SPECIAL SECTION ON EXPLOITING THE BENEFITS OF INTERFERENCE IN WIRELESS

NETWORKS: ENERGY HARVESTING AND SECURITY

Received January 18, 2017, accepted February 17, 2017, date of publication March 2, 2017, date of current version June 7, 2017.
Digital Object Identifier 10.1109/ACCESS.2017.2677521

Wireless Powered Communication Networks

Assisted by Backscatter Communication
BIN LYU, ZHEN YANG, GUAN GUI, (Member, IEEE), AND YOUHONG FENG
Key Laboratory of Ministry of Education in Broadband Wireless Communication and Sensor Network Technology, Nanjing University of Posts and
Telecommunications, Nanjing 21003, China
Corresponding author: Zhen Yang (yangz@njupt.edu.cn)
This work was supported in part by the National Natural Science Foundation of China under Grant 61671252, Grant 61471202, and
Grant 61401069, in part by the Key University of Science Research Project of Jiangsu Province under Grant 14KJA510003, and in part by
the Innovation Program of Graduate Education of Jiangsu Province under Grant KYLX15-0828.

ABSTRACT This paper proposes a backscatter-assisted wireless powered communication network that
includes a hybrid access point and multiple users. In conventional wireless powered communication networks
with only harvest-then-transmit (HTT) mode, urgent data transmission is not possible since users need
to first harvest sufficient energy before transmitting information. Backscatter communication depends on
instantaneous excitation energy such that the dedicated time for harvesting energy first is not required.
To improve the system performance, both HTT and backscatter modes are employed at the users in the
proposed model. An optimization problem is formulated to maximize the sum-throughput by finding the
optimal transmission policy, including the optimal users’ working mode permutation and time allocation.
Simulation results demonstrate the superiority of the proposed model.

INDEX TERMS Wireless powered communication network, harvest-then-transmit mode, backscatter

communication, resource allocation.

I. INTRODUCTION multiple users with rechargeable batteries or supercapacitors.

With the development of 5G communication systems,
a sustainable energy supply for energy-constrained elec-
tronic equipment has been receiving increasingly more
attention. To extend the lifetime of devices, e.g., sen-
sors, radio frequency (RF)-based energy harvesting (EH)
techniques [1], [2] have been introduced.
In wireless networks, the studies focusing on RF-based
EH systems have two main directions, one of which is to
study simultaneous wireless information and power trans-
fer (SWIPT) systems [3]–[6]. In [3], the authors studied
SWIPT in a multiple-input multiple-output (MIMO) wireless
broadcasting system, and the tradeoff between information
transmission and energy transfer was analyzed. In [4], SWIPT
was introduced in a noise-limited broadband system. Orthog-
onal frequency division multiplexing (OFDM) was employed
to partition the broadband channel into sub-channels to trans-
mit energy and information. In [5], [6], novel SWIPT schemes
in interference alignment networks were proposed.
The other direction is to study wireless powered com-
munication networks (WPCNs). WPCNs have recently been
extensively studied [7]–[14]. In a WPCN, there is typically
a hybrid access point (HAP) with fixed energy sources and
To make a WPCN work, a well-known mode named the
‘‘harvest-then-transmit (HHT)’’ mode was introduced [8]. In
this mode, the HAP first radiated energy to users in the down-
link and then received information transmitted by the users
based on the harvested energy. The users were scheduled
in time division multiple access (TDMA), and the optimal
time allocation policy was studied. To further improve the
system performance, multiple antennas were employed at
the HAP in [9]. In the downlink, energy beamforming was
designed to transfer energy to all users, whereas in the uplink,
all users simultaneously transmitted information to the HAP
in space division multiple access (SDMA), which provided
higher spectrum efficiency. In [10], wireless power transfer
(WPT) based on emerging massive MIMO with estimated
channel state information (CSI) was considered. To improve
the time efficiency, the uplink channel estimation and WPT
were operated simultaneously in [11]. In [12], a large-scale
WPCN modeled using a stochastic geometry was considered.
The frame partition and transmit power were jointly opti-
mized to maximize the nodes’ spatial throughput for both
cases of battery-free and battery-deployed wireless nodes.
However, only the half-duplex (HD) mode was adopted in

2169-3536
2017 IEEE. Translations and content mining are permitted for academic research only.
7254 Personal use is also permitted, but republication/redistribution requires IEEE permission. VOLUME 5, 2017
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
B. Lyu et al.: WPCNs Assisted by Backscatter Communication
[8]–[12]. To fully exploit time, the full-duplex (FD) mode
was introduced in [13] since the HAP had two antennas.
The power and time were jointly optimized for maximizing
the system throughput for considering non-causal energy
scenarios. In sensor networks, the harvested energy is typi-
cally stored in supercapacitors, which suffer from high self-
discharge [15]. Hence, during the given block, the users can
only use the harvested energy before its transmission slot
to transmit information. Considering the energy causality, a
more practical FD-WPCN was proposed in [14]. For the HTT
mode, users need to harvest energy first and then transmit
information. Hence, urgent data transmission for users, par-
ticularly for users that are far from the HAP, is not possible
because a longer time for harvesting energy is required [16].
There has recently been an emerging interest in backscat-
ter communication. In a backscatter communication system,
the users, also known as tags, receive instantaneous exci-
tation energy radiated by the HAP (known as reader) and
reflect modulated signals to the HAP simultaneously [17].
Hence, the time for harvesting energy before transmission
is not necessary. In [18], the authors first estimated the
backscatter channels, and then studied energy transfer in
the downlink based on the estimated CSI for multi-antenna
backscatter communication systems. The physical layer secu-
rity in MIMO backscatter systems was studied in [19].
In [17]–[19], the study focused on traditional RFID-style
backscatter systems, where a special-purpose power infras-
tructure, i.e., the reader, was required. In [20], ambient
backscatter was proposed. In the ambient backscatter com-
munication systems, the user harvested energy from envi-
ronmental sources, e.g., TV tower, rather than from specific
energy sources. In [21], the authors first introduced ambient
backscatter in RF-powered overlay cognitive radio networks.
The tradeoff between HTT and backscatter modes was ana-
lyzed, and the optimal transmission policy was derived. How-
ever, ambient backscatter is limited by the transmission range.
To overcome this drawback, a hybrid backscatter commu-
nication scheme that includes both ambient backscatter and
bistatic scatter radio was investigated [16]. Extending to [21],
novel RF-power underlay cognitive radio networks with
ambient backscatter were proposed, where the secondary
user could work in either HTT or backscatter mode [22].
Optimization problems to maximize the throughput for a
single secondary user by choosing the best mode were stud-
ied. However, multiple users are typically deployed in sensor
networks.
Motivated by utilizing the advantages of both [14]
and [22], in this paper, we propose a backscatter-assisted
WPCN (BAWPCN) that includes an HAP and multiple users,
where both HTT and backscatter modes are employed at the
users. To the best of our knowledge, no work has been con-
ducted on the proposed multi-user model. In the downlink,
the HAP broadcasts energy to all users continuously, whereas
each user transmits or backscatters information to the HAP in
the uplink via TDMA. As stated in [23], each user cannot
work in both HTT and backscatter modes simultaneously.
VOLUME 5, 2017
Hence, in contrast to [14], the tradeoff between HTT and
backscatter modes for each user needs to be studied during its
allocated time slot, which leads to a challenging problem of
finding the optimal working mode permutation for multiple
users. Moreover, to maximize the system throughput, the
dynamic time allocation policy is further studied.
The key contributions of this paper are summarized as
follows:
• We propose a new model that employs both HTT and
backscatter modes for a multi-user RF-powered wire-
less sensor network. We formulate one fundamental
optimization problem to maximize the sum-throughput
by studying the optimal transmission policy, including
both the optimal working mode permutation and time
allocation.
• The optimal transmission policy is first obtained for the
two-user case. We also investigate the impact of users’
schedule order on the system performance. The insights
about users’ working mode permutation are further
discussed.
• For the case with K (K ≥ 3) users, based on all possi-
ble optimal permutations of working mode, the original
optimization problem can be decomposed into K +1 sub-
problems. By sequentially solving these sub-problems
using convex optimization techniques [24], both the
optimal working mode permutation and time allocation
policy are obtained. Moreover, a special case in which
all users have the same backscatter rate is discussed to
further reduce the computational complexity.
The rest of this paper is organized as follows. In Section II,
the system model is proposed, and the expression for the
sum-throughput is presented. In Section III, we formulate an
optimization problem to maximize the sum-throughput and
obtain the optimal transmission policy. The simulation results
are presented in Section IV. Finally, Section V concludes this
paper.
II. SYSTEM MODEL
As shown in Figure 1, we consider a BAWPCN with one
two-antenna HAP and K single-antenna users. This system
can be used for environmental monitoring systems in wireless
sensor networks [25]. It is assumed that the HAP can work in
the FD mode by employing perfect successive interference
cancellation (SIC) techniques [26], [27]. Hence, one antenna
of the HAP is used for transferring energy to the users in
the downlink, and the other antenna is used for receiving
signals from the users in the uplink simultaneously. These
users with supercapacitors for storing energy, denoted as
Ui , i = 1, · · · , K , can switch between HTT and backscatter
modes adaptively and transmit or backscatter information to
the HAP in TDMA [14], [21]. The forward channel power
gain from the HAP to Ui and the backward channel power
gain from Ui to the HAP are denoted as hi and gi , respec-
tively. We assume that hi and gi are quasi-static flat fading
and remain constant during one block, denoted as T , and
7255
 B. Lyu et al.: WPCNs Assisted by Backscatter Communication

transmission period) with the previously harvested energy.

We assume that all harvested energy will be exhausted during
βi to avoid energy waste. Hence, the achievable throughput of
Ui for the HTT mode is given by
Ei gi
Rhi = βi W log2 (1 + )
βi σ 2
Pi−1
j=0 tj + αi
= βi W log2 (1 + γi ), i = 1, · · · , K ,
βi
where W is the bandwidth and γi = ηi Ph
σ2
i gi
.
Denote the backscatter rate of Ui as Bbi , which depends
on the design of RC circuit elements [20]. Hence, Bbi is
independent of the allocated time. The throughput of Ui for
the backscatter mode over ti − αi − βi is given by
Rbi = (ti − αi − βi )Bbi , i = 1, · · · , K .

FIGURE 1. Backscatter-assisted wireless powered communication Based on the above analysis, the sum-throughput of all
network. users is given by
K
X K
X
R= Rhi + Rbi .
i=1 i=1

III. SUM-THROUGHPUT MAXIMIZATION

In this section, we study the sum-throughput maximization
of a BAWPCN system by optimizing the transmission policy.
An optimization problem is first formulated. However, it is
not easy to solve the optimization problem with multiple
users, particularly when the number of users exceeds two.
Hence, we first consider a special case, i.e., K = 2, to
FIGURE 2. Block structure for BAWPCN. obtain some insights. Based on these insights, the general
case (K ≥ 3) is further solved. Denote t = [t0 , · · · , tK ],
may change from one block to another. It is assumed that α = [α1 , · · · , αK ], and β = [β1 , · · · , βK ]. The optimization
the channel power gains are perfectly known at the HAP. problem is formulated as
Moreover, Ui , i = 1, · · · , K does not have initial energy,
and its transmission relies on the harvested energy for the max R(t, α, β)
t,α,β
HTT mode or on instantaneous excitation energy for the
K
backscatter mode. X
s.t. ti ≤ T ,
The block structure for the BAWPCN system is described
i=0
in Figure 2. In the proposed system, the HAP continuously
αi + βi ≤ ti , i = 1, · · · , K ,
broadcasts energy with constant power, denoted as P, to all
users during one block. The block is divided into K + 1 main αi ≥ 0, i = 1, · · · , K ,
time slots, denoted as ti , i = 0, · · · , K . Ui is allocated with βi ≥ 0, i = 1, · · · , K ,
ti , i = 1, · · · , K . The main time slot ti is further divided into ti ≥ 0, i = 0, 1, · · · , K . (P1)
three parts, denoted as αi , βi and ti −αi −βi , i = 1, · · · , K . αi
and βi are utilized for the HTT mode, and the remaining part It can be proved that the objective function is concave since
is used for the backscatter mode. Due to the high discharge its Hessian is negative semidefinite. Additionally, all con-
characteristic of supercapacitors [15], we assume that each straints are affine. Hence, we obtain that Problem P1 is a con-
user can only harvest energy before its transmission but not vex optimization problem [24]. Denote the optimal solution
after P [14]. Hence, the harvested energy of Ui , denoted as Ei , for Problem P1 as t ∗ = [t0∗ , · · · , tK∗ ], α ∗ = [α1∗ , · · · , αK∗ ],
over i−1 and β ∗ = [β1∗ , · · · , βK∗ ]. First, a useful structure of the
j=0 tj + ai is given by
optimal solution for Problem P1 is given in the following
i−1
X lemma.
Ei = ηi Phi ( tj + αi ), i = 1, · · · , K , Lemma 1: The optimal solution for Problem P1 satisfies
j=0
K
where ηi is the efficiency of energy harvesting. Then,
X
ti∗ = T and αi∗ = 0, i = 1, · · · , K .
Ui transmits information to the HAP during βi (the data i=0

7256 VOLUME 5, 2017

B. Lyu et al.: WPCNs Assisted by Backscatter Communication

It is easy to prove Lemma 1 by contradiction. For sim- solution for Problem P4 is given as
plification, we briefly discuss the proof of Lemma 1 in the
z∗1 − 1 ¯ γ1
following remark. t0 = T̄ and β1 = T̄¯ ,
Remark 1: It can be seen that Rhi is an increasing function γ1 + z∗1 − 1 γ1 + z∗1 − 1
withP respect to t0 so that the entire block should be exploited, where z∗1 > 1 is the corresponding solution of f1 (z1 ) =
i.e., K ∗
i=0 ti = T , to maximize the sum-throughput. More- γ1 and f1 (z1 ) is defined as f1 (z1 ) = z1 ln z1 − z1 + 1.
over, αi is only used for harvesting energy. Compared with Let Bt1 = γ +z γ1 ∗
∗ −1 W log2 z1 , which is the normalized
αi 6 = 0, let αi∗ = 0 does not change the energy harvested by 1 1
throughput of U1 with the HTT mode. With the above result,
Ui , but increases Rbi−1 (i = 2, · · · , K ). Problem P3 is reduced to Problem P5, which is given by
With αi∗ = 0, i = 1, · · · , K , the sum-throughput is re-
max Bt1 T̄¯ + (T̄ − T̄¯ )Bb1
Pi−1
j=0 tj
PK
expressed as R(t, β) = β i W log (1 + βi γi ) +
PK i=1 2 T̄¯
i=1 (ti − βi )Bi . Problem P1 can be reduced to Problem P2,
b
s.t. 0 ≤ T̄¯ ≤ T̄ . (P5)
which is formulated as
max R(t, β) The solution for Problem P5 is straightforward to derive and
t,β is given as follows.
K (
T̄ , Bt1 ≥ Bb1 ,
X
s.t. ti = T , ¯
T̄ = (1)
i=0 0, Bt1 < Bb1 .
βi ≤ ti , i = 1, · · · , K ,
βi ≥ 0, i = 1, · · · , K , If T̄¯ = T̄ , then U1 works in the HTT mode; otherwise, U1
works in the backscatter mode.
ti ≥ 0, i = 0, 1, · · · , K . (P2)
Before solving Problem P2 directly, a two-user case is first 2) THE OPTIMAL SOLUTION FOR TWO-USER CASE
considered. With the working mode of U1 , Problem P2 is further studied
to find the optimal working mode of U2 . Consider two sub-
A. TWO-USER CASE cases, i.e., U1 works in the HHT mode or the backscatter
To solve Problem P2 with K = 2, we adopt a two-stage mode. If U1 works in the HTT mode, then the sum-throughput
approach. Let T̄ = t0 + t1 . From [22], we know that each user is rewritten as R(T̄ , t2 , β2 ) = Bt1 T̄ + β2 W log2 (1 + βT̄2 γ2 ) +
can work in either HTT mode or backscatter mode during the (t2 − β2 )Bb2 . With the updated R(T̄ , t2 , β2 ), Problem P2 is
given slot. First, given T̄ , we study the optimal working mode rewritten as
of U1 , corresponding to the single-user case [22]. Then, with
the derived result of U1 , we further study the working mode max R(T̄ , t2 , β2 )
T̄ ,t2 ,β2
of U2 . To facilitate the understanding of this paper, we still
present the details of solving the single-user case. s.t. T̄ + t2 = T ,
0 ≤ β2 ≤ t 2 ,
1) THE OPTIMAL SOLUTION FOR SINGLE-USER CASE T̄ ≥ 0. (P6)
With given T̄ , the optimization problem of the single-user
case is given as follows. To solve Problem P6, a method similar to that used for
t0 Problem P3 is adopted. Let T̂ = T̄ + β2 . First, we determine
max β1 W log2 (1 + γ1 ) + (t1 − β1 )Bb1 the relationship between T̄ and β2 by solving the following
t0 ,t1 ,β1 β1
problem:
s.t. t0 + t1 = T̄ ,
0 ≤ β1 ≤ t 1 , max Bt1 T̄ + β2 W log2 (1 +
T̄
γ2 )
t0 ≥ 0. (P3) T̄ ,β2 β2
s.t. T̄ + β2 = T̂ ,
To solve Problem P3, we first derive the relationship
between t0 and β1 . Let T̄¯ = t0 +β1 . The relationship between T̄ , β2 ≥ 0. (P7)
t0 and β1 can be derived by solving the following problem Using a method similar to that for Problem P4, we derive the
t0 optimal solution for Problem P7 with the given T̂ , which is
max β1 W log2 (1 + γ1 )
t0 ,β1 β1 as follows:
¯
s.t. t + β = T̄ , z∗2 − 1 γ2
0 1
T̄ = T̂ and β2 = T̂ ,
t0 , β1 ≥ 0. (P4) γ2 + z∗2 − 1 γ2 + z∗2 − 1
The solution for Problem P4 can be obtained using the where z∗2 is the corresponding solution of f2 (z2 ) = γ2 and
method presented in [8]. With the given T̄¯ , the optimal f2 (z2 ) is defined as f2 (z2 ) = z2 ln z2 − (1 + lnW2 Bt1 )z2 + 1. Let

VOLUME 5, 2017 7257

 B. Lyu et al.: WPCNs Assisted by Backscatter Communication

z∗ −1 γ2

Bt2 = γ +z
2 t ∗
∗ −1 B1 + γ +z∗ −1 W log2 z2 . Hence, Problem P6 is  0, if ‘‘0B’’,
2 2 2 2

 z∗−1
reduced to Problem P8, which is formulated as 
 3
T, if ‘‘BH’’,
t1 = γ2 + z∗3 − 1
max Bt2 T̂ + (T − T̂ )Bb2 
 γ1 z∗2 − 1
T , if ‘‘HH’’,

T̂ 
γ1 + z1 − 1 γ2 + z∗2 − 1
∗

s.t. 0 ≤ T̂ ≤ T . (P8) 
0, if ‘‘0B’’,
γ2

The result for Problem P8 is easily derived and is given as 
T , if ‘‘BH’’,


follows. β2 = γ2 + z∗3 − 1
(  γ2
T , Bt2 ≥ Bb2 ,

 T if ‘‘HH’’,
γ2 + z∗2 − 1

T̂ = (2)
0, Bt2 < Bb2 .
T,

if ‘‘0B’’
γ2

Combining the results from (1) and (2), the permutations of

T , if ‘‘BH’’,


two users’ working mode are summarized as follows. If Bt1 ≥ t2 = γ2 + z∗3 − 1
Bb1 and Bt2 ≥ Bb2 , then both U1 and U2 work in the HTT mode.
 γ2
T , if ‘‘HH’’.


If Bt1 ≥ Bb1 and Bt2 < Bb2 , then U1 cannot work and U2 works γ2 + z∗2 − 1


in the backscatter mode with the entire block. To simplify the Note that Proposition 1 only provides one of the optimal
description, we define the user cannot work, the user works in transmission policies for the two-user case. This is because
the backscatter mode and the user works in the HTT mode as when one of the special conditions, i.e., Bt1 = Bb1 , Bt2 = Bb2
‘‘0’’, ‘‘B’’ and ‘‘H’’, respectively. Therefore, the permutations and B̄t2 = Bb2 is satisfied, we can also find other solutions to
of this sub-case are given by ‘‘HH’’ and ‘‘0B’’. achieve the maximum sum-throughput. However, the result
For symbol simplification, the symbols defined in the pre- given in Proposition 1 can provide us with some interesting
vious sub-case are still used in this sub-case corresponding insights. Hereinafter, without specific description, we do not
to U1 works in the backscatter mode. Since U1 works in the consider the special conditions.
backscatter mode, R is rewritten as R(T̄ , t2 , β2 ) = Bb1 T̄ + In summary, for a BAWPCN with two users, there are
β2 W log2 (1 + βT̄2 γ2 ) + (t2 − β2 )Bb2 . In this sub-case, with three possible optimal permutations of the working mode, i.e.,
‘‘0B’’, ‘‘BH’’ and ‘‘HH’’.
the updated R(T̄ , t2 , β2 ), we also solve Problem P6 to derive
With Proposition 1 and the above analysis, we can derive
the optimal solution. Hence, T̄ , β2 and T̂ in this sub-case are
the following corollaries.
rewritten as
Corollary 1: In the optimal condition, the last user could
z∗3 − 1 γ2 be scheduled and work in only one mode.
T̄ = T̂ and β2 = T̂ ,
γ2 + z3 − 1
∗ γ2 + z∗3 − 1 For simplification, we briefly discuss the proof of Corol-
lary 1 in the following remark.
(
T , B̄t2 ≥ Bb2 ,
T̂ = (3) Remark 2: For the two-user case, it has been proved that
0, B̄t2 < Bb2 , the last user could be scheduled and work in only one mode in
where z∗3 is the corresponding solution of f3 (z3 ) = γ2 and the optimal condition. Considering the three-user case, with
f3 (z3 ) is defined as f3 (z3 ) = z3 ln z3 − (1 + lnW2 Bb1 )z3 + 1, the given working modes of the first two users, it can be
z∗3 −1 γ2 seen that the time allocation between the HTT and backscatter
B̄t2 = b
γ2 +z∗3 −1 B1 + γ2 +z∗3 −1 W log2 z∗3 .
modes of the third user is a linear problem, which is similar as
Consider both (1) and (3). If Bt1 < Bb1 and Bt2 ≥ Bb2 , then Problem P8. To solve the problem, we obtain the conclusion
U1 works in the backscatter mode and U2 works in the HTT that the third user could be scheduled and work in only one
mode. If Bt1 < Bb1 and Bt2 < Bb2 , then only U2 works and stays mode in the optimal condition. Similarly, for K -user case
in the backscatter mode. The permutations of this sub-case are (K ≥ 4), the same conclusion for the K -th user can be
summarized as ‘‘BH’’ and ‘‘0B’’. obtained.
Considering the above two sub-cases, the optimal time Corollary 2: In the optimal condition, if the last user
allocation policy for the two-user case is given by the fol- works in the backscatter mode, one optimal solution is to
lowing proposition. allocate the entire block time to UK , i.e., the first K − 1 users
Proposition 1: For the two-user case, one optimal solution are not scheduled.
for Problem P2 is given by Proof: Please refer to Appendix.
 From the above result, we observe that the schedule order
0, if ‘‘0B’’ or ‘‘BH’’,
t0 = z∗1 − 1 z∗2 − 1 of users may affect the time allocation policy. To explain
T , if ‘‘HH’’, this observation, the following example is given. For ‘‘0B’’
γ1 + z∗1 − 1 γ2 + z∗2 − 1

 corresponding to the condition Bt1 ≥ Bb1 and Bt2 < Bb2 , it
0, if ‘‘0B’’ or ‘‘BH’’, can achieve a larger sum-throughput with a reversed schedule
β1 = γ1 z∗2 − 1 order, i.e., U2 is first scheduled. Next, we analyze the sum-
T , if ‘‘HH’’,
γ1 + z∗1 − 1 γ2 + z∗2 − 1

throughput of the reversed schedule order scheme. According

7258 VOLUME 5, 2017

B. Lyu et al.: WPCNs Assisted by Backscatter Communication

to [14], as the number of users increases, the normalized ti ≥ 0, i = N , N + 1, · · · , K . (P9)

sum-throughput increases. Since Bb2 > Bt2 , we can derive
γ2 † γ2 † Using convex optimization techniques [24], the optimal time
that Bb2 > † W log2 z1 , where † W log2 z1 is the allocation policy for the second type of sub-problems is given
γ2 +z1 −1 γ2 +z1 −1
† in the following proposition.
normalized throughput of U2 with the HTT mode, and z1 is
the solution of f1 (z1 ) = γ2 . Additionally, with Bb2 > Bb1 , Proposition 2: The optimal solution for Problem P9
we can derive that the optimal permutation of the reversed satisfies
1

schedule order scheme is given as ‘‘BH’’. From [28], the opti- 
 T, i = K,
mal solution for the reversed schedule order scheme indeed  1 + vP

 i
K ∗
provides a larger sum-throughput. ti∗ = T − n=i+1 tn
 , i = K − 1, · · · , N + 1,

 1 + vi
T− K

n=N +1 tn , i = N ,
∗
P
B. MULTI-USER CASE (K ≥ 3)

In this sub-section, we propose an algorithm with lower γi −1
1 W ( eai +1 )+ai +1
complexity to solve Problem P2 if K ≥ 3. With the result where vi = γi (e − 1), i = N + 1, · · · , K ,
for the two-user case, we can derive all possible optimal BbN BbN Pi−1 γn
permutations of the working mode, one of which is the glob-
aN +1 = W ln 2, ai = W ln 2 + n=N +1 γn vn +1 , i = N +
2, · · · , K , and W(.) is the Lambert W-Function [29].
ally optimal permutation. For the two-user case, there are
Proof: This problem is similar to our previous work
three permutations, i.e., ‘‘0B’’, ‘‘BH’’ and ‘‘HH’’. When we
in [28]. For more details, please refer to [28].
consider the three-user case, from Corollary 1, there are six
For the third type of sub-problem, since all users work in
possible optimal permutations of the working mode, which
the HTT mode, βi = ti , i = 1, · · · , K . Problem P2 is reduced
are given as follows: ‘‘0BB’’, ‘‘0BH’’, ‘‘BHB’’, ‘‘BHH’’,
to Problem P10, which was studied in [14].
‘‘HHB’’ and ‘‘HHH’’. From Corollary 2, the permutations Pi−1
K
‘‘0BB’’, ‘‘BHB’’ and ‘‘HHB’’ can be reduced to the per- j=0 tj
X
mutation ‘‘00B’’ without affecting the optimal solution. The max ti W log2 (1 + γi )
t ti
updated permutations are summarized as ‘‘00B’’, ‘‘0BH’’, i=1
K
‘‘BHH’’, and ‘‘HHH’’. Similarly, for the four-user case, X
the possible optimal permutations are ‘‘000B’’, ‘‘00BH’’, s.t. ti = T ,
i=0
‘‘0BHH’’, ‘‘BHHH’’, and ‘‘HHHH’’. Hence, we can con-
clude that the K -user case finally has K + 1 possible opti- ti ≥ 0, i = 0, 1, · · · , K . (P10)
mal permutations. Since each permutation corresponds to a Using the algorithm proposed in [14], the optimal time allo-
sub-problem, Problem P2 can be decomposed into K + 1 cation policy for this type of sub-problem is given in the
sub-problems. By solving these sub-problems sequentially, following proposition.
the solution for the sub-problem with the maximum sum- Proposition 3: The optimal solution for Problem P10 sat-
throughput also solves Problem P2. isfies
The analysis for the K +1 sub-problems is given as follows. 
1
The K + 1 sub-problems can be classified into three types.
 T, i = K,
 1 + µi



The first type is that only the last user works and stays in the PK
ti∗ = T − n=i+1 tn
∗
backscatter mode. The second one is that the first working , i = K − 1, · · · , 1,
1P+ µi



user adopts the backscatter mode and the other working users  K
T − n=1 tn∗ ,

i = 0,
adopt the HTT mode. The third one is that all users work in
γi −1
the HTT mode. For the first type of sub-problem, it is easy W( )+ci +1
where µi = γ1i (e eci +1 − 1), i = 1, · · · , K , c1 = 0,
to derive that the entire block time will be allocated to UK . γn
and ci = i−1 , · · , K.
P
For the second type of sub-problems, assume that the number n=1 γn µn +1 i = 2, ·
of working users is M . Hence, the index of the first working After solving K +1 sub-problems, the solution corresponding
user, which is in the backscatter mode, is K − M + 1. Let to the maximum sum-throughput also solves Problem P2.
N = K −M +1. Since the permutation of the working mode is The algorithm to solve Problem P2 is summarized in
given, to maximize the sum-throughput, it is straightforward Algorithm 1. The computational complexity analysis of the
to derive that ti = 0, i = 0, 1, , · · · , K − M , βi = 0, i = proposed algorithm is given as follows. For the second
1, · · · , N and βi = ti , i = N + 1, · · · , K . Hence, Problem P2 type of sub-problems, the complexity is O(M − 1), where
is rewritten as M = 2, · · · , K [14]. Similarly, the complexity for the third
K Pi−1 type of sub-problem is O(K ). Hence, the total complexity of
j=N tj Algorithm 1 is given as O(1) + · · · + O(K ).
X
b
max tN BN + ti W log2 (1 + γi )
{ti }K
i=N
ti Remark 3: For the special case in which the backscatter
i=N +1
K rate of each user is the same, i.e., B1 = · · · = BK , the
computational complexity of solving Problem P2 can be
X
s.t. ti = T ,
i=N further reduced. The explanation is given by taking the three-
user case as an example. For the three-user case, all possible
VOLUME 5, 2017 7259
 B. Lyu et al.: WPCNs Assisted by Backscatter Communication

Algorithm 1 Algorithm for Problem P2.

• Step 1: Determine K + 1 possible optimal permutations.
Based on these permutations, Problem P2 is decom-
posed into K + 1 sub-problems;
• Step 2: Solve these sub-problems sequentially;
• Step 3: Comparing the sum-throughputs of K + 1 sub-
problems, the solution corresponding to the maximum
sum-throughput is also the solution for Problem P2.

FIGURE 3. Average sum-throughput vs. transmit power for two-user case

with κ = 3, d1 = 5 m, d2 = 10 m, Bb
1
= 20 kbits/s, and Bb
2
= 60 kbits/s.

optimal permutations are given as ‘‘00B’’, ‘‘0BH’’, ‘‘BHH’’,

and ‘‘HHH’’. From [28], with the increase of the number of
users, the sum-throughput of the second type of sub-problems
increases. Moreover, ‘‘00B’’ can be viewed as a special
second type of sub-problem. Hence, the sum-throughput of
‘‘BHH’’ is larger than that of ‘‘00B’’ or ‘‘0BH’’. In other
words, we only need to solve two sub-problems correspond-
ing to ‘‘BHH’’ and ‘‘HHH’’, respectively. Extending to the
case with K (K ≥ 3) users, we only need to consider
the two sub-problems in which all users are scheduled.
Hence, the computational complexity of the special case is
O(K − 1) + O(K ).
Remark 4: For the two-user case, the optimal permutation
can be derived from (1), (2) and (3) by taking two compar-
isons. Using the similar method, we can derive the optimal
permutation for K -user case (K ≥ 3) by taking K com-
parisons. With the derived optimal permutation, we easily
find the optimal solution for Problem P2. The complexity
FIGURE 4. Average probability of possible optimal permutations vs.
analysis of the new method is given as follows. Assume the D
transmit power for three-user case with κ = 3, di = 5 + KK ∗ i and
complexity of the bisection method used in each comparison Dk = 5 m. (a) Scenario one with Bb = 40 kbits/s, Bb = 40 kbits/s and
1 2
is O(Si ), i = 1, · · · , K . Considering the worst case, the total Bb3
= 40 kbits/s. (b) Scenario two with Bb
1
= 20 kbits/s, Bb
2
= 40 kbits/s
complexity of this method is K
P
i=1 O(S i ) + O(K ). Compare and Bb3
= 60 kbits/s. (c) Scenario three with B b = 60 kbits/s,
1
with this method, the complexity of Algorithm 1 is still lower Bb2
= 40 kbits/s and Bb 3
= 20 kbits/s.

when K is not large.

lation environment is set as follows. The bandwidth is set as
IV. SIMULATION W = 1 MHz. Assume that the forward and backward channel
In this section, simulation results are presented to evaluate power gains are reciprocal and modeled as hi = gi =
the proposed transmission policy for BAWPCNs. The simu- 10−3 diκi ρi2 , where di is the distance between the HAP and

7260 VOLUME 5, 2017

B. Lyu et al.: WPCNs Assisted by Backscatter Communication
FIGURE 5. Average sum-throughput vs. number of users with
P = 20 dBm, κ = 3, and Bb = 40 kbits/s.
i
Ui , κi is the pathloss exponent, ρi is the channel short-term
fading, and ρi2 is an exponentially distributed random variable
with unit mean. Without loss of generality, let κ = κ1 =
· · · = κK and T = 1 s. The noise power at the HAP is given
as σ 2 = −70 dBm. For simplicity, the energy harvesting
efficiency of each user is assumed to be the same and given
as η1 = · · · = ηK = 0.6. All simulation curves are obtained
by averaging 1000 Monte Carlo runs.
Figure 3 investigates the effect of schedule order for the
two-user case. Let κ = 3, d1 = 5 m, d2 = 10 m, Bb1 =
20 kbits/s, and Bb2 = 60 kbits/s. It can be observed that
the performance of the reversed order scheme is superior
to that of the given order scheme, which indicates that the
schedule order indeed affects the sum-throughput. Moreover,
as the transmit power increases, the sum-throughput of both
schedule order schemes increases.
Figure 4 shows the average probability of possible optimal
permutations versus the transmit power for the three-user case
with κ = 3 and di = 5 + DKK ∗ i, where Dk = 5 m.
Three scenarios in which users have different backscatter
rates are considered. Let Bb = [Bb1 , Bb2 , Bb3 ]. The three
scenarios are given as follows: Bb = [40, 40, 40] kbits/s,
Bb = [20, 40, 60] kbits/s, and Bb = [60, 40, 20] kbits/s,
which correspond to Figs. 4 (a), (b) and (c), respectively.
Figs. 4 (a) and (c) indicate that when the transmit power is
low, ‘‘BHH’’ is the optimal permutation. This result occurs
because with a low transmit power, the normalized through-
put of U1 adopting the HHT mode is lower than its backscatter
rate. Conversely, the optimal permutation of Figure 4 (b)
with a low transmit power is ‘‘00B’’ since Bb3 is quite large.
For the three scenarios, as the transmit power increases,
‘‘HHH’’ tends to be the optimal permutation with a high
probability.
Figure 5 depicts the average sum-throughput versus the
number of users with P = 20 dBm, κ = 3, di = 5 + DKK ∗ i,
Dk = 5 m, and Bbi = 40 kbits/s. With the increase of the
number of users, the sum-throughput increases. Compared
with only adopting the HHT mode or backscatter mode,
VOLUME 5, 2017
the proposed transmission policy can achieve a larger sum-
throughput.
V. CONCLUSION
This paper has proposed a new WPCN system model that
integrates backscatter communication. To maximize the sum-
throughput, an optimization problem has been formulated.
The insights about permutations of users’ working mode
have been discussed, and the optimal transmission policy has
been presented. The simulation results have confirmed that
the proposed model achieves a larger sum-throughput than
traditional schemes.
APPENDIX
PROOF OF COROLLARY 2
Define the optimal normalized throughput for the first K − 1
users as BtK −1 . Additionally, define z∗K as the optimal solution
for fK (zK ) = γK and fK (zK ) = zK ln zK − (1 + lnW2 BtK −1 )
zK +1. We derive that if UK can work in the backscatter mode
in the optimal condition, then the condition γK +z
z∗K −1 t
∗ −1 BK −1 +
γK
K
γK +z∗K −1 Wlog2 z∗K ≤ BbK is satisfied. Hence, in the optimal
condition, we can let the total time allocated to the first K − 1
users be zero, leading to the desired result.
REFERENCES
[1] C. K. Ho and R. Zhang, ‘‘Optimal energy allocation for wireless communi-
cations with energy harvesting constraints,’’ IEEE Trans. Signal Process.,
vol. 60, no. 9, pp. 4808–4818, Sep. 2012.
[2] M. Pinuela, P. D. Mitcheson, and S. Lucyszyn, ‘‘Ambient RF energy
harvesting in urban and semi-urban environments,’’ IEEE Trans. Microw.
Theory Techn., vol. 61, no. 7, pp. 2715–2726, Jul. 2013.
[3] R. Zhang and C. K. Ho, ‘‘MIMO broadcasting for simultaneous wireless
information and power transfer,’’ IEEE Trans. Wireless Commun., vol. 12,
no. 5, pp. 1989–2001, May 2013.
[4] K. Huang and E. Larsson, ‘‘Simultaneous information and power transfer
for broadband wireless systems,’’ IEEE Trans. Signal Process., vol. 61,
no. 23, pp. 5972–5986, Dec. 2013.
[5] N. Zhao, F. R. Yu, and V. C. M. Leung, ‘‘Opportunistic communications
in interference alignment networks with wireless power transfer,’’ IEEE
Wireless Commun., vol. 22, no. 1, pp. 88–95, Feb. 2015.
[6] N. Zhao, F. R. Yu, and V. C. M. Leung, ‘‘Wireless energy harvesting in
interference alignment networks,’’ IEEE Commun. Mag., vol. 53, no. 6,
pp. 72–78, Jun. 2015.
[7] S. Bi, Y. Zeng, and R. Zhang, ‘‘Wireless powered communication net-
works: An overview,’’ IEEE Wireless Commun., vol. 23, no. 2, pp. 10–18,
Apr. 2016.
[8] H. Ju and R. Zhang, ‘‘Throughput maximization in wireless powered
communication networks,’’ IEEE Trans. Wireless Commun., vol. 13, no. 1,
pp. 418–428, Jan. 2014.
[9] L. Liu, R. Zhang, and K.-C. Chua, ‘‘Multi-antenna wireless powered com-
munication with energy beamforming,’’ IEEE Trans. Commun., vol. 62,
no. 12, pp. 4349–4361, Dec. 2014.
[10] G. Yang, C. K. Ho, R. Zhang, and Y. L. Guan, ‘‘Throughput optimization
for massive MIMO systems powered by wireless energy transfer,’’ IEEE J.
Sel. Areas Commun., vol. 33, no. 8, pp. 1640–1650, Aug. 2015.
[11] J. Li, H. Zhang, D. Li, and H. Chen, ‘‘On the performance of wireless-
energy-transfer-enabled massive MIMO systems with superimposed
pilot-aided channel estimation,’’ IEEE Access, vol. 3, pp. 2014–2027,
Oct. 2015.
[12] Y. L. Che, L. Duan, and R. Zhang, ‘‘Spatial throughput maximization of
wireless powered communication networks,’’ IEEE J. Sel. Areas Commun.,
vol. 33, no. 8, pp. 1534–1548, Aug. 2015.
[13] H. Ju and R. Zhang, ‘‘Optimal resource allocation in full-duplex wireless-
powered communication network,’’ IEEE Trans. Commun., vol. 62, no. 10,
pp. 3528–3540, Oct. 2014.
7261

[14] X. Kang, C. K. Ho, and S. Sun, ‘‘Full-duplex wireless-powered commu-
nication network with energy causality,’’ IEEE Trans. Wireless Commun.,
vol. 14, no. 10, pp. 5539–5551, Oct. 2015.
[15] M. Kaus, J. Kowal, and D. U. Sauer, ‘‘Modelling the effects of charge redis-
tribution during self-discharge of supercapacitors,’’ Electrochim. Acta,
vol. 55, no. 25, pp. 7516–7523, Oct. 2010.
[16] S. H. Kim and D. I. Kim, ‘‘Hybrid backscatter communication for wireless
powered communication networks,’’ in Proc. ISWCS, Poznań, Poland,
Sep. 2016, pp. 265–269.
[17] C. Boyer and S. Roy, ‘‘Backscatter communication and RFID: Coding,
energy, and MIMO analysis,’’ IEEE Trans. Commun., vol. 62, no. 3,
pp. 770–785, Mar. 2014.
[18] G. Yang, C. K. Ho, and Y. L. Guan, ‘‘Multi-antenna wireless energy trans-
fer for backscatter communication systems,’’ IEEE J. Sel. Areas Commun.,
vol. 33, no. 12, pp. 2974–2987, Dec. 2015.
[19] Q. Yang, H.-M. Wang, Y. Zhang, and Z. Han, ‘‘Physical layer security
in MIMO backscatter wireless systems,’’ IEEE Trans. Wireless Commun.,
vol. 15, no. 11, pp. 7547–7560, Nov. 2016.
[20] V. Liu, A. Parks, V. Talla, S. Gollakota, D. Wetherall, and J. R. Smith,
‘‘Ambient backscatter: Wireless communication out of thin air,’’ in Proc.
ACM SIGCOMM, Hong Kong, Aug. 2013, pp. 39–50.
[21] D. T. Hoang, D. Niyato, P. Wang, D. I. Kim, and Z. Han. (Aug. 2016). ‘‘The
tradeoff analysis in RF-powered backscatter cognitive radio networks.’’
[Online]. Available: https://arxiv.org/abs/1608.01789
[22] D. T. Hoang, D. Niyato, P. Wang, D. I. Kim, and Z. Han, ‘‘Ambient
backscatter: A new approach to improve network performance for
RF-powered cognitive radio networks,’’ IEEE Trans. Commun., to be
published.
[23] P. Zhang and D. Ganesan, ‘‘Enabling bit-by-bit backscatter communication
in severe energy harvesting environments,’’ in Proc. NSDI, Seattle, WA,
USA, Apr. 2014, pp. 345–357.
[24] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge, U.K.:
Cambridge Univ. Press, 2004.
[25] V.-H. Duong, N. X. Hieu, H.-S. Lee, and J.-W. Lee, ‘‘A battery-assisted
passive EPC Gen-2 RFID sensor tag IC with efficient battery power man-
agement and RF energy harvesting,’’ IEEE Trans. Ind. Electron., vol. 63,
no. 11, pp. 7112–7123, Nov. 2016.
[26] D. Bharadia, E. McMilin, and S. Katti, ‘‘Full duplex radios,’’in Proc. ACM
SIGCOMM, Hong Kong, Aug. 2013, pp. 375–386.
[27] B. Radunovic et al., ‘‘Rethinking indoor wireless mesh design: Low
power, low frequency, full-duplex,’’ in Proc. WIMESH, Boston, MA, USA,
Jun. 2010, pp. 1–6.
[28] B. Lyu, Z. Yang, G. Gui, and Y. Feng, ‘‘Throughput maximization in
backscatter assisted wireless powered communication networks,’’ IEICE
Trans. Fundam., vol. E100-A, no. 6, pp. 1–5, Jun. 2017.
[29] Lambert W-Function. [Online]. Available: http://mathworld.wolfram.
com/LambertW-Function.html
BIN LYU received the B.E. degree in elec-
tronic and information engineering from the
Nanjing University of Posts and Telecommuni-
cations, Nanjing, China, in 2013, where he is
currently pursuing the Ph.D. degree in signal
and information processing. His research interests
include wireless power transfer and backscatter
communication.
7262
B. Lyu et al.: WPCNs Assisted by Backscatter Communication
ZHEN YANG received the B.E. and M.E. degrees
from the Nanjing University of Posts and Telecom-
munications, China, in 1983 and 1988, respec-
tively, and the Ph.D. degree from Shanghai Jiao
Tong University, China, in 1999, all in elec-
trical engineering. He was a Lecturer with the
Nanjing University of Posts and Telecommunica-
tions, where he was promoted to Associate Pro-
fessor in 1995, and then a Full Professor in 2000.
He was a Visiting Scholar with Bremen University,
Germany, from 1992 to 1993, and an Exchange Scholar with Maryland
University, USA, in 2003. He has authored over 200 papers in academic
journals and conferences. His research interests include various aspects of
signal processing and communication, such as communication systems and
networks, cognitive radio, spectrum sensing, speech and audio processing,
compressive sensing, and wireless communication.
Dr. Yang serves as the Vice Chairman and a fellow of the Chinese Institute
of Communications, the Chairman of the Jiangsu Institute of Internets. He is
also a member of Editorial Board for several journals, such as the Chinese
Journal of Electronics, the China Communications, and the Data Collection
and Processing. He was the Chair of the Asian Pacific Communication
Conference Steering Committee from 2013 to 2014.
GUAN GUI (M’11) received the Dr.Eng. degree in
information and communication engineering from
the University of Electronic Science and Tech-
nology of China, Chengdu, China, in 2011. From
2009 to 2012, with the financial support from the
China Scholarship Council and the Global Center
of Education of Tohoku University, he joined the
Wireless Signal Processing and Network Labora-
tory (Prof. Adachi’s laboratory), Department of
Communications Engineering, Graduate School of
Engineering, Tohoku University, as a Research Assistant and a Post-Doctoral
Research Fellow, where he was supported by the Japan Society for the
Promotion of Science Fellowship as a Post-Doctoral Research Fellow, from
2012 to 2014. From 2014 to 2015, he was an Assistant Professor with
the Department of Electronics and Information System, Akita Prefectural
University. Since 2015, he has been a Professor with the Nanjing University
of Posts and Telecommunications, Nanjing, China. He is currently involved
in the research of multidimensional system control, super-resolution radar
imaging, adaptive filter, compressive sensing, sparse dictionary designing,
channel estimation, and advanced wireless techniques.
Dr. Gui received the IEEE International Conference on Communications
Best Paper Award and the IEEE Vehicular Technology Conference Best
Student Paper Award in year 2014. He has been an Associate Editor of the
Wiley Journal of Security and Communication Networks since 2012.
YOUHONG FENG received the B.S. degree and
the M.S. degree in information engineering from
Chang’an University, Xi’an, China, in 2003 and
2006, respectively. He is currently pursuing the
Ph.D. degree with the Nanjing University of Posts
and Telecommunications. He is also an Asso-
ciate Professor with the College of Physics and
Electronic information Engineering, Anhui Nor-
mal University. His research, including wireless
security and energy-efficient communications.
VOLUME 5, 2017