This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS 1

A WPT-Enabled UAV-Assisted Condition

Monitoring Scheme for Wireless
Sensor Networks
Tharindu Dilshan Ponnimbaduge Perera , Graduate Student Member, IEEE, Stefan Panic, Member, IEEE,
Dushantha Nalin K. Jayakody , Senior Member, IEEE, P. Muthuchidambaranathan, Member, IEEE,
and Jun Li , Senior Member, IEEE

Abstract— In this paper, a resource allocation and data comparison and insights of the system performance results.
gathering scenario of an unmanned aerial vehicle (UAV) assisted Finally, the performance superiority of the proposed system is
wireless powered sensor network is investigated, in which the demonstrated with compare to identified baseline WSNs.
sensor nodes (SNs) are remotely powered by power beacons
(PBs) via radio-frequency wireless power transmission (RF- Index Terms— RF energy harvesting, RF wireless power
WPT). A time-block structure with two phases is proposed to transfer, UAV communication, wireless sensor networks and
accommodate operations in the proposed system. During Phase- 5G communication.
I, SNs harvest energy from PBs and periodically send its sensed
data to the selected cluster heads (CHs). In Phase-II, an UAV I. I NTRODUCTION
collects the data from CHs to be delivered to the data sink for
further processing avoiding the need for long range transmission
and multi hop communication to the data sink. Then, a closed-
form expression for outage probability of the proposed system
A DVANCES in fifth generation and beyond (5GB)
communication networks are designed to accommo-
date emerging technologies such as Internet-of-Things (IoT),
over Rayleigh and Rician fading channels is derived. Next, outage to increase new innovations across various industries. Oper-
probability minimization problem is formulated to obtain optimal ationally advanced IoT services are empowered by using
time allocation for RF-WPT energy harvesting to improve the
system performance. Due to the complexity of the problem, massive number of sensor nodes (SNs) in wireless sensor net-
Lagrangian duality method is used to develop an asymptotic works (WSNs). Multiple groups of SNs, which are positioned
optimal solution with less execution complexity avoiding complex in the field-of-interest are used to sense the environmental
brute force/ exhaustive search approach. Furthermore, a heuristic conditions of special events and hard-to-reach locations such
method is presented to further lower the computation complexity. as natural disasters, agricultural fields, smart cities, volcanic
Simulation results reveal the superiority of the proposed methods
compare to brute force/ exhaustive search approach via analysis, environments, natural gas reservoirs etc. A sensor node (SN)
is composed of multiple sensors to detect specific changes
Manuscript received March 31, 2020; revised June 23, 2020; accepted in the surrounding environment, which are later to be deliv-
July 30, 2020. This work was supported in part by the Russian Foundation for ered to data sinks for further processing. However, these
Basic Research (RFBR) under Grant 19-37-90037, in part by the framework of
the Competitiveness Enhancement Program of the National Research Tomsk SNs are significantly constrained in terms of its battery life,
Polytechnic University, Russia, under Grant VIU-ISHITR-180/2020, and in limiting the operational capabilities of the WSN and achiev-
part by the Scheme for Promotion of Academic and Research Collaboration able quality of service (QoS). Hence, a significant number
(SPARC), Ministry of Human Resource Development, India, under Grant
SPARC/2018 − 2019/P145/S L. The work of Jun Li was supported in of different energy-aware protocols have been proposed to
part by the National Key Research and Development Program under Grant minimize energy consumption in order to improve the life-
2018YFB1004800 and in part by the National Natural Science Foundation time of WSNs. Nevertheless, current energy-aware protocols
of China under Grant 61727802 and Grant 61872184. The Associate Editor
for this article was M. Guizani. (Corresponding authors: Dushantha Nalin unable to achieve uninterrupted long-term operations for the
K. Jayakody; Jun Li.) WSNs [1]. Thus, energy harvesting (EH) techniques have been
Tharindu Dilshan Ponnimbaduge Perera and Stefan Panic are with the applied to WSNs enabling SNs to powered-up or replenish its
School of Computer Science and Robotics, National Research Tomsk Poly-
technic University, 634050 Tomsk, Russia (e-mail: [email protected]; inbuilt batteries from ambient sources [2]. However, in general,
[email protected]). ambient energy sources i.e., solar, wind, are unable to power-
Dushantha Nalin K. Jayakody is with the School of Computer Science and up WSNs continuously, even after adopting effective energy
Robotics, National Research Tomsk Polytechnic University, 634050 Tomsk,
Russia, and also with the Centre for Telecommunication Research, School allocation and management schemes [3]. To overcome the
of Engineering, Sri Lanka Technological Campus, Padukka 10500, Sri Lanka aforementioned drawbacks, a new EH technique named radio
(e-mail: [email protected]). frequency wireless power transfer (RF-WPT) has emerged to
P. Muthuchidambaranathan is with the Department of Electronics and
Communication Engineering, National Institute of Technology Tiruchirappalli, power-up WSNs. RF-WPT is reliable and stable compare to
Tiruchirappalli 620015, India (e-mail: [email protected]). ambient energy sources. Thus, EH from RF-WPT is suitable
Jun Li is with the School of Electronic and Optical Engineering, Nanjing for low power energy constrained communication networks
University of Science and Technology, Nanjing 210094, China (e-mail:
[email protected]). such as WSNs, to increase the overall operational time unin-
Digital Object Identifier 10.1109/TITS.2020.3018493 terruptedly [1], [2]. RF-WPT is investigated in open literature
1524-9050 © 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: Auckland University of Technology. Downloaded on November 05,2020 at 16:17:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

along with recent advances in wireless communication sys- schemes i.e., static collection and direct trajectory. A novel
tems, cf. [2]. data acquisition framework in WSN using UAV is proposed
In recent decade, RF-WPT applied WSNs have been inves- in [15] to increase the efficiency of data gathering process.
tigated extensively to improve operational time of the com- The authors have introduced a priority based frame selection
munication systems [4]–[9]. The simulation for a complete scheme to reduce the number of redundant data transmission
system of RF-WPT is used to energized SNs of WSN in [4]. between SNs and the UAV. The UAV aided data collection
In addition, the authors of [4] were implemented a prototype to strategy for a SNs located in a straight line has been studied
identify the difference between simulation and the real world in [16]. Here the aviation time of the UAV can be minimized
applications. The authors of [5] implemented a RF energy by following classical water filling policy for optimal power
harvester made of compact ferrite rod antenna together with allocation.
rectenna, in principle, to power-up SNs within 120 km of If all SNs transmit the sensed data to the data sink with
150 kW transmitter. In [6], the authors aimed to optimize the aid of an UAV, energy efficiency of the entire system
the power allocation strategy in priority constraint RF-WPT can be decreased in a considerable amount causing decrease
system with a large number of SNs in cognitive IoT. The in QoS. Generally, neighbouring SNs have the similar data
authors proposed a critical path based decomposition strategy since each group of SNs collect information within a specific
with low computational complexity. A selective, tracking area [16]. In addition, collecting data from each SN in a
and power adaptive far-field RF-WPT system, which can large scale WSN using UAV can be practically infeasible
integrate into passive WSNs proposed in [7]. Furthermore, due to UAV’s limited inbuilt battery capacity. Thus, as a
the authors have shown that effective far-field RF-WPT links solution for the aforementioned problems, a cluster head (CH)
can be created with reasonable simplicity making RF-WPT is selected from each cluster of SNs in the WSN. CHs received
a competitive candidate for energy constrained problem in the data from SNs in the matching cluster and then UAV
WSNs. The authors of [8] presented RF-WPT and intelligent collects the data from the CHs in order to be conveyed to
routing modules to preserve energy consumption and life- the data sink. Thus, the selection of CH for a given cluster is
time of SNs. The authors exploited wireless power charging one of the vital problem in WSNs, which affects the overall
device to solve the multi-objective function for charging trails. system performance. In order to overcome unreasonable CH
By considering downlink RF-WPT and uplink information selections and excessive energy consumption, a modified CH
transmission, the authors of [9] established a task processing selection algorithm based on LEACH was proposed in [17].
strategy including uplink and downlink power information Furthermore, the authors managed to balance the network
dynamic time division. Furthermore, it is shown that the down energy burden and improved energy efficiency by leveraging
going energy transmission can effectively improve the average CH competitive mechanism in the proposed algorithm. In [18],
received power maximizing the throughput. firefly with cyclic randomization is proposed for CH selection
In most of the WSN applications, SNs are required to to improve communication network performance. In [19],
send collected sensor observations directly or through multiple the authors have proposed CH selection scheme that rotates
hops to data sinks for further processing. Thus, there is a the CH position among the SNs with higher energy levels as
requirement for an innovative new design of data collection compared to other.
strategy to enhance the scalability and the performance of the However, [4-16] have not focused on the combination of
WSN. Recently, unmanned aerial vehicle (UAV) is identified RF-WPT, UAV aided data collection strategy and the perfor-
as an effective application to collect data from SNs located in mance metrics at the data sinks in a WSN while considering
hard-to-reach environments, when there are no feasible con- possibility of occupying different types of fading channels
nections towards data sinks [10]. As Compared to conventional within the communication network, which can resemble a real
WSNs that depend on static data collecting nodes or multi- world practical scenario in IoT applications. RF-WPT is a
hop data relaying, the UAV can directly communicate with potential and effective candidate to reduce carbon footprint in
the SN exploiting favourable LoS dominant air-to-ground IoT applications [20]. Following are the identified advantages
channels. Furthermore, UAVs are help to increase coverage of RF-WPT for future IoT applications, i.e. 1). RF-WPT acts
area and sum-throughput of the WSNs while reducing the as a bridge between green energy sources and communication
overall energy consumption of the WSNs [11], [12]. devices, 2). The energy harvested via RF-WPT at a receiver
Various applications of UAV-aided WSNs have been inves- is foreseeable, 3). The amount of harvested energy depends
tigated in the literature [13]–[18]. A UAV-aided data collec- upon to transmit power, propagation loss, and wavelength.
tion scheme for wireless powered WSN over Rician fading In addition, the rapid advances in intelligent transportation
channels is developed in [13]. The authors analyzed the systems (ITS) have led to a rapid expansion in condition
outage probability of the proposed system under the assump- monitoring of communication systems, infrastructures, vehi-
tion of Rician fading channels and demonstrated that for cles, and machinery using SNs [21]. Thus, in this work,
the given information rate, outage probability threshold and an energy efficient UAV-aided data collection strategy for wire-
UAV location, the achievable service range can be uniquely less powered WSN is proposed and the system performance
determined. In [14], the authors have jointly optimized the is investigated in terms of outage probability and achievable
SNs’ wake up schedule and UAV’s trajectory to minimize throughput at the data sink. In particular, this work aims at
the energy consumption of each SN. This proposed scheme introducing a framework to realize the integration of WPT,
achieved significant energy savings compared to its benchmark UAV into WSNs to improve overall energy efficiency of

Authorized licensed use limited to: Auckland University of Technology. Downloaded on November 05,2020 at 16:17:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

PERERA et al.: WPT-ENABLED UAV-ASSISTED CONDITION MONITORING SCHEME FOR WIRELESS SENSOR NETWORKS 3

IoT applications such as ITS, smart cities, etc. while main-

taining the expected QoS. The main contributions of the paper
are fourfold, which are summarized as follows.
• First, a wireless powered WSN is designed, in which
SNs are remotely powered by the PBs using RF-WPT.
A new time-block structure is presented by dividing
entire operational time into two phases. In the first phase,
WSN is clustered, an optimal CHs are selected within
the powered-up SNs and each SNs periodically sends its
sensed data to CHs. In the second phase, an UAV used
to increase energy level of CH via RF-WPT and collects
the aggregated data from the CHs.
• Secondly, under the assumption of Rayleigh and Rician
fading channels, the outage probability at the data sink
Fig. 1. Illustration of reference UAV aided data collection in WSN.
is derived. Then an optimization problem is formulated (a) Sample WSN with randomly distributed SNs and PBs. (b) Sample output
to quantify the time allocation for EH that minimizes of the proposed data gathering scheme in Phase II with four clusters (separated
the outage probability subject to SNs transmit power, with dotted lines), selected CHs and projected UAV trajectory.
harvested energy and UAV’s trajectory constraints.
• Thirdly, an asymptotic optimal solution is proposed by
and landing locations, post-mission flying paths, the capacity
considering the duality of the formulated optimization of the UAV’s battery, etc.
problem to reduce the complexity as compare to brute- A time-block structure for the data collection process of
force or exhaustive search mechanism. Then, a heuristic the proposed WSN is illustrated in Fig. 2. Entire operational
solutions for clustering, CH selection and UAV’s trajec- time T is partitioned into two phases T α and T (1 − α),
tory is presented to further reduce the complexity. where 0 < α < 1. Phase-I is further divided into another
• Finally, the effectiveness of proposed asymptotic and
two fractions T ατ and T α(1 − τ ). During the time T ατ ,
heuristic solutions for the formulated problem, in compar- all SNs harvest energy from the RF-WPT signals received
ison to the existing traditional brute-force is demonstrated from PBs to recharge their batteries. Then, in the next time
by the aid of simulation results. Furthermore, the impacts fraction T α(1 − τ ), SNs are grouped into clusters, an optimal
of different system parameters to the overall system CH is selected for each clusters and then, SNs transmit the
performance are investigated. Finally, the performance sensed data to the respective CH. Time allocated for Phase-II
superiority of the proposed WSN is demonstrated com- is equally divided into M number of clusters which can be
pare to the benchmark WSNs. represented as T (1−α)/M. Then the time fraction T (1−α)/M
is further divided into two time slots Fβ and F(1 − β), where
The remainder of this paper is organized as follows. The
F = T (1 − α)/M and 0 ≤ β < 1. During the time slot
Section II depicts the system model and the time block
F(1 − β), CH harvests energy from RF-WPT signal of
structure. Section III offers detail derivation of the outage
the UAV. Then, CH transmits the aggregated data to the UAV
probability of the system. The problem formulation for outage
in order to be sent to the data sink for further processing.
minimization is provided in Section IV. The proposed asymp-
The channel coefficients between SNs-PBs and SNs-CHs are
totic and heuristic solutions are given in Section V and VI,
modeled as Rayleigh fading random variable h x y from node x
respectively. Simulation results are provided in Section VII,
to y, where x, y ∈ {n, j }. The LoS channel between the
followed the conclusion.
UAV and the CHs are modeled using Rician fading random
variable g x y , where x, y ∈ {m, u}.
II. S YSTEM M ODEL It is assumed that the data transmission from each SN to
A UAV-aided WSN is considered in this paper as illustrated CH is orthogonal1 as in [22] to avoid interference with each
in Fig. 1 with a UAV flying over N number of SNs and other’s transmission. Thus, it is possible to have multiple SNs
collecting data from the selected CHs within a given time to transmit data in same time slot to the CHs as long as
period of T . M PBs are randomly deployed within the WSN to they do not conflict with each other. Similarly, it is assumed
power-up the SNs. Initially, all SNs do not have energy in their that the CHs are not transmitting simultaneously to the UAV
rechargeable batteries to transmit sensed data. A reference avoiding interference with each other. We have consider two
WSN is illustrated in Fig. 1(a), where the UAV’s starting minimum SNR threshold values γ1 and γ2 for the received
and ending points are represented by S and D and SNs are signals at the CHs and the UAV to maintain the expected
denoted by circles, respectively. It is assumed that the location QoS, respectively. This is mainly due to the difference within
D represents the desired data sink where sensor observations the weight of received information at the CHs and the UAV.
are intended to be delivered for further processing. It is also At the beginning, the UAV works out its trajectory and
noteworthy that the S and D are not fixed and can position in communication schedule for CHs using the knowledge of its
different locations depending on the structure of the WSN. It is 1 Orthogonality can be achieved in various ways, i.e. by performing Orthog-
also noteworthy that in practice, the start and final locations onal frequency-division multiple access (OFDMA) or Time-division multiple
of the UAV depend on multiple factors such as UAV take-off access (TDMA) approach.

Authorized licensed use limited to: Auckland University of Technology. Downloaded on November 05,2020 at 16:17:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

Fig. 2. Time allocation diagram of the proposed WSN, where total operational time is divide into two phases. Time allocation for energy harvesting process
for SNs and CHs denote by τ and β.

locality with the CHs along with the trajectory. Furthermore, h mn denotes the channel coefficient between SN n and the
CHs’ locations and energy level are assumed to be known at CH of cluster m, sn denotes the normalized modulated signal
the UAV. The details of such approach can be found in [23]. transmitted from SN n with the value E[|sn |2 ] = 1 and
n mn represents the AWGN noise between the SN n and the
A. Phase-I respective CH. Thus, by considering (1) and (2), the effective
Let P j denotes the transmission power of the j t h PB during signal-to-noise ratio (SNR) at the CH in cluster m can be given
the time T ατ , where j ∈ {1, .., M}. Assuming that the as
transmitted RF-WPT signals by PBs are mutually independent J −1 P d ω |h |2 |h |2 τ
 j nj nj mn
as in [22], the total harvested energy at the n t h SN can be γm = . (4)
(1 − τ )σmn
2
expressed as n=1


M 
M
En = En j = P j dnωj ηn |h n j |2 T ατ, (1) B. Phase-II
j =1 j =1
At the beginning of the Phase-II, CHs possess the aggre-
where dn j denotes the distance between n t h SN and j t h PB, gated data received from the SNs during the Phase-I. During
ω represents the path-loss exponent and 0 < ηn < 1 is the time slot Fβ, CH in m t h cluster harvests energy from the
the energy conversion efficiency at the n t h SN. After SNs UAV’s RF-WPT signal. Thus, the amount of energy harvested
energized its rechargeable batteries, SNs sense data from the by the CH can be expressed as
surrounding environment. The transmit power of the each  
Pu ηn |gmu |2 ρu Fβ, if E mu < E max
SN during the Phase-I can be given by E mu = (5)
0, otherwise
M P d ω η |h |2 τ

En j nj n nj
where Pu denotes the RF-WPT transmission power of the
Pn = = . (2)
T α(1 − τ ) (1 − τ ) UAV, F = T (1 − α)/M, gmu represents the complex fading
j =1
coefficient between the UAV and CH, ρu = κdmu ω denotes
Therefore, the received signals by the CH at the end of
the pathloss between UAV and CH, where κ is the constant
Phase-I can be expressed as
coefficient of the pathloss and dmu is the LoS distance between
J −1 
 UAV and the CH. It is also noteworthy that the dmu = Hmin
ym = Pn |h mn |sn + n mn , (3) during the time when UAV is hovering on top of the CH. Also
n=1 it is assumed that the Hmin is the minimum altitude required
where m ∈ {1, . . . , M} is the index of the clusters in the for maneuvering without any terrain blockage and it complies
WSN, J = N/M denotes the number of SNs per each cluster, with the municipal rules and regulations of the area. Thus,

Authorized licensed use limited to: Auckland University of Technology. Downloaded on November 05,2020 at 16:17:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

PERERA et al.: WPT-ENABLED UAV-ASSISTED CONDITION MONITORING SCHEME FOR WIRELESS SENSOR NETWORKS 5

the total energy available in CH to use for data transmission cluster m, where m = {1, . . . , M} can be calculated as
can be calculated by adding (3) and (4).2 The received signal  ∞ ∞
m
at the UAV from the CH of cluster m can be written as Pout = FX 2
 0
 0 √ √
yum = Pnm |gum |ρu (ym ) + n um , (6) γ2 a3 X 1 + γ2 X 1 ρu σu2 + γ2 σu2
×
X 1 X 3 (a1 − a2 γ2 )
where Pnm denotes the transmission power of the CH in
× f X 1 (X 1 ) f X 3 (X 3 )d X 1 d X 3 , (11)
cluster m, gum represents the complex fading coefficient
between CH in m and the UAV and n um denotes the AWGN where cumulative distribution function (CDF) FX 2 (X 2 ) is fol-
noise between CH and the UAV. Then, the end-to-end SNR lows Rayleigh distribution. Thus, probability density function
can be written as (PDF) of Rayleigh distributed SNR at the CH can be expressed
 J −1 as
Pm |gum |2 ρu2 m=1 Pn |h mn |2 
γum = . (7) X2
Pm |gum | ρu σmn + σum
2 2 2 2
FX 2 (X 2 ) = 1 − exp − , (12)
mn
By substituting (2), (4) and (5) into (7) SNR at the UAV
at the end of data collecting from CH in cluster m can be where mn = E(X 2 ) denotes the average channel SNR
arrange as in (8) given in bottom of the page, where a1 = between SNs and the CH. Similarly, PDF function f X 3 (X 3 ) is
P j dnωj ατ Mηn2 +ρu2 Pu ρu (1−α)β ηn P j dnωj ατρu Nakagami-m distributed SNR at the SN and can be expressed
(1−α)(1−τ )(1−β) , a2 = a1κ 2, a3 = (1−α)(1−β) . as

III. O UTAGE P ROBABILITY A NALYSIS X 3M−1 M M exp (−M X 3 )

f X 3 (X 3 ) = , (13)
(M) M
nj
In this paper, a delay-limited transmission mode is con-
sidered.3 Thus, in this section, detailed derivation of outage where nMj = E(X 3 ) average channel SNR between PBs and
probability of the proposed system is provided. Let us observe SNs. Since each CHs receive multiple signals from SNs during
the random variables X 1 , X 2 and X 3 denoting squares of the second part of the
channel gain amplitudes represent in (8), where X 1 = |gu,m |4 , Phase
J −1
I, random variable X 2 can be
also written as X 2 = n=1 |h m,n |2 . Thus, sum of M number
X 2 = |h m,n |2 and X 3 = |h n, j |2 . In this work, the outage of distributed Rayleigh random variables can be expressed as
probability within a single cluster is defined as the probability Nakagami-m random process with parameter m = M [26]. The
that the end-to-end SNR at the UAV falls under a desired PDF of X 1 is considered as double distributed Rician random
threshold SNR value γ2 , which can be expressed as variable, thus, SNR at the UAV can be expressed as
a1 X 1 X 2 X 3 ∞ 
 ∞ ( p+q)
Pr ( √ √ ≤ γ2 ), (9) (1 + K m )( p+q+2) K m exp(−2K m )
a2 X 1 X 2 X 3 + a3 X 1 X 3 + X 1 ρu σu2 + σu2 f X 1 (X 1 ) = 2 p+q+2
p=0 q=0 um ( p + 1) (q + 1) p!q!
or equivalently 
√ √
p+q
(1 + K 1 ) 
× X1 2
K ( p−q) 2 X 1 , (14)
γ2 a3 X 1 + γ2 X 1 ρu σu2 + γ2 σu2
Pr (X 2 ≤ ), (10) um
X 1 X 3 (a1 − a2 γ2 ) where K m denotes the Rician K -factor, which is defined as the
with the assumption of (a1 − a2 γ2 > 0), which is feasible in ratio of the power of LoS component to the scatter component,
practical scenario [26]. Therefore, Outage probability of the um denotes the average channel SNR between CH and the
UAV, K v () represents the v t h order modified Bessel function
2 Note that SNs can choose to perform power control without exhausting
of the second kind and (x) denotes the complete Gamma
the harvested energy as in [1], [24], which is considered as out of the scope
of this paper
function. The double integral form in (11) could be further
3 Achievable sum-throughput can be calculated by evaluating the outage simplified into single integral form by integrating over X 3
probability of the system with fixed transmission rate [25]. variable as given in (15), shown at the bottom of the page.

 J −1  M
j =1 a1 |gu,m | |h n, j | |h m,n |
4 2 2
n=1
γum =  J −1  M  M
(8)
j =1 |gu,m | |h n, j | |h m,n | a2 + j =1 |gu,m | |h m,n | a3 + |gu,m |2 |ρu σu2 + σu2
4 2 2 2 2
n=1

∞
∞  ( p+q) M  √
∞ γ
√ M
 (1 + K i )( p+q+2) K exp(−2K i )M 2 X 1 + γ2 X 1 ρu σu2 + γ2 σu2 2
2 a3
i
m
Pout =1−4
p+q+2 M
M
X 1 (a1 − a2 γ2 )
n, j hi ( p + 1) (q + 1) p!q!
p=0 q=0 2 0
um
⎛ ⎞
 √ √ 
p+q
⎝ γ 2 a 3 X 1 + γ 2 X 1 ρu σu
2 + γ σ2
2 u ⎠ (1 + K 1 ) 
× X1 K M 2
2
M K ( p−q) 2 X1 d X1 (15)
X 1 (a1 − a2 γ2 ) um um

Authorized licensed use limited to: Auckland University of Technology. Downloaded on November 05,2020 at 16:17:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

However, to the best of the authors’ knowledge, this integral TABLE I

cannot be further simplified into a closed-form expression. A CRONYMS U SED IN S ECTION IV
Thus, we apply Laplace approximation over the double inte-
gral form in (11) as
 ∞ ∞
f 1 (X 1 , X 3 ) exp (−λ f 2 (X 1 , X 3 )) d X 1 d X 3
0 0
2π f 1 (X 1 (0), X 3 (0))
= √ exp (−λ f 2 (X 1 (0), X 3 (0))). (16)
λ |det (H )|

where X 1 (0) and X 3 (0) are considered as positive real values

obtained from the system:

d( f 1 (X 1 , X 3 ))
= 0,
d X1
d( f 1 (X 1 , X 3 ))
= 0, (17)
d X3

while H denotes Hessian matrix of function f 1 (X 1 , X 3 ) over

X 1 and X 3 at points X 1 (0) and X 3 (0). Then, substituting
Hessian matrix of function f 1 (X 1 , X 3 ) into (16) and con-
this section are provided in Table I.
sidering (17), the final closed-form expression of the outage
probability at the UAV after receiving data from CH in cluster n ∈ A,
E n ≥ E min , (22)
m can be expressed as in (18), shown at the bottom of the n∈/ A, otherwise,
page.  N
Thus, the outage probability at the data sink can be written Cmn , ≤ ∀m ∈ B, (23)
M
as n∈ A

Cmn = 1, ∀n ∈ A, (24)

M
m∈B
m
Pout = Pout . (19) γ1
i=1 Pj ≥ j = 1, . . . , M. (25)
max[D]−ω
Based on the outage probability closed-form expression The constraint (22) asserts that after performing energy
in (18), the achievable sum-throughput at the UAV for cluster harvesting from PBs, each SN should have at least E min
m can be written as amount of energy in the SN’s battery/energy storage in order
 to be considered for clustering process. Otherwise, SN will
(1 − α)(1 − β) be considered as an inactive SN and will not consider for
Rm = (1 − Pout
m
)RT , (20)
M clustering process. The constraint (23) enforces that no SN is
belongs to more than one cluster. The symbol Cm,n ∈ {0, 1}
where R denotes the fixed transmission rate of the CHs. indicates whether the SN n is a member of cluster m and
Considering (19), the sum throughput at the data sink can also asserts that each SN only belongs to one cluster. The
be expressed as constraint (25) enforces that the transmission power of the PB
j is sufficient enough to harvest energy to maintain expected
Rt = (1 − Pout )RM T (1 − α)(1 − β). (21) QoS during the information transmission to CH. Following
constraints are used in the CH selection process and make sure
that the optimal CH is selected in terms of system performance
IV. O UTAGE M INIMIZATION P ROBLEM F ORMULATION
and UAV’s trajectory.
In this section, the problem of clustering, CH selection, Cnm ≤ Cmn , ∀m ∈ B, ∀h ∈ A, (26)
UAV’s optimal trajectory and optimal power allocation that 
minimizes the outage probability are considered to obtain Cnm = 1, ∀m ∈ B, (27)
h∈ A
optimal values for time ratio α, τ, β. For the proposed system,
four constraints are required for the clustering process of the
CH
E min < E nm , ∀m ∈ B, (28)
WSN to maintain the expected QoS. The acronyms used in Tn = 1, ∀m ∈ B.
m
(29)

∞  p−1  

p+1 p
m (1 + K m ) 2 K m exp(−K m ) a3m γ2 2 (1 + K m ))a3m
Pout =1−2 × K ( p−1) 2 (18)
um ( p + 1) p! (a1m − a2m γ2 ) n,m (a1m − a2m γ2 ) u,m mn
p=0

Authorized licensed use limited to: Auckland University of Technology. Downloaded on November 05,2020 at 16:17:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

PERERA et al.: WPT-ENABLED UAV-ASSISTED CONDITION MONITORING SCHEME FOR WIRELESS SENSOR NETWORKS 7

The constraint (26) enforces that a SN cannot be selected This optimization problem is a non linear mixed integer
as a CH if it is not in cluster m. The constraints (27) and (28) programming problem involving finding optimal clustering,
assure that there is only one CH per each cluster and a SN CH selections with joint computation of time ratios α, τ, βm
cannot be selected as CH if the harvested energy is less than and optimal energy allocations. Due to the complexity of the
the provided threshold value, respectively. The symbol Tnm formulated problem, an asymptotic solution with less com-
indicates the UAV’s trajectory and constraint (29) enforces plexity as compared to highly complex Brute-force/exhaustive
that the CH n is selected for UAVs trajectory. Following search method is presented in the following section.
trajectory constraints ensure that the UAV starts from the
selected starting point, fly over all CHs to collect data and V. A SYMPTOTIC O PTIMAL S OLUTION
finally arrive at the destination location as in [24], [25]. The discrete nature of the binary variables involved in
  clustering, CH selection and trajectory of the UAV, causes dif-
Yz,q = 1, (30)
ficulties in solving the formulated problem in (35), especially
z∈Bm (zq)∈V
 when the amount of harvested energy and transmission power
Y Sz = 1, (31) of the SNs and the UAV are considered. Therefore, a method
z∈V that offers asymptotically optimal values for variables with

Yq D = 1, (32) minimum time complexity is required for the proposed system.
q∈V By using the Lagrangian duality method adhering to the
⎧ ⎫ requirements of the proposed system, the formulated (P1)
  ⎨1, z = S ⎬
Yzq − Yqz = −1, z = D , (33) in (35) can be solved as follows. First, a set D is defined
⎩ ⎭ with all possible clustering c1 = Cm,n , selected CHs c2 =
(z,q)∈V (q,z)∈V 0
Cnm , UAV’s trajectory information y = Yz,q and time ratio
Rz − Rq + Yzq M <= M − 1. (34)
β = βm . The set D satisfies the all constraints (C6), (C7) and
The constraints (30), (31), (32) and (33) enforce UAV to fly (C8). Thus, the corresponding Lagrangian dual function of the
from selected starting point to selected destination flying over problem (P1) is given by
only one CH of each cluster. Finally, constraint (34) asserts
g()  min L( p, c1 , c2 , y, β, α, τ ; ), (36)
that the UAV will not get into a flying loop during the data p ∈ P(c1 , c2 , y, β)
collection. Then, the corresponding optimization problem can (c1 , c2 , y, β) ∈ D
be mathematically expressed as
where p denotes the set of power allocation for given set P

M
of clusters, CHs, UAV trajectory and time ratio βm . The
m
(P1) : min Pout
{m∈M,E m ,Pu α,τ,β} corresponding Lagrangian expression of (36) can be expressed
m=1
γ2 T (1 − α)(1 − β) as
subject to: (C1) : E nm ≥ , 
M 
M
M Hmin m
L( p, c1 , c2 , y, β, α, τ ; ) = Pout + λu (E u − Em )

M
m=1 m=1
(C2) : Eu ≥ Em ,
m=1

J M
+ λmn (E nm − E n j ),
γ1 T ατ
(C3) : En ≥ , n=1 j =1
max[D]−ω
(37)
(C4) : 0 < α < 1,
(C5) : 0 < τ < 1, where  = (λu , . . . , λn,m ) ≥ 0 representing the vector of dual
variables adhering to the energy constraints of the proposed
(C6) : 0 ≤ βm < 1, ∀m ∈ B,
WSN. Therefore, the dual problem can be formulated as
(C7) : (22) − (29),
(P2) : max g()
(C8) : (30) − (34), (35) 
subject to:  ≥ 0. (38)
where m = {1, . . . , M}, E nm denotes the amount of energy
harvested by the CH, E u represents the energy allocated for Dual function g() is a concave function by definition
UAV’s RF-WPT link to power-up CHs and E n denotes the since it has a point wise minimum of affine functions.
amount of energy harvested by SNs from RF-WPT signal (L( p, c1 , c2 , y, β, α, τ ; ) is affine, i.e. linear, in ()) [27].
received from PBs. The objective of (35) is to minimize the Therefore, gradient or subgradient based methods can be used
outage probability at the UAV by identifying optimal time to solve the (P2) with guaranteed convergence. Then the
ratios. Constraint (C1) make sure that each CH has harvested subgradient of (36) can be written as
sufficient energy to maintain the expected QoS between CH 
M
T (1 − α)(1 − β)
and the UAV. Constraint (C2) represents the required amount λu = E u − Pm∗ () , (39)
of energy by the UAV for RF-WPT. The constraint (C3) M
m=1
ensures that the PBs’ RF-WPT transmission power is ade- M
quate enough for SNs to harvest energy to maintain expected λmn = E nm − P j∗ ()dnωj |h n, j |2 T ατ, (40)
QoS during information transmission between SNs and CHs. j =1

Authorized licensed use limited to: Auckland University of Technology. Downloaded on November 05,2020 at 16:17:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

where  = (λu , λm,1 , . . . λm,J ). This updated method of the dual function can be expressed as
consists of polynomial computational complexity similar to the

M 
J
number of dual primal variables ( J + 1), where J = N/M. g()  min Cm,n Cnm L3(m,n) + λu E u
Determining the optimal optimization variables in (36) realize p ∈ P(c1 , c2 , y, β) m=1 n=1
the computation of the dual function. Thus, the dual function (c1 , c2 , y, β) ∈ D
in (36) can be re-written as

M
+ λmn E nm . (47)

M 
J
m=1
g()  min L2(m,n) + λu E u
p ∈ P(c1 , c2 , y, β) m=1 n=1 Then, it is easily perceived that the optimal clustering and
(c1 , c2 , y, β) ∈ D CH selection should be the one having minimum value for
as L3(m,n) as

M
 
+ λmn E nm , (41) ∗ 1, (m, n) = arg min L3(m,n)
m=1 Cmn = , (48)
0, otherwise
 
where 1, n = arg min L3(m,n)
Cnm = , (49)
0, otherwise

M   
T (1 − α)(1 − β) 1, Cnm = arg min L3(m,n)
L2(n,m)  m
Pout − λu Pm () Tnm = . (50)
M 0, otherwise
m=1
J 
M Hence, the function L3(m,n) works as the optimal criteria for
ω 2
− λmn P j ()dn, j |h n, j | T ατ. (42) clustering and CH selection under the threshold SNR values
m=1 j =1 γ1 , γ2 provided in the proposed system to maintained the
expected QoS. Finally, by considering the values obtained
The optimal transmission power allocation of PB and CH in (48) and (49) substitute in L3(m,n) and adhering to the
over a given cluster of SNs can be ascertained by finding constraints (C1)(C3)(C6) in (35), the optimal value for βm
solutions for following formulated problem can be expressed as
∞ 
p−1
(P3) : min L2(m,n)  γ 2
γ1 K m2 (1 − α ∗ )τ ∗ α ∗
 βm∗ = 2
subject to: Pm > 0 & P j > 0. (43) M Hmin ηn (1 − γ1 κm2 )
p=0

The function L2(m,n) is a concave function of P j , Pm , where γ1 γ2 (1 + p)κm2 exp(−2K m )
× K (v) 2 . (51)
P j denotes the transmission power of the PB and Pm denotes ηn um ( p + 1) p!
the transmission power of the CH in the cluster m. Hence, Finally, substituting(45) and (51) into (44) and consider-
the optimal power allocations can be obtained by applying ing (39),(40) and constraints (C1)(C2)(C3)(C4)(C5) in (35),
Karush-Kuhn-Tucker (KKT) conditions as in [27], [28] the optimal values for τ and α can be expressed as
  −ω 2 ∗
∗ ∗ Pu |gmu |2 ηn ρu γ2 βm 1 γ1 dn j σn j α T α ∗ + |h n j |2
Pm = P j 1 + , (44) τ∗ = + , (52)
((1 − βm )σu2 + βm κ 2 ) γ1 P j |h n j | 2 Mmax(D)−ω
and and
M 

p−1
 2 d −ω ∗ [γ1 γ2 ]( 2 )
+ T |h n j |2 |h mn |2
σn2j σmn α = 
P j∗ =
nj
max[D]ω Mj =1 P j ηn |h n j | M
2
x(1 − τ )|h n j |2 |h mn |2 p=0

 (1 + K m ) 
(1 − τ )γ1 α|h n j |2 |h mn |2
− λu T ατ |h n j |2 × K (v) × . (53)
× 2 Md −ω
, (45) mn
mn
λu x + (1 − τ )ασnm mn
By analyzing (53), it can be observed that γ1 and γ2
where play a significant role in deciding α ∗ . Also it is noteworthy
that the α ∗ is independent from the time ratios τ ∗ and β ∗ .
λnm (1 − τ )α|h mn |2 σn2j However, τ ∗ in (52) depends on α ∗ , while β ∗ in (51) depends
x= . (46) on both α ∗ and τ ∗ . This phenomena can be easily under-
λu |h n j |2 σmn
2
stand by analyzing (43) and (44) with respect to the Fig. 2.
It is also noteworthy that the dual variables λu and λm,n The computational complexity of the proposed asymptotic
affects the pricing scheme of the WSN in terms of economic method is O(N · M 5 · J 2 ) considering the traveling salesman
perspective of the energy usage at the UAV and the PBs for problem (TSP) for UAV trajectory. It is noteworthy that
RF-WPT technique. Next, substituting the optimal transmis- the the proposed asymptotic method is having a less time
sion power in (44) and (45) in (42), the transmission power complexity compared to the traditional Brute-force search
variables can be eliminated from (42) and is denoted by L3(m,n) . method of O(N 3 · (N − J )! · J · M 6 ) N! which is a NP-Hard
Then, by applying the changes into (41), an alternative version problem [29].

Authorized licensed use limited to: Auckland University of Technology. Downloaded on November 05,2020 at 16:17:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

PERERA et al.: WPT-ENABLED UAV-ASSISTED CONDITION MONITORING SCHEME FOR WIRELESS SENSOR NETWORKS 9

VI. L OW C OMPLEXITY H EURISTIC M ETHOD Algorithm 1 Clustering

In this section, a low complexity sub optimal heuristic Data:
solution is proposed to solve (P1) in (35), while reducing • Sets N, M, A and T
the travelling distance of the UAV. Therefore, clustering and • simulation based values of α, τ
CH selection criteria that minimize the equivalent of (47) is • Optimal transmission power of PBs as in (44)
given by the proposed algorithms to further reduce the time Result: sets A, A set of clusters Q m , m = 1, 2, . . . , M,
complexity. where Q m ∈ B
The proposed clustering algorithm is illustrated in 1 Randomly placed PBs with locations r1 , . . . , rm Initialize
Algorithm 1. The proposed clustering technique presented as a Qm
modified version of K-mean clustering, where SNs are grouped 2 SNs harvest energy from PBs
around the PBs depending on the amount of energy harvested 3 Phase I
and minimum Euclidean distance to PBs. K-mean algorithm is 4 for each SN n ∈ S do
a well known method used to partition N number of SNs into 5 if n is active then
given number of clusters resulting the network in to Voronoi 6 Add n into set A
cells [30]. Apart from the nearest mean to the PBs, the amount 7 else
of harvested energy and the uniform distribution of SNs 8 Add n into set A
within each clusters are considered in proposed Algorithm. 9 end
After performing modified version of K-mean algorithm in 10 end
Algorithm 1, we update the clusters to be end up with nearly 11 while convergence do
equal in size by distorting the shape of the Veronoi cells. 12 for each SN n ∈ A do
The algorithm starts by placing M number of PBs randomly 13 for each PB m do
and obtaining optimal transmission power allocations for PBs. 14 if E n,m is max Em then
Then PBs are considered as a center point for clustering (for 15 if cluster size ≤ J & dm,n ≤ Q m then
K-mean clustering K = M) inside WSN. Next, by jointly 16 νn − rm 2 = min{λ3(m,n) }
considering the maximum amount of energy harvested out of 17 Qm ← − (n)
PBs and the mean of the horizontal and vertical Euclidean 18 else
distances to the PBs from the SN, each SN identifies its cluster. 19 νn − rm 2 = mi n{λ3(m,n) }
Even though SN satisfies the aforementioned requirements for 20 Qm ← − (n)
clustering, if the selected cluster has already J number of SNs 21 Q max ←− Add(n)
and a larger value for mean Euclidean distance to the PB, SN 22 end
will be put into set Q m , which will be used later to balance 23 else
the size of the each cluster. This step is repeated until the size 24 next iteration;
of each cluster is equals to J or J + 1. Remaining part of the 25 end
algorithm balance the size of each cluster until the cluster size 26 end
is equal to J or J + 1. The running time of the Algorithm 1 27 end
can be calculated in two phases. The running time of proposed 28 for all clusters m do
modified K-mean algorithm is O(N M Ta ), where Ta denotes 29 rm = meanof Q m
the number of iterations until the convergence. The running 30 end
time of the second phase of the algorithm is O(M J G log(G)). 31 end
Thus, the time complexity of the Algorithm 1 can be given as 32 Phase II while until cluster size J or J + 1 do
O(N M Ta + M J G log(G)). 33 for all clusters m do
Convergence Proof of Algorithm 1: In time, the algorithm 34 while Cluster size < J do
terminates after some iterations for the first phase, which 35 if selected SN n ∗ ∈ Q max & size(m) > J then
performs initial clustering of SNs depending on the amount 36 Q max ←− Add(n ∗ ) increase restricted level
of energy harvested and characteristics of K-means algorithm. 37 else if selected SN n ∗ ∈ Q max & restricted level
Due to the facts that the PBs act as the fixed central points = 0 then
for each clusters and after some iterations, completion of SN 38 Q max ←− Add(n ∗ ) increase restricted level
assignment are resulting convergence of the algorithm. For 39 else
the phase two of the algorithm, if cluster size is not equals 40 selected SN n ∗ = Q max restricted level
to J or J + 1, then borrowing SNs from large size clusters ≤ Rmax Q max ← − Add(n ∗ ) increase
algorithm terminates by balancing the clusters’ sizes within the restricted level
WSN. Also it is noteworthy that for each small size cluster 41 end
there is at least one large size cluster according to the first 42 end
phase of the algorithm. 43 end
The proposed heuristic method for CH selection and UAV 44 end
trajectory is given in Algorithm 2. The clustering results

Authorized licensed use limited to: Auckland University of Technology. Downloaded on November 05,2020 at 16:17:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

10 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

obtained in Algorithm 1 is required to execute Algorithm 2. TABLE II

The algorithm considers the starting and ending points of the S IMULATION PARAMETERS
UAV when deciding the optimal CH for the given cluster.
First, for a given q number of SNs, which are having the
first highest amount of energy is added to the set E and the
Euclidean distance from each SN to UAV’s direct trajectory
is calculated. Then, the SN having the minimum distance to
UAV’s direct trajectory and SN which minimizes the (47) is
selected as the CH. To decide the UAV trajectory, a well known
Travelling salesman Problem (TSP) is used [29].

Algorithm 2 Cluster Head Selection & UAV Trajectory

Data:
• Sets S, A, B, E, and Q m
• UAV starting(S) and ending points (D)
Result:
• Sets E and Q C H
• Cnm , n = 1, 2, . . . , N ∈ A and m = 1, . . . , M ∈ B
Thus, the total time complexity of the Algorithm 2 is can
• Trajectory of the UAV
be given as O(N M). Similarly, the time complexity of the
1 Phase I Algorithm 1 can be given as O(N M Tα ) since modified K-
2 Add q no of SNs with highest energy within Q m to set mean algorithm run time is higher than the cluster balancing.
E m , where n ∈ (S ∩ Q m ∩ A ∩ E). Therefore, the proposed heuristic method shows a greater
3 L U S ,U D ←
− Draw a straight line between U S and U D advantage of computational time complexity of O(N M Tα +
4 for each cluster m ∈ B do N M) compare to traditional Brute-force search method of
5 for each SN n ∈ (A ∩ E m ∩ Q m ) do O(N 3 · (N − J )! · J · M 6 ) and also better than the proposed
6 Find d L U S ,U D asymptotic method of O(N · M 5 · J 2 ).
7 end Convergence Proof of Algorithm 2: The convergence of
8 for each SN n ∈ (A ∩ E ∩ Q m ) in do CH selection phase in the algorithm is easy to prove since
9 Cnm = min(d L U S ,U D ) computing euclidean distance to UAV’s direct trajectory and
10 if Cnm = mi n{λ3(m,n) } then selecting SN from max energy vector(E m ) are straightforward
11 Cnm =1 mathematical computations. Thus, the phase I of the algorithm
12 QC H ← − Cnm terminates easily when the one CH is selected. For each
13 else iteration of phase II, one CH is assigned to the UAV’s
14 Cnm =0 trajectory adhering to the constraints given in (30)-(34). Thus,
15 end the phase II of the algorithm will eventually terminates since
16 end the number of CHs is finite and equal to number of PBs in
17 end the proposed context.
18 Phase II
19 Q = Q C H VII. S IMULATION R ESULTS
20 Z ← − Find nearest CH to S from Q
In this section, simulation results are provided to evaluate
21 Y S,Z ← − (S, Z )
the performance of the proposed wireless powered UAV-
22 delete Z from Q
aided WSN. Unless otherwise stated, 10 different WSN
23 while Q = ∅ do
setups have been used for the simulations, where N =
24 Z ← − find nearest CH to Z
50, 100, 150, . . . , 500 and SNs are randomly distributed over
25 Y Z ,Z ← − link(Z , Z )
the area of N × Nm 2 . In addition, M number of PBs are
26 Z=Z
randomly placed within the area of WSN. The number of PBs
27 delete Z from Q
are set to N/0.2, which is proved and analyzed via subsequent
28 end
simulation results. The both starting and ending of the UAV
29 Y Z ,D ← − link(Z , D)
are placed randomly within 1 − 50 m away from the edges
of the WSN such that the direct line between the points is
The running time complexity of the Algorithm 2 can crossing the WSN. Unless otherwise stated, the values used
be calculated in two phases, clustering and UAV trajectory, in simulations are given in Table II. The maximum battery
respectively. The proposed method for clustering has the capacity of the SNs E max is set to ≈ 15 Wh [32].4
running time complexity of O(N M) and O(M 2 ) of time 4 SNs have high capacity batteries such as dry cell (D) batteries (15000
complexity similar to TSP in [29], where clustering takes more mAh) comparing to with the widely used AA batteries ( 2000 ∼ 2400 mAh)
time to run than the time taken for finalizing UAV’s trajectory. [32].

Authorized licensed use limited to: Auckland University of Technology. Downloaded on November 05,2020 at 16:17:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

PERERA et al.: WPT-ENABLED UAV-ASSISTED CONDITION MONITORING SCHEME FOR WIRELESS SENSOR NETWORKS 11

TABLE III
O UTAGE P ROBABILITY OF THE P ROPOSED S YSTEM W ITH R ESPECT TO T IME R ATIO α, τ, β

Fig. 4. Achievable throughput of the proposed system with respect to α, β, τ ,

where N = 100. Due to the limitations in representation, τ is presented in
Fig. 3. Outage probability of the proposed system with respect to α, β, τ , the figure.
where N = 100. Due the limitations in representation, τ is presented
in TABLE III.
simulation results shown in Table III and Fig. 3 proving the
Fig. 3 depicts the influence of time ratios α, τ, β on the accuracy of proposed asymptotic method, and it achieves local
end-to-end outage probability of the proposed system. The optimality.
objective of this simulation is to identify optimal time ratios Fig. 4 demonstrates the influence of time ratios on the
that minimize the outage probability. Due to the limitations achievable sum throughput at the data sink. It is can clearly
in representation, τ is not presented in the Fig. 3. Thus, be seen from Fig. 4 that there exists an optimal values
Table III is provided with the corresponding outage probability for time ratios similar to Fig. 3. This observation makes
values with respect to α, τ, β. It can be observed from the sense since the outage probability in (18) directly affects the
both Table III and Fig. 3 that the combination of optimal sum throughput as in (20). According to the results shown
values of α, τ, β that achieves lowest outage probability at in Fig. 4, maximum sum throughput is always achieved when
0.0553 equals to 0.6, 0.4, 0.2. Also, it is observed from Table the combination of time ratios (α, τ, β) equals to 0.6, 0.1, 0.4.
III that when τ = 0.4, the outage probability increase for However, it is worth mentioning that the value of β = 0.1
every combination of α, β. This observation makes sense is different from the optimal value of β = 0.2 in Fig. 3.
since τ decides the amount of time allocation for EH via Understanding of this result is straightforward since the weight
RF-WPT in Phase-I. If τ is larger, there is not enough time of the aggregated data transmitted by CHs is heavier than the
for information transmission between SNs and CHs and if τ is data transmitted by SN. Thus, more time need to be allocated
smaller, not enough energy is harvested at SNs leading to poor for data transmission between CH and the UAV to achieve
transmission power. In either case, outage probability increases higher sum-throughput according to (20). Jointly considering
regardless of the combination of α, β. Moreover, it is also Fig. 3 and Fig. 4 it is worth claiming that the optimal value for
noted from the Table III that regardless of τ , when β increases α, τ should be 0.6 and 0.4, respectively and the value of β has
α also increases up to 0.8 and then decreases to maintain low to be selected out of the values 0.1 and 0.2, when E nm < E max
outage probability. This make sense since the time allocation depending on the QoS requirement of the system.
for information transmission between CH and UAV is critical The outage probability at the data sink for different number
to maintain lower outage probability. The analytical values of SNs versus number of clusters in proposed system is
of α, β, τ obtained from (53),(51) and (52) of the proposed illustrated in Fig. 5. Both analytical and simulation based
asymptotic method are 0.6458, 0.2114, 0.4499, respectively. optimal values of time ratios are used to obtain the results
Analytically obtained time ratios approximately equals to the shown in Fig. 5. Further, it can be clearly observed from Fig. 5,

Authorized licensed use limited to: Auckland University of Technology. Downloaded on November 05,2020 at 16:17:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

12 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

Fig. 5. Outage probability for different number of SNs versus number of Fig. 6. Outage probability for different transmission powers of PBs versus
clusters in proposed WSN. The both asymptotic and simulation based optimal number SNs in WSN.
values for time ratios (α, β, τ ) are used in the simulation.

that regardless of the number of SNs, the minimum outage

probability is always achieved when the number of clusters
equals to 5, claiming that 20% of the SNs should be in one
cluster. Thus, in order to achieve optimal results for the outage
probability, M = N × 0.2 PBs are needed in the proposed
system setup. In addition, analytically obtained optimal time
ratios are shown superior outage probability compared to
numerically obtained optimal values and the outage probability
difference between analytical and numerical optimal values
are increasing with the increase in number of SNs in the
proposed WSN.
The outage probability for different transmission powers of
PBs with respect to the number of SNs in WSN is depicted
in Fig. 6. As it can be clearly seen from Fig. 6, regardless of the
number of SNs in the system, P j = 80 dBm always achieves
the lowest outage probability. Moreover, similar to Fig. 5, opti- Fig. 7. Execution time for proposed asymptotic and heuristic methods versus
number of SN in the WSN.
mal time ratios obtained analytically show the superior outage
performance compared to optimal values obtained through
the simulations. However, with the increase in transmission Fig. 8 depicts the input and output of the proposed WSN.
power of PBs and number of SNs in the proposed system, Fig. 8a shows the initial stage of the proposed WSN, where
the outage probability decrease illustrating maximum capacity 100 SNs are randomly distributed in the area of 100 × 100m 2
of the number of SNs in the proposed system, which is equals with M number of randomly distributed PBs. Fig. 8b illustrates
to around 250 − 300. the output result of the proposed system after the Phase I,
Execution time5 of the proposed asymptotic and heuristic where SNs are clustered into M number of clusters having
methods are shown in the Fig. 7. The proposed asymptotic 20% SNs per each cluster. And then, CH per each cluster is
method obtains optimal solutions in 24 to 478 seconds depend selected according to the proposed optimal methods consider-
on the number of SNs, while low complexity heuristic method ing the UAVs direct trajectory. Fig. 8c shows the same scenario
obtains solutions under 15 seconds. Conversely, the Brute- similar to Fig. 8a with different starting and ending locations
force method takes more than 19 hours to obtain results for of the UAV. It is clearly can be seen from Fig. 8a that the
the given WSN. Thus, these results confirm that the proposed selected CHs are different as compared to Fig. 8b in order to
asymptotic and heuristic method is much lower in complexity accommodate the changes in UAV’s trajectory.
than the traditional Brute-force search method. It also note- The sum throughput versus transmission power of PBs
worthy that the proposed heuristic method is much better than are shown in Fig. 9 to compare the performance of pro-
the proposed asymptotic method in terms of execution time posed asymptotic and heuristic methods with the traditional
complexity. Brute-force search. Due to the execution complexity of
Brute-force search for larger number of SNs (N ≥ 20),
5 Note that the execution time can be varying according to the system spec- two simulation results are given for better comparison of the
ification of the computer and the versions of application software(i.e,MatLab) proposed asymptotic and heuristic methods. It can be observed
running the simulation. from the Fig. 9a that the proposed asymptotic method shows

Authorized licensed use limited to: Auckland University of Technology. Downloaded on November 05,2020 at 16:17:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

PERERA et al.: WPT-ENABLED UAV-ASSISTED CONDITION MONITORING SCHEME FOR WIRELESS SENSOR NETWORKS 13

Fig. 8. (a) Sample input network topology for low complexity heuristics method, where N = 100 and M ∗ = 5. (b) Output network topology of low
complexity heuristics method when S and D of the UAV are located in west and east side of the network, respectively. (c) Output network topology of low
complexity heuristics method when S and D of the UAV are located in north and south side of the network, respectively.

Fig. 9. Sum throughput achieved by the proposed system with respect to difference transmission power of the PB for comparison of the algorithms.
(a) WSN with N = 20 due to the high complexity of Brute-force method, which is unable to realize for higher number of SNs. (b) WSN with N = 100
using the general parameters as in TABLE II.

approximately close performance as Brute-force search with a fare comparison, i.e. (1). UAV-aided WSN without WPT,
higher transmission power of PBs. However, in lower transmis- (2) WSN without UAV and RF-WPT, in which CHs are
sion powers of PBs primal solution of the Brute-force method directly communicating with the data sink. As it can
shows significant superiority having a duality gap between clearly seen from the Fig. 10, the both proposed asymp-
primal and proposed asymptotic method. On the other hand, totic and heuristic methods show superiority in achievable
the proposed heuristic method shows inferior performance sum-throughput at data sink compared to baseline WSNs.
compared to the proposed asymptotic method. Since Fig. 9a The proposed asymptotic method shows 41.44% performance
alone is not enough to claim the performance difference gain as compared to traditional WSN without UAV and RF-
between proposed asymptotic and heuristic methods, sum WPT, while the proposed heuristic method shows 39.63%
throughput versus transmission power of PBs in WSN with performance gain in terms of achievable sum-throughput.
100 SNs is given in Fig. 9b. The proposed asymptotic methods In addition, the proposed asymptotic and heuristic method
still show the superior performance compared to the proposed shows 8.27% and 6.89% performance gain in terms of sum-
heuristic method. However, increase in transmission power throughput at data sink, respectively with comparison to UAV-
of PB reduces the performance gap between the proposed assisted WSN without RF-WPT. The relative sum-throughput
asymptotic and heuristic methods. performance gain of the proposed asymptotic and heuristic
Finally, the sum-throughput performance of the proposed methods compared to baseline WSNs is given in Table IV.
asymptotic and heuristic methods with respect to baseline Jointly considering all the simulation results provided
WSNs are illustrated in Fig. 10. Two baseline systems are in this section, it is worth claiming that the proposed
used with similar parameters as mentioned in Table II to make RF-WPT enabled UAV-assisted WSN is showing significant

Authorized licensed use limited to: Auckland University of Technology. Downloaded on November 05,2020 at 16:17:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

14 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS

problems; clustering and CH selection to minimize the compu-

tational complexity. Simulation results revealed the superiority
of the proposed methods as compared to other benchmark
schemes in general. The proposed RF-WPT enabled UAV-
aided WSN can be used to collect more information from
infrastructures of transportation systems, i.e, railways, roads,
etc. to have better condition monitoring for ITS applications.

R EFERENCES

[1] T. Liu, X. Wang, and L. Zheng, “A cooperative SWIPT scheme for

wirelessly powered sensor networks,” IEEE Trans. Commun., vol. 65,
no. 6, pp. 2740–2752, Jun. 2017.
[2] T. D. Ponnimbaduge Perera, D. N. K. Jayakody, S. K. Sharma,
S. Chatzinotas, and J. Li, “Simultaneous wireless information and
power transfer (SWIPT): Recent advances and future challenges,” IEEE
Commun. Surveys Tuts., vol. 20, no. 1, pp. 264–302, 1st Quart., 2018.
Fig. 10. Sum-throughput versus transmission SNR of CH for comparison of [3] Y. Zhang, S. He, J. Chen, Y. Sun, and X. S. Shen, “Distributed sampling
the proposed asymptotic and heuristic methods with baseline systems. rate control for rechargeable sensor nodes with limited battery capac-
ity,” IEEE Trans. Wireless Commun., vol. 12, no. 6, pp. 3096–3106,
TABLE IV Jun. 2013.
[4] S. Riviere, A. Douyere, F. Alicalapa, and J. D. L. S. Luk, “Study
T HE R ELATIVE P ERFORMANCE G AIN IN T ERMS OF A CHIEVABLE of complete WPT system for WSN applications at low power level,”
S UM -T HROUGHPUT OF THE P ROPOSED A SYMPTOTIC AND Electron. Lett., vol. 46, no. 8, pp. 597–598, Apr. 2010.
H EURISTIC M ETHODS . C ONVENTIONAL WSN I S
[5] T. Ajmal, D. Jazani, and B. Allen, “Design of a compact RF energy
T REATED AS THE BASELINE S CHEME
harvester for wireless sensor networks,” in Proc. IET Conf. Wireless
Sensor Syst. (WSS), Jun. 2012, pp. 1–5.
[6] L. Sun, L. Wan, K. Liu, and X. Wang, “Cooperative-evolution-based
WPT resource allocation for large-scale cognitive industrial IoT,” IEEE
Trans. Ind. Informat., vol. 16, no. 8, pp. 5401–5411, Aug. 2020.
[7] D. Belo, D. C. Ribeiro, P. Pinho, and N. B. Carvalho, “A selective,
tracking, and power adaptive far-field wireless power transfer system,”
IEEE Trans. Microw. Theory Techn., vol. 67, no. 9, pp. 3856–3866,
Sep. 2019.
[8] N. Aslam, K. Xia, and M. U. Hadi, “Optimal wireless charging inclusive
of intellectual routing based on SARSA learning in renewable wireless
sensor networks,” IEEE Sensors J., vol. 19, no. 18, pp. 8340–8351,
Sep. 2019.
[9] L. Wang, H. Shao, J. Li, X. Wen, and Z. Lu, “Optimal multi-user
computation offloading strategy for wireless powered sensor networks,”
performance gain compared to baseline WSN system mod- IEEE Access, vol. 8, pp. 35150–35160, 2020.
[10] P. Kumar, S. Garg, A. Singh, S. Batra, N. Kumar, and I. You,
els within the given simulation setup presented in Table II. “MVO-based 2-D path planning scheme for providing quality of ser-
It is also identified that the maximum limit for sum through- vice in UAV environment,” IEEE Internet Things J., vol. 5, no. 3,
put achieved when the transmission power of PBs P j = pp. 1698–1707, Jun. 2018.
[11] S. Garg, A. Singh, S. Batra, N. Kumar, and L. T. Yang, “UAV-
80d Bm. Further increase in P j has no effects on the sys- empowered edge computing environment for cyber-threat detection in
tem performance of the proposed system. The proposed smart vehicles,” IEEE Netw., vol. 32, no. 3, pp. 42–51, May 2018.
RF-WPT enabled UAV-assisted WSN is a well designed candi- [12] S. Garg, G. S. Aujla, N. Kumar, and S. Batra, “Tree-based attack–
defense model for risk assessment in multi-UAV networks,” IEEE
date for any UAV-assisted IoT applications such as intelligent Consum. Electron. Mag., vol. 8, no. 6, pp. 35–41, Nov. 2019.
transportation systems in automotive industry, smart cities, [13] T. Shen and H. Ochiai, “A UAV-aided data collection for wireless
agriculture automation etc. powered sensor network over Rician fading channels,” in Proc. 16th
IEEE Annu. Consum. Commun. Netw. Conf. (CCNC), Jan. 2019, pp. 1–5.
[14] C. Zhan, Y. Zeng, and R. Zhang, “Energy-efficient data collection in
VIII. C ONCLUSION UAV enabled wireless sensor network,” IEEE Wireless Commun. Lett.,
vol. 7, no. 3, pp. 328–331, Jun. 2018.
In this paper, a new resource allocation and UAV aided [15] S. Say, H. Inata, J. Liu, and S. Shimamoto, “Priority-based data gathering
data collection is proposed for wireless powered UAV-assisted framework in UAV-assisted wireless sensor networks,” IEEE Sensors J.,
WSN over multiple fading channels, i.e. Rayleigh and Rician vol. 16, no. 14, pp. 5785–5794, Jul. 2016.
[16] J. Gong, T.-H. Chang, C. Shen, and X. Chen, “Aviation time mini-
fading. All the SNs are powered using randomly placed PBs mization of UAV for data collection from energy constrained sensor
and then SNs periodically send its sensed data to selected networks,” in Proc. IEEE Wireless Commun. Netw. Conf. (WCNC),
CHs to be collected by the UAV. A new time block structure Apr. 2018, pp. 1–6.
is proposed to accommodate all the operations of the pro- [17] L. Zhao, S. Qu, and Y. Yi, “A modified cluster-head selection algo-
rithm in wireless sensor networks based on LEACH,” EURASIP J.
posed WSN. Due to the complexity of the formulated outage Wireless Commun. Netw., vol. 2018, no. 1, p. 287, Dec. 2018, doi:
minimization problem, Lagrangian duality is used to solve the 10.1186/s13638-018-1299-7.
problem with polynomial complexity compared to traditional [18] A. Sarkar and T. Senthil Murugan, “Cluster head selection for energy
efficient and delay-less routing in wireless sensor network,” Wireless
NP-hard Brute-force search method. Additionally, a heuristic Netw., vol. 25, no. 1, pp. 303–320, Jan. 2019, doi: 10.1007/s11276-017-
algorithm is proposed by decomposing the problem into sub 1558-2.

Authorized licensed use limited to: Auckland University of Technology. Downloaded on November 05,2020 at 16:17:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

PERERA et al.: WPT-ENABLED UAV-ASSISTED CONDITION MONITORING SCHEME FOR WIRELESS SENSOR NETWORKS 15

[19] T. M. Behera, S. K. Mohapatra, U. C. Samal, M. S. Khan, Dushantha Nalin K. Jayakody (Senior Member,
M. Daneshmand, and A. H. Gandomi, “Residual energy-based cluster- IEEE) received the Ph.D. degree in electronics,
head selection in WSNs for IoT application,” IEEE Internet Things J., electrical, and communications engineering from
vol. 6, no. 3, pp. 5132–5139, Jun. 2019. University College Dublin, Ireland, in 2013. From
[20] H.-V. Tran and G. Kaddoum, “RF wireless power transfer: Regreening 2014 to 2016, he was a Postdoctoral Research Fel-
future networks,” IEEE Potentials, vol. 37, no. 2, pp. 35–41, Mar. 2018. low with the Institute of Computer science, Univer-
[21] V. J. Hodge, S. O’Keefe, M. Weeks, and A. Moulds, “Wireless sensor sity of Tartu, Estonia, and the Department of Infor-
networks for condition monitoring in the railway industry: A sur- matics, University of Bergen, Norway. Since 2016,
vey,” IEEE Trans. Intell. Transp. Syst., vol. 16, no. 3, pp. 1088–1106, he has been a Professor with the School of Computer
Jun. 2015. Science and Robotics, National Research Tomsk
[22] N. P. Le, “Throughput analysis of power-beacon-assisted energy harvest- Polytechnic University (TPU), Russia. In addition,
ing wireless systems over non-identical Nakagami-m fading channels,” since 2019, he also serves as the Dean of the School of Postgraduate and
IEEE Commun. Lett., vol. 22, no. 4, pp. 840–843, Apr. 2018. Research, Sri Lanka Technological Campus (SLTC), Padukka, Sri Lanka,
[23] D. Ebrahimi and C. Assi, “On the interaction between scheduling and and the Founding Director of the Centre of Telecommunication Research,
compressive data gathering in wireless sensor networks,” IEEE Trans. SLTC. He has received the Best Paper Award from the IEEE International
Wireless Commun., vol. 15, no. 4, pp. 2845–2858, Apr. 2016. Conference on Communication, Management and Information Technology
[24] C. You and R. Zhang, “3D trajectory optimization in Rician fading for (ICCMIT) in 2017 and International Conference on Emerging Technologies
UAV-enabled data harvesting,” IEEE Trans. Wireless Commun., vol. 18, of Information and Communications, Bhutan, in March 2019. In July 2019,
no. 6, pp. 3192–3207, Jun. 2019. he received the Education Leadership Award from the World Academic
[25] D. N. K. Jayakody, T. D. P. Perera, A. Ghrayeb, and M. O. Hasna, “Self- Congress in 2019. In 2017 and 2018, he received the Outstanding Faculty
energized UAV-assisted scheme for cooperative wireless relay networks,” Award by National Research Tomsk Polytechnic University, Russia. He also
IEEE Trans. Veh. Technol., vol. 69, no. 1, pp. 578–592, Jan. 2020. received Distinguished Researcher in Wireless Communications, Chennai,
[26] S. Panic, M. Stefanovic, J. Anastasov, and P. Spalevic, Fading and India, in 2019. He has published over 140 international peer reviewed journals
Interference Mitigation in Wireless Communications. Boca Raton, FL, and conference papers and books. His research interests include PHY and NET
USA: CRC, 2013. layer prospective of 5G communications technologies such as NOMA for 5G
[27] D. Knowles, “Lagrangian duality for dummies,” Stanford.edu, Stanford, etc., cooperative wireless communications, device to device communications,
CA, USA, Tech. Rep., Nov. 2010. [Online]. Available: https://www- LDPC codes, and Unmanned Ariel Vehicle. He has organized or co-organized
cs.stanford.edu/~davidknowles/lagrangian_duality.pdf more than 20 workshops and special sessions of various IEEE conferences.
[28] S. Boyd, S. P. Boyd, and L. Vandenberghe, Convex Optimization. He also served as the Chair, Session Chair or Technical Program Committee
Cambridge, U.K.: Cambridge Univ. Press, 2004. Member for various international conferences, such as IEEE PIMRC 2013-
[29] J. Ny, E. Feron, and E. Frazzoli, “On the dubins traveling salesman 2019, IEEE WCNC 2014-2018, and IEEE VTC 2015-2018. He currently
problem,” IEEE Trans. Autom. Control, vol. 57, no. 1, pp. 265–270, serves as an Area Editor of Elsevier Physical Communications Journal, MDPI
Jan. 2012. Information journal, and Wiley Internet of Technology Letters. Also, he serves
[30] M. Mahajan, P. Nimbhorkar, and K. Varadarajan, “The planar k- as a reviewer for various IEEE Transactions and other journals.
means problem is NP-hard,” Theor. Comput. Sci., vol. 442, pp. 13–21,
Jul. 2012.
[31] Series 1000-Watt NDB Transmitter. (2020). Southern Avionics
Company, Leaders in Non Directional Beacon and Differential P. Muthuchidambaranathan (Member, IEEE)
GPS Transmitters. Accessed: Jun. 10, 2020. [Online]. Available: received the B.Eng. degree in electronics and com-
https://www.southernavionics.com/se-series-1000-watt-ndb munication engineering from the Government Col-
[32] G. Tuna and V. C. Gungor, “Energy harvesting and battery technologies lege of Technology, Coimbatore, India, in 1992,
for powering wireless sensor networks,” in Industrial Wireless Sensor the M.Eng. degree in microwave and optical engi-
Networks. Amsterdam, The Netherlands: Elsevier, 2016, pp. 25–38. neering from the A. C. College of Engineering and
Technology, Karaikudi, India, in 1994, and the Ph.D.
degree in optical communication from the National
Tharindu Dilshan Ponnimbaduge Perera (Grad- Institute of Technology (NIT), Tiruchirappalli, India,
uate Student Member, IEEE) received the B.Sc. in 2009. He is currently working as an Associate
degree (Hons.) in software engineering and the Professor with the Department of Electronics and
M.Sc. degree from the Department of Computing, Communication Engineering, National Institute of Technology (NIT), Tiruchi-
Faculty of ACES, Sheffield Hallam University, U.K. rappalli. His research interests include wireless communications and optical
He is currently pursuing the Ph.D. degree in com- communications. He published his research papers in refereed international
puter science and wireless communications with journals, and international and national conferences. He is an author of the
the Infocomm Lab, School of Computer Science textbook Wireless Communications (published by Prentice Hall of India).
and Robotics, National Research Tomsk Polytechnic
University, Russia. He has been awarded a full-time
scholarship from the Ministry of Education, Russian
Federation for Ph.D. studies. Also, he is working as a Research Engineer with
the Department of IT, School of Computer Science and Robotics, National Jun Li (Senior Member, IEEE) received the Ph.D.
Research Tomsk Polytechnic University. His research interests include simul- degree in electronic engineering from Shanghai Jiao
taneous wireless information and power transfer, interference exploitation in Tong University, Shanghai, China, in 2009. From
RF energy harvesting, and UAV-assisted communication. January 2009 to June 2009, he worked at the
Department of Research and Innovation, Alcatel
Stefan Panic (Member, IEEE) received the M.Sc. Lucent Shanghai Bell, as a Research Scientist. From
and Ph.D. degrees in electrical engineering from June 2009 to April 2012, he was a Postdoctoral
the Faculty of Electronic Engineering, Nis, Serbia, Fellow with the School of Electrical Engineering
in 2007 and 2010, respectively. He is currently an and Telecommunications, University of New South
Associated Professor with the Department of Infor- Wales, Australia. From April 2012 to June 2015, he
matics, Faculty of Natural Science and Mathematics, was a Research Fellow with the School of Electrical
University of Pristina. He has published over 40 SCI Engineering, The University of Sydney, Australia. Since June 2015, he has
indexed articles. His research interests in mobile been a Professor with the School of Electronic and Optical Engineering,
and multichannel communications include statistical Nanjing University of Science and Technology, Nanjing, China. He was
characterization and modeling of fading channels, a Visiting Professor with Princeton University from 2018 to 2019. His
performance analysis of diversity combining tech- research interests include network information theory, game theory, distributed
niques, and outage analysis of multiuser wireless systems subject to interfer- intelligence, multiple agent reinforcement learning, and their applications in
ence. Within digital communication, his current research interests include the ultradense wireless networks, mobile edge computing, network privacy and
information theory, source and channel coding, and signal processing. security, and industrial Internet of things.

Authorized licensed use limited to: Auckland University of Technology. Downloaded on November 05,2020 at 16:17:11 UTC from IEEE Xplore. Restrictions apply.