IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 22, NO.

7, JULY 2023 4401

Throughput Maximization of Wireless-Powered

Communication Network With
Mobile Access Points
Xiaoying Liu , Member, IEEE, Bin Xu, Kechen Zheng , Member, IEEE,
and Haifeng Zheng , Senior Member, IEEE

Abstract— In order to mitigate the double near-far effect, Index Terms— Wireless-powered communication network,
we focus on a mobile wireless-powered communication network mobile access points, energy harvesting, throughput
(WPCN), where sensor nodes harvest energy from the radio maximization.
frequency (RF) signal of the mobile energy access point (EAP),
and transmit data to the mobile data access point (DAP) by using
the harvested energy. Only the sensor nodes with energy larger I. I NTRODUCTION
than a threshold, which is mainly determined by the energy
consumption of one transmission, have opportunities to transmit
data. Due to the mobility of the EAP and DAP, the distance
between the EAP and DAP changes over time. When the DAP
N OWADAYS, Internet of Things (IoT) nodes are widely
applied in the society [1], [4], and try their best to
provide high performance communications. The achievement
moves into the operation region of the EAP, the EAP and DAP of high performance such as throughput, however, is at the
could not work simultaneously due to the severe interference, expense of more energy supply [5]. Traditional IoT nodes are
and an energy harvesting probability is employed to denote the
probability that the EAP works in this scenario. The purpose powered by the embedded finite capacity batteries, resulting
of this paper is to identify the optimal transmission policy, i.e., in a limited lifetime. Accordingly, energy supply becomes a
the optimal pairing of the energy consumption of one transmis- bottleneck that limits the performance and mass deployment
sion and the energy harvesting probability, that maximizes the of IoT nodes [6]. Although rechargeable batteries and regular
throughput of the WPCN under an energy causality constraint. replacement could prolong the lifetime of IoT nodes, the
By analyzing the energy causality constraint, we show that the
WPCN could be divided into an energy-sufficient state and an problems of inconvenience, high cost, and danger exist at
energy-limited state by the pairing of the energy consumption of hard-to-reach places such as toxic areas and within the human
one transmission and the energy harvesting probability. Since body [7].
the energy consumption of one transmission and the energy Providing energy harvesting capability for IoT nodes is
harvesting probability are jointly intertwined with the energy an effective approach to achieve energy self-sufficiency and
causality constraint, making the joint optimization problem
intractable, we divide the throughput maximization problem into perpetual operation, thereby addressing the problem of limited
two layers. In the inner problem, we investigate the optimal lifetime resulting from the energy shortage. Energy harvest-
energy consumption of one transmission with a given energy ing technology collects and transforms environmental renew-
harvesting probability. In the outer problem, we derive the able energy into electrical energy [8]. The renewable energy
optimal energy harvesting probability based on the obtained sources (RES) could be roughly divided into two categories:
optimal energy consumption of one transmission. According
to the aforementioned investigations, we propose a two-layer natural RES and man-made RES. Compared with the nature
algorithm to obtain the specific solution. Numerical results are RES such as solar, wind, and thermoelectric energy, man-made
conducted to validate the theoretical results and the efficiency of RES, in particular radio frequency (RF) energy transmitted by
the proposed two-layer algorithm. a dedicated energy source, have the advantages of stability and
controllability. Due to the attractive advantages of RF-based
Manuscript received 18 May 2022; revised 8 September 2022; accepted
22 November 2022. Date of publication 5 December 2022; date of current energy harvesting, a new framework called wireless-powered
version 12 July 2023. This work was supported in part by the National communication network (WPCN) has been proposed [9].
Natural Science Foundation of China under Grant 61902351, Grant 61902353, A typical WPCN consists of an energy access point (EAP),
and Grant 61971139; in part by the Zhejiang Provincial Natural Science
Foundation of China under Grant LY21F020023 and Grant LY21F020022; a data access point (DAP) [10], and multiple nodes with energy
and in part by the Fundamental Research Funds for the Provincial Universities harvesting capability. Particularly, when a DAP also delivers
of Zhejiang under Grant RF-A2022005. The associate editor coordinating the energy signal as an EAP, it is called a hybrid access point
the review of this article and approving it for publication was D. Yu.
(Corresponding author: Kechen Zheng.) (HAP). In general, the nodes first harvest energy from the RF
Xiaoying Liu, Bin Xu, and Kechen Zheng are with the School of Computer signal transmitted by the EAP in the downlink, and then utilize
Science and Technology, Zhejiang University of Technology, Hangzhou the harvested energy to transmit data to the DAP in the uplink.
310023, China (e-mail: [email protected]; [email protected];
[email protected]). Hence, in order to achieve high performance, the allocation of
Haifeng Zheng is with the College of Physics and Information Engineering, time and energy resources for the downlink energy transfer
Fuzhou University, Fuzhou 350116, China (e-mail: [email protected]). and uplink data transmission is essential in WPCNs.
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/TWC.2022.3225085. There have been some efforts on investigating the through-
Digital Object Identifier 10.1109/TWC.2022.3225085 put by optimizing resource allocation of both HAP-based
1536-1276 © 2022 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: Zhejiang University of Technology. Downloaded on September 10,2023 at 11:56:34 UTC from IEEE Xplore. Restrictions apply.
4402 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 22, NO. 7, JULY 2023

WPCNs and non-HAP-based WPCNs where EAP and Most of the previous works [11], [14], [16], [21], however,
DAP work separately [11], [21]. Various scenarios on focused on the scenario with static access points (APs), hence
the HAP-based WPCNs have been considered in [11] the efficiency of energy harvesting and data transmission is
and [15]. Considering the features of the severe wireless limited by the distance between the nodes and APs, which
path loss and diverse data traffic demands from different leads to the double near-far effect [22], i.e., the nodes far
sensors in the wireless-powered underground sensor network away from APs harvest less energy, but conversely need to
(WPUSN), Liu et al. [11] proposed to use the multiple-input compensate for larger distance-based path loss with higher
multiple-output (MIMO) technique to eliminate the unreason- transmission power to ensure a certain data rate. In this regard,
able distribution of harvested energy resources, and improve with the aim of throughput maximization, time resources of
the throughput performance. By investigating a WPCN the nodes far away from APs would be exploited by the nodes
where cellular users and device-to-device users coexist, Mao close to APs, which causes a serious fairness problem [23].
et al. [12] proposed a cooperation protocol among different Unlike [11], [14], [16], and [21], in order to mitigate the
users to maximize the ratio of the throughput to the corre- double near-far effect, we consider a WPCN with mobile
sponding energy consumption through optimizing the energy APs. Due to the mobility of APs, the nodes are not always
beamformer of the HAP and time allocation. Applying the far away from the mobile APs, which enhances the fairness
intelligent reflecting surface (IRS) technology to the WPCN, among the nodes. Note that, although [15] has investigated
Lyu et al. [13] employed a wireless-powered IRS to act as a a mobile HAP-based WPCN, the mobile non-HAP-based
hybrid relay to enhance the energy transfer efficiency from the WPCNs, which could perform energy harvesting and data
HAP to the nodes and assist the data transmission. In addition transmission simultaneously, and work more efficiently, have
to the long packet communication in the uplink, the short not been studied.
packet communication, which suffers from the transmission Compared with the mobile HAP-based WPCN, there are
rate degradation and significant packet error rate, has been two potential challenges in the mobile non-HAP-based WPCN.
studied in [14]. Chen et al. [14] aimed to maximize the defined First, due to the mobility of the EAP and DAP, the scenario
effective-throughput by the joint optimization of the packet of the WPCN varies over time. In particular, when the EAP
error rate and transmission time. Different from previous and DAP are so close that they could not work simultane-
works, Zheng et al. [15] studied a mobile HAP-based WPCN, ously, the mobile non-HAP-based WPCN degenerates into
and proposed an energy threshold approach for throughput. the mobile HAP-based WPCN. Therefore, one challenge is
Compared with HAP-based WPCNs, non-HAP-based how to categorize the different scenarios, and analyze the
WPCNs work more efficiently, since the latter could per- harvested/consumed energy of the WPCN for the different
form energy harvesting and data transmission simultane- categories. Second, in the mobile HAP-based WPCN, the
ously, while the former could not [6]. Different scenarios on energy harvesting and data transmission could not occur
non-HAP-based WPCNs have been analyzed in [16] and [21]. simultaneously, hence the throughput is maximized mainly
Zewde et al. [16] focused on a WPCN where the non- by optimizing the tradeoff between the energy harvesting and
orthogonal multiple access (NOMA) scheme is applied in the data transmission, such as [15]. Since the mobile HAP-based
uplink information transfer, and studied the energy-efficient WPCN is a special scenario of the mobile non-HAP-based
time allocation strategies. Abd-Elmagid et al. [17] optimized WPCN, there may exist other potential intertwined tradeoffs.
the sum-throughput of the WPCN under two NOMA decod- Therefore, another challenge is how to optimize the potential
ing schemes: Successive Interference Cancellation Decoding intertwined tradeoffs for the throughput maximization.
(SICD) and Low Complexity Decoding (LCD). Aboelwafa In this paper, we consider a WPCN consisting of a mobile
et al. [18] further considered the fairness among the users EAP, a mobile DAP, and multiple sensor nodes. There is an
in the WPCN, and maximized the min-throughput of all users energy harvesting zone (EZ) around the EAP and a trans-
under the SICD and LCD schemes. Li et al. [19] formulated mission zone (TZ) around the DAP. A sensor node in the
the process of energy harvesting and data transmission as EZ could harvest energy from the EAP. While a sensor node
an infinite-horizon discounted Markov decision process to in the TZ could transmit data to the DAP when it is in the
maximize the long-term throughput of the WPCN under the active mode, i.e., its energy is larger than a threshold mainly
modes of time division multiplexing (TDM) and frequency determined by the energy consumption of one transmission
division multiplexing (FDM). Instead of investigating the et . The active mode is a manifestation of the energy causality
WPCN relying only on the dedicated downlink energy transfer, constraint requiring that the expected energy consumption
Wu et al. [20] considered a WPCN where the nodes harvest must not exceed the expected harvested energy. Due to the
energy from not only a dedicated energy source, i.e., EAP, but mobility of the EAP and DAP, the scenario of the WPCN
also ambient energy sources. Under this WPCN, they derived changes over time. Based on the distance between the EAP
the conditions for the activation of downlink energy transfer. and DAP, we divide the WPCN into three typical scenarios:
Unlike the aforementioned works considering one pair of EAP separation scenario (SS), overlap scenario (OS), and cover
and DAP, Liang et al. [21] investigated the WPCN consisting scenario (CS). To be specific, in the SS, the EZ and TZ do
of one DAP and multiple EAPs, and proposed a non-uniform not overlap with each other, hence the EAP and DAP could
deployment scheme for EAPs to minimize total average power work simultaneously. In the OS, the EZ and TZ overlap with
consumption. each other, while the DAP is outside of the EZ. Hence the

Authorized licensed use limited to: Zhejiang University of Technology. Downloaded on September 10,2023 at 11:56:34 UTC from IEEE Xplore. Restrictions apply.
LIU et al.: THROUGHPUT MAXIMIZATION OF WPCN WITH MOBILE ACCESS POINTS 4403

EAP and DAP could still work simultaneously. In the CS, the and the area of the TZ, on the design of the optimal
DAP is in the EZ, hence the EAP and DAP could not work transmission policy, and give directions on how to dynam-
simultaneously. If the EAP works in the CS, the WPCN har- ically adjust the energy consumption of one transmission
vests more energy but has less transmission opportunities, and and energy harvesting probability to reach the maximum
vice versa. Since both energy and transmission opportunities throughput when network parameters change. The effi-
are important for the throughput, we consider that, in the CS, ciency of the two-layer algorithm is also validated.
the EAP works with probability pe , called energy harvesting The remainder of this paper is organized as follows.
probability, that is, the DAP works with probability 1 − pe . Section II introduces the system model. Section III shows that
Actually, the energy consumption of one transmission et the WPCN could be divided into an energy-sufficient state and
and the energy harvesting probability pe have strong effects an energy-limited state. The design of the optimal transmission
on the throughput. et represents the tradeoff between the data policy is given in Section IV. Section V provides the numerical
rate and active probability that a sensor node is in the active results and discussions, and Section VI concludes this paper.
mode. pe represents the tradeoff between the active probability
and transmission probability that the DAP works. It is quite II. S YSTEM M ODEL
straightforward to know that the data rate, active probability, We introduce the system model from three parts: network
and transmission probability have positive correlations with the model, transmission model, and energy harvesting model.
throughput. However, the aforementioned two tradeoffs make In the first part, we introduce the network topology. In the
the throughput optimization problem much more perplexed other two parts, we introduce the details of data transmission
and complicated, since et and pe are jointly intertwined with and energy harvesting by the sensor nodes.
the active probability. The problem is to come up with a
proper pairing of et and pe based on the two interwoven
A. Network Model
tradeoffs. Therefore, it is challenging to design the opti-
mal transmission policy (et , pe ) that maximizes the expected We study a WPCN, where a mobile EAP, a mobile DAP,
achievable throughput of the WPCN. The main contributions and N uniformly distributed wireless-powered sensor nodes
are summarized as follows. coexist in the region with area A. The EAP, DAP, and each
sensor node are equipped with a single antenna. Time is
• By designing an optimal transmission policy, in terms of
divided into slots with unit length. The mobile EAP and DAP
the pairing of the energy consumption of one transmission
complete the movement at the beginning of each time slot.
and the energy harvesting probability, we address the
Due to the short movement duration, we neglect it at each
throughput maximization problem in a mobile WPCN
time slot, and consider that the data transmission and energy
that mitigates the double near-far effect.
harvesting do not occur during the movement [24]. During the
• By analyzing the energy causality constraint, we show
remaining time of each time slot, the locations of the EAP and
that the WPCN could be characterized in terms of
DAP remain unchanged. To ensure the fairness and the data
an energy-sufficient state and an energy-limited state
diversity of the WPCN, we consider the mobility model of
depending on the pairing of the energy consumption of
the EAP/DAP satisfies the requirement that the EAP/DAP has
one transmission and the energy harvesting probability.
uniform probability distribution at every location in the region
In the energy-sufficient state, the harvested energy is
from a long-term perspective.
sufficient for the WPCN to perform data transmission
The sensor nodes have no energy sources other than the RF
continuously, and the key factors that limit the throughput
signal of the EAP. Each sensor node harvests energy from the
are transmission probability and data rate. While in
RF signal of the EAP, and stores the harvested energy in a
the energy-limited state, the data transmission is strictly
rechargeable battery. Considering the requirement of energy
limited by the harvested energy, and the key factor that
harvesting circuits in each sensor node, the received power
limits the throughput becomes the energy supply.
of the RF signal at the sensor nodes should be larger than a
• To overcome the challenge that the energy consumption
certain threshold to ensure the efficient energy harvesting [15].
of one transmission and the energy harvesting proba-
Considering the large-scale path loss, we define EZ as a disk
bility are jointly intertwined with the energy causality
centered at the EAP with radius re . Only when the distance
constraint, we divide the joint throughput maximization
between the sensor node and the EAP is less than re , the
problem into an inner optimization problem and an outer
received power of the RF signal from the EAP at the sensor
optimization problem. To be specific, we first investigate
node is larger than the threshold, and the sensor node is
the optimal energy consumption of one transmission with
able to harvest energy. The value of re is determined by the
a given energy harvesting probability to address the inner
transmission power of the EAP and the threshold. During each
problem, and then derive the optimal energy harvesting
time slot, only the sensor nodes in the EZ could harvest energy
probability based on the obtained optimal energy con-
from the RF signal of the EAP. Since the sensor nodes are
sumption of one transmission to address the outer prob-
uniformly distributed, the probability that a given sensor node
lem. According to the above investigations, we propose
is in the EZ is expressed as
a two-layer algorithm for the specific solution.
• Numerical results show the impacts of key network re2 π
PEZ = . (1)
parameters, such as the transmission power of the EAP A

Authorized licensed use limited to: Zhejiang University of Technology. Downloaded on September 10,2023 at 11:56:34 UTC from IEEE Xplore. Restrictions apply.
4404 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 22, NO. 7, JULY 2023

Fig. 1. Three typical network scenarios.

Each sensor node transmits data to the DAP by using the the area of the OZ, and Aoz (red ) is expressed as
energy stored in the battery. In order to study the maximum re2 −rt2 + red
2
r2 −re2 + red
2
achievable throughput of the WPCN, we consider that the Aoz (red ) = re2 arccos + rt2 arccos t
2re red 2rt red
sensor nodes have sufficient data to transmit [25]. With consid-  2 2 2

r − rt + red
eration of the limited transmission power of the sensor nodes, − re red sin arccos e . (5)
the TZ is defined as a disk centered at the DAP with radius 2re red
rt , and only the data transmissions of the sensor nodes in the According to Aoz (red ) in (5), the average area of the OZ,
TZ satisfy the minimum data rate requirement of the WPCN. denoted by Aoz , is expressed as
Thus only the sensor nodes in the TZ have the opportunity to  rt +re
fpdf (red )
transmit data. The probability that a given sensor node is in Aoz = Aoz (red ) dred , (6)
re POS
the TZ is expressed as
and for a given sensor node, the probability that it is in the
r2 π
PT Z = t . (2) OZ is expressed as
A
Aoz
Due to the mobility of the EAP and DAP, the distance POZ = . (7)
A
between the EAP and DAP varies over time, and so does the
Cover scenario (CS): In the CS, the distance relationship
network scenario, which could be divided into three typical
0 ≤ red < re holds. The probability that the WPCN is in the
scenarios. As shown in Fig. 1, according to the distance
CS at each time slot is expressed as
between the EAP and DAP, the WPCN would be in one of  re
the following three typical scenarios at each time slot. PCS = fpdf (red ) dred . (8)
Separation scenario (SS): In the SS, the distance relation- 0
ship re + rt ≤ red holds, where red denotes the distance In the CS, since the DAP is in the EZ, the DAP receives
between the EAP and DAP. Since the interference from the strong interference from the EAP. Thus the EAP and DAP
EAP to the DAP is negligible, the EAP and DAP could work could not work simultaneously, i.e., the WPCN in the CS has
simultaneously during one time slot. Since the EAP/DAP has to choose between the energy harvesting and data transmis-
uniform probability distribution at every location in the region, sion. We consider that, when the WPCN is in the CS, the
the probability that the WPCN is in the SS at each time slot energy harvesting occurs with probability pe , and the data
is expressed as transmission occurs with probability 1 − pe . It is obvious
 rt +re that if neither of the energy harvesting and data transmission
PSS = 1 − fpdf (red ) dred , (3) occurs, the WPCN gains no benefit for the throughput. From
0
the long-term perspective, when pe increases, the number of
where fpdf (red ) is the probability density function of red , and the time slots allocated to the energy harvesting increases,
depends on the mobility models of the EAP and DAP. which allows sensor nodes to harvest more energy for the data
Overlap scenario (OS): In the OS, the distance relationship transmission. While the number of the time slots allocated to
re ≤ red < re + rt holds. Since the DAP is still outside of the the data transmission decreases. In other words, the increase
EZ, the EAP and DAP could still work simultaneously during of pe provides a higher active probability for one sensor
one time slot. The probability that the WPCN is in the OS at node, but sacrifices the transmission probability. The active
each time slot is expressed as probability for one sensor node is defined as the probability
 rt +re that the residual energy in its battery is larger than an energy
POS = fpdf (red ) dred . (4) threshold, and the transmission probability is defined as the
re probability that the data transmission occurs, i.e., 1 − PCS pe .
To be specific, in the OS, the area of the overlap zone (OZ) Hence, there is a tradeoff between the active probability and
between the EZ and TZ varies with red . Let Aoz (red ) denote the transmission probability resulting from pe .

Authorized licensed use limited to: Zhejiang University of Technology. Downloaded on September 10,2023 at 11:56:34 UTC from IEEE Xplore. Restrictions apply.
LIU et al.: THROUGHPUT MAXIMIZATION OF WPCN WITH MOBILE ACCESS POINTS 4405

After the EAP and DAP complete the movement, they C. Energy Harvesting Model
broadcast their respective locations to the sensor nodes in When the WPCN performs energy harvesting, the sensor
the EZ and TZ. Moreover, if the DAP receives the location nodes in the EZ harvest energy from the RF signal of the
message broadcasted by the EAP, the WPCN is in the CS, EAP during the whole time slot. To be specific, when the
otherwise the WPCN is in the SS or OS. If the WPCN is WPCN is in the OS, each active node in the OZ needs to
in the CS, the DAP decides whether the energy harvesting send a report to the DAP within the duration τr . Therefore,
or data transmission occurs based on the energy harvesting if the active node in the OZ is not chosen to transmit data,
probability pe , and sends the decision to the EAP. After that, it harvests energy during τ − τr . Due to the short duration
if the decision is energy harvesting, the DAP broadcasts the of τr , we approximate the energy harvesting duration of the
decision to the sensor nodes in the TZ to avoid wasting energy active node as τ . If the active node in the OZ is chosen to
in sending the report. If the decision is data transmission, the transmit data, it harvests energy during τc − τr .1 Since the
EAP broadcasts the decision to the sensor nodes in the EZ duration of τc − τr is relatively short compared with that of τ ,
to avoid wasting time for energy harvesting. Due to the short we neglect the harvested energy during τc − τr . Namely, if the
duration of broadcasting the location messages and sending the active node in the OZ is chosen to transmit data, it harvests
decision, we neglect the duration of broadcasting the location negligible energy.
messages and sending the decision as [26]. Then we present the average energy harvested by one sensor
node in the EZ during one time slot. The energy harvesting
B. Transmission Model
power of the sensor node is a nonlinear function of the
Let τ denote the duration of each time slot. When the data input power, i.e., the power of the received RF signal from
transmission occurs, the sensor node in the TZ with residual the EAP at the sensor node [27]. We adopt the nonlinear
energy larger than er + et sends a report to the DAP one energy harvesting model called constant-linear-constant (CLC)
by one within the duration τc , where er denotes the energy model [28], [29] due to its practicality, accuracy and mathe-
consumption for one sensor node sending a report, and et matical simplicity. The energy harvesting power of the sensor
denotes the energy consumption of one transmission for a node is expressed as
sensor node. In order to ensure the DAP receives all the reports ⎧
successfully, we set τc as τc = N τr , where τr denotes the ⎪
⎨0, re < le ;
duration of one sensor node sending a report. According to the PH = PE le−α − Pmin , re ≤ le ≤ re ; (11)
⎪
⎩ 
received reports, the DAP randomly selects one sensor node Pmax − Pmin , le < re ,
that sends the report to transmit data during the remaining
time of one time slot τt with τt + τc = τ . where le denotes the distance between the sensor node
When et increases, i.e., sensor nodes transmit data with and EAP, α denotes the path loss exponent with α > 2,
more energy, the data rate increases, which improves the and PE denotes the transmission power of the EAP. Pmin and
throughput. However, the increase of et reduces the active Pmax respectively denote the sensitivity power threshold and
probability of the sensor node, and then decreases the prob- saturation power threshold of the energy harvesting circuits.
ability that there exists at least one active node in the TZ. When le = re , the power of the received RF signal from the
Namely, the increase of the data rate sacrifices the active EAP at the sensor node equals Pmin . When le = re , the power
probability. Therefore, there exists another tradeoff in the of the received RF signal from the EAP at the sensor node
WPCN, i.e., the tradeoff between the data rate and active equals Pmax . Since the sensor nodes are uniformly distributed,
probability. To formulate the tradeoff between the data rate and the average energy harvested by one sensor node in the EZ
active probability, we employ an energy threshold approach, during one time slot is expressed as
and define the energy threshold as er +et . We adjust the energy  re
2le
threshold by optimizing et to balance the data rate and active eh = ητ (Pmax − Pmin ) 2 dle
0 re
probability, and define the active node as the sensor node that  re
2le
the residual energy in the battery is larger than the energy + ητ (PE le−α − Pmin ) 2 dle , (12)
threshold. When the data transmission occurs, only the active 
re re
nodes in the TZ send the reports to the DAP. where η denotes the energy harvesting efficiency of each
As both et and pe affect the energy states of sensor sensor node with η ∈ (0, 1).
nodes, for a given sensor node, we let Pa (et , pe ) denote the
probability that the sensor node is active. Let M (et , pe ) denote
III. E NERGY S TATE A NALYSIS OF THE WPCN
the average number of active nodes in the TZ, and M (et , pe )
is expressed as In this section, we first formulate the energy causality con-
straint in Section III-A. Then we derive the active probability
M (et , pe ) = N PT Z Pa (et , pe ). (9)
Pa (et , pe ) by the energy causality constraint, and show that
The probability that there exists at least one active node in the the WPCN could be divided into an energy-sufficient state and
TZ is expressed as
 N PT Z  1 Note that in the OS, when the chosen active node is in the OZ, it could not

PA = 1 − 1 − Pa (et , pe ) , (10) harvest energy even if it is in the EZ. However, in the SS, each sensor node
in the EZ harvests energy. Thus from the perspective of the harvested energy,
where . denotes the floor operation. the OS is different from the SS, and we analyze the SS and OS separately.

Authorized licensed use limited to: Zhejiang University of Technology. Downloaded on September 10,2023 at 11:56:34 UTC from IEEE Xplore. Restrictions apply.
4406 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 22, NO. 7, JULY 2023

TABLE I
T HE E NERGY C ONSUMPTION AND H ARVESTED E NERGY OF A S ENSOR N ODE D URING O NE T IME S LOT

an energy-limited state depending on the energy consumption Lemma 1: For a given sensor node, the energy consump-
et and the energy harvesting probability pe in Section III-B. tion during one time slot is
⎧ et
A. Energy Causality Constraint ⎨ er + , (1−PCS pe )PT Z Pa (et , pe ),
Ec = M (e t , pe ) (13)
We summarize the possible cases of the energy consumption ⎩ 0, otherwise.
and harvested energy of a sensor node during one time slot,
and the corresponding probability of each case in Table I. The expected energy consumption of a sensor node during one
The probability that an active node in the TZ is chosen to time slot is expressed as
transmit data is M(e1t ,pe ) . Hence, for a given active node, the et
E(Ec ) = (1−PCS pe )PT Z Pa (et , pe )er +(1−PCS pe ) . (14)
energy consumption for data transmission during one time slot N
is M(eett,pe ) . Thus, from the perspective of energy consumption Proof: Corresponding to the 8 rows of results from
calculation during one time slot, the case that one of the active top to bottom in Table I, we define 8 cases as case 1,
nodes in the TZ transmits data is equivalent to the case that case 2, …, and case 8. The other cases not shown in Table I
each active node in the TZ transmits data with the energy are collectively referred to case 9. Let ci denote the number
consumption M(eett,pe ) in turn, i.e., in a time division multiple of time slots in which case i occurs during time slots [1, T ],
access (TDMA) fashion. Here, in order to generally express where T is a positive integer, and i ∈ {1, 2, . . . , 9}. The
all the possible cases in Table I in a clear manner, we employ number of time slots, where Ec = er + M(eett,pe ) holds,
the case that each active node in the TZ transmits data in turn. is (c2 + c4 + c6 + c8 ). The number of time slots, where
In Table I, Ec represents the amount of energy consumed Ec = 0 holds, is (c1 + c3 + c5 + c7 + c9 ). Based on the
by a sensor node during one time slot, and Eh represents law of large numbers, we obtain limT →∞ cTi = Po,i , where
the amount of energy harvested by a sensor node during one Po,i denotes the occurrence probability of case i. Hence,
time slot. Occurrence probability represents the probability as T → ∞, the probability of Ec = er + M(eett,pe ) is given
of occurrence for the corresponding case in the same row. by limT →∞ c2 +c4 +c
T
6 +c8
= Po,2 + Po,4 + Po,6 + Po,8 . The
EZ−OZ represents the region of the EZ except the OZ, and probability of Ec = 0 is given by limT →∞ c1 +c3 +cT5 +c7 +c9 =
TZ−OZ represents the region of the TZ except the OZ. When Po,1 + Po,3 + Po,5 + Po,7 + Po,9 . According to the occurrence
the WPCN is in the OS, the action decision (energy harvesting probability of each case, we obtain the energy consumption in
or data transmission) of an active node in the OZ depends on (13) and the expected energy consumption in (14).
whether the active node is chosen to transmit data or not, Lemma 2: For a given sensor node, the amount of energy
Pa (et ,pe )−1
and the probability that it is not chosen is N PT ZN PT Z . harvested during one time slot is
The cases in which both Ec and Eh equal 0 are not shown ⎧
⎨ e , 1 − P (1 − p )P − POS POZ ,
in Table I. Moreover, according to Section II-B, the energy h CS e EZ
Eh = N PT Z (15)
consumption of one transmission et and the energy harvesting ⎩ 0, otherwise.
probability pe in the CS are directly related to the active
probability Pa (et , pe ). However, it is hard to understand the The expected harvested energy of a sensor node during one
impacts of et and pe on the action decision of a sensor node time slot is expressed as
within a single time slot, since the action decision is affected  
  POS POZ
by the prior decisions. Therefore, we consider the energy E(Eh ) = 1 − PCS (1 − pe ) PEZ − eh . (16)
N PT Z
causality constraint and active probability from a long-term
perspective. Similar as the proof of Lemma 1, Lemma 2 could be
To investigate the energy causality constraint, we present proved. Due to the limited space, we omit the proof of
the expected energy consumption and the expected harvested Lemma 2. Based on Lemmas 1-2, we provide the energy
energy of a sensor node in Lemmas 1-2. causality constraint of a sensor node by Lemma 3.

Authorized licensed use limited to: Zhejiang University of Technology. Downloaded on September 10,2023 at 11:56:34 UTC from IEEE Xplore. Restrictions apply.
LIU et al.: THROUGHPUT MAXIMIZATION OF WPCN WITH MOBILE ACCESS POINTS 4407

Lemma 3: For a given sensor node, the energy causality the sensor node. Hence after a sufficient number of time
constraint could be formulated as slots, the residual energy accumulated in the battery would
be more than sufficient to exceed the energy threshold, and
E(Eh ) ≥ E(Ec ). (17)
always larger than the energy threshold during the following
Proof: We consider that each sensor node is equipped time slots. Therefore, Pa (et , pe ) = 1 holds. By substituting
with a battery with large capacity. Let E0 denote the initial Pa (et , pe ) = 1 into Lemmas 1-2, we obtain
energy of a sensor node, and the residual energy of the sensor E(Eh ) > E(Ec )
node at the beginning of time slot (T + 1) is expressed as  
9 9
  POS POZ
ET +1 = E0 − i=1 Ec,i ci + i=1 Eh,i ci , where ci is defined ⇒ 1 − PCS (1 − pe ) PEZ − eh
N PT Z
as the proof in Lemma 1, Ec,i denotes the amount of energy et
consumed by the sensor node during one time slot in case > (1 − PCS pe )PT Z er + (1 − PCS pe )
 N
i, and Eh,i denotes the amount of energy harvested by the 
sensor node during one time slot in case i. Due to ET +1 ≥ 0, 1 − PCS (1 − pe ) PEZ − PN OS POZ
PT Z eh
⇒
we obtain (1 − PCS pe )PT Z er
9 9 et
> 1+
E0 + Eh,i ci ≥ Ec,i ci . (18) N PT Z er
i=1 i=1 ⇒ Pa (et , pe ) = 1 ≤ Ψ(et , pe ), (23)
By dividing both sides of (18) by factor T , we obtain where Ψ(et , pe ) is expressed as (21).

E0
9
ci
9
ci When E(Eh ) = E(Ec ), Pa (et , pe ) < 1 holds as T → ∞.
lim + Eh,i ≥ lim Ec,i . (19) By substituting Lemmas 1-2 into E(Eh ) = E(Ec ), we obtain
T →∞ T T T →∞ T
i=1 i=1
E(Eh ) = E(Ec )
Based on the law of large numbers, we obtain  
9 9   POS POZ
i=1 Eh,i Po,i ≥ i=1 Ec,i Po,i . Accordingly, we obtain ⇒ 1 − PCS (1 − pe ) PEZ − eh
(17), and the proof is completed. N PT Z
1 − PCS pe
= (1 − PCS pe )PT Z Pa (et , pe )er + et
B. Active Probability by Energy Causality Constraint N
⇒ Pa (et , pe ) = Ψ(et , pe ). (24)
The active probability Pa (et , pe ) originates from
the energy causality constraint, which requires that the By combining the results (23) and (24) in the two cases, i.e.,
expected energy consumption should not exceed the expected E(Eh ) > E(Ec ) and E(Eh ) = E(Ec ), we obtain Pa (et , pe ) =
harvested energy. Consequently, we derive Pa (et , pe ) by the min{1, Ψ(et , pe )}. By comparing Ψ(et , pe ) with 1, the spe-
energy causality constraint in Theorem 1. cific expression of Pa (et , pe ) could be determined. To be
Theorem 1: For a given sensor node, the active probability specific, let Ψ(et , pe ) = 1, and we have et = eth (pe ), where
of the sensor node Pa (et , pe ) is eth (pe ) is expressed as (22). By analyzing the monotonicity
 of Ψ(et , pe ) in (21) with respect to et , it is easy to obtain that
Ψ(et , pe ), eth (pe ) < et ; Ψ(et , pe ) is a decreasing function of et . Consequently, when
Pa (et , pe ) = (20)
1, 0 ≤ et ≤ eth (pe ), 0 ≤ et ≤ eth (pe ), Ψ(et , pe ) ≥ 1 and Pa (et , pe ) = 1 hold.
When eth (pe ) < et , Ψ(et , pe ) < 1 and Pa (et , pe ) = Ψ(et , pe )
where Ψ(et , pe ) and eth (pe ) are expressed as hold. The proof is completed.
   
According to Theorem 1, we divide the WPCN into two
N 1 − PCS (1 − pe ) PEZ − POS POZ
eh
Ψ(et , pe ) =
PT Z long-term states: an energy-sufficient state with Pa (et , pe ) = 1
N (1 − PCS pe )PT Z er (i.e., E(Eh ) > E(Ec )) and an energy-limited state with
et Pa (et , pe ) = Ψ(et , pe ) < 1 (i.e., E(Eh ) = E(Ec )). After a
− , (21)
 NPT Z er   sufficient long time, in the energy-sufficient state, the harvested
N 1 − PCS (1 − pe ) PEZ − POS POZ
PT Z eh energy of the WPCN would be sufficient to carry out data
eth (pe ) = transmission for every time slot, i.e., Ψ(et , pe ) ≥ 1. However,
1 − PCS pe in the energy-limited state, the number of data transmissions is
− N PT Z er . (22)
limited by the harvested energy, and the WPCN is prohibited
When T → ∞, the equality (20) holds. from data transmission for some time slots due to the energy
Proof: According to Lemma 3, the long-term energy shortage, i.e., Ψ(et , pe ) < 1. Based on the aforementioned
relationship between E(Eh ) and E(Ec ) could be divided into analysis, it is straightforward to obtain the following corollary.
two cases, i.e., E(Eh ) > E(Ec ) and E(Eh ) = E(Ec ), where the Corollary 1: [Conditions for Energy-Sufficient State
WPCN has different characteristics on the active probability. and Energy-Limited State] The WPCN falls into the
When E(Eh ) > E(Ec ), the residual energy of a sensor energy-sufficient state or the energy-limited state based on the
node would be always larger than the energy threshold as energy consumption of one transmission et and the energy
T → ∞. The reason is that, from the long-term perspective, harvesting probability pe in the CS as follows:
the amount of the residual energy accumulates with expec- • If Ψ(et , pe ) ≥ 1, the WPCN falls into the
tation E(Eh ) − E(Ec ) at each time slot in the battery of energy-sufficient state, and the set of (et , pe ) in

Authorized licensed use limited to: Zhejiang University of Technology. Downloaded on September 10,2023 at 11:56:34 UTC from IEEE Xplore. Restrictions apply.
4408 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 22, NO. 7, JULY 2023

White Gaussian Noise (AWGN) channels, and σ 2 denotes the

noise power at the DAP. ld denotes the distance between the
sensor node and DAP, (1 + ld )−α represents the bounded path
−α
loss [31], and et (1+l d)
τt σ 2 represents the signal-to-noise-ratio
(SNR) at the DAP.
With the aim to maximize the expected achievable through-
put of the WPCN, we find that both the energy consumption
of one transmission et and the energy harvesting probability
pe have impacts on the throughput based on (25). To be
specific, the increase of et leads to the increase of data
rate, while reduces the active probability Pa (et , pe ) as well
as the probability that there exists at least one active node
in the TZ, i.e., PA . On the other hand, the increase of et
indicates that the sensor nodes need to harvest more energy
Fig. 2. A sketch for the energy states of the WPCN divided jointly by
for the data transmission, and then the sensor nodes are more
the energy consumption of one transmission et and the energy harvesting likely to harvest energy rather than transmit data when the
probability pe in the CS. WPCN is in the CS. Hence, the increase of et results in the
increase of pe , which reduces the transmission probability
(1 − PCS pe ), but in turn increases Pa (et , pe ) as well as
the
 energy-sufficient state is denoted by Ss = PA . Based on the aforementioned analysis, it is challenging
(et , pe )|Ψ(et , pe ) ≥ 1, et > 0, 0 ≤ pe ≤ 1 . to characterize the impacts of et and pe on the throughput,
• If Ψ(et , pe ) < 1, the WPCN falls into the energy-limited
since et makes the tradeoff between the data rate and active
state, and the set of(et , pe ) in the energy-limited state probability much more perplexed, and pe makes the tradeoff
 by Sl = (et , pe )|Ψ(et , pe ) < 1, et > 0, 0 ≤
is denoted between the transmission probability and active probability
pe ≤ 1 . much more perplexed. Therefore, optimizing to achieve the
Fig. 2 depicts the energy state of the WPCN versus the maximum throughput of the WPCN requires jointly and prop-
energy consumption of one transmission et and the energy erly designing et and pe with consideration of the aforemen-
harvesting probability pe in the CS. The WPCN falls into the tioned two tradeoffs. The optimization problem of designing
energy-sufficient state when 0 ≤ et ≤ eth (pe ), and falls into the optimal transmission policy (e∗t , p∗e ) that maximizes the
the energy-limited state when et > eth (pe ). It is worth noting expected achievable throughput could be formulated as
that eth (0) may also be less than 0, and its specific value is
determined by several network parameters, such as the number P0 : max R(et , pe )
et ,pe
of sensor nodes N and the energy consumption for sending a s.t. emin ≤ et , 0 ≤ pe ≤ 1, (27)
report er .
where emin denotes the minimum energy consumption of one
IV. P ROBLEM F ORMULATION AND D ESIGN OF THE transmission required for the minimum data rate.
O PTIMAL T RANSMISSION P OLICY
B. Optimal Transmission Policy
We first formulate the optimization problem for the design
According to Section IV-A, it is clear that, in order to
of transmission policy in Section IV-A. Then we design the
maximize the expected achievable throughput, et and pe
optimal transmission policy in Section IV-B. Finally, we pro-
should be jointly optimized. However, they are tightly inter-
pose a two-layer algorithm to obtain the specific solution in
twined with the active probability Pa (et , pe ), which makes the
Section IV-C.
joint optimization problem intractable. Instead of deriving the
optimal solution of P0 directly, we first investigate the optimal
A. Problem Formulation et with a given value of pe , called the inner optimization
According to Theorem 1, the expected achievable through- problem, and then investigate the optimal pe based on the
put is expressed as obtained optimal et , called the outer optimization problem,
τt as a method to maximize R(et , pe ). Mathematically, since et
R(et , pe ) = (1 − PCS pe )PA G(et ), (25) and pe are independent of each other in the constraints of P0 ,
τ
where G(et ) is given by we rewrite the joint optimization problem P0 as a two-layer
 rt   optimization problem as
et (1 + ld )−α 2ld
G(et ) = log2 1 + dld . (26) max max R(et , pe )
0 τt σ 2 rt2 pe et

In (25), (1 − PCS pe ) represents the transmission probability, s.t. 0 ≤ pe ≤ 1, emin ≤ et . (28)

PA defined in (10) denotes the probability that there exists The inner optimization problem P1 is formulated as
at least one active node in the TZ, and G(et ) denotes the
P1 : max R(et , pe )
average data rate of a sensor node. In (26), we model the et
channels between the sensor nodes and DAP as the Additive s.t. emin ≤ et , (29)

Authorized licensed use limited to: Zhejiang University of Technology. Downloaded on September 10,2023 at 11:56:34 UTC from IEEE Xplore. Restrictions apply.
LIU et al.: THROUGHPUT MAXIMIZATION OF WPCN WITH MOBILE ACCESS POINTS 4409

where we optimize et with given pe . The outer optimization Γ1 is positive, and decreases with et . We deduce that Γ1
problem P2 is formulated as decreases with et . Based on (26), the average data rate G(et )
in Γ2 increases with et . We obtain that Γ2 increases with et .
P2 : max R (pe )
pe By combining the monotonicity analysis of Γ1 and Γ2 with
(et ,pe )
s.t. 0 ≤ pe ≤ 1. (30) respect to et , we deduce that ∂R2∂e t
in (35) decreases
with et .
R (pe ) in (30) is expressed as Then we analyze the monotonicity of R2 (et , pe ) in (34) with
τt  ∗  respect to et . Based on (21), (26), and (38), Ψ(et , pe )|et =0 > 0,
R (pe ) = (1 − PCS pe )PA G et (pe ) , (31)
τ ∂G(et )
∂et |et =0 > 0, and G(et )|et =0 = 0 hold. According to (35),
where e∗t (pe ) is the optimal solution of P1 . In P2 , we optimize (et ,pe )
we deduce Γ1 |et =0 > 0, Γ2 |et =0 = 0, and ∂R2∂e |et =0 > 0.
pe based on e∗t (pe ). t

1) The Optimal Solution of P1 : In this part, we focus on the Moreover, based on (21), (26), and (38), Ψ(et , pe )|et →+∞ < 0,
∂G(et )
analysis of the inner optimization problem P1 . According to ∂et |et →+∞ > 0, and G(et )|et →+∞ → +∞ hold. According
Section III-B, the WPCN is divided into two long-term states. to (35), we deduce Γ1 |et →+∞ < 0, Γ2 |et →+∞ → +∞, and
∂R2 (et ,pe )
Thus we divide the analysis of P1 with respect to the energy- ∂et |et →+∞ < 0.
(et ,pe )
sufficient state and energy-limited state. Based on the monotonicity of ∂R2∂e t
with respect to
(i) Energy-sufficient state: In this state, 0 ≤ et ≤ eth (pe ) et and the values of ∂R2∂e (et ,pe )
| and ∂R2 (et ,pe )
|et →+∞ ,
et =0 ∂et
and Pa (et , pe ) = 1 hold. Based on (10), we obtain PA = 1.
t

we deduce that there exists a solution e0 (pe ) ∈ (0, +∞)

By substituting PA = 1 into (25), R(et , pe ) is reformulated as (et ,pe )
satisfying ∂R2∂e = 0. Moreover, when et < e0 (pe ),
τt ∂R2 (et ,pe )
t
R1 (et , pe ) = (1 − PCS pe ) G(et ). (32) ∂et > 0 holds, and R2 (et , pe ) in (34) increases with
τ (et ,pe )
et . When et > e0 (pe ), ∂R2∂e < 0 holds, and R2 (et , pe )
According to (26), G(et ) in (32) increases with et . Based t
decreases with et .
on the value range of et in the energy-sufficient state, i.e.,
By combining the value range of et in the energy-limited
0 ≤ et ≤ eth (pe ), R1 (et , pe ) in (32) increases with et when
state, i.e., eth (pe ) < et , and the monotonicity analysis of
0 ≤ et ≤ eth (pe ).
R2 (et , pe ) with respect to et , we find that:
(ii) Energy-limited state: In the energy-limited state,
- If eth (pe ) ≥ e0 (pe ), R2 (et , pe ) decreases with et when
eth (pe ) < et and Pa (et , pe ) = Ψ(et , pe ) hold. According to
eth (pe ) < et .
(10), PA is expressed as
- If eth (pe ) < e0 (pe ), R2 (et , pe ) increases with et when
 N PT Z  eth (pe ) < et ≤ e0 (pe ), and decreases with et when et >
PA = 1 − 1 − Ψ(et , pe ) . (33)
e0 (pe ).
For the monotonicity analysis of R(et , pe ) with respect to In summary, with the findings about the monotonicities
et , we make a simplification of PA . By substituting N  = of R1 (et , pe ) and R2 (et , pe ) in the two long-term states,
N PT Z  and x = 1 − Ψ(et , pe ) with 0 < x < 1 into (33), we conclude the monotonicity of R(et , pe ) in (25) with respect
N N  −1 i
we obtain PA = 1−x  = (1−x)
 i=0 x . Due to 0 < x < to et as
N  −1 i
1, we deduce that i=0 x is an increasing function of x, • If eth (pe ) ≥ e0 (pe ), R(et , pe ) increases with et when
 
both 1 − x N
and (1 − x) are decreasing functions of x, and 0 ≤ et ≤ eth (pe ), and decreases with et when eth (pe ) < et .
N  −1 i   • If eth (pe ) < e0 (pe ), R(et , pe ) increases with et when
0 < i=0 x < N  . Thus 1 − xN could be approximated
 0 ≤ et ≤ e0 (pe ), and decreases with et when e0 (pe ) < et .
 n(1 − x), where 1 < n < N holds. Replacing x with
by
Then we have the optimal solution of inner optimization
1 − Ψ(et , pe ) , we obtain PA ≈ nΨ(et , pe ). By substituting
PA ≈ nΨ(et , pe ) into (25), R(et , pe ) is reformulated as problem P1 in (29) by Theorem 2.
Theorem 2: For a given value of energy harvesting prob-
τt
R2 (et , pe ) = (1 − PCS pe )nΨ(et , pe ) G(et ). (34) ability pe in the CS, the optimal energy consumption of one
τ
transmission that maximizes the expected achievable through-
The first-order partial derivative of R2 (et , pe ) with respect to put e∗t (pe ) is expressed as
et is given by
e∗t (pe ) = max{emin , eth (pe ), e0 (pe )}, (39)
∂R2 (et , pe )
= Γ1 − Γ2 . (35) where eth (pe ) in (22) is the boundary energy consumption
∂et
τt ∂G(et ) that distinguishes the energy states of the WPCN, and e0 (pe ),
Γ1 = (1 − PCS pe )nΨ(et , pe ) , (36) (et ,pe )
τ ∂et the solution of equation ∂R2∂e t
= 0, denotes the local
n(1 − PCS pe )τt maximum resulting from the tradeoff between the data rate
Γ2 = G(et ), (37) and active probability.
NP e τ
 rt T Z r Proof: According to the maximum among emin , eth (pe ),
∂G(et ) 2ld (1 + ld )−α
=   dld . (38) and e0 (pe ), we analyze the optimal et with a given value of pe ,
∂et 0 rt2 τt σ 2 + et (1 + ld )−α ln 2
i.e., the optimal solution of P1 , in the following three cases.
Based on (21), Ψ(et , pe ) in Γ1 decreases with et , and the Case 1: When emin ≥ max{eth (pe ), e0 (pe )}, by combining
t)
value range of Ψ(et , pe ) is (0, 1). Based on (38), ∂G(e
∂et in the constraint about et of P1 in (29), i.e., emin ≤ et , and the

Authorized licensed use limited to: Zhejiang University of Technology. Downloaded on September 10,2023 at 11:56:34 UTC from IEEE Xplore. Restrictions apply.
4410 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 22, NO. 7, JULY 2023

highest data rate with Pa (et , pe ) = 1. It is worth noting that

when emin ≥ max{eth (pe ), e0 (pe )}, the WPCN falls into
the energy-limited state, since et ≥ emin ≥ eth (pe ) holds.
Hence the throughput decreases with et when et > emin ,
and the maximum throughput is achieved at emin . Up to
now, we complete the analysis of the inner optimization
problem P1 .
2) The Optimal Solution of P2 : In this part, we focus on the
analysis of the outer optimization problem P2 in (30). Based
on the three potential optimal et in Theorem 2, we divide
the analysis of P2 into the corresponding three cases for the
optimal transmission policy, i.e., (e∗t , p∗e ).
(i) Case 1: When emin ≥ max{eth (pe ), e0 (pe )}, i.e.,
e∗t (pe ) = emin , R (pe ) is reformulated as
τt
R1 (pe ) = (1 − PCS pe )PA (emin , pe ) G(emin ). (40)
τ
Fig. 3. The expected achievable throughput R versus the energy consumption
of one transmission et with different values of the harvested energy eh . The optimal solution of P2 in case 1 is given by
p1e = arg max R1 (pe ). (41)
emin ≥max{eth (pe ),e0 (pe )},0≤pe ≤1
monotonicity of R(et , pe ) with respect to et , we deduce that In case 1, the average data rate G(emin ) is fixed and inde-
R(et , pe ) decreases with et when emin ≤ et . Thus we obtain pendent of pe . The increase of pe leads to the decrease of the
e∗t (pe ) = emin . Case 2: When eth (pe ) ≥ max{emin , e0 (pe )}, transmission probability (i.e., (1−PCS pe )), and the increase of
similar to the analysis in case 1, we deduce that R(et , pe ) the active probability (i.e., Pa (et , pe )), which further increases
increases with et when emin ≤ et ≤ eth (pe ), and decreases PA . Therefore, the optimal solution p1e denotes the local
with et when et > eth (pe ). Thus we obtain e∗t (pe ) = eth (pe ). maximum resulting from the tradeoff between the transmission
Case 3: When e0 (pe ) ≥ max{emin , eth (pe )}, similar to the probability and active probability under the constraints emin ≥
analysis in case 1, we deduce that R(et , pe ) increases with max{eth (pe ), e0 (pe )} and 0 ≤ pe ≤ 1.
et when emin ≤ et ≤ e0 (pe ), and decreases with et when (ii) Case 2: When eth (pe ) ≥ max{emin , e0 (pe )}, i.e.,
et > e0 (pe ). Thus we obtain e∗t (pe ) = e0 (pe ). By combining  
the optimal solution e∗t (pe ) in the aforementioned three cases, e∗t (pe ) = eth (pe ), Pa eth (pe ), pe = 1 and PA = 1 hold.
the proof is completed. Hence R (pe ) is reformulated as
τt  
In order to clarify the scenarios where the three values R2 (pe ) = (1 − PCS pe ) G eth (pe ) . (42)
of emin /eth (pe )/e0 (pe ) are the optimal energy consumption, τ
respectively, we employ Fig. 3 as a supplementary illustration. The optimal solution of P2 in case 2 is given by
We observe that, when eh = 0.035mJ, the WPCN always falls
into the energy-limited state, and then the tradeoff between p2e = arg max R2 (pe ). (43)
eth (pe )≥max{emin ,e0 (pe )},0≤pe ≤1
the data rate and active probability is the critical factor that
affects the throughput. Hence the maximal throughput is In case 2, PA is fixed with PA = 1 and independent of
achieved when the tradeoff between the data rate and active pe . The increase of pe leads to the decrease of the trans-
probability is optimal, i.e., e∗t (pe ) = e0 (pe ). With the increase mission probability (i.e., (1 − PCS pe )), and the increase of
of eh , e.g., eh = 0.04mJ, the WPCN falls into the energy- the boundary energy  consumption
 eth (pe ) as well as the
sufficient state for a small value of et (i.e., et < eth (pe )), average data rate G eth (pe ) . Therefore, the optimal solution
and changes into the energy-limited state for a large value p2e denotes the local maximum resulting from the tradeoff
of et (i.e., et > eth (pe )). When et < eth (pe ), the data rate between the transmission probability and average data rate
is the critical factor that limits the throughput, but there is under the constraints eth (pe ) ≥ max{emin , e0 (pe )} and
superfluous energy not fully utilized. Thus increasing data 0 ≤ pe ≤ 1.
rate by increasing et is a straightforward manner to improve (iii) Case 3: When e0 (pe ) ≥ max{emin , eth (pe )}, i.e.,
the throughput. When et > eth (pe ), the tradeoff between e∗t (pe ) = e0 (pe ), R (pe ) is rewritten as
the data rate and active probability is still the critical factor   τt  
that affects the throughput, which means e∗t (pe ) = e0 (pe ). R3 (pe ) = (1 − PCS pe )PA e0 (pe ), pe G e0 (pe ) . (44)
τ
Nevertheless, when eh = 0.045mJ, the throughput could not The optimal solution of P2 in case 3 is given by
increase with et for et > eth (pe ), since the critical factor that
limits the throughput becomes the active probability instead p3e = arg max R3 (pe ). (45)
e0 (pe )≥max{emin ,eth (pe )},0≤pe ≤1
of the tradeoff between the data rate and active probability,
and the throughput is reduced arising from the low active In case 3, the increase of pe means the increase of the har-
probability. Therefore, the maximum throughput is achieved vested energy, the effect of which is equivalent to the increase
at the boundary energy consumption eth (pe ), which provides of eh in Fig. 3. Hence the increase of pe leads to the increase

Authorized licensed use limited to: Zhejiang University of Technology. Downloaded on September 10,2023 at 11:56:34 UTC from IEEE Xplore. Restrictions apply.
LIU et al.: THROUGHPUT MAXIMIZATION OF WPCN WITH MOBILE ACCESS POINTS 4411

of e0 (pe ), which represents the optimal tradeoff between the TABLE II

data rate and active probability. On the other hand, the increase S IMULATION PARAMETERS
of pe results in the decrease of transmission probability and
the increase of the active probability. Therefore, the optimal
solution p3e in (45) denotes the local maximum resulting
from the tradeoffs among the transmission probability, active
probability, and data rate under the constraints e0 (pe ) ≥
max{emin , eth (pe )} and 0 ≤ pe ≤ 1.
In summary, the optimal solution of the joint optimization
problem P0 in (27) is expressed as
(e∗t , p∗e ) = arg max R(et , pe ), (46)
(et ,pe )∈C

where R(et , pe ) is defined as (25). C, a set of (et , pe ), includes

three potential optimal pairings of (et , pe ), and is given by
     
C = emin , p1e , eth (p2e ), p2e , e0 (p3e ), p3e . (47)

However, it is hard to obtain the value range of pe in each

case, hence it is challenging to derive the optimal solution of
the joint optimization problem P0 . To this end, we propose a
two-layer algorithm to obtain the specific solution of P0 .
solution (e∗t , p∗e ) in lines 5-7 is also completed within the
Algorithm 1 Two-layer algorithm for the specific solution of complexity
 1 O(1).
 Thus the complexity of Algorithm 1 is
the joint optimization problem P0 O  Δp e
 + 1 .
Input: emin , fpdf (red ), rt , re , and other network parameters. Last but not least, we analyze the gap between the specific
Output: The specific solution (e∗t , p∗e ) of the joint optimiza- solution (e∗t , p∗e ) obtained by Algorithm 1 and the theoretical
tion problem P0 . optimal solution (e∗t , p∗e ). In line 4, with a given value of
1: Rmax = 0; e∗ ∗
t = 0; pe = 0.
pe , the solution e∗t (pe ) of P1 is obtained by Theorem 2, and
2: for pe = 0; pe ≤ 1; pe = pe + Δpe do
  e∗t (pe ) is indeed the optimal solution of P1 . As a result, the
3: Obtain the numerical solution e0 (pe ) of equation gap between (e∗t , p∗e ) and (e∗t , p∗e ) is only caused by the gap
∂R2 (et ,pe )
= 0. between p∗e and p∗e , which is determined by the search unit
Δpe . Therefore, a smaller value of Δpe makes (e∗t , p∗e ) closer
∂et
4: e∗ (p
t 

e ) = max{e  
 min , eth (pe ), e0 (pe )}.
∗  
5: if R et (pe ), pe > Rmax then to (e∗t , p∗e ), but leads to a larger size of search space, resulting
  in a longer running time.
6: Rmax = R e∗t (pe ), pe ; e∗t = e∗t (pe ); p∗e = pe .
7: end if
8: end for V. N UMERICAL R ESULTS AND D ISCUSSIONS
In this section, we evaluate the active probability and the
C. Two-Layer Algorithm expected achievable throughput versus the energy consumption
of one transmission et and the energy harvesting probability
The basic idea of the proposed two-layer algorithm is pe in the CS. Meanwhile, we discuss the important factors,
to address the outer optimization problem P2 in (30) with i.e., the number of sensor nodes N , the transmission power
the one-dimensional search based on the theoretical solu- of the EAP PE , the minimum energy consumption of one
tion of the inner optimization problem P1 . In Algorithm 1, transmission emin , and the radius of the TZ rt , in the design of
we divide the feasible range of pe into equal intervals the optimal transmission policy (e∗t , p∗e ). Finally, we evaluate
in units of Δpe , i.e., pe ∈ {0, . . . , jΔpe , . . . , 1}, where the efficiency of Algorithm 1. To clearly present the numerical
1
j ∈ {0, 1, 2, . . . ,  Δp e
}. We search the solution of P2 in results, we specify the i.i.d. mobility model as the mobile
the search space {0, . . . , jΔpe , . . . , 1}. In line 4, based on models of both the EAP and DAP. According to [15] and [27],
Theorem 2, we obtain the solution e∗t (pe ) of P1 for pe = pe . the network parameters are set as shown in Table II unless
In lines 5-7, the update of solution is performed. otherwise specified.
Then we analyze the complexity of Algorithm 1, where
the optimization problem P0 is divided into two layers.
To address P2 , we adopt the one-dimensional search with the A. Energy States of the WPCN
1

size of the search space is  Δp  + 1 , thus the complexity Fig. 4 describes the active probability Pa (et , pe ) versus the
 1 2 e
is O  Δpe  + 1 . In line 4, based on Theorem 2, P1 energy consumption of one transmission et and the energy
could be solved within the complexity O(1). The update of harvesting probability pe . We observe that, under the parame-


ter settings, Pa (et , pe ) equals 1 (i.e., energy-sufficient state)
2 f (n) f (n)
= O g(n) means limn→∞ sup g(n) < ∞. for a small value of et , and is independent of pe , since the

Authorized licensed use limited to: Zhejiang University of Technology. Downloaded on September 10,2023 at 11:56:34 UTC from IEEE Xplore. Restrictions apply.
4412 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 22, NO. 7, JULY 2023

Fig. 6. The optimal transmission policy (e∗t , p∗e ) versus the number of sensor
nodes N .

Fig. 4. The active probability Pa (et , pe ) versus the energy consumption of expected achievable throughput first increases with et , and
one transmission et and the energy harvesting probability pe .
then decreases with it. This is due to the reason that, when et is
small, the key factor that limits the throughput is the data rate,
which increases with et . When et is large, the key limiting fac-
tor becomes the active probability Pa (et , pe ), which decreases
with et as shown in Fig. 4. In order to clearly illustrate the
impact of pe on the throughput, we give the illustration in three
cases with respect to the value of et . For a small value of et ,
we observe that the throughput decreases with pe , since the
WPCN always falls into the energy-sufficient state, and the key
factor that limits the throughput is the transmission probability
(1 − PCS pe ), which decreases with pe . For a large value of
et , the throughput increases with pe , since the WPCN always
falls into the energy-limited state, and the key limiting factor
becomes the active probability Pa (et , pe ), which increases
with pe . For an intermediate value of et , the throughput first
increases with pe , and then decreases with it. The reason is
that, the WPCN turns from the energy-limited state to the
energy-sufficient state with the increase of pe , and accordingly
the key limiting factor changes from the active probability
Fig. 5. The expected achievable throughput R versus the energy consumption into the transmission probability. Moreover, we observe that
of one transmission et and the energy harvesting probability pe .
the optimal (e∗t , p∗e ) is obtained on the boundary between the
two energy
 states, and  corresponds to the optimal transmission
policy eth (p2e ), p2e .
harvested energy in the SS and OS is sufficient for the energy
consumption of data transmissions. However, for a given
value of pe , when et is larger than the critical value eth (pe ), B. Impacts of Key Network Parameters on the Optimal
Pa (et , pe ) is smaller than 1, and decreases with et . We notice Transmission Policy
that the critical value eth (pe ) increases with pe , since a larger Fig. 6 shows the optimal transmission policy (e∗t , p∗e ) versus
value of pe means more harvested energy, which could support the number of sensor nodes N . It is easy to observe that, e∗t
higher data rate on the premise of Pa (et , pe ) = 1. Moreover, increases with N , while p∗e decreases with N . The reason is
it is observed that, when Pa (et , pe ) is smaller than 1 (i.e., that, on the one hand, the increase of N leads to the increase
energy-limited state), Pa (et , pe ) increases with pe for a given of the number of sensor nodes in the EZ. This means that
value of et , resulting from that more energy could be harvested more sensor nodes harvest energy during one time slot, and
in the CS so as to alleviate the energy shortage of the WPCN. the harvested energy of the WPCN increases, while the energy
On the whole, the WPCN turns from the energy-sufficient state consumption of the WPCN during one time slot is fixed. Hence
to the energy-limited state with the increase of et and the the active probability increases, resulting in the increase of
decrease of pe . the probability that there exists at least one active node in the
Fig. 5 depicts the expected achievable throughput R versus TZ, i.e., PA . On the other hand, the increase of N leads to
the energy consumption of one transmission et and the energy the increase of the number of sensor nodes in the TZ, which
harvesting probability pe . With a given value of pe , the also brings about the increase of PA . Overall, the increase of

Authorized licensed use limited to: Zhejiang University of Technology. Downloaded on September 10,2023 at 11:56:34 UTC from IEEE Xplore. Restrictions apply.
LIU et al.: THROUGHPUT MAXIMIZATION OF WPCN WITH MOBILE ACCESS POINTS 4413

Fig. 7. The optimal transmission policy (e∗t , p∗e ) versus the transmission
power of the EAP PE . Fig. 8. The optimal transmission policy (e∗t , p∗e ) versus the minimum energy
consumption of one transmission emin .

N mitigates the energy causality constraint and the limitation

resulting from the small number of sensor nodes in the TZ.
Therefore, as the number of sensor nodes N increases, e∗t
becomes larger, and p∗e becomes smaller.
Fig. 7 shows the optimal transmission policy (e∗t , p∗e ) versus
the transmission power of the EAP PE . We observe that,
e∗t increases with PE , and p∗e nearly decreases with PE .
The reason is that, the increase of PE is equivalent to the
increase of the average harvested energy of the WPCN. As the
average harvested energy becomes larger, e∗t becomes larger,
and p∗e becomes smaller. We notice that p∗e equals 1 for
PE ≤ 0.72mW, and p∗e equals 0 for PE ≥ 1.02mW. This is
due to the reason that, the WPCN suffers from a severe energy
shortage for PE ≤ 0.72mW. Therefore, when the WPCN is
in the CS, it is the best to perform energy harvesting (i.e.,
p∗e = 1). For the large value of PE with PE ≥ 1.02mW, the
harvested energy of the WPCN is superfluous, and the WPCN Fig. 9. The optimal transmission policy (e∗t , p∗e ) versus the radius of
suffers from a serious shortage of transmission opportunities. the TZ rt .
Therefore, when the WPCN is in the CS, it is the best to
perform data transmission (i.e., p∗e = 0). Note that similar as
PE , the increase of the area of the EZ also makes the harvested Fig. 9 shows the optimal transmission policy (e∗t , p∗e ) versus
energy of the WPCN become larger. Therefore, the trend of the radius of the TZ rt . We first observe that e∗t decreases with
the impact of the area of the EZ on the optimal transmission rt , and p∗e equals 0 for rt ≤ 3.61. The reason is that, a small
policy could also be described by Fig. 7. value of rt implies a small number of sensor nodes in the TZ as
Fig. 8 shows the optimal transmission policy (e∗t , p∗e ) ver- well as a small probability that there exists at least one active
sus the minimum energy consumption of one transmission node in the TZ (i.e., PA ), which results in a serious shortage of
emin . We observe that, e∗t and p∗e remain unchanged for transmission opportunities for the WPCN, and then limits the
emin ≤ 0.215mJ, and increase with emin for emin > throughput. Therefore, when the WPCN is in the CS, it is the
0.215mJ. This is due to the reason that, for a small
 value of best to perform data transmission without doubt (i.e., p∗e = 0).
emin , the3 maximum
 throughput is achieved at eth (p2e ), p2e We know that the increase of rt means the increase of the
or e0 (pe ), p3e , since the small value of emin would not number of sensor nodes in the TZ. As the number of sensor
be the maximum among emin , eth (pe ), and e0 (pe ) based nodes in the TZ increases, e∗t and p∗e become smaller, since
on Theorem 2. Namely, the minimum energy consumption the WPCN requires more transmission opportunities. Yet, the
constraint, et ≥ emin , would not affect the design of the minimum value of p∗e is 0, hence p∗e could not become smaller,
optimal transmission policy (e∗t , p∗e ). However, a large value and p∗e keeps the value as 0. In addition, we observe that both
of emin , i.e., emin ≥ max{eth (pe ), e0 (pe )}, means a high e∗t and p∗e increase with rt for rt > 3.61. This is due to the
requirement for the minimum data rate, which has a vital reason that, the large value of rt greatly alleviates the shortage
impact on the design of the optimal transmission policy, of transmission opportunities, and the key factor that limits
and then the maximum throughput is achieved at (emin , p1e ). the throughput becomes the data rate. Consequently, we see
Certainly, a large value of emin results in high demands on that with the increase of rt , e∗t and p∗e become larger, since
the energy supply, hence both e∗t and p∗e become larger with the WPCN requires more harvested energy to support higher
the increase of emin for emin > 0.215mJ. data rate.

Authorized licensed use limited to: Zhejiang University of Technology. Downloaded on September 10,2023 at 11:56:34 UTC from IEEE Xplore. Restrictions apply.
4414 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 22, NO. 7, JULY 2023

the energy causality constraint, which makes the through-

put maximization problem intractable. Instead of addressing
the problem directly, we divided the problem into an inner
optimization problem and an outer optimization problem,
and further proposed a two-layer algorithm to obtain the
specific solution. Numerical results showed that the optimal
pairing of the energy consumption of one transmission and
the energy harvesting probability become higher and smaller
as the increase of the area of the EZ, which is equivalent to
the transmission power of the EAP. While they first become
lower and smaller, and then become higher and larger as the
increase of the area of the TZ. Future works include the joint
optimization of the mobility model of the EAP/DAP along
with the energy consumption of one transmission and the
Fig. 10. The running time of Algorithm 1 and the expected achievable energy harvesting probability.
throughput R obtained by Algorithm 1 versus the logarithm of the one-
dimensional search unit log1/2 (Δpe ).
R EFERENCES

C. Efficiency of the Two-Layer Algorithm [1] J. Liu, C. Zhang, and Y. Fang, “EPIC: A differential privacy framework
to defend smart Homes against internet traffic analysis,” IEEE Internet
Fig. 10 shows the expected achievable throughput R Things J., vol. 5, no. 2, pp. 1206–1217, Apr. 2018.
[2] K. Zheng et al., “Secrecy capacity scaling of large-scale networks
obtained by Algorithm 1 and the running time of Algorithm 1 with social relationships,” IEEE Trans. Veh. Technol., vol. 66, no. 3,
versus the logarithm of the one-dimensional search unit pp. 2688–2702, Mar. 2017.
log1/2 (Δpe ). As shown in Fig. 10, R increases with [3] J. Fei, K.-W. Chin, C. Yang, and M. Ros, “Charge-and-activate policies
for targets monitoring in RF-harvesting sensor networks,” IEEE Trans.
log1/2 (Δpe ) at the cost of the running time. It is easy to under- Veh. Technol., vol. 69, no. 7, pp. 7835–7846, Jul. 2020.
stand that a smaller search unit leads to the higher throughput, [4] M.-L. Ku, W. Li, Y. Chen, and K. J. R. Liu, “Advances in energy
since the obtained optimal throughput by Algorithm 1 is more harvesting communications: Past, present, and future challenges,” IEEE
precise and closer to the optimal solution. This observation Commun. Surveys Tuts., vol. 18, no. 2, pp. 1384–1412, 2nd Quart. 2016.
[5] A. Biason and M. Zorzi, “Battery-powered devices in WPCNs,” IEEE
validates the analysis of Algorithm 1 in Section IV-C. We also Trans. Commun., vol. 65, no. 1, pp. 216–229, Jan. 2017.
observe that, the expected achievable throughput remains [6] K. Chi, Y. Zhu, Y. Li, L. Huang, and M. Xia, “Minimization of trans-
unchanged for log1/2 (Δpe ) ≥ 6, since when log1/2 (Δpe ) = mission completion time in wireless powered communication networks,”
IEEE Internet Things J., vol. 4, no. 5, pp. 1671–1683, Oct. 2017.
6, the throughput obtained by Algorithm 1 is almost the [7] N. Xie, H. Tan, L. Huang, and A. X. Liu, “Physical-layer authentication
optimal solution. Therefore, in order to avoid long running in wirelessly powered communication networks,” IEEE/ACM Trans.
time of Algorithm 1, it is appropriate to set Δpe as the value Netw., vol. 29, no. 4, pp. 1827–1840, Aug. 2021.
[8] S. Hu, X. Chen, W. Ni, X. Wang, and E. Hossain, “Modeling and analy-
close to 216 . sis of energy harvesting and smart grid-powered wireless communication
networks: A contemporary survey,” IEEE Trans. Green Commun. Netw.,
vol. 4, no. 2, pp. 461–496, Jun. 2020.
VI. C ONCLUSION [9] X. Lu, P. Wang, D. Niyato, D. I. Kim, and Z. Han, “Wireless networks
In this paper, we studied a mobile non-HAP-based WPCN with RF energy harvesting: A contemporary survey,” IEEE Commun.
Surveys Tuts., vol. 17, no. 2, pp. 757–789, 2nd Quart., 2015.
which could mitigate the double near-far effect. In the WPCN, [10] X. Zhou, C. K. Ho, and R. Zhang, “Wireless power meets energy
sensor nodes harvest energy from the RF signal of the mobile harvesting: A joint energy allocation approach in OFDM-based sys-
EAP, and have opportunities to transmit data to the mobile tem,” IEEE Trans. Wireless Commun., vol. 15, no. 5, pp. 3481–3491,
May 2016.
DAP when their energy is larger than a threshold determined [11] G. Liu, Z. Sun, and T. Jiang, “Joint time and energy allocation for
by the energy consumption of one transmission. Based on the QoS-aware throughput maximization in MIMO-based wireless powered
distance between the EAP and DAP, we divided the WPCN underground sensor networks,” IEEE Trans. Commun., vol. 67, no. 2,
pp. 1400–1412, Feb. 2019.
into three typical scenarios including the SS, OS, and CS. [12] S. Mao, S. Leng, J. Hu, and K. Yang, “Energy-efficient transmission
In particular, when the WPCN is in the CS, the EAP and DAP schemes for cooperative wireless powered cellular networks,” IEEE
could not work simultaneously, and we employed the energy Trans. Green Commun. Netw., vol. 3, no. 2, pp. 494–504, Jun. 2019.
[13] B. Lyu, P. Ramezani, D. T. Hoang, S. Gong, Z. Yang, and A. Jamalipour,
harvesting probability to denote the probability that the EAP “Optimized energy and information relaying in self-sustainable IRS-
works in the CS. empowered WPCN,” IEEE Trans. Commun., vol. 69, no. 1, pp. 619–633,
To reach the goal of maximizing the throughput of the Jan. 2021.
[14] J. Chen, Z. Zhang, Y.-C. Liang, X. Kang, and R. Zhang, “Resource
WPCN through identifying the optimal pairing of the energy allocation for wireless-powered IoT networks with short packet commu-
consumption of one transmission and the energy harvesting nication,” IEEE Trans. Wireless Commun., vol. 18, no. 2, pp. 1447–1461,
probability, we first formulated the energy causality constraint, Feb. 2019.
[15] K. Zheng, X. Liu, B. Wang, H. Zheng, K. Chi, and Y. Yao, “Through-
and divided the WPCN into an energy-sufficient state and put maximization of wireless-powered communication networks: An
an energy-limited state by the energy consumption of one energy threshold approach,” IEEE Trans. Veh. Technol., vol. 70, no. 2,
transmission and the energy harvesting probability. We found pp. 1292–1306, Feb. 2021.
[16] T. A. Zewde and M. C. Gursoy, “NOMA-based energy-efficient wireless
that the energy consumption of one transmission and the powered communications,” IEEE Trans. Green Commun. Netw., vol. 2,
energy harvesting probability are jointly intertwined with no. 3, pp. 679–692, Sep. 2018.

Authorized licensed use limited to: Zhejiang University of Technology. Downloaded on September 10,2023 at 11:56:34 UTC from IEEE Xplore. Restrictions apply.
LIU et al.: THROUGHPUT MAXIMIZATION OF WPCN WITH MOBILE ACCESS POINTS 4415

[17] M. A. Abd-Elmagid, A. Biason, T. ElBatt, K. G. Seddik, and Xiaoying Liu (Member, IEEE) received the B.E.
M. Zorzi, “Non-orthogonal multiple access schemes in wireless powered degree in electronic engineering from the Nanjing
communication networks,” in Proc. IEEE Int. Conf. Commun. (ICC), University of Science and Technology, Nanjing,
Paris, France, May 2017, pp. 1–6. China, in 2013, and the Ph.D. degree in electronic
[18] M. M. N. Aboelwafa, M. A. Abd-Elmagid, A. Biason, K. G. Seddik, engineering from Shanghai Jiao Tong University,
T. ElBatt, and M. Zorzi, “Towards optimal resource allocation in wireless Shanghai, China, in 2018. She is currently an Asso-
powered communication networks with non-orthogonal multiple access,” ciate Professor with the School of Computer Science
Ad Hoc Netw., vol. 85, pp. 1–10, Mar. 2019. and Technology, Zhejiang University of Technology,
[19] Q. Li, J. Gao, H. Liang, L. Zhao, and X. Tang, “Optimal power Hangzhou, China. Her research interests include
allocation for wireless sensor powered by dedicated RF energy wireless communications, cognitive radio networks,
source,” IEEE Trans. Veh. Technol., vol. 68, no. 3, pp. 2791–2801, and energy harvesting networks. She was a recipient
Mar. 2019. (as the corresponding author) of the Best Paper Award from the International
[20] Q. Wu, G. Zhang, D. W. K. Ng, W. Chen, and R. Schober, “Generalized Conference on Networking and Network Applications in 2021.
wireless-powered communications: When to activate wireless power
transfer?” IEEE Trans. Veh. Technol., vol. 68, no. 8, pp. 8243–8248, Bin Xu received the B.S. degree in computer science
Aug. 2019. and technology from the Zhejiang University of
[21] K. Liang, L. Zhao, G. Zheng, and H.-H. Chen, “Non-uniform deploy- Technology, Hangzhou, China, in 2020, where he
ment of power beacons in wireless powered communication net- is currently pursuing the M.S. degree in computer
works,” IEEE Trans. Wireless Commun., vol. 18, no. 3, pp. 1887–1899, science and technology. His current research inter-
Mar. 2019. ests include cognitive radio networks and energy
[22] H. Tang, Q. Wu, W. Chen, J. Wang, and B. Li, “Mitigating the doubly harvesting networks.
near–far effect in UAV-enabled WPCN,” IEEE Trans. Veh. Technol.,
vol. 70, no. 8, pp. 8349–8354, Aug. 2021.
[23] Y. Liu, X. Chen, L. X. Cai, Q. Chen, and R. Zhang, “Nonorthogonal
multiple access for wireless-powered IoT networks,” IEEE Internet
Things J., vol. 8, no. 1, pp. 112–128, Jan. 2021.
[24] Y. Zhang, Z. Xiong, D. Niyato, P. Wang, and D. I. Kim, “Toward a Kechen Zheng (Member, IEEE) received the B.E.
perpetual IoT system: Wireless power management policy with threshold and Ph.D. degrees in electronic engineering from
structure,” IEEE Internet Things J., vol. 5, no. 6, pp. 5254–5270, Shanghai Jiao Tong University, Shanghai, China,
Dec. 2018. in 2013 and 2018, respectively. He is currently an
[25] K. H. Li and K. C. Teh, “Optimal spectrum access and energy sup- Associate Professor with the School of Computer
ply for cognitive radio systems with opportunistic RF energy har- Science and Technology, Zhejiang University of
vesting,” IEEE Trans. Veh. Technol., vol. 66, no. 8, pp. 7114–7122, Technology, Hangzhou, China. He has published
Aug. 2017. more than 30 technical papers in journals and confer-
[26] S. Chatterjee, S. P. Maity, and T. Acharya, “Energy-spectrum efficiency ences, including IEEE T RANSACTIONS ON M OBILE
trade-off in energy harvesting cooperative cognitive radio networks,” C OMPUTING, IEEE T RANSACTIONS ON W IRELESS
IEEE Trans. Cognit. Commun. Netw., vol. 5, no. 2, pp. 295–303, C OMMUNICATIONS, and IEEE T RANSACTIONS ON
Jun. 2019. C OMMUNICATIONS. His research interests include the IoT, performance
[27] D. Xu and H. Zhu, “Sum-rate maximization of wireless powered primary evaluation in cognitive networks and social networks, and energy harvesting
users for cooperative CRNs: NOMA or TDMA at cognitive users?” wireless communication networks. He was a recipient of the Best Paper Award
IEEE Trans. Commun., vol. 69, no. 7, pp. 4862–4876, Jul. 2021. from the International Conference on Networking and Network Applications
[28] P. N. Alevizos and A. Bletsas, “Sensitive and nonlinear far-field RF in 2021.
energy harvesting in wireless communications,” IEEE Trans. Wireless
Commun., vol. 17, no. 6, pp. 3670–3685, Jun. 2018. Haifeng Zheng (Senior Member, IEEE) received
[29] Q. Gu, G. Wang, R. Fan, N. Zhang, H. Jiang, and Z. Zhong, “Optimal the B.Eng. and M.S. degrees in communication
resource allocation in wireless powered relay networks with nonlin- engineering from Fuzhou University, Fuzhou, China,
ear energy harvesters,” IEEE Wireless Commun. Lett., vol. 9, no. 3, and the Ph.D. degree in communication and infor-
pp. 371–375, Mar. 2020. mation system from Shanghai Jiao Tong University,
[30] X. Liu, K. Zheng, K. Chi, and Y.-H. Zhu, “Cooperative spectrum sens- Shanghai, China. He is currently a Professor with
ing optimization in energy-harvesting cognitive radio networks,” IEEE the College of Physics and Information Engineering,
Trans. Wireless Commun., vol. 19, no. 11, pp. 7663–7676, Nov. 2020. Fuzhou University, China. He was a Visiting Scholar
[31] H. Inaltekin, M. Chiang, H. V. Poor, and S. B. Wicker, “On unbounded at the State University of New York at Buffalo
path-loss models: Effects of singularity on wireless network perfor- from October 2015 to September 2016. His research
mance,” IEEE J. Sel. Areas Commun., vol. 27, no. 7, pp. 1078–1092, interests include tensor theory, machine learning, the
Sep. 2009. Internet of Things, and wireless networks.

Authorized licensed use limited to: Zhejiang University of Technology. Downloaded on September 10,2023 at 11:56:34 UTC from IEEE Xplore. Restrictions apply.