence of multi-state sensor
nodes. To quantify such a capability, this paper proposes a
quantitative measure, called Area Coverage Reliability (ACR )
for WSNs. ACR brings together WSN reliability, area coverage,
energy efficiency, mobility of data collector or sink, random duty cycle of nodes, and multi-state nature of sensor nodes
under a common umbrella. This paper proposes a Monte Carlo simulation approach that utilizes an energy matrix to
evaluate the effect of energy-depleted nodes and energy-oriented data transfer capability on ACR. The energy matrix
reflects the residual energy of sensors, the energy required to transmit data to the neighboring nodes, connectivity, and
the multi-state nature of the sensors. The proposed approach is illustrated through a series of random examples. The
ACR information allows the network designers to achieve a better understanding of the impact of random duty cycle,
node energy, node/link reliability, and randomly deployed sensors on reliability.
Index Terms — Area-coverage, Monte Carlo simulation, multistate nodes, network reliability, wireless sensor networks.

I. I NTRODUCTION WSNs are randomly distributed, i.e., geographically deployed

in a random manner, in the field of interest and remain
W IRELESS Sensor Networks (WSNs) comprise a large
number of inexpensive sensor nodes, each capable of
sensing, processing, and transmitting environmental informa-
unattended after deployment. These sensor nodes are vulner-
able to different kinds of failure due to external causes such
tion. WSNs find applications in fields such as health care, as variability in environmental conditions, including rainfall,
home security, military surveillance, monitoring of railways, humidity, foliage, and internal reasons like noise, lack of
agriculture field, smart buildings, and remote monitoring battery power, hardware failure, etc.
applications that include biodiversity, rainforest monitoring, The sensors, after deployment, remain static throughout
detecting forest fire. Deployment of sensor nodes can be their lifetime. However, the connectivity and data-flow capac-
preplanned [1] or random [2], [3], depending on the types of ity vary due to various node-states, communication link failure,
application environment. In remote monitoring applications, limited hardware resources, and battery power. Hence, it is
extremely important to carry out a reliability analysis as a
Manuscript received October 8, 2019; revised January 2, 2020; precursor to WSN deployment. Since long, the reliability
accepted January 7, 2020. Date of publication January 10, 2020; analysis of WSNs has focused on connectivity among the
date of current version April 3, 2020. The associate editor coordi-
nating the review of this article and approving it for publication was sensors and data-driving capacity of the network. A rich liter-
Dr. Qammer H. Abbasi. (Corresponding author: Suparna Chakraborty.) ature [4]–[9] can be found on such approaches, which focus
Suparna Chakraborty is with the School of Electrical Engineering, on reliable routing [9] by selecting a single multi-constrained
Computing and Mathematical Sciences, Curtin University, Perth, WA
6102, Australia, and also with the Subir Chowdhury School of Quality and optimal path leading to reliable data transfer. However, besides
Reliability, Indian Institute of Technology Kharagpur, Kharagpur 721302, connectivity and data-driven capacity [10], coverage serves as
India (e-mail: [email protected]). one of the major objectives of sensor networks. Coverage is
Neeraj Kumar Goyal is with the Subir Chowdhury School of Quality and
Reliability, Indian Institute of Technology Kharagpur, Kharagpur 721302, concerned with successful monitoring of the area of interest
India (e-mail: [email protected]). in the presence of node failures.
Sieteng Soh is with the School of Electrical Engineering, Computing There are various coverage problems such as area
and Mathematical Science, Curtin University, Perth, WA 6102, Australia
(e-mail: [email protected]). coverage [11], point coverage [12], [13], K -coverage [14],
Digital Object Identifier 10.1109/JSEN.2020.2965592 [15], and m-connected K -coverage [16]. Two types of area

1558-1748 © 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: University of Canberra. Downloaded on April 28,2020 at 15:52:49 UTC from IEEE Xplore. Restrictions apply.
CHAKRABORTY et al.: ON AREA COVERAGE RELIABILITY OF mWSNs WITH MULTISTATE NODES 4993

coverage problems have been studied: full area coverage [17] sensor node is represented by a color. Each node receives a
and partial area coverage [18]–[20]. Irrespective of the cover- color i with probability ci . Then, a link between sensor nodes
age type, reliable monitoring of an area depends on both sens- of color i and j occurs with probability Pi j . However, for use
ing coverage and transfer of the sensed data to the processing in ACR, probability Pi j must consider the combined effect
center or sink. In other words, for successful operation, a WSN of communication range and energy required to transmit data
is required to provide sensing coverage of the monitored as well as the residual energy of each sensor node and link
area that satisfies a given application-specific coverage-area reliability.
requirement [21] and successful transfer of collective sensor This paper proposes a Monte Carlo simulation approach for
data to the sink. evaluating ACR. The proposed simulation approach randomly
Sensors are battery-constrained devices. Sensors deployed generates states for all nodes considering random duty-cycle,
near a sink node get heavily utilized by other nodes to transfer hardware components’ failure, and node energy states while
their sensed data to the sink node. Hence, these border nodes the sink moves along the periphery of the monitoring region.
die out earlier, leading to disconnection and failure of the These generated nodes-states are random and independent
entire network. To cope with this problem, the concept of of each other. As sensors are changing states and they are
multiple static sinks was introduced [22]. This reduced the not always operating, the area-covered by the mWSN varies
hop count from source(s) to sink(s) and hence resulted in with time. Therefore, the non-overlapping area sensed by
smaller average energy dissipation per node as compared to each sensor is evaluated and aggregated to calculate the
the case of deploying a single static sink. However, these static total area sensed by the mWSN at each instance of data
sinks should be deployed inside the monitored region in such collection. Finally, the capability of the mWSN to transfer
a way that load amongst the nodes is balanced. To address the application-specific coverage-oriented data to the mobile
the problems associated with the multiple-sink deployment, sink is checked through a connectivity matrix that accounts
the concept of a mobile sink was introduced [23]. A wide for both residual energies of all nodes and link reliability.
variety of literature can be found to evaluate the reliability of This connectivity matrix evaluates the availability of link L i j
a WSN with a mobile sink (mWSN) [24]–[26]. between sensors i and j, including not only the communica-
However, such approaches considered only two states of tion range but also the energy availability of each node to
nodes and thus were inadequate for evaluating the coverage transmit data to its neighbor as well as node and link reliability.
area reliability of mWSN with multi-state nodes. Due to The effectiveness of our proposed approach is shown through
hardware architecture and/or node energy, a sensor may exist several performance comparisons on mWSNs with various
in various states. Men and Chen [27] present an approach to sizes.
identify the states but do not provide any way to evaluate Rest of the paper is structured as follows: Section II
mWSN reliability in the presence of multi-state nodes. There- describes the preliminaries required for development of the
fore, appropriate modeling of multiple states of nodes for reli- proposed approach. Section III enlightens the proposed metric
ability evaluation of mWSN remains an open area for research. ACR. Section IV discusses the proposed approach and method-
Consideration of multiple states of nodes allows more practical ology. Section V provides simulation results, and finally,
reliability assessment of mWSNs, as it addresses not only Section VI concludes the paper.
hardware failures of nodes but also limitations imposed on
nodes to achieve high-energy efficiency. This paper aims at II. B ASIC B UILDING B LOCKS
proposing an approach that quantifies the application-specific This section explains the modeling process of a typical
coverage-oriented multi-source single-sink mWSN reliability WSN system. It first explains the characteristics of the mWSNs
with multi-state nodes and mobile sink. The quantifying mea- with modeling assumptions. Then, it discusses state modeling
sure, coined as Area Coverage Reliability (ACR), considers of a network followed by the definition and evaluation of
i) multi-state nature of nodes, ii) coverage-area requirement area-coverage.
iii) Euclidean distance between two sensors, iv) node’s trans-
mission range, v) energy required to transmit data, vi) node’s A. Assumptions
residual energy, and vii) node and link reliability. 1) All states of a sensor node are statistically independent
Traditionally WSNs have been modeled through Random of each other.
Geometric Graphs (RGG) [28], where a link exists between 2) Once a sensor node fails, it stays failed for the remaining
any two sensor nodes if their Euclidian distance is less period of the mission time.
than or equal to communication range. Percolation theory 3) All links are bidirectional without any constraint on their
concepts have been widely used to evaluate the connectiv- load-carrying capacity.
ity [29]–[31] and coverage of WSNs [32]. In such a theory, 4) Sensor nodes are resource-constrained, and the sink node
a probabilistic connectivity matrix [29], [33], is used as a tool has sufficient resources.
to measure the quality of network connectivity and coverage.
However, such concepts do not consider i) energy availability B. Mobile Wireless Sensor Networks (mWSNs) Model
of a node to transmit the sensed data to any node within An mWSN is composed of a set of N randomly deployed
its communication range, and ii) multi-state nature of nodes. sensor nodes S = {s1 , s2 , . . . , s N }, and a mobile sink v that
To address limitation (ii), Cannings and Penman [28] proposed moves along the periphery of a simulation region with a speed
Random Randomly Coloured Graphs wherein each state of a of u m/s. The mWSN is modeled as an undirected graph

Authorized licensed use limited to: University of Canberra. Downloaded on April 28,2020 at 15:52:49 UTC from IEEE Xplore. Restrictions apply.
4994 IEEE SENSORS JOURNAL, VOL. 20, NO. 9, MAY 1, 2020

Evaluation of probabilities of a node being in any of the above

states is discussed in Section IV-A.
At any point of time, the network’s connectivity is
determined by the graph G = {{v} ∪ S, E}. A communication
link exists between any two sensor nodes si and s j with
reliability λi j , for 0 ≤ λi j ≤ 1, if and only if, i) nodes si
and s j are in either ACTIVE or RELAY state, ii) nodes
si and s j are within communication range (r ) of each other,
and iii) node si has sufficient amount of energy required to
transmit its collected data to node s j . A link that complies
with these three requirements is called an available link. Such
link formation model is proposed owing to the fact that even
if the Euclidian distance d (i, j ) between sensors si and s j ,
is less than the communication range, i.e., d (i, j ) ≤ r , a link
between nodes si and s j will not be formed if sensor node
si has insufficient amount of energy to establish a link and
send data to sensor node s j . Moreover, in a wireless scenario,
weather, hills, high-rise buildings, noise, interference, etc. also
affect a link’s quality, which in turn affects the link’s reliability.
Thus, while determining a link’s existence it does not suffice
only to consider node-energy.
Fig. 1. A generic example of mWSN for the area-coverage problem. Formally, a link between sensors si and s j is available,
if and only if, the following conditions are satisfied:
Condition (i ): nodes si and s j are in either
ACTIVE or RELAY state,
Condition (ii): d (i, j ) ≤ min(d En,si , r ), and the link
between two nodes si and s j is operating with a reliability
of λi j ,
where d En,si denotes the maximum distance afforded by the
battery-power of sensor si , up to which it can transmit data.
This distance, i.e., d En,si , is termed as the energy-oriented
Fig. 2. Typical architecture of a sensor node.
transmission range of sensor si . The evaluation of energy-
oriented transmission range will be discussed later in this
G = {{v} ∪ S, E} where E is the set of communication section. The incorporation of link’s reliability while deter-
links. After deployment, the sensors remain geographically mining its availability is discussed in Section IV-C, while
static throughout their lifetime. Fig. 1 shows a generic mWSN the node state determination for Condition (i ) is discussed
with 21 sensors deployed randomly in the monitoring field, in Section IV-A.
and a mobile sink that travels along the periphery of the We follow the energy model as in [29] to evaluate the
simulation region. energy consumed by a sensor node si in sensing (E sense,si ),
Each sensor node is comprised of four major compo- transmitting (E T x,si ), and receiving (E Rx,si ) a data bit.
nents: a sensing unit, a power device, a processing unit, Equations (1) and (2) respectively give expressions for cal-
and a transceiver unit. Fig. 2 shows the architecture of a culating the energy required to transmit and receive data of
typical sensor node. Owing to a hardware component’s fail- size k bits to a distance d:
ure or energy shortage, at any point of time, a sensor node 
may be in one of the following four states: ACTIVE, RELAY, k E elec +kε f s d 2 , d <d0
E T x,si (k, d) = (1)
SLEEP, or FAIL. An ACTIVE state is a state in which a node k E elec +kεmp d 4 , d ≥d0
can sense, transmit, and receive data as its sensing, powering, E Rx,si (k) = k E elec (2)
and transceiver units are operating reliably. Failure of the
processing unit leads to complete node failure. Therefore, where E elec is the energy consumed by the transmitter circuitry
in our study, the processing unit is assumed to be perfectly per data bit, ε f s and εmp are respectively the energy consumed
reliable. A sensor node in a RELAY state can transmit and by the power amplifier per data bit for the free-space
 (fs) and
receive data; however, it cannot sense due to the failure of its multi-path (mp) fading channel model, and d0 = ε f s /εmp is
sensing unit. In SLEEP state, a node can neither sense nor the threshold distance that determines the channel model.
transmit or receive data (temporarily) as all its circuitry is At any point of time, residual energy E Res,si of a sensor
temporarily turned off, depicting the effect of random duty- node si determines its communication capability, i.e., the max-
cycle. In contrast, a sensor in FAIL state can neither sense imum distance up to which a node can communicate. There-
nor transmit or receive data (permanently) due to the failure fore, assuming the residual energy to be the currently available
of either the transceiver unit or battery depletion or both. energy that the sensor node si can spend on transmitting data,

Authorized licensed use limited to: University of Canberra. Downloaded on April 28,2020 at 15:52:49 UTC from IEEE Xplore. Restrictions apply.
CHAKRABORTY et al.: ON AREA COVERAGE RELIABILITY OF mWSNs WITH MULTISTATE NODES 4995

from (1), we have, ACTIVE, SLEEP, RELAY, or FAIL states depending on hard-
2 ware failure or working states, random duty cycle, and energy
E Res,si =k E elec +kε f s d En,s i
(3) states of the communicating node. Fig. 1a shows an mWSN
Solving (3) for d En,si , we get the energy-oriented transmis- that satisfies its area-coverage requirement of 50% of total
sion range of sensor si as monitoring region, i.e., A (S) > Areq . However, the mWSN
 in Fig. 1a is unreliable as it is not connected to the mobile sink,
( E Res
k −E elec )
and thus fails to transfer the collected data to the sink. Fig. 1b
d En,si = (4) shows another instance of the mWSN, where the network is
εfs
unreliable because it fails to satisfy Areq i.e. ( A (S) = 45% of
total monitoring region) < ( Areq = 50% of total monitoring
C. Area-Coverage region). Thus, merely satisfying Areq or being connected to
Area-Coverage refers to the region of interest under surveil- the mobile sink does not make any mWSN a reliable network.
lance by a sensor network. In most applications, a network’s In other words, an mWSN is reliable iff it satisfies Areq
overall capability to monitor the region of interest determines and is capable of transferring A (S) ≥ Areq to the mobile
its performance. Given a region to be monitored, a network’s sink. Fig. 1c accounts for a reliable network as (i) A (S) >
area-coverage is the total area sensed by the network through Areq , and (ii) the network is connected to the mobile sink.
its multi-state sensor nodes. The total area covered by an The mWSN in Fig. 1c thus satisfies the application-specific
mWSN, A (S), is given as coverage-area requirement and can successfully transfer the
required amount of sensor data to the mobile sink.
|T a |

A (S) = A (si ) (5)
i=1 IV. P ROPOSED M ETHODOLOGY AND A PPROACH
where Ta denotes the set of nodes in the ACTIVE state, | ∗ | This section discusses the proposed approach, shown in
represents the number of elements in “∗”, and A (si ) is area Algorithm 1, for ACR evaluation of an mWSN. The pro-
sensed by sensor si . Note that A (si ) refers to the disjoint-area posed approach addresses quantitative evaluation of the area-
sensed by sensor si ; Section IV-B describes the evaluation coverage oriented reliability, ACR, of mWSNs. Algorithm 1
of A (si ). starts by deploying the sensor nodes randomly in the field
to be monitored (Line 1) with length W L and width W B ,
III. A REA -C OVERAGE R ELIABILITY (ACR) OF AN M WSN both in meter. Once deployed, the sensors, except the sink,
remain geographically static and free from human intervention
To model an mWSN’s ability to satisfy the application- throughout their lifetime. Thus, randomness in the network
specific area-coverage requirement in the presence of multi- configuration of an mWSN is an effect of the random node-
state sensor nodes, a new performance index, Area-Coverage states, energy-availability of nodes and link reliability. The
Reliability (ACR) is introduced in this paper. ACR quantifies mobile sink moves along the simulation boundary and collects
mWSN’s (with multi-state nodes) capability of data after every t seconds each time from a newly generated
(i) satisfying the application-specific area-coverage network configuration. For each iteration q = 1, 2, 3, . . . , Q,
requirement Areq , i.e., A (S) ≥ Areq and, Line 3 generates M = W L /t number of sink positions.
(ii) successfully transmitting the sensed coverage-oriented For each sink position, Lines 4-20 simulate one network
data to the mobile sink node. configuration. Thus, there are Q × M network configurations,
Note that any mWSN is considered to be working reliably iff and Algorithm 1 in total performs Q × M number of sim-
it satisfies both the conditions. Therefore, based on the above ulation runs. To be more specific, Lines 5-6 enumerate the
discussion, the ACR of mWSN is defined as follows. current state of each node. The current state of each node is
Definition 1 (ACR): Given an application-specific coverage- affected by its random duty-cycle, hardware components’ fail-
area monitoring requirement Areq of an mWSN, ACR is ure or working states, and battery energy state of each sensor
defined as the probability that the mWSN can transmit at least node. Recall that at any instant of time, a sensor node can
the application-specific required amount of area monitoring be in ACTIVE, RELAY, SLEEP, or FAIL state. Section IV-A
data, Areq , to the mobile sink node for a given period of time elaborately discusses the random node-states generation of a
under given environmental conditions. sensor node and the current state determination of any sensor.
Definition 2 (ACR Problem): Given a graph G = {{v}∪S, E} The proposed approach then evaluates the disjoint area sensed
of an mWSN with mobile sink node v, a set of ACTIVE and by each node in ACTIVE state (Lines 7-9), and the total area
RELAY nodes in S, and a set of communication links E, sensed by the random network configuration (see Line 11 and
compute the probability that there exists an operational path Section IV-B). Each random network configuration is checked
from all source (ACTIVE) nodes to the mobile sink node, iff for being a successful network state. A network configuration
the total area sensed by the ACTIVE nodes is at least Areq . is in a successful state iff it satisfies Areq and successfully
Fig. 1 gives a typical example of the ACR Problem. It shows transmits the sensed data to the mobile sink node. If any
an mWSN of 21 sensor nodes randomly deployed in the network configuration satisfies Areq , Line 13 enumerates the
region of interest with an Areq = 50% of total monitoring energy-available link matrix L (see Section IV-C). This matrix
region. As depicted in Fig. 1, the sensor nodes can be in is then used to check the connectivity i.e., the existence of

Authorized licensed use limited to: University of Canberra. Downloaded on April 28,2020 at 15:52:49 UTC from IEEE Xplore. Restrictions apply.
4996 IEEE SENSORS JOURNAL, VOL. 20, NO. 9, MAY 1, 2020

a path from each of the ACTIVE nodes to the sink for Algorithm 1 ACR Evaluation
the current network configuration in Line 14. To mention Inputs: S, E in , Areq , W L , W B , Q, t, u, αsi
here, Line 14 uses the node-fusion method [31] to check S: A set of nodes.
the existence of a path between each ACTIVE node and the E in : Initial energy of each node.
mobile sink. As discussed in Section IV-D, a Boolean variable Areq : Application specific coverage area requirement.
χq,m , whose value is determined in Lines 15-17, keeps track W L : Length of the simulation region.
of the reliability/unreliability of the m th (m ∈ M) network W B : Width of the simulation region.
configuration in the q th iteration. After every simulation run, Q : number of iterations.
the residual energy of each node is updated in Line 18. The t: data collection time interval.
residual energy is updated by subtracting the energy spent in u: speed of the mobile sink.
the current simulation run from the energy the node had before αsi : duty cycle of each sensor node.
starting the current simulation run. Finally, in Line 22, after Output: ACR
Q × M number of simulation runs, ACR of the mWSN is ACR: Area-Coverage Reliability.
evaluated (see Section IV-D). 1. Generate random node positions within the simulation
The major steps of Algorithm 1 are discussed in the region (W L , W B )
following sub-sections. // generateQ × Mnetwork configurations
2. For each simulation q = 1, 2, 3, . . . , Q
A. Random Node States Generation 3. Find sink position m = 1, 2, …, M // generated after
This section discusses the process of random node-states every t seconds
generation at each sink position, i.e., after every t seconds. 4. For each sink position m //node-state generation
The sensor nodes are random in nature and exhibit multiple 5. For each node si of S
states due to the random duty-cycling approach followed by 6. Enumerate the current state of the sensor node si
the sensors, random failure of the sensor node’s hardware // refer Section IV-A
components, and limited battery life of the sensors. It is //disjoint-area evaluation
assumed that the sensor nodes follow a random duty-cycling 7. If si is in ACTIVE state
approach to conserve energy where sensors turn on and turn off 8. Evaluate the disjoint area A(s i ) // refer Section
in a random fashion independent of each other. The duty cycle IV-B
(α si ) of each sensor node si is the ratio between the average 9. End If
time a sensor node spends in working state to the total time 10. End For
of working or sleep states [32]. Thus, the probability that any 11. Evaluate the total area sensed A (S) by the network
sensor node remains in SLEEP state is // by following(5)
12. If A (S) ≥ Areq
pslp (si ) = (1 − αsi ). (6) // Link-State matrix enumeration
13. Find the link-state matrix (L) of the network //
Any node in working state can be in ACTIVE or RELAY
refer Section IV-C
state depending on the success or failure states of the sensing
14. Check if the network is connected to the mobile
and transceiver unit. Let pcomm (si ) and psense (si ) denote the
sink
probabilities that the transceiver unit and the sensing unit of a
15. If connected
sensor si are working respectively. Further, let pa (si ) denote
16. χq,m = 1 // Network configuration(q,m)is reliable
the probability that any sensor si is in ACTIVE state and pr (si )
17. End If
denote the probability that any sensor si is in RELAY state.
18. Update residual energy of each node and go to #
Therefore, (7) and (8), as shown below, respectively give the
Step 3.
reliability of a node si in ACTIVE and RELAY state:
19. End If
pa (si ) = pcomm (si )× psense (si )×αsi . (7) 20. End For
pr (si ) = pcomm (si )×(1 − psense (si ))×αsi . (8) 21. End For
//ACR evaluation from Q × M network configura-
A node can be in FAIL state due to hardware failure and/or tions
energy depletion. Each node starts operating with an initial 22. Evaluate ACR using (12) // refer Section IV-D.
energy E in and gradually depletes its battery due to sensing
and/or transmission of data. Thus, at the onset of operation,
the probability that any sensor si is in FAIL state is
p f (si ) = (1− pcomm (si )) ×αsi . (9)
value is then compared against the following conditions to
In due course, if any node runs out of energy, it remains in determine the current state of any sensor node si :
FAIL state for the entire network lifetime. Condition 1: 0 < test si ≤ pa (si )
To determine the current state of any sensor node si , If the generated test si value falls in this category, then a
a random value, test si , following a uniform distribution sensor node si is in ACTIVE state.
(0 < test si < 1) is generated. This randomly generated test si Condition 2: pa (si ) < test si ≤ pa (si ) + pr (si )

Authorized licensed use limited to: University of Canberra. Downloaded on April 28,2020 at 15:52:49 UTC from IEEE Xplore. Restrictions apply.
CHAKRABORTY et al.: ON AREA COVERAGE RELIABILITY OF mWSNs WITH MULTISTATE NODES 4997

node. Disjoint area A(si ), covered by a sensor si for given

sink location, is determined by
A(si ) = Asense (si )∩V (si )∩(W L ×W B ) (10)
where  ∗  is the area of “∗”, Asense (si ) represents the
circular region covered by sensor si , and V (si ) represents
the Voronoi polygon containing sensor si . Note that the areas
formed by (10) are not generally polygons. As an example,
in Fig. 4, the disjoint area sensed by the sensor s1 is rep-
resented by a non-polygonal area (shown by yellow-colored
dots). While evaluating the disjoint area sensed by any sensor
node si , two special cases may arise:
Fig. 3. Overlapping area sensed by sensors. Case 1: The Voronoi polygon encircles the circular area
sensed by a sensor node.
In such a case, the disjoint area assigned to the sensor node
is the circular area, i.e., A(si ) = |Asense (si )|.
Case 2: The Voronoi polygon is encircled by the circular
area sensed by a sensor node.
In such type of cases, the disjoint area assigned to the sensor
node is the polygonal area, i.e., A(si ) = |V (si )|.

C. Enumeration of the Link Matrix

This section shows enumeration of the available links in
set E of an mWSN G = {{v} ∪ S, E}. We use an adjacency
matrix L of size N × (N+ 1) to represent the availability of
Fig. 4. Evaluating the disjoint area sensed by a sensor.
each link in E. Let L i, j denote the entry in L at row i and
column j . L i, j is set to ‘1’ if the link is available; otherwise
If the generated test si value falls in this category, then a L i, j is set to ‘0’. More specifically, we have L i, j = 1 when
sensor node si is in RELAY state. the link satisfies condition (i ) and (ii) of Section II-B. In other
Condition 3: pa (si ) + pr (si ) < test si ≤ pa (si ) + pr (si ) + words, each entry L i, j ∈ L signifies the capability of sensor
p f (si ) si to transfer the sensed data to its neighbor node s j . It also
If the generated test si value falls in this category, then a signifies that despite node si being in ACTIVE or RELAY
sensor node si is in FAIL state. state, it may not have sufficient energy to transfer data to
Condition 4: test si > pa (si ) + pr (si ) + p f (si ) its next node s j . Further, while the energy is adequate,
If the generated test si value falls in this category, then a the link is operational with reliability λi j . To account for
sensor node si is in SLEEP state. a link’s functioning with reliability λi j , while constructing
the matrix L, a random value test L i j , following a uniform
B. Area Sensed by Sensor Nodes distribution (0 < test L i j ≤ 1), is generated. This test L i j is
This section explains the evaluation of the area covered, generated when a link’s end nodes, si and s j , are either in
A(S) for a random network configuration, where S = ACTIVE or RELAY, and they have sufficient energy for data
{s1 , s2 , . . . , s N } denote the set of N deployed sensor nodes. transmission. Formally, we have
Assuming a unit disk sensing model, A(si ) = πr 2 is the area 1 t est L i j ≤λi j ∩(n i =1)∩(n j =1)
sensed by each sensor node si . One would consider evaluating L i, j ={0 ot herwise (11)
the area A(S) by simply summing up all circular areas of where n i = 1 and n j = 1 symbolize that the sensor nodes si
the nodes in ACTIVE state. However, merely summing up and s j are either in ACTIVE or RELAY state, and they have
the areas would lead to erroneous calculation of the sensed sufficient energy for data transmission.
area as two or more sensors can have a common overlapping Note that the last column of matrix L signifies the con-
area, as shown in Fig. 3. As an example, let us consider nectivity of any node si in ACTIVE / RELAY state with the
calculation of the area-sensed by the sensor s1 in Fig. 4; mobile sink node v.
the circular area centered at node s1 shows its coverage area.
As shown in Fig. 4, the coverage area of sensor s1 partially
covers the areas sensed by s10 , s14 , s19 , s25 , s29 , and s32 . This D. ACR Evaluation
is undesired as it leads to an over-estimation of coverage- The enumerated matrix L is now analyzed to determine
oriented reliability. Therefore, it is necessary to evaluate the the coverage-oriented data flow capacity of the mWSN.
disjoint area sensed by each sensor and then sum them up. The A randomly generated network configuration is said to satisfy
evaluation of disjoint area starts by dividing the monitored the application-specific area-coverage requirement if the area
region into Voronoi cells with each cell containing a sensor sensed by the network configuration, A (S) , is at least Areq .

Authorized licensed use limited to: University of Canberra. Downloaded on April 28,2020 at 15:52:49 UTC from IEEE Xplore. Restrictions apply.
4998 IEEE SENSORS JOURNAL, VOL. 20, NO. 9, MAY 1, 2020

TABLE I
PARAMETER S ETTINGS

If the ACTIVE sensors of any network configuration at number of nodes N = 5, 6, 7, and 8. The CS approach
q th iteration and m th sink position satisfy the area-coverage generates all 4 N possible network states that the mWSN may
requirement and have a path to the mobile sink, then we set reside in its entire life cycle. It then analyzes each of these 4 N
χ q,m = 1. On the other hand, if the network configuration fails states for being a successful network state. More specifically,
to meet the required area-coverage, then it is dropped and not the CS approach comprises of four steps:
checked for its connectivity with the sink node. In this case, 1. Generate a set N S containing all possible 4 N network
we set χ q,m = 0. Finally, ACR is evaluated by the Monte states. Note that each state N S m ∈ N S consists of
Carlo simulation approach as nodes in ACTIVE, RELAY, SLEEP, and FAIL states.
 Q M Therefore, N S m = {Ta ∪ Tr ∪ Tslp ∪ T f , where Ta ,
q=1 m=1 χq,m
AC R = (12) Tr , Tslp , and T f denote the set of sensors in ACTIVE,
Q×M RELAY, SLEEP, and FAIL states, respectively.
V. S IMULATION R ESULTS 2. For each state N S m ∈ N S, check if N S m satisfies
condition (i ) of Section III. If so, store it in T .
In this section, the simulation results are presented. All sim- 3. For each Tn ∈ T
ulations use the parameters given in Table I, unless otherwise
3.1 Construct a connectivity matrix C M. Each entry
specified. We assume that for a square meter of the sensed
C M i j ∈ C M is set to “1” when any sensor s j
area, a sensor transmits 1 bit of data. For example, a sensor
lies within the transmission range of a sensor si ,
with a sensed area of 100m 2 will transmit data in a packet
otherwise, we have C M i j = 0. The connectivity
of size 100 bits. Failing to find a set of benchmark networks
of each ACTIVE node in Tn with the sink is then
in the literature, we generate various random mWSNs (called
checked by performing the node-fusion [31] on
problems) and use our approach to evaluate their ACR. Each
C M. Note that this step addresses the condition
problem refers to one spatial location for the sensors. Since
(ii) of Section III.
we could not find any comparable approach in the literature
3.2 If all nodes in Tn are connected to the sink node,
to gauge the performance of our approach, in Section V-A,
then the mWSN is in a successful state, and the
we compare its fidelity and performance against a Com-
probability that Tn arises is
plete State enumeration (CS) approach. As described in
Section V-A, a CS approach enumerates all possible network
states for each mWSN. Thus, the CS approach is applicable Pr ob (Tn )
   
only for use in small-sized networks because it generates = pa (si ) pr (si ) pslp (si ) p f (si ).
an exponential number of network states, in terms of the i∈Ta i∈Tr i∈Tslp i∈T f
number of nodes. Section V-B presents the applicability of the (13)
proposed approach on 24 comparatively large-sized networks.
Then we study the effect of duty-cycle and Areq on ACR 4. The reliability of the mWSN is then calculated as
in Section V-C. Finally, in Section V-D, the effect of link
reliability on ACR is studied. |Tn |

AC R = Pr ob (Tn ). (14)
A. Comparison and Fidelity Analysis of the Proposed n=1
Approach It is important to mention here that the AC R in (14) is the
In this section, we explore the performance of the proposed ACR when links are perfectly reliable, and each node always
approach against a CS approach for small-sized networks with has sufficient energy to transmit data to its next node. In other

Authorized licensed use limited to: University of Canberra. Downloaded on April 28,2020 at 15:52:49 UTC from IEEE Xplore. Restrictions apply.
CHAKRABORTY et al.: ON AREA COVERAGE RELIABILITY OF mWSNs WITH MULTISTATE NODES 4999

TABLE II
N ETWORK S IMULATION R ESULTS

words, in the CS approach, the existence of a link between sen- the time complexity of the CS approach increases dramatically
sors si and s j is always dominated by the transmission range r from 2.06 to 8226.47 CPU seconds when the problem size
of the sensor si . More specifically, for the CS approach, we set increases from N = 5 to N = 8; see the seventh column.
d (i, j ) ≤ r , and λi j = 1 for condition (ii) in Section II-B. Therefore, the CS approach is limited for use in only small
For a fair comparison of the proposed approach against the networks.
CS approach, Algorithm 1 determines the link-state matrix in To show the effects of node-energy constraint on ACR,
Line 13 with d (i, j ) ≤ r , and λi j = 1. we use the proposed approach to generate ACR values for
For both the CS approach and proposed approach, mon- each problem in Table II with each node having initial energy
itoring area is set to 10×10 m 2 , duty-cycle is 0.8, and the E in = 2 J; note that we still consider reliable links. The results
minimum required fraction of total area (θ ) is 50%. In both are shown in Table III; for convenience, Table III also shows
approaches, the mobile sink collects data every t = 5 seconds, the AC R values of Table II. As can be seen from Table III,
and it moves along the x-axis of the monitoring field until it energy constraint in each node reduces the mWSN’s reliability,
reaches the simulation boundary. The assumed speed of the i.e., the reliability values when nodes have energy constraints
mobile sink is 1 m/sec. In Algorithm 1, for each problem, are less than the reliability values for nodes without energy
we set Q = 10000, and thus it has 20000 simulation runs per constraints. More specifically, across all problems, i.e., for
problem. N = 5 to N = 8, the ACR is reduced by up to 43.67%; see its
The compared results of the proposed approach and the CS AC R and ACR values. This depicts a realistic aspect of the
approach for N = 5, 6, 7, 8 are given in Table II. The second fact that despite sensors being in ACTIVE or RELAY state,
column of Table II shows the number of random problems after few rounds of communication, the sensors might not have
generated corresponding to each N. A total of 10 random sufficient amount of energy to transmit the sensed data to their
problems were developed. The third and fourth columns of neighbors, thus resulting in unsuccessful communication. This
Table II depict ACR values evaluated by the proposed approach leads to a reduction in the reliability value. Table III also shows
and CS approach, denoted by AC R . Recall that AC R con- the required CPU time to generate each ACR value. As shown,
siders each node to always have sufficiently high energy for generating ACR is faster as compared to generating AC R ; see
data transmission, and each link is reliable. The fifth column the sixth columns in Table II and Table III. This is because
gives the simulation error of our proposed approach relative the link-state matrix, generated while evaluating the AC R , has
to the CS approach. By observing the third, fourth, and fifth more number of connected nodes as compared to the link-state
columns, we can conclude that the ACR value computed by matrix generated while evaluating the ACR. The node-fusion
the proposed approach is almost equal to that generated by the method for checking the connectivity with the sink node takes
CS approach, i.e., with error ranging from 0.001% and 0.37%. a longer time when the matrix contains a larger number of
The simulation error, as observed in the fifth column, confirms connected nodes.
the accuracy and fidelity of the proposed approach. Table II
also shows the CPU time required by our approach and the
CS approach. As shown in the sixth column of the table, B. Application of ACR to Larger Networks
the proposed approach is scalable, i.e., its time complexity To show its scalability, the proposed approach is applied to
increases insignificantly when the number of nodes increases, network of sizes N = 50, 60, 70, 80, 90, and 100. For each N,
ranging between 1.29 and 3.12 CPU seconds. In contrast, the proposed approach is run four times, giving 24 randomly

Authorized licensed use limited to: University of Canberra. Downloaded on April 28,2020 at 15:52:49 UTC from IEEE Xplore. Restrictions apply.
5000 IEEE SENSORS JOURNAL, VOL. 20, NO. 9, MAY 1, 2020

TABLE III
ACR R ESULTS ON S MALL N ETWORKS

TABLE IV
ACR R ESULTS ON L ARGE N ETWORKS

generated networks or problems. For these simulations, we set energy consumption. Rather, stochasticity is observed. For
θ = 50%; see Table I for the remaining parameters. example, for N = 90, the energy consumed by the mWSN
Table IV shows the resulting ACR value, and the average of problem2 with ACR= 0.91456 is 2.20738 mJ, whilst the
energy consumed to transmit sensed data to the mobile sink per energy used by the mWSN in problem3 with ACR= 0.99967 is
simulation run (shown in a bracket, and given in milliJoule). 0.85014 mJ. This shows the combined effect of random duty-
One would expect a higher value of ACR when the number cycle, random node-states, node energy, mWSN’s topology,
of nodes within the same simulation region is increased. and/or link reliability.
However, the results in Table IV are against the intuition,
i.e., more nodes do not always increase ACR value. More C. Effect of Duty Cycle and Areq on ACR
specifically, while increasing N = 70 to N = 80 in problem 4 Table V and Table VI show the impact of duty cycle and
improves ACR from 0.69190 to 0.90018, increasing N = Areq on ACR. While performing this analysis, the links con-
60 to N = 70 in problem 3, decreases the ACR value structed are considered to be always reliable, i.e., λi j = 1. The
from 0.99952 to 0.73776. This is because with more nodes, proposed approach is applied on six large arbitrary networks
the number of successful communication increases. In some with N = 200, 300, 400, and 500 sensor nodes. We set the
cases, this causes the sensor nodes to die out earlier, leading duty cycle α si = 0.8 for Table V and α si = 0.9 for Table 6,
to unsuccessful communication in later simulation runs. This and θ varies from 50% to 90% for both tables; note that θ =
causes a reduction in ACR value. 50% is equivalent to Areq = 100 × 100 × 50% = 5000 m 2
By analyzing the energy consumption values in Table IV, of the sensed area. All other parameters are given in Table I.
one cannot conclude that larger ACR values imply larger As can be seen from Table V and Table VI, increasing the

Authorized licensed use limited to: University of Canberra. Downloaded on April 28,2020 at 15:52:49 UTC from IEEE Xplore. Restrictions apply.
CHAKRABORTY et al.: ON AREA COVERAGE RELIABILITY OF mWSNs WITH MULTISTATE NODES 5001

TABLE V
ACR VALUES FOR WSN W ITH D UTY C YCLE (αsi ) = 0.8

TABLE VI
ACR VALUES FOR WSN W ITH D UTY C YCLE (αsi ) = 0.9

duty cycle from 0.8 to 0.9 enables more number of nodes to the ACR values of Table V and Table VI for problem 1 with
stay awake and thus the ACR values in Table VI is greater N = 300, and θ = 50%. An increase in the duty cycle from
than the ACR values in Table V. For example, let us consider 0.8 to 0.9 leads to an increase of ACR value by 11%.

Authorized licensed use limited to: University of Canberra. Downloaded on April 28,2020 at 15:52:49 UTC from IEEE Xplore. Restrictions apply.
5002 IEEE SENSORS JOURNAL, VOL. 20, NO. 9, MAY 1, 2020

TABLE VII
ACR VALUES WITH I NCREASING L INK R ELIABILITY

Table V and Table VI also show the effect of the values of θ network reliability stand out as critical challenges. In this
on ACR. As shown in Table V, increasing θ from 50% to 90% work, a simulation scheme to approximate the WSN coverage-
for each problem decreases the ACR values. As an example, for oriented reliability under multi-state nodes is proposed. The
problem 1 with N = 200, increasing θ from 50% to 90% leads proposed approach accounts for node-states, node and link
to a decrease in ACR value by 2.1%. This is because the same reliability, and nodes performance on network’s capability of
mWSN now has to satisfy a larger area-coverage requirement. fulfilling an application-specific area-coverage requirement.
Analyzing the results in Table VI, it can be concluded that It should be pointed out here that the proposed approach
Table VI follows the trend of Table V, in terms of the considers non-rechargeable sensors. Further, investigations are
effect of θ . needed to evaluate ACR for mWSN with rechargeable nodes.
Computing ACR of such mWSNs is challenging because the
battery-discharge process does not follow a linear relationship
D. Effect of Link Reliability on ACR
with time. We aim to investigate this direction in our future
This section depicts the effect of link reliability on ACR. research.
A total of 18 random problems are generated for N = 50, 60,
70, 80, 90, and 100. In this section, we set θ = 50%, while R EFERENCES
all other parameters obey Table I. As evident from Table VII, [1] G. Tolle et al., “A macroscope in the redwoods,” SenSys, vol. 5,
increasing the link reliability from 0.5 to 1, consistently pp. 51–63, Nov. 2005.
improves the ACR value. This is because, with higher link [2] Y. Zou and K. Chakrabarty, “Sensor deployment and target localization
in distributed sensor networks,” Trans. Embedded Comput. Sys., vol. 3,
reliability, more links are formed amongst the sensors as well no. 1, pp. 61–91, Feb. 2004.
as between the sensors and the mobile sink. As a result, [3] J. Li, L. Cui, and B. Zhang, “Self-deployment by distance and orientation
the possibility of each ACTIVE node being connected to the control for mobile sensor networks,” in Proc. Int. Conf. Netw., Sens.
Control (ICNSC), Apr. 2010, pp. 549–553.
mobile sink increases; this eventually increases the ACR value. [4] H. Mostafaei and M. S. Obaidat, “A distributed efficient algorithm for
self-protection of wireless sensor networks,” in Proc. IEEE Int. Conf.
Commun.(ICC), May 2018, pp. 1–6.
VI. C ONCLUSION [5] S. Chakraborty, S. K. Chaturvedi, and N. K. Goyal, “Comments on
an efficient method based on self-generating disjoint minimal cut-sets
Among numerous challenges, when designing a new area- for evaluating reliability measures of interconnection networks,” Int.
monitoring scheme, maintaining connectivity and maximizing J. Performability Eng., vol. 10, no. 7, pp. 771–774, 2014.

Authorized licensed use limited to: University of Canberra. Downloaded on April 28,2020 at 15:52:49 UTC from IEEE Xplore. Restrictions apply.
CHAKRABORTY et al.: ON AREA COVERAGE RELIABILITY OF mWSNs WITH MULTISTATE NODES 5003

[6] S. Chakraborty and N. K. Goyal, “Subset cut enumeration of flow [30] Q. Zhang, D. Li, R. Kang, E. Zio, and P. Zhang, “Reliability analysis of
networks with imperfect nodes,” Int. J. Performability Eng., vol. 11, interdependent networks using percolation theory,” in Proc. Int. Conf.
no. 1, 2015. Signal-Image Technol. Internet-Based Syst., Dec. 2013, pp. 626–629.
[7] S. Chakraborty and N. K. Goyal, “Irredundant subset cut enumeration [31] D. Li, Q. Zhang, E. Zio, S. Havlin, and R. Kang, “Network reliability
for reliability evaluation of flow networks,” IEEE Trans. Rel., vol. 64, analysis based on percolation theory,” Rel. Eng. Syst. Saf., vol. 142,
no. 4, pp. 1194–1202, Dec. 2015. pp. 556–562, Oct. 2015.
[8] S. Chakraborty and N. K. Goyal, “An efficient reliability evalua- [32] L. Liu, X. Zhang, and H. Ma, “Percolation theory-based exposure-path
tion approach for networks with simultaneous multiple-node-pair flow prevention for wireless sensor networks coverage in Internet of Things,”
requirements,” Qual. Reliab. Engng. Int., vol. 33, no. 5, pp. 1067–1082, IEEE Sensors J., vol. 13, no. 10, pp. 3625–3636, Oct. 2013.
Jul. 2017. [33] S. Dasgupta, G. Mao, and B. Anderson, “A new measure of wireless
[9] H. Mostafaei, “Energy-efficient algorithm for reliable routing of wire- network connectivity,” IEEE Trans. Mobile Comput., vol. 14, no. 9,
less sensor networks,” IEEE Trans. Ind. Electron., vol. 66, no. 7, pp. 1765–1779, Sep. 2015.
pp. 5567–5575, Jul. 2019. [34] R. R. Rout, M. S. Krishna, and S. Gupta, “Markov decision process-
[10] H. M. Aboelfotoh, E. S. Elmallah, and H. S. Hassanein, “A flow-based based switching algorithm for sustainable rechargeable wireless sensor
reliability measure for wireless sensor networks,” Int. J. Sensor Netw., networks,” IEEE Sensors J., vol. 16, no. 8, pp. 2788–2797, Apr. 2016.
vol. 2, nos. 5–66, p. 311, 2007. [35] N. Deo, Graph Theory With Applications to Engineering and Computer
[11] G. Fan and S. Jin, “Coverage problem in wireless sensor network: Science. Upper Saddle River, NJ, USA: Prentice-Hall, 1974.
A survey,” J. Netw., vol. 5, no. 9, pp. 1033–1040, 2010. [36] C.-F. Hsin and M. Liu, “Randomly duty-cycled wireless sensor net-
[12] C.-F. Huang and Y.-C. Tseng, “The coverage problem in a wireless works: Dynamics of coverage,” IEEE Trans. Wireless Commun., vol. 5,
sensor network,” Mob. Netw. Appl., vol. 10, no. 4, pp. 519–528, 2005. no. 10, pp. 3182–3192, Nov. 2006.
[13] A. E. Zonouz, L. Xing, V. M. Vokkarane, and Y. Sun, “Applica-
tion communication reliability of wireless sensor networks support-
ing K-coverage,” in Proc. IEEE Int. Conf. Distrib. Comput. Sensor
Syst. (DCoSS), May 2013, pp. 430–435.
[14] A. Shrestha, L. X, and H. Liu, “Modeling and evaluating the reliability Suparna Chakraborty received the
of wireless sensor networks,” in Proc. Annu. Rel. Maintainability Symp. B.Tech. degree in electronics and communication
(RAMS), 2007, pp. 186–191. engineering from WBUT, in 2010, and the M.S.
[15] C. Wang, L. Xing, A. E. Zonouz, V. M. Vokkarane, and Y. L. Sun, degree in network reliability from the Indian
“Communication reliability analysis of wireless sensor networks using Institute of Technology Kharagpur, India, in 2014.
phased-mission model,” Qual. Reliab. Engng. Int., vol. 33, no. 4, She is currently pursuing the Ph.D. degree with
pp. 823–837, Jun. 2017. the School of Electrical Engineering, Computing
[16] D. Li and H. Liu, “Sensor coverage in wireless sensor networks,” in and Mathematical Sciences, Curtin University,
Wireless Networks: Research Technology and Applications. Commack, Perth, WA, Australia, and also with the Subir
NY, USA: Nova, 2009, pp. 1–30. Chowdhury School of Quality and Reliability,
[17] M. Ma and Y. Yang, “Adaptive triangular deployment algorithm for Indian Institute of Technology Kharagpur, India.
unattended mobile sensor networks,” IEEE Trans. Comput., vol. 56, Her research interests include wireless sensor network reliability,
no. 7, pp. 846–847, Jul. 2007. adhoc networks, communication network reliability, and performance
[18] Y. Liu and W. Liang, “Approximate coverage in wireless sensor evaluation of network protocols and algorithms.
networks,” in Proc. IEEE Conf. Local Comput. Netw. 30th Anniver-
sary (LCN), 2005, pp. 68–75.
[19] H. Mostafaei, A. Montieri, V. Persico, and A. Pescape, “An efficient
partial coverage algorithm for wireless sensor networks,” in Proc. IEEE
Symp. Comput. Commun. (ISCC), Jun. 2016, pp. 501–506. Neeraj Kumar Goyal received the Bachelor of
[20] H. Mostafaei, A. Montieri, V. Persico, and A. Pescapé, “A sleep schedul- Engineering (Hons.) degree in electronics and
ing approach based on learning automata for WSN partialcoverage,” communications engineering from MREC Jaipur,
J. Netw. Comput. Appl., vol. 80, pp. 67–78, Feb. 2017. Rajasthan, India, in 2000, and the Ph.D. degree
[21] M. Shazly, E. S. Elmallah, J. Harms, and H. M. Aboelfotoh, “On area in reliability engineering from the Indian Institute
coverage reliability of wireless sensor networks,” in Proc. IEEE 36th of Technology Kharagpur in 2006.
Conf. Local Comput. Netw., Oct. 2011, pp. 580–588. He has served as an Executive in M/s Secure
[22] J. Luo, J. Panchard, M. Piórkowski, M. Grossglauser, and J.-P. Hubaux, Meters Ltd., Udaipur, from 2000 to 2001. He has
“MobiRoute: Routing towards a mobile sink for improving lifetime in been serving IIT Kharagpur as a Faculty Member
sensor networks,” in Distributed Computing in Sensor Systems. Springer, since 2006. He is currently serving as an Asso-
2006, pp. 480–497. ciate Professor with the Subir Chowdhury School
[23] E. Lee, S. Park, F. Yu, and S.-H. Kim, “Communication model and of Quality and Reliability, Indian Institute of Technology Kharagpur, India.
protocol based on multiple static sinks for supporting mobile users He is actively involved in providing consultancy services to national
in wireless sensor networks,” IEEE Trans. Consum. Electron., vol. 56, defense and nuclear power organizations in the area of reliability and
no. 3, pp. 1652–1660, Aug. 2010. safety engineering. He has coauthored books titled Early Software
[24] S. Vupputuri, K. K. Rachuri, and C. Siva Ram Murthy, “Using mobile Reliability Prediction: A Fuzzy Logic Approach published by Springer and
data collectors to improve network lifetime of wireless sensor networks Artificial Neural Network Applications for Software Reliability Prediction
with reliability constraints,” J. Parallel Distrib. Comput., vol. 70, no. 7, published by John Wiley & Sons.
pp. 767–778, Jul. 2010.
[25] M. Di Francesco, S. K. Das, and G. Anastasi, “Data collection in
wireless sensor networks with mobile elements,” ACM Trans. Sen. Netw.,
vol. 8, no. 1, pp. 1–31, Aug. 2011.
[26] S. Prasadkori and R. K. Baghel, “Reliability analysis of the energy Sieteng Soh (Member, IEEE) received the
efficient cluster technique for WSN for mobile sink,” Int. J. Comput. B.S. degree in electrical engineering from the
Appl., vol. 82, no. 5, pp. 40–44, Nov. 2013. University of Wisconsin-Madison in 1987, and
[27] M. Men and L. Chen, “Node energy and state probabilities in linear the M.S. and Ph.D. degrees in electrical engi-
wireless sensor networks,” Int. J. Performability Eng., vol. 12, no. 1, neering from Louisiana State University-Baton
pp. 95–99, 2016. Rouge, in 1989 and 1993, respectively. From
[28] C. Cannings and D. Penman, “Models of random graphs and their 1993 to 2000, he was with Tarumanagara Uni-
applications,” in Handbook of Statistics, vol. 21. Amsterdam, The versity, Indonesia. He is a Senior Lecturer with
Netherlands: Elsevier, 2003, pp. 51–91. the Department of Computing, Curtin Univer-
[29] R. R. Brooks, B. Pillai, S. Racunas, and S. Rai, “Mobile network sity, Perth, WA, Australia. His research interests
analysis using probabilistic connectivity matrices,” IEEE Trans. Syst., include computer networks, network reliability,
Man, Cybern. C, Appl. Rev., vol. 37, no. 4, pp. 694–702, Jul. 2007. wireless mesh networks, and wireless sensor networks.

Authorized licensed use limited to: University of Canberra. Downloaded on April 28,2020 at 15:52:49 UTC from IEEE Xplore. Restrictions apply.