See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/379497051

Outage Probability Analysis for Relay-Aided Self-Energy Recycling Wireless

Sensor Networks Over INID Rayleigh Fading Channels

Article in IEEE Sensors Journal · April 2024

DOI: 10.1109/JSEN.2024.3365698

CITATIONS READS

3 95

6 authors, including:

Nguyen Nhat Tan Trinh Van Chien

Ton Duc Thang University Hanoi University of Science and Technology
128 PUBLICATIONS 1,387 CITATIONS 154 PUBLICATIONS 2,247 CITATIONS

SEE PROFILE SEE PROFILE

Thanh Lam Tu Miroslav Vozňák

École Supérieure d'Electricité VŠB-Technical University of Ostrava
83 PUBLICATIONS 1,319 CITATIONS 542 PUBLICATIONS 4,943 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Nguyen Nhat Tan on 09 May 2024.

The user has requested enhancement of the downloaded file.

11184 IEEE SENSORS JOURNAL, VOL. 24, NO. 7, 1 APRIL 2024

Outage Probability Analysis for Relay-Aided

Self-Energy Recycling Wireless Sensor
Networks Over INID Rayleigh
Fading Channels
Tan N. Nguyen , Member, IEEE, Trinh Van Chien , Member, IEEE, Viet Quang Dinh , Lam-Thanh TU ,
Miroslav Voznak , Senior Member, IEEE, and Zhiguo Ding , Fellow, IEEE

Abstract—Communication reliability is one of the key chal-

lenging issues in future communications due to massive
connections, especially for wireless sensor networks (WSNs)
with low-cost devices. This article studies the communication
reliability of wireless systems in the presence of multiple
sensor relays, which carry out energy harvesting to prolong
the network lifetime. By exploiting the deep shadow fading
model, the three sensor selection methods are investigated
based on the different prior information of the propagation
channels. We then derive the analytical expressions of the
outage probability (OP) for each sensor selection, which only
depends on the statistical channel knowledge that can be applied for multiple coherence intervals whenever the channel
statistics remain the same. Since the obtained analytical OPs are interpreted based on several coupled integrals that
are costly to compute, we further propose a learning framework to predict the OP with low computational complexity via
exploiting supervised learning. Numerical results indicate that the two suboptimal sensor selection solutions provide a
competitive OP with each other. In contrast, the optimal solution outperforms the remaining benchmarks by many folds.
Besides, the deep-learning-based approach performs almost the same performance as the analytical-based framework.
Index Terms— Deep neural networks (DNNs), independent but not necessarily identically distributed (INID) Rayleigh
fading channels, outage probability (OP), relay-aided wireless sensor networks (WSNs), self-energy recycling.

Manuscript received 17 December 2023; accepted 25 January 2024. I. I NTRODUCTION

Date of current version 2 April 2024. This work was supported in part by HE fifth-generation (5G) wireless communication tech-
the European Union through the Research Excellence For Region Sus-
tainability and High-Tech Industries (REFRESH) Project of the European
Just Transition Fund under Grant CZ.10.03.01/00/22_003/0000048; in
T nology has brought about a wireless data transfer
andreliable communications shift in the field of telecommu-
part by the Ministry of Education, Youth and Sports of the Czech nications. Thanks to its enhanced capabilities, 5G has paved
Republic (MEYS CZ) through the e-Infrastructure Czechia (e-INFRA
CZ) Project under Grant 90254; and in part by the MEYS CZ through
the way for innovative applications, namely the Internet of
the Project Student Grant Scheme (SGS) conducted by VSB-Technical Things (IoT) [1], wireless sensor networks (WSNs) [2], and
University of Ostrava under Grant SP 7/2023. This research is funded autonomous vehicles [3], to deep-dive into virtual reality
by Hanoi University of Science and Technology (HUST) under project
number T2022-TT-001 for Trinh Van Chien. The associate editor coor- experiences [4]. However, as technologies continue to evolve,
dinating the review of this article and approving it for publication was researchers and industry experts have already started looking
Dr. Qammer H. Abbasi. (Corresponding author: Trinh Van Chien.) forward to the next sixth-generation (6G) wireless network.
Tan N. Nguyen and Lam-Thanh TU are with the Communication
and Signal Processing Research Group, Faculty of Electrical and 6G has been foreseen as the next generation of wireless
Electronics Engineering, Ton Duc Thang University, Ho Chi Minh City communication systems [5] that will further revolutionize the
70000, Vietnam (e-mail: [email protected]; tulamthanh@ digital age connection [6]. Building upon the groundwork
tdtu.edu.vn).
Trinh Van Chien and Viet Quang Dinh are with the School of Informa- established by 5G, 6G has been devised to overcome the
tion and Communication Technology (SoICT), Hanoi University of Sci- constraints and deficiencies of its forerunner while introducing
ence and Technology, Hanoi 100000, Vietnam (e-mail: chientv@soict. a host of new features and capabilities. The development
hust.edu.vn; [email protected]).
Miroslav Voznak is with the VSB-Technical University of Ostrava, 708 of 6G networks is driven by the need to meet the ever-
00 Ostrava, Czech Republic (e-mail: [email protected]). increasing demands for higher data rates, massive connectivity,
Zhiguo Ding is with the Department of Electrical Engineering and
Computer Science, Khalifa University, Abu Dhabi, United Arab Emirates, and ultralow latency of diverse devices and applications [7].
and also with the Department of Electrical and Electronic Engineer- The goal of full coverage and reliable wireless infrastruc-
ing, University of Manchester, M13 9PL Manchester, U.K. (e-mail:
[email protected]).
tures with sensors and relays has emerged and been expected
Digital Object Identifier 10.1109/JSEN.2024.3365698 to be supported in 6G systems [8], [9]. For such, the networks
1558-1748 © 2024 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: Technical University of Ostrava. Downloaded on April 02,2024 at 03:09:54 UTC from IEEE Xplore. Restrictions apply.
NGUYEN et al.: OUTAGE PROBABILITY ANALYSIS FOR RELAY-AIDED SELF-ENERGY RECYCLING WSNs 11185

will likely increase the speed and data throughput achieved in (GPUs). Second, it establishes a strong starting point for future
5G wireless communication by, at least, an order of magnitude endeavors and serves as a reliable reference for subsequent
by using the same time and frequency resources [10]. Relay activities. Ha et al. [23] exploited a deep neural network
has been a spectrally efficient technology for expanding a radio (DNN) to support the estimation of channels within an MIMO-
network’s coverage and quality of service (QoS) without hav- OFDM system. This approach focuses on two distinct scenar-
ing to pay more for wired backhaul [11]. The 3GPP 5G new ios, both involving multipath fading channel models. These
radio work items entitled integration of access and backhaul models are based explicitly on the TDL-A model defined
(IAB) in Rel-16 demonstrated the cost-effective deployment of within the context of the 5G network [23]. This research
cellular networks through wireless backhauling and conducted has successfully reduced channel estimation errors through the
relay support standardization [12], [13]. However, it should be study of channel properties [24]. It is crucial to note that deep
noted that only fixed relays are considered in Rel-16 IAB. The learning in communication often requires large amounts of
sensor and relay selections can be helpful to deliver a better data for training [25], which may pose challenges in specific
user experience under high mobility scenarios, for example, scenarios. Moreover, we should carefully consider the compu-
high-mobility trains and cars, to meet various beyond 5G net- tational complexity and the real-time processing requirements
work standards under complicated deployment scenarios [14]. for practical model deployment in communication systems as
The randomness influences of fast-fading channels on wire- machine-learning algorithms are deployed.
less communication systems are well known to be successfully Deep learning has recently emerged as a powerful tool
mitigated via cooperation between devices [15]. To establish for analyzing and predicting outage probabilities in various
spatial variety in a cooperative system, sensor nodes operated domains with low computational complexity [26]. Its ability
as relays can support and compensate for the propagation to automatically learn complex patterns and relationships from
loss when the radio waves are sent from the source to the data is the reason for its being well-suited for OP analysis.
destination [16]. Several cooperation protocols have been Here, we explore the application of deep learning in OP
developed in the literature, for example, amplify-and-forward analysis and its potential benefits. Deep-learning algorithms,
(AF) and decode-and-forward (DF), and they have drawn the such as DNNs, can be trained on historical outage data,
most attention both in academia and industry due to their network performance metrics, and other relevant parameters
efficiency and simplicity [17]. The signals received from the to model and predict outage probabilities. By learning from
source and relays must be appropriately combined at the massive datasets, DNNs can capture intricate dependencies
destination to maximize the benefits of relaying under specific and nonlinearities, enabling more accurate outage predictions
measurement metrics. If relays act as sensors, opportunistic compared to traditional analytical models [27]. However,
sensor selection is straightforward but efficient, combining no related work applied DNNs for the OP analysis in WSNs.
methods in which only one relay cooperates [18]. These relay In this article, we consider a data transmission protocol,
selections ought to be one of the groups that are capable of where multiple relays are employed as sensor nodes to help
appropriately decoding the source signal as reported in [19] the transmission between the source and destination, where
and reference therein. In addition, the relay selection should we aim to investigate the influence of sensor selections on
opt for the destination having the highest signal-to-noise power the OP. In summary, our main contributions are listed as
ratio (SNR). Designing a fair relaying strategy, or a method in follows.
which the average power utilized by the relays is roughly the 1) We propose a novel protocol that integrates self-energy
same, is a crucial but understudied subject in relay networks. recycling, energy harvesting, and sensor selection meth-
To the best of the authors’ knowledge, there is no existing ods to improve the OP. Our framework is based on a
work to analyze and compare the outage probability (OP) general fading model that is practically independent but
under different relay selection methods. not necessarily identically distributed (INID) Rayleigh
Machine learning and deep learning are subfields of artificial fading channels.
intelligence that focus on developing algorithms learning from 2) We rigorously analyze the optimal and suboptimal relay
provided datasets and making predictions without the need for selection methods based on channel statistics. Specif-
explicit programming. Despite their complicated computation ically, we provide the exact analytical expression of
and time-consuming training phase, both of them have had the OP and evaluate the impacts of important system
significant impacts across applications in communications, parameters, such as the number of devices, on the
enabling breakthroughs in various areas such as speech recog- performance.
nition [20], channel coding and error correction [21], and 3) We provide a learning-based method that can only pre-
QoS optimization. A convolutional neural network architec- dict the OP based on statistical information. The learned
ture (CNN) with residual dense connections was proposed neural network can be applied for multiple coherence
in [22] to deal with the joint pilot and data power allocation intervals whenever the channel statistics are unchanged.
problems in dynamic cellular massive MIMO systems. For 4) Monte-Carlo simulations match well with analytical
these nonconvex optimization problems, using a supervised results. Besides, the optimal relay selection can bring
learning method yields accurate predictions and is valuable for significant benefits to multirelay FD networks and avoid
two reasons. First, it enables the development of a simplified the outage floor by increasing the number of relays.
implementation, taking advantage of the progress made in In contrast, in the case of using the suboptimal relay
implementing neural networks on graphic processing units selections, the OP reaches an outage floor.

Authorized licensed use limited to: Technical University of Ostrava. Downloaded on April 02,2024 at 03:09:54 UTC from IEEE Xplore. Restrictions apply.
11186 IEEE SENSORS JOURNAL, VOL. 24, NO. 7, 1 APRIL 2024

are exponential random variables (RVs) whose cdf are given

as
FγSRm (x) = 1 − exp −λSRm x

FγRm D (x) = 1 − exp −λRm D x .

(3)
To take path loss into account, we can model the parameters
in (3) as follows:

β
β
λSRm = dSRm , λRm D = dRm D (4)
where dSRm and dRm D are the distances between the S →
Rm and Rm → D, respectively. The RV f m is also modeled
as complex Gaussian RV, and hence ϕm = | f m |2 is also an
exponential RV, whose cdf is expressed as
Fϕm (x) = 1 − exp (−m x) . (5)
Then, the PDFs of γSRm , γRm D and ϕm can be expressed as
f X m (x) = ξm exp (−ξm x) (6)

Fig. 1. Considered system model including the information transmis- where Xm ∈ {γSRm , γRm D , ϕm }, and ξm ∈ {λSRm , λRm D , m }.
sion and energy harvesting. Using the recycle approach [30], the harvested energy obtained
at Rm can be obtained as
E m = ηρT P0 γSRm + Pm ϕm


(7)
The rest of this article is organized as follows: Section II
presents in detail the system and signal models. Meanwhile, where η is conversion efficiency (0 ≤ η ≤ 1) and 0 < ρ < 1
Section III provides the different relay selection methods is the power splitting factor. Following the same methodology
and the corresponding OP. Numerical results confirming the as in [31], the transmit power of Rm can be formulated as
analytical achievements are given in Section V and the main follows:
conclusions are drawn in Section VI. Em
= ηρ P0 γSRm + Pm ϕm = κ P0 γSRm + Pm ϕm (8)



Pm =
T
where κ = ηρ. In addition, from (8), we obtain
II. C OMMUNICATION S YSTEM AND S IGNAL M ODELS
Fig. 1 presents a system model of the proposed protocol, κ P0 γSRm
Pm = . (9)
in which a source S communicates with a destination D via 1 − κϕm
multiple FD relays denoted by R1 , R2 , . . . , R M . It is assumed It is worth noting from (9) that Pm = 0 as ϕm ⩾ 1/κ.
that the direct link between S and D does not exist due to In practice, ϕm is much less than 1 due to passive interference
the far distance between S and D, and deep shadow fading. cancellation (IC) such as antenna isolation, which means that
Moreover, one of these relays is selected to forward the source the denominator in (9) is positive [30]. Assume that Rm is the
data to the destination. Because of the limited energy, the selected relay, and S sends the signal x[n] to Rm . Therefore,
relays have to harvest the wireless energy from S and use the received signal at the receive antenna of Rm can be written
the harvested energy to power its relaying transmission, using as
the DF mode. p
yRm = 1 − ρ P0 h SRm x [n]
p
Let us denote h SRm and h Rm D as the channel coefficients
p
of the S → Rm and Rm → D links, respectively, where + 1 − ρ Pm f m x [n − n 0 ] + u m (10)
p
m = 1, 2, . . . , M. We also denote f m as the self-interference
between the transmit antenna and the receive antenna of where n = 1, 2, . . . , x[n − n 0 ] is a signal transmitted
Rm . Assume that h SRm and h Rm D are INID Rayleigh fading by the transmit antenna of Rm , n 0 (n 0 > 0) is a positive
channels, channel gains1 integer number indicating the delay time, Pm is transmit
power of Rm , u m is zero mean additive white Gaussian
γSRm = |h SRm |2 ∀m = 1, . . . M (1) noise (AWGN) with variance [32] σ02 . It is noted from (7)
that (1 − ρ)1/2 (Pm )1/2 f m x[n − n 0 ] is a loopback interference
γRm D = |h Rm D | 2
∀m = 1, . . . M (2)
component. In addition, for ease of presentation and analysis,
we assume that all AWGNs have the same variance, that is,
1 The influences of propagation channels over NIID Rayleigh fading have σ02 . From (9) and (10), we can formulate the instantaneous
not been studied in the satellite-relay cooperated literature yet. Note that NIID signal-to-interference-plus-noise (SINR) obtained at Rm as
Rayleigh fading channels are of practice under rich scattering environments,
where scatterers are distributed, for example, around the relays and users [28]. (1 − ρ) P0 γSRm
As future work, the investigation of networks under different traffic using ψSRm = . (11)
dense sensor deployments should be of interest [29]. (1 − ρ) Pm ϕm + σ02

Authorized licensed use limited to: Technical University of Ostrava. Downloaded on April 02,2024 at 03:09:54 UTC from IEEE Xplore. Restrictions apply.
NGUYEN et al.: OUTAGE PROBABILITY ANALYSIS FOR RELAY-AIDED SELF-ENERGY RECYCLING WSNs 11187

M
κ1γSRa γRa D
 
By substituting (9) into (11), one can obtain that X 1−κϕa
− Pr γSRa > Z a , > γth , > γth =1
(1 − ρ) (1 − κϕm ) P0 γSRm κϕa 1−κϕa
a=1
ψSRm =
(1 − ρ) κ P0 γSRm ϕm + σ02 XM 
1 γth −κγth ϕa

(1 − ρ) (1 − κϕm ) γSRm − Pr ϕa < , γSRa >Z a , γSRa > .
= . (12) κ (1 + γth ) κ1γRa D
σ2 a=1 |
(1 − ρ) κγSRm ϕm + P00
{z }
Q
(18)
Since σ02 ≪ P0 , (12) can be approximated as follows:
At the first, we will find the cdf and pdf of Z a as follows:
(1 − κϕm )
ψSRm ≈ . (13) M
κϕm Y
FZ a (x) = Pr (Z a < x) =



1 − exp −λSRm x
After successfully decoding x[n], Rm will re-encode, and m=1,m̸=a
forward the encoded signal to D, and the received signal at D M−1
X M
X u
X
!
can be expressed as =1+ (−1) u
exp − λSRiv x
p u=1 i 1 =...=i u =1, v=1
yD = Pm h Rm D x [n] + u D (14) i 1 <...<i u ,
i 1 ,...,i u ̸=a
where u D is zero mean AWGN at D with variance σ02 . M−1
X M
X
(−1) u sum


From (14), the instantaneous SINR received at D can be =1+ exp −λa,u x (19)
calculated as follows: u=1 i 1 =...=i u =1,
i 1 <...<i u ,
Pm γRm D κ1γSRm γRm D i 1 ,...,i u ̸=a
ψRm D = = (15)
σ02 1 − κϕm sum =
where λa,u
Pu
v=1 λSRiv . Then, the pdf of Z a can be found
by the following formulation:
where 1 = P0 /σ02 denotes transmit SNR. Then, the instan-
M−1 M
taneous end-to-end (e2e) SINR, with respect to x[n], can be X X
f Z a (x) = (−1)u+1 sum sum
x . (20)


expressed as λa,u exp −λa,u
u=1 i 1 =...=i u =1,
i 1 <...<i u ,
ψme2e = min ψSRm , ψRm D .


(16) i 1 ,...,i u ̸=a

From the SINR expressions in (11), (15), and (16), we will Next, the expression Q in (18) can be calculated as
γth − κγth ϕa
 
present the selection methods considered in this article. 1
Q = Pr ϕa < , γSRa > Z a , γSRa >
κ (1 + γth ) κ1γRa D
1
III. D IFFERENT R ELAY S ELECTIONS Z
κ (1+γth )
Z ∞
a exp (−a x) λRa D exp −λRa D y


In this section, we present the three relay selection methods =
0
based on the different SINRs. After that, the OP is analyt- "Z γth −κγ0th x
λSRa (γth − κγth x)
 
κ9y
ically obtained, which is independent of small-scale fading × f Z a (u) exp − du
coefficients.2 0 κ9y
Z +∞ #
+ γ −κγ x exp −λSRa u f Z a (u) du d xdy


A. Conventional Partial Relay Selection th th
κ9y
In this method, the relay providing the highest channel gain Z 1 Z ∞
κ (1+γth )
at the first hop is selected for the cooperation, that is, a exp (−a x) λRa D exp −λRa D y


=
0 0
γSRa = max γSRm (17)

m=1,2,...,M 





 M−1
 M 
where Ra denotes the selected relay (a ∈ {1, 2, . . . , M}). The
X X 
× (−1)u+1
γ (x, y) d xdy
OP in this method can be computed as follows: 

 u=1 i 1 =...=i u =1,



i 1 <...<i u ,

 

OPC - PRS i 1 ,...,i u ̸=a
M
X = Q1 − Q2 (21)
Pr γSRa > Z a , min ψSRa , ψRa D > γth = 1



=1 − where a is the exponential scalar of the RVs and the
a=1
definitions hold for γ (x, y), Q 1 , and Q 2 as follows:
M
1−κϕa κ1γSRa γRa D
   
γ (x, y)
X
− Pr γSRa > Z a , min , > γth = 1
κϕa 1−κϕa 
λSRa (γth − κγth x)

a=1
= exp −
2 Once the channel state information is available, all the considered relay
κ9y
!
SRa (γth − κγth x)
sum + λ


selections can be implemented. The channel estimates are obtained during λSRa λa,u
the pilot training phase. We note that the accuracy of the channel estimates − sum exp −
is controlled by the pilot power and the coherence time dedicated to the pilot
λa,u +λSRa κ9y
training phase. (22)

Authorized licensed use limited to: Technical University of Ostrava. Downloaded on April 02,2024 at 03:09:54 UTC from IEEE Xplore. Restrictions apply.
11188 IEEE SENSORS JOURNAL, VOL. 24, NO. 7, 1 APRIL 2024

1
   
M−1 M

42 (0) + 42 1 N −1 42 n
Z
κ (1+γth )
κ(1+γth ) κ(1+γth )N
X X
(−1)u+1 a exp (−a x)
X
Q1 = × + 
u=1 i 1 =...=i u =1, 0 2κ (1 + γth ) N κ (1 + γth ) N
n=1
i 1 <...<i u ,
i 1 ,...,i u ̸=a (28)
∞ λSRa (γth −κγth x)
Z   
× λRa D exp −λRa D y − dy d x
0 κ9y where the following definitions hold for 41 (x) and 42 (x) as
(23) follows:
λSRa λRa D (γth − κγth x)
r
Q2
M−1 M 1 41 (x) = a exp (−a x)
κ9 !
Z
X X λSRa κ (1+γth )
= (−1)u+1 a exp (−a x)
λSRa λRa D (γth − κγth x)
r
sum +λ
λa,u SRa 0
u=1 i 1 =...=i u =1, × K1 2 (29)
i 1 <...<i u , κ9
i 1 ,...,i u ̸=a s
! 
SRa (γth−κγth x)
sum +λ


∞
SRa λRa D (γth −κγth x)
λa,u
Z

sum +λ
λa,u
× λRa D exp −λRa D y−
κ9y
dy d x. 42 (x) = a exp (−a x)
0 κ9
s 
(24) λa,u +λSRa λRa D (γth − κγth x)
sum


× K1   . (30)
With the help of [33, 3.324.1], Q 1 and Q 2 can be thus, κ9
respectively, reformulated as
M−1 M Z 1 Finally, the OP in this case can be formulated as
X X κ (1+γth )
Q1 = 2 (−1)u+1 a exp (−a x)
u=1 i 1 =...=i u =1, 0 OPC−PRS
i 1 <...<i u , M M−1 M
i 1 ,...,i u ̸=a X X X
=1−2 (−1)u+1
λSRa λRa D (γth − κγth x)
r
× K1 a=1 u=1 i 1 =...=i u =1,
κ9 i 1 <...<i u ,
! i 1 ,...,i u ̸=a
λSRa λRa D (γth − κγth x)
r
   
× 2 dx (25)

κ9 41 (0) + 41 1
κ(1+γth )
N
X −1 41 n
κ(1+γth )N
× + 
M−1 M
λSRa 2κ (1 + γth ) N κ (1 + γth ) N
n=1
X X
Q2 = 2 (−1) u+1
sum + λ
λa,u SRa M M−1 M
u=1 i 1 =...=i u =1, X X X λSRa
i 1 <...<i u , +2 (−1)u+1 sum + λ
i 1 ,...,i u ̸=a λa,u SRa
1
a=1 u=1 i 1 =...=i u =1,
i 1 <...<i u ,
Z
κ (1+γth )
× a exp (−a x) 
i 1 ,...,i u ̸=a
  
s0

42 (0)+42 1
κ(1+γth )
N −1 42 n
κ(1+γth )N
SRa λRa D (γth − κγth x)


sum + λ
X
λa,u × +  (31)
× K1 2κ (1+γth ) N κ (1 + γth ) N
s κ9  n=1

(γ κγ


sum + λ
λa,u λ − x)
SRa Ra D th th which demonstrates the complicated dependence of the OP on
×  dx (26)
κ9 the number of relay nodes. However, we emphasize that the
OP obtained in (31) is only a function of the channel statistics
where K 1 (.) is the first-order modified Bessel function of the instead of instantaneous realizations.
second kind. Next, by applying the trapezoidal rule by dividing
[0; (1/κ(1 + γth ))] by xk = (k/N ), k = 1, 2, . . . , N , where N
is a sufficiently large positive integer, (20) and (21) can be, B. Second-Hop-Based Partial Relay Selection
respectively, rewritten as follows: Different from the conventional partial relay selection
M−1
X XM (CPRS) method, the second-hop-based partial relay selection
Q1 = 2 (−1)u+1 (SPRS) method selects the relay to obtain the highest channel
u=1 i 1 =...=i u =1, gain at the second hop (i.e., the relay-destination links), that is,
i 1 <...<i u ,
i 1 ,...,i u ̸=a γRb D = max γRm D (32)
     m=1,2,...,M
41 (0) + 41 κ(1+γ 1
th )
N −1 41 n
κ(1+γth )N
where Rb denotes the selected relay (b ∈ {1, 2, . . . , M}).
X
× + 
2κ (1 + γth ) N κ (1 + γth ) N We emphasize that (32) implies the highest channel gain
n=1
(27) between Rb and D. The OP, in this case, can be expressed by
M−1 M OPS−PRS
X X λSRa
Q2 = 2 (−1)u+1 M
λsum + λSRa
i 1 =...=i u =1, a,u
X
u=1 Pr γRb D > Z b , min ψSRb , ψRb D > γth



i 1 <...<i u ,
=1−
i 1 ,...,i u ̸=a b=1

Authorized licensed use limited to: Technical University of Ostrava. Downloaded on April 02,2024 at 03:09:54 UTC from IEEE Xplore. Restrictions apply.
NGUYEN et al.: OUTAGE PROBABILITY ANALYSIS FOR RELAY-AIDED SELF-ENERGY RECYCLING WSNs 11189

1 − κϕm κ9γSRm γRm D

 
M
1 − κϕb κ1γSRb γRb D
   
= 1 − Pr > γth , > γth
X
= 1− Pr γRb D > Z b , min , >γth κϕm 1 − κϕm
κϕb 1−κϕb
b=1 γth − κγth ϕm
 
1
M = 1 − Pr ϕm < , γSRm >
γth −κγth ϕb κ (1 + γth ) κ9γRm D
 
X 1
= 1− Pr ϕb < , γRb D > , γRb D>Z b
κ (1+γth ) κ1γSRb Z 1
κ (1+γth )
Z ∞
b=1 | {z } =1− f ϕm (x) d x f γRm D (y) dy
Q̃
Z ∞0 0
(33) × γ −κγ x f γSRm (z) dz (38)
th th
where Z b = max (γRb D ). By following the same κ9y
m=1,2,...,M,m̸=b Z 1 Z ∞
κ (1+γth )
approach for (26), OPS−PRS can be obtained as =1− f ϕm (x) d x λRm D exp
OPS−PRS  0 0
γth − κγth x

M M−1 M × −yλRm D − λSRm dy. (39)
X X X κ9y
=1−2 (−1)u+1
b=1 u=1 i 1 =...=i u =1, With the help of [33, 3.324.1], (32) can be reformulated by
i 1 <...<i u ,
i 1 ,...,i u ̸=b Z 1
κ (1+γth )
DF
OPm = 1 − 2 m exp (−m x)
    
41 (0) + 41 1
κ(1+γth )
N
X −1 41 n
κ(1+γth )N 0
× + 
2κ (1 + γth ) N κ (1 + γth ) N λSRm λRm D (γth − κγth x)
r
n=1 × K1
M M−1 M κ9
X X X λRb D !
(−1)u+1 λSRm λRm D (γth − κγth x)
r
+2
λsum + λRb D × 2 dx
b=1 u=1 i 1 =...=i u =1, b,u κ9
i 1 <...<i u ,
i 1 ,...,i u ̸=b     
    ϒ (0) + ϒ κ(1+γ 1 N −1 ϒ n
κ(1+γth )N
th )

43 (0)+43 κ(1+γ 1 −1 43 n

N
X
th) κ(1+γth )N =1−2 +
2κ (1+γth ) N κ (1 + γth ) N 
X
× +  (34)
2κ (1 + γth ) N κ (1 + γth ) N

n=1
n=1
(40)
where 43 (x) in (34) is defined as follows:
v
u  where ϒ(x) is given as
Rb D λSRb (γth −κγth x)
u λsum +λ
λSRm λRm D (γth − κγth x)
t b,u r
43 (x) = b exp (−b x) ϒ (x) = m exp (−m x)
κ9 κ9 !
v  
λSRm λRm D (γth − κγth x)
u  r
Rb D λSRb (γth −κγth x) 
u λsum +λ
t b,u × K1 2 . (41)
× K1   . (35) κ9
 κ9 
Due to the complicated propagation environment, the opti-
mally full relay selection (ORS) demonstrates its benefits
C. Optimally Full Relay Selection by offering the best system performance via exploiting full
The optimal relay selection can be written as follows: channel state information. The remaining benchmarks only
  exploit partial channel state information, which results in the
ψoe2e = max ψme2e (36) suboptimal relay selection.
m=1,2,...,M
Remark 1: By assuming the deep shadow fading model,
where Ro denotes the selected relay (o ∈ {1, 2, . . . , M}). The we have illustrated the three relay selection methods based on
OP in this case can be expressed by3 the different available SINR levels of the propagation links.
  Y M   The OP obtained analytically for each relay selection method
OPORS = Pr ψoe2e < γth = Pr ψme2e < γth (37) is independent of small-scale fading coefficients, which can
m=1 | {z } be applied for multiple coherence intervals where the channel
OPDF
m statistics remain unchanged. Consequently, the analytical OP
where OPDF obtained in this article offers much lower computational com-
m is further computed as follows:
plexity than Monte Carlo simulations, which require averaging
m = 1 − Pr ψSRm > γth , ψRm D > γth
OPDF


out the fluctuations of many different realizations of small-
scale fading coefficients.
3 Regarding connection efficiency, the OP is a good metric to evaluate
the disconnection probability, and its compensation measures the coverage
probability. Regarding data-gathering efficiency, we stress that the OP guar- IV. DNN D ESIGN
antees that each receiver can successfully collect data from the transmitters.
Moreover, it allows us to gather data and train a neural network for learning- In this section, we establish and propose a DNN framework
based purposes. for predicting the OP of CPRS, SPRS, and ORS schemes [34]

Authorized licensed use limited to: Technical University of Ostrava. Downloaded on April 02,2024 at 03:09:54 UTC from IEEE Xplore. Restrictions apply.
11190 IEEE SENSORS JOURNAL, VOL. 24, NO. 7, 1 APRIL 2024

TABLE II
PARAMETER S ETTING FOR S IMULATION

(i − 1)th layer by the expression

 
X
g ij = w  cijl gli−1 + s ij  (42)
l∈Ni−1
Fig. 2. Proposed DNN architecture to learn the OP based on statistical
channel state information. where w(.) denotes the activation function, for example, ReLu,
cijl , where Ni−1 denote the number of neurons of the i − 1th
TABLE I layer, cijl denote the weight from the jth neuron of the
I NPUT PARAMETERS FOR DNN T ESTING AND T RAINING
ith layer to the lth neuron, and s ij denote the bias in the
jth neuron in the ith layer.

C. Performance for DNN Network

We study the DNN model’s accuracy with 105 samples
created, and 80% for training, 10% for validation, and 10% for
testing. The developed DNN model, in particular, comprises
5 hidden layers, each hidden layer has 128 neurons, and the
hidden neurons use the rectified linear unit (ReLU) as the
with short run-time and low computing cost, toward practi- nonlinear activation function. In addition, the Adam optimizer
cal applications. Furthermore, the DNN model will be used is used to determine the appropriate weights and biases for
to reveal the relationship between network parameters and each connection. The mean-square error (MSE), which is used
secrecy performance by utilizing a compact mapping function to assess the effectiveness of the suggested DNN approach,
and lowering execution time. is defined as
L
1X pre pre

A. Architecture of DNN MSE = OPout [n] − OP̃out [n] (43)
L
We design a DNN model with an input layer, hidden layers, n=1
and the output layer as depicted in Fig. 2. For the input layer, where L denotes the number of training samples used for
pre
data are sent to the input layer so that the DNN model may the training phase. OPout [n] denotes the expected values and
pre
determine the relationship between system parameters and the OP̃out [n] denotes the predicted values. We use the root-MSE
appropriate OP. As a result, the number of neurons in the (RMSE) to calculate the difference between the natural OP
input layer is equal to the number of network parameters values and predicted ones throughout the whole test set in
with seven neurons and shown in Table I. The hidden layers the OP prediction, where the RMSE is defined as RMSE =
primarily determine the relationship between input and output (MSE)1/2 . In this article, we use a quasistatic channel model
data. To accurately calculate the relation, each link in each and perfect buffer so that one can evaluate the potential of
hidden neuron has a separate weight and bias. To enhance a learning-based approach. The main purpose is to evaluate
computational effectiveness, each hidden neuron also has a the learning-based performance in comparison to the model-
nonlinear activation function. The layer makes up the output, based approach. The effective experience reply buffer should
which uses the linear activation function to produce the be of interest for future work as investigating the limitations
pre
projected OP value, OPout , as the goal of the regression issue is of neural networks under practical constraints [35].
to predict an output value devoid of any additional conversions.
The learning-based method represented has low computational V. N UMERICAL R ESULTS
complexity since the fully connected neural network designed This section verifies the analytical results obtained in
in this article only exploits basic arithmetic operators and this article by using with the parameter setting given in
simple activation functions. Table II [36]. All three relay selection methods described
in Section III are considered benchmarks for comparison to
B. Training for DNN evaluate the OP performance. We first observe that the analyt-
We consider a fully connected layer in which activation g ij ical results match perfectly with the Monte Carlo simulations
of the jth neuron in the i layer is linked to activation in the for the considered parameter settings, therefore validating the

Authorized licensed use limited to: Technical University of Ostrava. Downloaded on April 02,2024 at 03:09:54 UTC from IEEE Xplore. Restrictions apply.
NGUYEN et al.: OUTAGE PROBABILITY ANALYSIS FOR RELAY-AIDED SELF-ENERGY RECYCLING WSNs 11191

Fig. 5. OP of the different number of relays and the different relay

Fig. 3. OP versus the different transmit SNR [dB] and the different relay selection methods with ρ = 0.2, η = 0.8, and Ψ = 2 [dB].
selection methods with ρ = 0.2, M = 3, and η = 0.8 and different the
target rate Cth .

Fig. 6. OP of the different Ωm and the different relay selection methods

Fig. 4. OP of the different ρ and the different relay selection methods with Ψ = 2 [dB], M = 3, η = 0.8, and ρ = 0.2.
with M = 3, η = 0.8, and Ψ = 2 [dB].

accuracy of the developed analytical results. The analysis for is still the baseline, but its performance is very close to the
the OP corresponding to a given set of different parameters is SPRS. At the minimum OP, the performance gap between the
presented in detail hereafter.4 two benchmarks is by a factor of about 1.41. In addition,
In Fig. 3, we plot the OP as a function of the transmit the superiority of the optimal relay selection is demonstrated
SNR [dB] with the different values Cth . We observe that by the lowest OP of about 1.89 × 10−4 regarding the power
the CPRS is the baseline with the OP is affected much by splitting factor ρ = 0.25. The higher ρ leads to less power
adjusting Cth . It exemplifies the advantages of employing more dedicated to the data transmission, so the OP is dramatically
transmit power to improve coverage performance. In addition, increased, especially approaching one as ρ ≥ 0.8.
the SPRS produces a lower OP than the baseline depending In Fig. 5, we show the OP as a function of relays in
on the data from the second hop to choose the appropriate the network. As expected, the communication reliability is
relay. In particular, the OP typically improves by around 1.5× improved when many relays are available in the coverage area
with Cth = 0.5. The OPS produces the highest communication providing more degree of freedom to select the best relay to
dependability among the benchmarks, outperforming the SPRS support the data transmission. Both the suboptimal methods,
on average. that is, CPRS and SPRS, produce the OP competitive to each
In Fig. 4, we illustrate the OP versus the different values other. In particular, the OP gets reduced from 0.07 to 0.04 as
of the power splitting factor ρ. Interestingly, the OP is a the number of relays increases from one to five. Despite
nonconvex function of ρ, where the lowest OP depends on the sharing the same trend, the significant reductions of the OP
relay selection methods. However, the power splitting factor compared to the other benchmarks are attained by exploiting
narrows in [0.2, 0.3] regarding the minimum OP. The CPRS the ORS, thanks to the SINR of both the links expressed
in (16). The gains are more dominant if the number of relays
4 Our numerical implementation and the source code are available for becomes larger, which unveils the contributions of the global
reproducibility at: https://github.com/lamthanh0/IEEE_Sensors. relay selection solution to the network performance.

Authorized licensed use limited to: Technical University of Ostrava. Downloaded on April 02,2024 at 03:09:54 UTC from IEEE Xplore. Restrictions apply.
11192 IEEE SENSORS JOURNAL, VOL. 24, NO. 7, 1 APRIL 2024

hop to select the relevant relay. Specifically, the improvement

of the OP is about 50% on average compared to the baseline.
Among the benchmarks, the OPS produces the best commu-
nication reliability, outperforming the SPRS by about 3.8× on
average. Also, take note of the fact that the predicted OP for
CPRS, SPRS, and ORS from the DNN model closely matches
the results of the Monte-Carlo simulation and the mathematical
analysis, allowing us to conclude that the data-driven approach
is a good way to estimate system performance when it is too
complex to be expressed using a closed-form expression.

VI. C ONCLUSION
We have investigated the OP of relay-aided WSNs where
the communication reliability between the source and the
Fig. 7. MSE with the different epochs with training and validating of the
DNN.
destination is affected by sensors reacting as relay selections.
In particular, we have considered the three different sensor
selection methods and their OP in analytical formulation that
operates over a long period and therefore is of practical
interest. Despite the complicated expressions, the OP can be
theoretically obtained as a function of channel statistics that
can operate over multiple coherence intervals. We have, there-
fore, proposed to exploit a deep-learning structure. Numerical
results demonstrate the critical roles of exploiting the full
SINR information to obtain the optimal relay selection that
minimizes the OP and therefore offers better reliable commu-
nications. The investigation of the OP utilizing different relay
selection methods under imperfect channel state information
and different data traffic is a potential research direction for
future work.

R EFERENCES
[1] K. Shafique, B. A. Khawaja, F. Sabir, S. Qazi, and M. Mustaqim,
Fig. 8. OP versus the different transmit SNRs [dB] and the different
“Internet of Things (IoT) for next-generation smart systems: A review
relay selection methods with ρ = 0.2, M = 3, and η = 0.8.
of current challenges, future trends and prospects for emerging 5G-IoT
scenarios,” IEEE Access, vol. 8, pp. 23022–23040, 2020.
[2] T. N. Nguyen et al., “Physical layer security in AF-based cooperative
SWIPT sensor networks,” IEEE Sensors J., vol. 23, no. 1, pp. 689–705,
Furthermore, in Fig. 6, we plot the OP versus the different Jan. 2023.
values of the exponential scalar m . Both the CPRS and [3] S. Hakak et al., “Autonomous vehicles in 5G and beyond: A survey,”
SPRS offer almost equal performance, where the OP gradually Veh. Commun., vol. 39, Feb. 2023, Art. no. 100551.
[4] S. Hudson, S. Matson-Barkat, N. Pallamin, and G. Jegou, “With or
degrades as m is increased. Specifically, the enhancement without you? Interaction and immersion in a virtual reality experience,”
of the OP is by a factor of about 9.6. Besides, the ORS J. Bus. Res., vol. 100, pp. 459–468, Jul. 2019.
improves the OP about 463× as m changes from 0.25 to 4. [5] M. H. Alsharif, M. S. Hossain, A. Jahid, M. A. Khan, B. J. Choi, and
S. M. Mostafa, “Milestones of wireless communication networks and
In addition, the network with the ORS will provide the OP technology prospect of next generation (6G),” Comput., Mater. Continua,
intensification of 463× when m upgrades in the same order. vol. 71, no. 3, pp. 4803–4818, 2022.
The training and validation datasets are exploited in Fig. 7 [6] I. Tomkos, D. Klonidis, E. Pikasis, and S. Theodoridis, “Toward the
6G network era: Opportunities and challenges,” IT Prof., vol. 22, no. 1,
to assess the training’s correctness and effectiveness of our pp. 34–38, Jan. 2020.
proposed learning framework. As can be seen, the MSE con- [7] W. Saad, M. Bennis, and M. Chen, “A vision of 6G wireless systems:
verges and falls below 10−4 after 30 epochs. In the considered Applications, trends, technologies, and open research problems,” IEEE
DNN, we set four hidden layers, each comprising 150 neurons. Netw., vol. 34, no. 3, pp. 134–142, Oct. 2019.
[8] H. Sedjelmaci, M. A. Messous, S. M. Senouci, and I. H. Brahmi,
The results unveil that the DNN can learn and predict the “Toward a lightweight and efficient UAV-aided VANET,” Trans. Emerg.
OP with very high accuracy and can be applied for practical Telecommun. Technol., vol. 30, no. 8, p. e3520, Aug. 2019.
purposes. [9] J. Jia, L. Ma, J. Chen, X. Wang, and A.-H. Aghvami, “Joint optimization
on power allocation and splitting for WSNs with SWIPT-based relay,”
In Fig. 8, we plot the OP as a function of the transmit IEEE Sensors J., vol. 22, no. 3, pp. 2877–2888, Feb. 2022.
SNR [dB]. The CPRS is the baseline with the OP reducing [10] T. N. Nguyen et al., “Security-reliability tradeoff analysis for SWIPT-
37.1× as the SNR increases from −15 to 5 [dB]. It demon- and AF-based IoT networks with friendly jammers,” IEEE Internet
strates the benefits of using higher transmit power in enhancing Things J., vol. 9, no. 21, pp. 21662–21675, Nov. 2022.
[11] M. Mohammadi et al., “Full-duplex non-orthogonal multiple access for
the coverage performance. Besides, the SPRS yields a lower next generation wireless systems,” IEEE Commun. Mag., vol. 57, no. 5,
OP than the baseline based on the information of the second pp. 110–116, May 2019.

Authorized licensed use limited to: Technical University of Ostrava. Downloaded on April 02,2024 at 03:09:54 UTC from IEEE Xplore. Restrictions apply.
NGUYEN et al.: OUTAGE PROBABILITY ANALYSIS FOR RELAY-AIDED SELF-ENERGY RECYCLING WSNs 11193

[12] Study on Integrated Access and Backhaul (Release 16), docu- [35] P. K. Donta, S. N. Srirama, T. Amgoth, and C. S. R. Annavarapu,
ment TR 38.874, 3GPP, Dec. 2018. “ICoCoA: Intelligent congestion control algorithm for CoAP using deep
[13] G. Noh, H. Chung, and I. Kim, “Mobile relay technology for 5G,” IEEE reinforcement learning,” J. Ambient Intell. Humanized Comput., vol. 14,
Wireless Commun., vol. 27, no. 3, pp. 6–7, Jun. 2020. no. 3, pp. 2951–2966, Mar. 2023.
[14] T. N. Nguyen, D.-H. Tran, T. Van Chien, V.-D. Phan, M. Voznak, and [36] R. Y. Rubinstein and D. P. Kroese, Simulation and the Monte Carlo
S. Chatzinotas, “Security and reliability analysis of satellite-terrestrial Method. Hoboken, NJ, USA: Wiley, 2016.
multirelay networks with imperfect CSI,” IEEE Syst. J., vol. 17, no. 2,
pp. 2824–2835, Jun. 2023, doi: 10.1109/JSYST.2022.3201128.
[15] A. C. Cirik, Y. Rong, and Y. Hua, “Achievable rates of full-duplex
MIMO radios in fast fading channels with imperfect channel estima-
tion,” IEEE Trans. Signal Process., vol. 62, no. 15, pp. 3874–3886,
Aug. 2014.
[16] A. Ranjha, G. Kaddoum, and K. Dev, “Facilitating URLLC in UAV-
assisted relay systems with multiple-mobile robots for 6G networks: Tan N. Nguyen (Member, IEEE) was born in
A prospective of Agriculture 4.0,” IEEE Trans. Ind. Informat., vol. 18, 1986, Nha Trang City, Vietnam. He received
no. 7, pp. 4954–4965, Jul. 2022. the B.S. degree in electronics from the Ho Chi
[17] S. Kurma, P. K. Sharma, S. Dhok, K. Singh, and C.-P. Li, “Adap- Minh University of Natural Sciences, Ho Chi
tive AF/DF two-way relaying in FD multiuser URLLC system with Minh City, Vietnam, in 2008, the M.S. degree
user mobility,” IEEE Trans. Wireless Commun., vol. 21, no. 12, in telecommunications engineering from the
pp. 10224–10241, Dec. 2022. Vietnam National University, Ho Chi Minh City,
[18] K. N. Le, “Distributions of even-degree-of-freedom generalized-Rician in 2012, and the Ph.D. degree in commu-
fading for opportunistic-relay selection employing MRC under outdated- nications technologies from the Faculty of
CSI conditions,” IEEE Trans. Commun., vol. 67, no. 9, pp. 6441–6455, Electrical Engineering and Computer Science,
Sep. 2019. VSB–Technical University of Ostrava, Ostrava,
[19] M. Sadeghi and A. M. Rabiei, “A fair relay selection scheme for a DF Czech Republic, in 2019.
cooperative network with spatially random relays,” IEEE Trans. Veh. He joined the Faculty of Electrical and Electronics Engineering, Ton
Technol., vol. 66, no. 12, pp. 11098–11107, Dec. 2017. Duc Thang University, Ho Chi Minh City, in 2013, and since then has
[20] A. B. Nassif, I. Shahin, I. Attili, M. Azzeh, and K. Shaalan, “Speech been lecturing. His major interests are cooperative communications,
recognition using deep neural networks: A systematic review,” IEEE cognitive radio, signal processing, satellite communication, UAV, and
Access, vol. 7, pp. 19143–19165, 2019. physical layer security.
Dr. Nguyen started as the Editor-in-Chief of Advances in Electrical and
[21] N. Shribala, P. Srihari, and B. C. Jinaga, “Multiple error correction
Electronic Engineering (AEEE) journal in 2023.
binary channel coding scheme,” in Proc. Int. Conf. Green Comput.,
Commun. Conservation Energy (ICGCE), Dec. 2013, pp. 10–16.
[22] T. Van Chien, T. Nguyen Canh, E. Björnson, and E. G. Larsson, “Power
control in cellular massive MIMO with varying user activity: A deep
learning solution,” IEEE Trans. Wireless Commun., vol. 19, no. 9,
pp. 5732–5748, Sep. 2020.
[23] A. L. Ha, T. van Chien, T. H. Nguyen, W. Choi, and V. D. Nguyen,
“Deep learning-aided 5G channel estimation,” in Proc. 15th Int. Conf. Trinh Van Chien (Member, IEEE) received
Ubiquitous Inf. Manage. Commun. (IMCOM), Jan. 2021, pp. 1–7. the B.S. degree in electronics and telecom-
[24] L. Tang, K.-C. Wang, Y. Huang, and F. Gu, “Channel characterization munications from the Hanoi University of Sci-
and link quality assessment of IEEE 802.15.4-compliant radio for factory ence and Technology (HUST), Hanoi, Vietnam,
environments,” IEEE Trans. Ind. Informat., vol. 3, no. 2, pp. 99–110, in 2012, the M.S. degree in electrical and
May 2007. computer engineering from Sungkyunkwan Uni-
[25] Y. Liu, S. Bi, Z. Shi, and L. Hanzo, “When machine learning meets big versity (SKKU), Seoul, South Korea, in 2014,
data: A wireless communication perspective,” IEEE Veh. Technol. Mag., and the Ph.D. degree in communication sys-
vol. 15, no. 1, pp. 63–72, Mar. 2020. tems from Linköping University (LiU), Linkoping,
[26] T.-H. Vu, T.-V. Nguyen, and S. Kim, “Wireless powered cognitive Sweden, in 2020.
NOMA-based IoT relay networks: Performance analysis and deep learn- He was a Research Associate with the Uni-
ing evaluation,” IEEE Internet Things J., vol. 9, no. 5, pp. 3913–3929, versity of Luxembourg, Esch-sur-Alzette, Luxembourg. He is currently
Mar. 2022. with the School of Information and Communication Technology (SoICT),
[27] C. She et al., “A tutorial on ultrareliable and low-latency communications Hanoi University of Science and Technology (HUST). His research inter-
in 6G: Integrating domain knowledge into deep learning,” Proc. IEEE, est lies in convex optimization problems and machine-learning applica-
vol. 109, no. 3, pp. 204–246, Mar. 2021. tions for wireless communications and image and video processing.
[28] T. Van Chien, C. Mollén, and E. Björnson, “Large-Scale-Fading decod- Dr. Van Chien received the Award of Scientific Excellence in the first
ing in cellular massive MIMO systems with spatially correlated chan- year of the 5GWireless Project funded by European Union Horizon
nels,” IEEE Trans. Commun., vol. 67, no. 4, pp. 2746–2762, Apr. 2019. 2020. He was an Exemplary Reviewer of IEEE W IRELESS C OMMUNI -
[29] D. K. Sah, K. Cengiz, P. K. Donta, V. N. Inukollu, and T. Amgoth, CATIONS L ETTERS 2016, 2017, and 2021.
“EDGF: Empirical dataset generation framework for wireless sensor
networks,” Comput. Commun., vol. 180, pp. 48–56, Dec. 2021.
[30] H. Liu, K. J. Kim, K. S. Kwak, and H. V. Poor, “Power splitting-based
SWIPT with decode-and-forward full-duplex relaying,” IEEE Trans.
Wireless Commun., vol. 15, no. 11, pp. 7561–7577, Nov. 2016.
[31] Z. Wang, X. Yue, and Z. Peng, “Full-duplex user relaying for
NOMA system with self-energy recycling,” IEEE Access, vol. 6,
pp. 67057–67069, 2018. Viet Quang Dinh is currently a Research Assis-
tant at the School of Information and Commu-
[32] A. Thangaraj, G. Kramer, and G. Böcherer, “Capacity bounds for
nication Technology (SoICT), Hanoi University
discrete-time, amplitude-constrained, additive white Gaussian noise
of Science and Technology, Hanoi, Vietnam.
channels,” IEEE Trans. Inf. Theory, vol. 63, no. 7, pp. 4172–4182,
He is a member of the Modern Computer Net-
Jul. 2017.
works and Communications Technology Labora-
[33] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and
tory, SoICT. His research interests include the
Products. New York, NY, USA: Academic, 2014.
optimization and applications of machine learn-
[34] K. Sultan and F. A. Alhaidari, “Relay selection in dual-hop under- ing for wireless communications and satellite
lay relay-assisted cognitive radio networks: A comprehensive survey,”
systems.
J. Comput. Theor. Nanosci., vol. 16, no. 1, pp. 1–13, Jan. 2019.

Authorized licensed use limited to: Technical University of Ostrava. Downloaded on April 02,2024 at 03:09:54 UTC from IEEE Xplore. Restrictions apply.
11194 IEEE SENSORS JOURNAL, VOL. 24, NO. 7, 1 APRIL 2024

Lam-Thanh TU received the B.Eng. degree in Dr. Voznak served as the General Chair for the 11th IFIP Wireless and
electronics and telecommunications engineering Mobile Networking Conference, in 2018, and the 24th IEEE/ACM Inter-
from the Ho Chi Minh City University of Tech- national Symposium on Distributed Simulation and Real-Time Applica-
nology, Ho Chi Minh City, Vietnam, in 2009, tions, in 2020.
the M.Sc. degree in telecommunications engi-
neering from the Posts and Telecommunica-
tions Institute of Technology, Ho Chi Minh City,
in 2014, and the Ph.D. degree from the Univer-
sity of Paris Sud, Paris-Saclay University, Bures-
sur-Yvette, France, in 2018.
From 2015 to 2018, he was with the French
National Center for Scientific Research (CNRS), Paris, France, as an
Early Stage Researcher of the European-funded project H2020 ETN-
5GWireless. From 2019 to 2021, he was with the XLim Research Insti-
tute, University of Poitiers, Poitiers, France, as a Postdoctoral Research
Fellow. Since 2022, he has been with the Faculty of Electrical and
Electronics Engineering, Ton Duc Thang University, Ho Chi Minh City.
His research interests include stochastic geometry, LoRa networks,
reconfigurable intelligent surfaces, covert communications, and artificial
intelligence applications for wireless communications.
Dr. TU was a recipient of the 2017 IEEE SigTelCom and the 2022
RICE Best Paper Award. Since 2023, he has been an Associate Editor
of the IEEE C OMMUNICATIONS L ETTERS and the Managing Editor of
the Advances in Electrical and Electronic Engineering. He has been a
member of the Technical Program Committee of several conferences
such as IEEE Globecom, IEEE ICC, IEEE SPAWC, EuCNC, IEEE ATC,
and IEEE NICS. He was an Exemplary Reviewer of IEEE T RANSACTION Zhiguo Ding (Fellow, IEEE) received the B.Eng.
ON C OMMUNICATIONS in 2016. degree from the Beijing University of Posts and
Telecommunications, Beijing, China, in 2000,
and the Ph.D. degree from Imperial College Lon-
Miroslav Voznak (Senior Member, IEEE) don, London, U.K., in 2005. He is currently a
received the Ph.D. degree in telecommuni- Professor of Communications, Khalifa University,
cations and the Habilitation degree, from the Abu Dhabi, United Arab Emirates, and has also
Faculty of Electrical Engineering and Computer been affiliated with Princeton University, Prince-
Science, VSB–Technical University of Ostrava, ton, NJ, USA. His research interests include
Ostrava, Czech Republic, in 2002 and 2009, 6G networks, multiple access, energy harvesting
respectively. networks, and statistical signal processing.
He was appointed as a Full Professor of Dr. Ding received the EU Marie Curie Fellowship, from 2012 to 2014;
Electronics and Communications Technologies, the Top IEEE TVT Editor 2017; the IEEE Heinrich Hertz Award
in 2017. Currently, he is a Principal Investigator 2018; the IEEE Jack Neubauer Memorial Award 2018; the IEEE Best
in the research project QUANTUM5 funded Signal Processing Letter Award 2018; the Friedrich Wilhelm Bessel
by NATO, which focuses on the application of quantum cryptography Research Award 2020, the IEEE SPCC Technical Recognition Award
in 5G campus networks. He participated in six projects funded by 2021; and the IEEE VTS Best Magazine Paper Award 2023. He is
the EU in programs managed directly by the European Commission. a Distinguished Lecturer of IEEE ComSoc, and a Web of Science
He has authored and coauthored over one hundred articles indexed in Highly Cited Researcher in two categories in 2022. He is serving as
SCI/SCIE journals. According to a Stanford University study released an Area Editor for IEEE T RANSACTIONS ON W IRELESS C OMMUNICA -
in 2020, he is one of the World’s Top 2% of scientists in Networking TIONS , and IEEE O PEN J OURNAL OF THE C OMMUNICATIONS S OCI -
and Telecommunications and Information and Communications ETY , an Editor for IEEE T RANSACTIONS ON V EHICULAR T ECHNOLOGY .
Technologies. His research interests generally focus on ICT, especially He was an Editor for IEEE W IRELESS C OMMUNICATION L ETTERS,
quality of service and experience, network security, wireless networks, IEEE T RANSACTIONS ON C OMMUNICATIONS, IEEE C OMMUNICATION
and big data analytics. L ETTERS, from 2013 to 2016.

Authorized licensed use limited to: Technical University of Ostrava. Downloaded on April 02,2024 at 03:09:54 UTC from IEEE Xplore. Restrictions apply.

View publication stats