2082 IEEE COMMUNICATIONS LETTERS, VOL. 25, NO.

6, JUNE 2021

Residual-Energy Aware Modeling and Analysis of Time-Varying

Wireless Sensor Networks
Zhaoming Ding , Graduate Student Member, IEEE, Lianfeng Shen , Senior Member, IEEE,
Hongyang Chen , Senior Member, IEEE, Feng Yan , and Nirwan Ansari , Fellow, IEEE

Abstract— In this letter, the residual-energy aware feature of to prolong network lifetime by balancing residual energy and
a sensor node in time-varying wireless sensor networks (WSNs) regional average energy of nodes in a cluster.
is analyzed and modeled as a Markov chain, upon which the In node-scheduling schemes for duty-cycle WSNs,
state-transition probability (STP) about the energy level of any
node with undetermined and deterministic residual energy can
Agarwal et al. [6] developed a stochastic energy model to
be evaluated. Based on Markov chain and energy-efficient relay capture the expected energy consumption, which can be
search region models, an energy-efficient routing algorithm is used to evaluate expected lifetime and cycle lifetime of
proposed to further analyze the impact of STP with known any node. Zhang et al. [7] derived a joint probability of
residual energy on extending network lifetime of time-varying data-packet numbers and node phases for a new duty-cycle
WSNs. Simulation results show that the proposed algorithm can WSN model by constructing a multi-dimensional Markov
effectively extend network lifetime even more than twice while
holding a better energy efficiency as compared with the algorithm process, so that energy consumption of nodes switching from
without considering node-residual energy changes. one phase to another can be obtained. Based on an analytical
hierarchical process with the consideration of node-residual
Index Terms— Energy efficiency, Markov chain, network life-
time, node-residual energy, wireless sensor networks.
energy, distance to the sink, and node degree, Chang et al. [8]
proposed an optimization routing algorithm to balance energy
I. I NTRODUCTION consumption between nodes.
Relay selection (RS) is another approach to save energy
E NERGY consumption of sensor nodes in energy-
constrained wireless sensor networks (WSNs) has an
important impact on the network lifetime, which depends
in multi-hop routing by exploiting spatial diversity [9].
Ding et al. [10] analyzed characteristics of energy-efficient
RS, and proposed the best RS policy-based energy-efficient
on the lifetimes of single nodes that constitute the sensor
routing algorithm, referred to as RS-EERA, to improve
network [1]. Various techniques [2] such as hierarchical clus-
network energy efficiency based on the presented energy-
tering, node scheduling, and energy-efficient relaying have
efficient relay search region (EE-RSR) model. A round-robin
been proposed to extend the network lifetime.
scheduling-based energy-efficient routing algorithm, referred
Clustering is particularly useful for relay-based WSNs.
to as RR-EERA, is further proposed to sequentially select
Heinzelman et al. [3] proposed a self-organizing adaptive pro-
each available relay node for any source node through a
tocol called low energy adaptive clustering hierarchy (LEACH)
scheduling mechanism with the consideration of link-energy
that uses a distributed clustering formation algorithm. How-
load balance [11].
ever, cluster heads in LEACH may become overloaded, thus
With the development of smart sensors and other enabling
causing uneven energy consumption. A hybrid energy-efficient
technologies, WSNs evolve into Internet of Things (IoT)
distributed clustering (HEED) was proposed to generate clus-
with much higher dynamics and diversities, in which energy
ter heads based on energy and communication cost [4].
consumption of each node becomes much larger and nondeter-
Leu et al. [5] proposed an energy-efficient clustering scheme
ministic [12]. As a result, time-varying characteristics in such
Manuscript received December 17, 2020; revised February 6, 2021; WSN stem from uneven energy consumption between nodes
accepted March 6, 2021. Date of publication March 10, 2021; date of current since workloads on any sensor node may vary over time. How-
version June 10, 2021. This work was supported in part by National Key R&D ever, the above energy consumption modeling of nodes does
Program of China (No. 2018YFB1500800), the Specialized Development
Foundation for the Achievement Transformation of Jiangsu Province not reflect the node-residual energy state, which may change
(No. BA2019025), the Pre-research fund of Science and Technology on more frequently in a time-varying WSN for IoT applications.
Near-Surface Detection Laboratory (No. 6142414190405), the Open Project Thus, the related algorithms based on these models may
of Key Laboratory of Wireless Sensor Network & Communication, Shanghai
Institute of Microsystem and Information Technology, Chinese Academy of not be suitable for time-varying WSNs. Motivated by these
Sciences (No. 20190907), and the High-Level Innovation and Entrepreneurial investigations, in this letter, we analyze node-residual energy
Research Team Program in Jiangsu. The associate editor coordinating the state changes during data transmissions in time-varying WSNs.
review of this letter and approving it for publication was M. Erol-Kantarci.
(Corresponding author: Feng Yan.)
Our main contributions include: 1) construct a node-residual
Zhaoming Ding, Lianfeng Shen, and Feng Yan are with the National energy state model for time-varying WSNs, and formu-
Mobile Communications Research Laboratory, Southeast University, late state-transition probabilities (STPs) under unknown and
Nanjing 210096, China (e-mail: [email protected]; [email protected]; known node-residual energy cases, respectively; 2) propose
[email protected]).
Hongyang Chen is with the Research Center for Intelligent Network, an energy-efficient routing algorithm based on the EE-RSR
Zhejiang Laboratory, Hangzhou 311121, China (e-mail: [email protected]). model by employing STP with known node-residual energy
Nirwan Ansari is with the Department of Electrical and Computer Engi- to validate the state model and its analytical results. Key
neering, New Jersey Institute of Technology, Newark, NJ 07102 USA (e-mail:
[email protected]). parameters and notations used in this letter are summarized in
Digital Object Identifier 10.1109/LCOMM.2021.3065062 TABLE I.
1558-2558 © 2021 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: Aditya Engineering College. Downloaded on December 02,2021 at 09:49:42 UTC from IEEE Xplore. Restrictions apply.
DING et al.: RESIDUAL-ENERGY AWARE MODELING AND ANALYSIS OF TIME-VARYING WSNs 2083

TABLE I The above best RS policy does not consider residual-energy

L IST OF K EY PARAMETERS AND N OTATIONS state changes due to varying workloads. As a result, rapid
energy depletion of selected best Aj leads to the reduction
of T . So, both RS-EERA and RR-EERA algorithms based on
the EE-RSR model are not suitable for time-varying WSNs.

III. M ODELING AND A NALYSIS OF T IME -VARYING WSN

AND THE P ROPOSED A LGORITHM
A. Residual-Energy Aware Modeling and Analysis of
Time-Varying WSN
∀n ∈ VG , denote en (t) and ln (t) as residual energy and its
level of node n at time t, respectively. At time t, let ln (t) = l,
where l ∈ Le = {0, 1, · · · , L}, and ln (0) = L which is the
maximum energy level. Thus, residual-energy state of node n
in a time-varying WSN, denoted by sn , can be represented by
sn = ln , where ln ∈ Le . Let EL (l) be initial energy of node n
entering energy-level l, where EL (0) = 0 and EL (L) = en (0).
Fig. 1. EE-RSR RSi for a given node pair Si , D. So, ln (t) = l implies EL (l − 1) < en (t)  EL (l).
The packet length Lpkt follows an exponential distribution
II. S YSTEM M ODEL AND P ROBLEM S TATEMENT (k)
with parameter μL , i.e., Lpkt ∼ E(μL ). Let tn be the time of
A. Energy Consumption Model and EE-RSR Model node n starting to send the kth packet. During data transmis-
(k)
For a given node pair S, D, considering both circuit sions, residual energy of node n at time t = tn +τk can be
(k)
energy and transmit energy, the total power consumption expressed as en (t) = en (tn )− ecn (τk ), where ecn (τk ) denotes
(h) (h)
QS→D in path ZS→D is given by [10] energy consumed by node n in sending the kth packet  during
c
τk interval. Let k = K at time t; en (t) = en (0)− K k=1 en (τk )

h−1
  (1)
(h)
QS→D = Si ,Ai · vSi ,Ai ,
1 + 2 (α)dα (1) can be iteratively calculated for given tn = 0. ∀n ∈ VG ,
i=0 due to Lpkt ∼ E(μL ), τk also follows τk ∼ E(μL vn,m ), where
vn,m is the link rate allocated for node n in sending the kth
where α is the path-loss exponent, 1 and 2 (α) are 1-bit
packet on given link (n, m). Hence, state sn can be modeled
energy spent by electronic circuits of receiver or transmitter,
as a Markov chain.
and the amplifier of transmitter, respectively. While the total
(h) (h) ∀n ∈ VG and m ∈ Nn , energy efficiency of link (n, m),
energy consumption ES→D in ZS→D per bit transmission can
denoted by ηn,m , is defined as the number of transmitted bits
be obtained by substituting link rate vSi ,Ai = 1 into (1).
(h) per Joule on link (n, m) during unit time [bits/J], i.e., ηn,m =
∀Si ∈ ZS→D , from [10], EE-RSR RSi is a circle region vn,m Tx
. Let ηn,m be energy efficiency of link (n, m)
Tx + v
Pn,m 1 n,m
with the optimal relay position (ORP) PSi as its center point Tx v
(Si ) without circuit energy consumption, i.e., ηn,m = Pn,mTx .
and rSi as its radius, where rmax is the maximum value of n,m

rSi and d0 = 2dSi ,PSi , as shown in Fig. 1. For the jth relay ∀n ∈ VG , given K − 1 packets that have been sent by node
(2) (1) n before a time, then STP of WSN for node n from current
node Aj within RSi (xAj ∈ RSi ), ESi →Si+2 = ESi →Aj + sn to next sn is evaluated by Proposition 1.
(1) (1) (K)
EAj →Si+2 < ESi →Si+2 , where i ∈ [0, h − 2], S0 = S, Sh = SD , Proposition 1: ∀n ∈ VG , let t = tn + τK , τK > 0,
and Aj ∈ NSi , i.e., dSi ,Aj  Rtx . (K)
and given current ln (tn ) = h as well as next ln (t) = l
(K)
while satisfying h  l, then Case 1) unknown en (tn ), STP
B. Problem Statement 
qn (sn (t)|sn (t)) is approximately calculated by (2), shown
To make the analysis simple and tractable, the lifetime of at the bottom of the next page, and in (2) K  2; δ =
WSN, denoted by T , is defined as the cumulative active-time EL (h − 1) − EL(l) if h > l, and δ = 0 otherwise; Case 2) known
(K)
of the network until the energy of the first node is exhausted en (tn ), STP qn (sn (t)|sn (t)) is calculated by
[2]. ∀Si ∈ VG , given a node pair Si , D, Si is allowed
qn (sn (t)|sn (t)) = exp (−μL ηn,m δ ∗ )
to forward a packet to Aj , where xAj ∈ RSi . From Fig. 1,   
(1) (2)
PSi and PSi are the current ORP and next ORP of Si , − exp −μL ηn,m en (t(K) n ) − E L (l − 1) , (3)
(1) (1)
respectively. Similarly, PA1 and PA2 are current OPRs of (K) (K)
A1 and A2 , respectively. Based on the EE-RSR model and where en (tn ) ∈ (EL (h−1), EL (h)); δ ∗ = en (tn )−EL (l) if
the best RS policy [10], A1 is chosen by Si to relay the h > l, and δ ∗ = 0 otherwise. In (2) and (3), ηn,m and ηn,m
(1) are link energy efficiencies obtained by node n in sending the
packet for being nearest to PSi . The same RS process will be
performed for A1 when it receives the packet. Consequently, Kth packet and the (K − 1)th packet, respectively, where both
(h) m and m ∈ Nn .
energy-efficient routing ZSi →D can be achieved. Since the
(2) (1) Proof: See Appendix A. 
nodes are deployed randomly, and PSi usually offsets to PAj To facilitate the study, channel fading coefficient hc fol-
(h)
during data transmissions, ZSi →D is sub-optimal. lows an identical exponential distribution with unit-mean,

Authorized licensed use limited to: Aditya Engineering College. Downloaded on December 02,2021 at 09:49:42 UTC from IEEE Xplore. Restrictions apply.
2084 IEEE COMMUNICATIONS LETTERS, VOL. 25, NO. 6, JUNE 2021

i.e., hc ∼ E(1). Given any link (n, m) and its Algorithm 1 MC-EERA Algorithm
expected link outage probability pout , from (1), at least Input: source-destination S, D, ∀S ∈ VG
Tx Tx
Pn,m   (α)dα vn,m . Also, Pn,m needs to satisfy (h)
P Tx h (d 2 /d n,m) α
Output: path ZS→D
n,m c ref n,m SNR SNR SNR
P PN < βmin  pout , where PN and βmin 1 initialization: PN , βmin , dref , Rtx , pout , α, and en (0),
are noise power and the minimum recipient signal-to-noise ∀n ∈ VG ;
(h)
ratio (SNR), respectively. Since random variable hc ∼ E(1), 2 n ← S, ZS→D ← {S};
Tx
substituting the lower bound of Pn,m into above the proba- 3 while eS (t) > 0 do
bility inequality yields 4 compute wS,Aj (t) based on (6) and (3), and allocate
min
SNR
PN βmin vS,A j
by (4), where Aj ∈ NS , j = 1,· · ·, |NS |;
min
vn,m = −1 , (4) 5 set S ← Aj , where wS,Aj (t) = max∀i {wS,Ai (t)};
ref ln (1 − pout )
2 (α)dα (h) (h) 
min
6 ZS→D = ZS→D {Aj };
where vn,m is the minimum of vn,m and dref is a reference 7 if S == D then
distance to the antenna far field, satisfying dn,m  dref . (h) (h)
Tx Tx 8 output ZS→D , and reset S ← n, ZS→D ← {n};
If the minimum Pn,m and vn,m are allocated, then ηn,m =
1/(2 (α)dn,m ). For ease of theoretical analysis, ∀n ∈ VG , let
α

xn = x0 ∈ R2 at the origin. ∀m ∈ Nn , by modeling the TABLE II

position of node m as a homogeneous Poisson point process S IMULATION PARAMETERS
(PPP) Φ = {xi }∞ 2
i=1 ⊂ R with intensity λp = 1/f (An ) where
1 Tx
area f (An ) = πR2 , E 1/ηn,m  in (2) can be estimated by
tx

Tx 22 (α) α dα+2 ref
E 1/ηn,m = α
2 (α)xm λp dxm = Rtx − 2 ,
An α+2 Rtx Fig. 2 depicts the impact of ηn,m and en on qn (sn |sn ) at
(5) a time, ∀n ∈ VG and m ∈ Nn . In fact, ηn,m is limited by
Tx
where An is the communication region of node n and xm is dref and Rtx when Pn,m = 2 (α)dα
n,m vn,m . Assume node n
position coordinate vector of node m , satisfying xm ∈ An . has sent one packet and current ln = 5, but en is unknown.
From the upper sub-figure in Fig. 2, in the case of unknown
B. Markov Chain and EE-RSR Model-Based Energy-Efficient en indicated by en = x, STP of node n from ln = 5 to
Routing Algorithm (MC-EERA) ln = 4 is larger than its SHP when ηn,m is smaller. With
∀n, m ∈ VG , based on the EE-RSR model, the weight of the increase of ηn,m , node n may stay at ln = 5 or enter into
link (n, m) at time t, denoted by wn,m (t), is defined as ln = 4. Since the last ηn,m and K are undetermined, evaluated
 d  qn (sn |sn ) is rough. Different from that, in the case of known
1− n,m em (t)
qn (sn (t)|sn (t)), xm ∈ Rn en from the upper sub-figure, qn (sn |sn ) distinctly shows that
rmax em (0)
(n)
wn,m (t) = (6) node n has a larger possibility to stay at current ln = 5 when
0, xm ∈
/ Rn
ηn,m increases for the case of en = 9.5J. This is because the
where qn (sn (t)|sn (t)) is the state-holding probability (SHP) larger ηn,m helps to reduce energy consumption of the packet
(k) transmission. From the lower sub-figure of Fig. 2, on the other
of node n with known en (tn ) under current state sn when
sending the kth packet on given link (n, m). hand, qn (sn |sn ) in the case of known en is also affected by
By using a Markovian approach with the consideration of Δen (the energy difference between en and the lower bound
residual-energy state changes, an improved algorithm based on of an energy-level range), e.g., Δen = en − EL (h − 1) when
the EE-RSR model, called MC-EERA, is proposed to further ln = h. From the lower sub-figure, node n has a higher
analyze and compare the performance of different algorithms qn (sn |sn ) when Δen becomes larger, and vice versa.
on network energy efficiency and network lifetime. Basic idea For ease of comparison, both RR-EERA and RS-EERA use
of MC-EERA (Algorithm 1) is for any node n to select the the same power and rate allocations as those of MC-EERA.
best relay node which is closer to its ORP while keeping a The difference is the design  of link weight in (6). RS-EERA
d
higher SHP when node n starts to send a packet. only concerns 1− n,m (n) while RR-EERA replaces SHP by
rmax
the scheduling mechanism to select each valid relay node
IV. P ERFORMANCE E VALUATION in turn. Moreover, network energy efficiency, denoted by η,
K K (k,h)
Numerical simulations are conducted to validate the state is evaluated by k=1 Nk,bit / k=1 QS→D [bits/J], where
model and analytical results by comparing different algorithms Nk,bit is the bit number of the kth packet received at the sink,
(k,h) (h) (h)
using Network Simulator 3 (NS3), where a given source node and QS→D is QS→D obtained by the kth packet on ZS→D .
and the sink are at (50m, 50m) and (950m, 950m), respectively, Fig. 3 depicts performance comparison among different
and other major simulation parameters are listed in TABLE II. algorithms about η and T , where the reference value is the one

 K−2
exp (−μL ηn,m δ) − exp (−μL ηn,m (EL (h) − EL (l − 1))) 1 − 1 ηn,m
qn (sn (t)|sn (t)) ≈ Tx
, (2)
exp (−μL ηn,m (en (0) − EL (h)))−exp (−μL ηn,m (en (0) − EL (h − 1))) 1+ηn,m E 1/ηn,m 

Authorized licensed use limited to: Aditya Engineering College. Downloaded on December 02,2021 at 09:49:42 UTC from IEEE Xplore. Restrictions apply.
DING et al.: RESIDUAL-ENERGY AWARE MODELING AND ANALYSIS OF TIME-VARYING WSNs 2085

From (7),
qn (sn |sn )
P {EL (h − 1) − EL (l) < ecn (τK )  EL (h) − EL (l − 1)}
= 
(K)
P EL (h − 1) < en (tn )  EL (h)
(K)  (K)
if en (tn ) is unknown, and qn (sn |sn ) = P en (tn ) −
c (K) 
EL (l)  en (τK ) < en (tn ) − EL (l − 1) otherwise.
Tx
Given Pn,m and vn,m when node n sends the kth packet on
link (n, m), energy consumption during τK can be formulated
Fig. 2. STP qn (sn |sn ) vs. ηn,m . α = 2, L = 5, K = 2. by ecn (τK ) = Pn,m
Tx
τK + 1 vn,m τK . Since τK ∼ E(μL vn,m ),
Tx
by replacing vn,m /(Pn,m + 1 vn,m ) with ηn,m , we obtain the
 (K)
joint probability in qn (sn (t)|sn (t)) with unknown en (tn )
below
P {EL (h − 1)−EL (l) < ecn (τK )  EL (h)−EL (l − 1)}
= e−μL ηn,m (EL (h−1)−EL (l)) −e−μL ηn,m (EL (h)−EL (l−1)) . (8)
Since ecn (τK )  0, and EL (h − 1) − EL (l) < 0 when h = l,
replacing EL (h − 1) − EL (l) with δ in (8) yields the joint
probability shown in (2), where δ = EL (h − 1) − EL (l) if
h > l, and δ = 0 otherwise. Similarly, for qn (sn |sn ) with
(K) (K)
Fig. 3. Performance comparison on η and T vs. N . α = 2, pout = 0.2. known en (tn ), letting δ ∗ = en (tn ) − EL (l) if h > l, and
∗
δ = 0 otherwise, then (3) follows. 
(K)
obtained by RS-EERA in the case of network size N = 100. For qn (sn |sn ) with unknown en (tn ), P EL (h − 1) <
(K)   (K)   (K)
From the figure, RS-EERA achieves the largest η but the least en (tn )  EL (h) = P en (tn )  EL (h) − P en (tn ) 
T , and RR-EERA is the opposite. Although η obtained by  (K) (K)
EL (h−1) . Since K −1 packets sent at time tn , en (tn ) =
MC-EERA is slightly less than that by RS-EERA, its T can K−1 c K−2 c
en (0)− k=1 en (τk ) = en (0)− k=1 en (τk )− ecn (τ(K−1) ).
approach the longest one obtained by RR-EERA, which is Tx 
Given Pn,m  and vn,m on selected link (n, m ) for the
much larger than that by RS-EERA. As N increases, η and T
(K − 1)th packet. Also, due to τ(K−1) ) ∼ E(μL vn,m ),
obtained by each algorithm improve except T by RS-EERA,  (K) 
in which the optimal path for a given S, D remains the P en (tn )  EL (h) can be calculated by
same for each fixed N . Last, from the figure, MC-EERA  
P en (t(K)
n )  EL (h) = exp − μL ηn,m
generally outperforms RR-EERA, and can effectively extend
K−2 
T as compared with RS-EERA, especially in a dense WSN  
c
even more than twice, e.g., in the case of N  260. × (en (0) − EL (h)) × E exp (μL ηn,m en (τk )) .
k=1
V. C ONCLUSION (9)

In this letter, we have analyzed the node-residual energy In considering τk ∼ E(μL vn,j ) and τk is independent iden-
state of time-varying WSNs that is modeled as a Markov tically distributed, substituting formulated ecn (τk ) into the
chain. Hence, STPs about the energy level of any node can be expectation in (9) yields
evaluated. Based on the EE-RSR model, a routing algorithm Tx
E 1/(1 − ηn,m /ηn,j )
called MC-EERA by employing STP on the same level has E [exp (μL ηn,m ecn (τk ))] = , (10)
1 − 1 ηn,m
been proposed to further analyze the impact of STP with
Tx
known residual energy on T . Simulation results show that where ηn,j is a random variable, which indicates energy
MC-EERA can effectively extend T even more than twice efficiency of link (n, j) without considering circuit energy for
Tx
while maintaining a better η as compared with the algorithm node n in sending the kth packet. Owing to ηn,m /ηn,j ∈
Tx Tx
without considering node-residual energy changes. (0, 1), we can obtain 1/(1 − ηn,m /ηn,j ) ≈ 1 + ηn,m /ηn,j ,
and apply this approximation into (10) to yield
A PPENDIX A Tx Tx
E 1/(1 − ηn,m /ηn,j ) ≈ 1 + ηn,m /E ηn,j , (11)
P ROOF OF P ROPOSITION 1
Tx Tx
(K) where E ηn,j can be replaced with E ηn,m because
Since qn (sn (t)|sn (t)) = P{ln (t) = l| ln (tn ) = h}, by using 

(K) selected link (n, j) is random. By substituting (10) with the

en (t) = en (tn )−ecn (τK ) and omitting symbol t in sn (t) for
approximation
 as in (11)into (9), we obtain the approximation

brevity without confusion, qn (sn |sn ) is rewritten as (K) (K)
of P en (tn )  EL (h) , and thus P en (tn )  EL (h− 1) .
  (K) 
qn (sn |sn ) = P EL (l − 1) < en (t(K) c Consequently, P EL (h−1) < en (tn )  EL (h) is achieved.
n )−en (τK )  EL (l)
 Finally, by (7) and (8)-(11), qn (sn |sn ) with unknown
 EL (h − 1) < en (t(K)
n )  EL (h) . (7) (K)
en (tn ) follows that in (2). The proof is complete.

Authorized licensed use limited to: Aditya Engineering College. Downloaded on December 02,2021 at 09:49:42 UTC from IEEE Xplore. Restrictions apply.
2086 IEEE COMMUNICATIONS LETTERS, VOL. 25, NO. 6, JUNE 2021

R EFERENCES [7] Y. Zhang and W. W. Li, “Energy consumption analysis of a duty cycle
wireless sensor network model,” IEEE Access, vol. 7, pp. 33405–33413,
[1] I. Dietrich and F. Dressler, “On the lifetime of wireless sensor networks,” 2019.
ACM Trans. Sensor Netw., vol. 5, no. 1, pp. 1–39, Feb. 2009. [8] Y. Chang et al., “Distributed joint optimization routing algorithm based
[2] H. Yetgin et al., “A survey of network lifetime maximization techniques on the analytic hierarchy process for wireless sensor networks,” IEEE
in wireless sensor networks,” IEEE Commun. Surveys Tuts., vol. 19, Commun. Lett., vol. 21, no. 12, pp. 2718–2721, Dec. 2017.
no. 2, pp. 828–854, 2nd Quart., 2017. [9] B. Li et al., “Energy-effective relay selection by utilizing spacial
[3] W. R. Heinzelman et al., “Energy-efficient communication protocol for diversity for random wireless sensor networks,” IEEE Commun. Lett.,
wireless microsensor networks,” in Proc. 33rd Annu. Hawaii Int. Conf. vol. 17, no. 10, pp. 1972–1975, Oct. 2013.
Syst. Sci., Maui, HI, USA, Jan. 2000, pp. 1–10. [10] Z. Ding et al., “Energy-efficient relay selection with blockage for
[4] O. Younis and S. Fahmy, “HEED: A hybrid, energy-efficient, distributed LOS transmissions in wireless sensor networks,” in Proc. IEEE
clustering approach for ad hoc sensor networks,” IEEE Trans. Mobile 88th Veh. Technol. Conf. (VTC-Fall), Chicago, IL, USA, Aug. 2018,
Comput., vol. 3, no. 4, pp. 366–379, Oct. 2004. pp. 1–5.
[5] J.-S. Leu et al., “Energy efficient clustering scheme for prolonging the [11] Z. Ding et al., “An interference-aware energy-efficient routing algorithm
lifetime of wireless sensor network with isolated nodes,” IEEE Commun. with quality of service requirements for software-defined WSNs,” IET
Lett., vol. 19, no. 2, pp. 259–262, Feb. 2015. Commun., vol. 13, no. 18, pp. 3105–3116, Nov. 2019.
[6] V. Agarwal et al., “Modeling energy consumption and lifetime of a [12] X. Liu and N. Ansari, “Toward green IoT: Energy solutions and
wireless sensor node operating on a contention-based MAC protocol,” key challenges,” IEEE Commun. Mag., vol. 57, no. 3, pp. 104–110,
IEEE Sensors J., vol. 17, no. 16, pp. 5153–5168, Aug. 2017. Mar. 2019.

Authorized licensed use limited to: Aditya Engineering College. Downloaded on December 02,2021 at 09:49:42 UTC from IEEE Xplore. Restrictions apply.