5092 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 17, NO.

8, AUGUST 2018

UAV-Enabled Wireless Power Transfer: Trajectory

Design and Energy Optimization
Jie Xu , Member, IEEE, Yong Zeng , Member, IEEE, and Rui Zhang , Fellow, IEEE

Abstract— This paper studies a new unmanned aerial I. I NTRODUCTION

vehicle (UAV)-enabled wireless power transfer system, where a
UAV-mounted mobile energy transmitter is dispatched to deliver
wireless energy to a set of energy receivers (ERs) at known
locations on the ground. We investigate how the UAV should
R ADIO frequency (RF) transmission enabled wireless
power transfer (WPT) is a promising solution to provide
perpetual and cost-effective energy supplies to low-power
optimally exploit its mobility via trajectory design to maximize electronic devices, and it is anticipated to have abundant appli-
the amount of energy transferred to all ERs during a finite cations in future Internet-of-things (IoT) wireless networks
charging period. First, we consider the maximization of the sum
energy received by all ERs by optimizing the UAV’s trajectory (see, [3], [4] and the references therein). In conventional
subject to its maximum speed constraint. Although this problem WPT systems, dedicated energy transmitters (ETs) are usually
is non-convex, we obtain its optimal solution, which shows that deployed at fixed locations to send RF signals to charge
the UAV should hover at one single fixed location during the distributed energy receivers (ERs) [5] such as low-power
whole charging period. However, the sum-energy maximization sensors or IoT devices. However, due to the severe prop-
incurs a “near-far” fairness issue, where the received energy by
the ERs varies significantly with their distances to the UAV’s agation loss of RF signals over long distance, the perfor-
optimal hovering location. To overcome this issue, we consider mance of practical WPT systems for wide coverage range
a different problem to maximize the minimum received energy is fundamentally constrained by the low end-to-end power
among all ERs, which, however, is more challenging to solve than transmission efficiency. As a consequence, in order to pro-
the sum-energy maximization. To tackle this problem, we first vide ubiquitous wireless energy accessibility for massive low-
consider an ideal case by ignoring the UAV’s maximum speed
constraint, and show that the relaxed problem can be optimally power ERs distributed in a large area, fixed-location ETs
solved via the Lagrange dual method. The obtained trajectory need to be deployed in an ultra-dense manner. This, however,
solution implies that the UAV should hover over a set of fixed would tremendously increase the cost, and thus hinder the
locations with optimal hovering time allocations among them. large-scale implementation of future WPT systems. In the
Then, for the general case with the UAV’s maximum speed literature, various approaches have been proposed aiming to
constraint considered, we propose a new successive hover-and-fly
trajectory motivated by the optimal trajectory in the ideal case alleviate this issue by enhancing the WPT efficiency at the link
and obtain efficient trajectory designs by applying the successive level, including multi-antenna energy beamforming [6]–[12],
convex programing optimization technique. Finally, numerical energy scheduling [13], [14], and energy waveform optimiza-
results are provided to evaluate the performance of the proposed tion [15], [16]. Different from these prior studies, in this
designs under different setups, as compared with benchmark paper we tackle this problem from a fundamentally new
schemes.
perspective at the system level, i.e., we propose a radically
Index Terms— Wireless power transfer (WPT), unmanned novel architecture for WPT systems by utilizing unmanned
aerial vehicle (UAV), trajectory optimization, energy fairness. aerial vehicles (UAVs) as mobile ETs.
Manuscript received September 22, 2017; revised March 3, 2018; accepted UAV has drawn significant research interests recently due
May 14, 2018. Date of publication May 25, 2018; date of current version to its wide range of applications, including surveillance and
August 10, 2018. This work was supported in part by the National University monitoring, aerial radar and camera, cargo delivery, commu-
of Singapore under Grant R-263-000-D12-114 and in part by the National
Natural Science Foundation of China under Grant 61628103. This paper was nication platforms, etc. Particularly, mounted with miniatur-
presented in part at the IEEE Global Communications Conference Workshop, ized communication transceivers, low-altitude UAVs can be
Singapore, December 4–8, 2017 [1], and at the Asia–Pacific Conference on used as aerial mobile base stations (BSs) or relays to help
Communications Workshop, Perth, WA, Australia, December 11–13, 2017 [2].
The associate editor coordinating the review of this paper and approving it enhance the performance of terrestrial wireless communication
for publication was S. Kompella. (Corresponding author: Yong Zeng.) systems (see, [17] and the references therein). By optimizing
J. Xu is with the School of Information Engineering, Guangdong University the UAV’s trajectory jointly with communication scheduling,
of Technology, Guangzhou 510006, China (e-mail: jiexu@gdut.edu.cn).
Y. Zeng was with the Department of Electrical and Computer Engineering, the air-to-ground link distances between the UAV and its
National University of Singapore, Singapore 117583. He is now with the served ground users can be effectively shortened, thus sig-
School of Electrical and Information Engineering, The University of Sydney, nificantly improving the system throughput [18], [19].
Sydney, NSW 2006, Australia (e-mail: yong.zeng@sydney.edu.au).
R. Zhang is with the Department of Electrical and Computer Engi- Motivated by UAV-assisted wireless communications, in this
neering, National University of Singapore, Singapore 117583 (e-mail: paper we propose a new UAV-enabled WPT architecture as
elezhang@nus.edu.sg). illustrated in Fig. 1. With the proposed architecture, a group
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. of UAVs are dispatched as mobile ETs that fly above the
Digital Object Identifier 10.1109/TWC.2018.2838134 serving area to cooperatively charge a set of distributed
1536-1276 © 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
XU et al.: UAV-ENABLED WPT: TRAJECTORY DESIGN AND ENERGY OPTIMIZATION 5093

variables, we derive its optimal solution, which shows that the

UAV should hover at one single fixed optimal location during
the whole charging period, and the optimal hovering location
can be obtained via a 2D exhaustive search. In particular,
we obtain the optimal hovering location in closed-form for
the special case with K = 2 ERs. It is shown that when
the distance between the two ERs is smaller than a certain
threshold, the optimal hovering location is exactly above the
middle point between the two ERs; whereas when their dis-
Fig. 1. Illustration of a UAV-enabled WPT system.
tance is larger than the threshold, the optimal hovering location
is closer to one ER than the other. In general, the sum-energy
ERs on the ground. We assume that these ERs have fixed maximization with the optimal fixed hovering location incurs
locations that are a priori known for the UAV trajectory a severe “near-far” fairness issue, especially for a network
design. By exploiting the fully controllable mobility of the spanning over a large area, as the near ERs in close proximity
UAVs via trajectory design, the proposed system is expected to the UAV can receive significantly more energy than the
to significantly improve the WPT performance, while reducing far-away ERs.
the number of required ETs as compared to the conventional Next, to resolve the fairness issue, we consider an alternative
WPT systems with ETs deployed at fixed locations. Notice problem to maximize the minimum received energy among
that there have been some prior works (e.g., [20], [21]) that all ERs via trajectory optimization. This problem is more
considered the use of moving ground vehicles as mobile challenging to solve than the previous sum-energy maximiza-
charging stations to wirelessly charge sensor nodes. Different tion. To obtain useful insight and a performance upper bound,
from ground vehicles that have only limited mobility in a we first consider an ideal case by assuming that the UAV
two-dimensional (2D) area usually with a large number of maximum speed constraint can be ignored.1 In this case,
ground obstacles, UAVs can be more flexibly deployed and the problem is shown to satisfy the so-called time-sharing
moved in the three-dimensional (3D) free space. Furthermore, condition in [24], and thus can be optimally solved via the
compared to terrestrial wireless channels that typically suffer Lagrange dual method. The obtained optimal solution reveals
from various impairments such as shadowing and fading in that the UAV should hover over an optimal set of fixed
addition to path loss, UAVs usually possess better channels locations, with optimal hovering time allocations among them.
to ground ERs due to the higher chance of having moderate- For the special case of K = 2 ERs, the closed-form solution
distance line-of-sight (LoS) links with them. is also obtained and compared against that for sum-energy
A fundamental question to be addressed for the proposed maximization.
UAV-enabled WPT systems is as follows: how to jointly Last, we consider the above min-energy maximization prob-
optimize the trajectories of multiple UAVs so as to maximize lem for the general case with the UAV’s maximum speed
the energy transferred to all ERs in a fair manner? This constraint considered. Inspired by the optimal multi-location-
question, however, has not yet been studied in the literature to hovering solution in the ideal case, we propose a successive
our best knowledge, and it is non-trivial to be addressed, even hover-and-fly trajectory design, where the UAV successively
for the simplest scenario with one UAV and two ERs [1]. hovers at a given set of hovering locations (e.g., using the
Notice that in this basic setup, the transferred power from optimal set of hovering locations obtained in the ideal case if
the UAV to the two ERs critically depends on the UAV’s the charging duration is sufficiently large) each for a certain
trajectory or locations at different time. For example, when the duration, and flies with the maximum speed between these
UAV moves from one ER to the other, their received power hovering locations. The total flying time is minimized by
will decrease and increase, respectively, thus resulting in an finding the path with the shortest traveling distance to visit all
interesting power trade-off between them. of these hovering locations. The proposed trajectory is proved
For the purpose of exposition, in this paper we focus on to be optimal for the case of K = 2, and also asymptotically
the UAV-enabled multiuser WPT system with one UAV/ET optimal for K > 2 if the charging duration is sufficiently
and K > 1 ERs, while leaving the general scenario with more large so that the total flying time becomes asymptotically
than one UAVs/ETs for our future work. Under the considered negligible. Furthermore, we also propose a successive con-
setup, we aim to find the optimal UAV trajectory to maximize vex programming (SCP)-based algorithm to obtain a locally
the amount of energy transferred to the K ERs during a optimal solution to the min-energy maximization problem with
finite charging period, subject to the UAV’s maximum speed K > 2. By employing the successive fly-and-hover trajectory
constraint. To our best knowledge, this work is the first that as the initial point, the SCP-based algorithm iteratively refines
explores the UAV’s trajectory design for WPT performance the UAV trajectory to improve the max-min energy of all ERs
optimization. The main results of this paper are summarized until convergence.
as follows. It is worth noting that in the preliminary work [1], we con-
First, we consider the UAV’s trajectory optimization in the sidered the UAV-enabled WPT system in the case with K = 2
horizontal plane with a fixed altitude above the ground to
maximize the sum received energy of all ERs. Despite that 1 Notice that this assumption holds approximately if the charging duration
this problem is non-convex and involves an infinite number of and/or the UAV maximum speed are sufficiently large.
5094 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 17, NO. 8, AUGUST 2018

ERs and characterized the achievable region of the received The total energy received by each ER k ∈ K over the
energy by the two ERs via UAV trajectory optimization; whole charging period is a function of the UAV’s trajectory
while in [2], we studied the min-energy maximization for the {x(t), y(t)}, which can be written as
case with K > 2 ERs. Different from the above two prior  T
studies, this paper studies both the sum-energy and min-energy Ek ({x(t), y(t)}) = Qk (x(t), y(t))dt. (3)
maximization problems for the general case with K ≥ 2 ERs, 0
and provides more comprehensive analysis, simulations, and Note that at each ER, the received RF signals are con-
discussions. verted into direct current (DC) signals for energy harvesting
The remainder of this paper is organized as follows. via rectifiers [7]. In practice, the RF-to-DC conversion is
Section II presents the system model of the UAV-enabled WPT generally non-linear and the conversion efficiency critically
system. Section III presents the optimal solution to the sum- depends on the received RF power and waveform at the ER
energy maximization problem. Section IV presents the optimal (see, [4], [15], [22]). To the authors’ best knowledge, a generic
solution to the min-energy maximization problem when the model to accurately characterize the non-linear RF-to-DC
UAV maximum speed constraint is ignored. Section V presents conversion efficiency is not available yet, but in general the
the proposed solutions to the min-energy maximization prob- harvested DC power monotonically increases with the received
lem with the maximum UAV speed constraint considered. RF power. Therefore, for simplicity, in this paper we consider
Section VI provides numerical results to validate the effective- the received RF power in (2) and the resultant received energy
ness of our proposed trajectory designs. Finally, Section VII in (3) by ERs prior to RF-to-DC conversion as the performance
concludes this paper. metrics.

II. S YSTEM M ODEL

III. S UM -E NERGY M AXIMIZATION
We consider a UAV-enabled multiuser WPT system, where
a UAV is dispatched to deliver wireless energy to K ≥ 2 ERs In this section, we consider the maximization of the sum
located on the ground. Let K  {1, . . . , K} denote the set received energy of all ERs over the charging period, by opti-
of ERs. Each ER k ∈ K has a fixed location on the ground, mizing the UAV’s trajectory {x(t), y(t)} subject to the speed
denoted by (xk , yk , 0) in a 3D Euclidean coordinate, which is constraints in (1). The problem can be expressed as
assumed to be known to the UAV a priori for its trajectory 
(P1) : max Ek ({x(t), y(t)})
design. We consider a finite charging period with duration T , {x(t),y(t)}
k∈K
denoted by T  [0, T ]. At each time instant t ∈ T , the UAV
s.t. (1).
is assumed to fly at a fixed altitude H > 0 above the ground,
whose time-varying location is denoted as (x(t), y(t), H). Problem (P1) involves an infinite number of optimization
We assume that the initial and final UAV locations at time variables, i.e., x(t)’s and y(t)’s over continuous time t.
t = 0 and t = T are not pre-determined, but can be freely Furthermore, (P1) is a non-convex optimization problem as the
optimized.2 Denote by V in meter/second (m/s) the maximum objective function is a non-concave function with respect to
possible speed of the UAV. We then have the maximum speed the trajectory {x(t), y(t)}. Therefore, (P1) is generally difficult
constraint at each time instant expressed as to be optimally solved. In the following, we first present the
 optimal solution to (P1) for the general case with arbitrary
ẋ2 (t) + ẏ 2 (t) ≤ V, ∀t ∈ T , (1)
value of K ≥ 2 in Section III-A, and then consider (P1) for
with ẋ(t) and ẏ(t) denoting the time-derivatives of x(t) and the special case with K = 2 in Section III-B to draw more
y(t), respectively. insights.
We assume that the wireless channel between the UAV
and each ER is LoS-dominated, so that the free-space path
loss model similarly as in [17] and [18] is adopted. At time A. Optimal Solution to Problem (P1)
t ∈ T , the channel power gain from the UAV to ER For the ease of description, given the UAV’s location x(t)
−2
 ∈ K is modeled as hk (t) = β0 dk (t), where dk (t) =
k and y(t) at a given time t, we define the sum-power received
2 2 2
(x(t) − xk ) + (y(t) − yk ) + H is their distance and β0 by all the K ERs as
denotes the channel power gain at a reference distance of 
d0 = 1 m. Assuming that the UAV has a constant transmit ψ(x(t), y(t))  Qk (x(t), y(t))
k∈K
power P , the received RF power by ER k at time t is thus 
given by β0 P
= . (4)
(x(t) − xk )2 + (y(t) − yk )2 + H 2
k∈K
Qk (x(t), y(t)) = hk (t)P
β0 P Accordingly, the sum-energy received by the K ERs over the
= . (2)
(x(t) − xk )2 + (y(t) − yk )2 + H 2 whole charging period is
  T
2 This assumption is made simply for the purpose of exposition, while our
proposed algorithms in Section V can be easily modified to incorporate the Ek ({x(t), y(t)}) = ψ(x(t), y(t))dt. (5)
k∈K 0
initial and/or final location constraints.
XU et al.: UAV-ENABLED WPT: TRAJECTORY DESIGN AND ENERGY OPTIMIZATION 5095

Let x and y  denote an optimal UAV location that maxi- Towards this end, we first re-express the received power by
mizes the function ψ(x, y), i.e., ER 1 and ER 2 as follows, given the UAV’s location (x, 0, H).
(x , y  ) = arg max ψ(x, y). (6) β0 P
x,y Q̂1 (x) = , (9)
(x + D/2)2 + H 2
As the function ψ(x, y) is non-concave with respect to x and β0 P
Q̂2 (x) = . (10)
y, it is generally difficult to find the closed-form expression (x − D/2)2 + H 2
of x and y  . Fortunately, ψ(x, y) in problem (6) only has Accordingly, the function ψ(x, y) in (4) can be simplified as
two variables x and y. Besides, it is not difficult to show that
2

x and y  should satisfy x ≤ x ≤ x and y ≤ y  ≤ y,
respectively, where ψ̂(x)  Q̂k (x)
k=1
 
x = min xk , x = max xk , y = min yk , y = max yk . (7) 1 1
k∈K k∈K k∈K k∈K = β0 P + . (11)
(x+D/2)2 + H 2 (x − D/2)2 +H 2
This is because if (x , y  ) lies outside the box specified above,
As a result, finding x is equivalent to determining the
we can always increase the energy transferred to all the K ERs
maximizer of ψ̂(x), i.e., x = arg maxx ψ̂(x).
by moving (x , y  ) into the box. As a result, we can adopt
Lemma 2: The function ψ̂(x) has the following properties:
a 2D exhaustive search over the box region [x, x] × [y, y] to
• It is symmetric over x = 0, i.e., ψ̂(−x) = ψ̂(x),
find (x , y  ).3 Notice that the optimal solution of x and y 
to problem (6) is generally non-unique. ∀x ∈ (−∞, ∞). √
• When D ≤ 2H/ 3, ψ̂(x) is monotonically increasing
Given x and y  , we have the following proposition.
Proposition 1: The optimal trajectory solution to prob- over x ∈ (−∞, 0) and decreasing over x ∈ (0, ∞);
lem (P1) is given as hence, x = 0 is √the unique maximizer of ψ̂(x).
• When D > 2H/ 3, there exists a value
x (t) = x , y  (t) = y  , ∀t ∈ T . (8)  
ξ  −(D2 /4 + H 2 ) + D4 /4 + H 2 D2 < D/2,
Proof: See Appendix A.
(12)
Proposition 1 indicates that the UAV should hover at one
single fixed location (x , y  , H) during the whole charging such that ψ̂(x) is monotonically increasing, decreasing,
period, referred to as single-location hovering. Due to the non- increasing and decreasing over x ∈ (−∞, −ξ), (−ξ, 0),
uniqueness of the optimal solution x and y  to problem (6), (0, ξ) and (ξ, ∞), respectively; hence, together with the
such an optimal hovering location (x , y  , H) is non-unique symmetry of the function ψ̂(x) over x = 0, it follows that
in general, as will be shown in an example given in the x = −ξ and x = ξ are the two equivalent maximizers
next subsection for the case of K = 2. However, this of ψ̂(x).
single-location-hovering solution can lead to a severe “near- Proof: See Appendix B.
far” fairness issue in multiuser WPT, as the near ERs in Based on Lemma 2, the optimal solution x to maximize
close proximity to the optimal hovering location can receive ψ̂(x) is found. By using this result together with Proposition 1,
significantly more energy than the far ERs, especially in a we have the following proposition to solve (P1) for K = 2.
large network with many ERs that are sufficiently separated Proposition 3: In the special case of K = 2, the optimal
from each other. To overcome this issue, in Section IV we solution to (P1) is given as follows.
will consider an alternative problem formulation to maximize √  
• When D ≤ 2H/ 3, we have x (t) = 0, y (t) =
the minimum received energy among all ERs to ensure their 0, ∀t ∈ T , i.e., the UAV should hover at the fixed location
fairness. Before that, in the following subsection, we derive (0, 0, H) above the middle point between the two ERs
the optimal solution to problem (P1) in closed-form for the during the whole charging
special case of K = 2 ERs to provide more insights on the √ period.
• When D > 2H/ 3, there are two symmet-
sum-energy maximization problem. ric optimal solutions to (P1), given by x (t) =
−ξ, y  (t) = 0, ∀t ∈ T , and x (t) = ξ, y  (t) = 0,
B. Special Case With K = 2 ERs ∀t ∈ T , respectively. In other words, the UAV should
In the special case with K = 2, without loss of generality hover at either (−ξ, 0, H) near ER 1 or (ξ, 0, H) near
we assume x1 = −D/2, x2 = D/2, and y1 = y2 = 0, ER 2 during the whole charging period.
where D denotes the distance between the two ERs. Based on For illustration, Fig. 2 shows the optimal hovering location
Proposition 1, in this case, the UAV should hover at a fixed x , with x (t) = x , ∀t ∈ T , versus the distance D between
location (x , y  , H) above the line between the two ERs with the two ERs in the case of√H = 5 m. It is observed that
y  = 0. Therefore, it only remains to find x . x = 0 when D ≤ 2H/ 3 = 5.77 m and 0 < x <
D/2 (or equivalently −D/2 < −x < 0 for the other √
3 Actually, (x , y  ) should lie within the convex hull of all the ERs’ symmetric optimal hovering location) when D > 2H/ 3.
locations (x1 , y1 ), . . . , (xK , yK ), which is generally a subset of the box It is also observed that as D increases, x becomes closer
[x, x] × [y, y]. Furthermore, denoting the solution accuracy as ε, then the
complexity of the 2D exhaustive search method increases inversely with and eventually converges to D/2. These observations are con-
respect to ε2 and linearly with respect to the number of ERs, K. sistent with Proposition 3. Furthermore, notice that it follows
5096 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 17, NO. 8, AUGUST 2018

A. Optimal Solution to Problem (P3)

Problem (P3) can be equivalently expressed in the following
problem by introducing an auxiliary variable E.
(P3.1) : max E
{x(t),y(t)},E
s.t. Ek ({x(t), y(t)}) ≥ E, ∀k ∈ K. (13)
Though problem (P3.1) is non-convex, it can be shown that
it satisfies the so-called time-sharing condition in [24]. There-
fore, the strong duality holds between (P3.1) and its Lagrange
dual problem. As a result, we can optimally solve (P3.1) by
using the Lagrange dual method [23].
Let λk ≥ 0, k ∈ K, denote the dual variable associated with
the constraint in (13) for the kth ER. Then the Lagrangian
Fig. 2. The optimal hovering location x versus the distance D between the associated with (P3.1) is given as
two ERs in the case of H = 5 m.
L({x(t), y(t)}, E, {λk })
  T 

from (12) that limD→∞ x = limD→∞ ξ = D/2, which can 
= 1− λk E + λk Qk (x(t), y(t))dt. (14)
be intuitively explained as follows. When the distance between 0
k∈K k∈K
the two ERs is sufficiently large, the UAV should hover above
one of the two ERs to maximize the transferred energy to Accordingly, the dual function of (P3.1) is given by
it, which in turn maximizes the sum-energy transferred to f ({λk }) = max L({x(t), y(t)}, E, {λk }), (15)
both ERs. In this case, the energy transferred to the other {x(t),y(t)},E
ER becomes negligible. This thus leads to a severe near-far for which the following lemma holds.
fairness issue between the two ERs. Lemma 4: In order for the dual function f ({λk }) to be
upper-bounded

from above (i.e., f ({λk }) < ∞), it must hold
IV. M IN -E NERGY M AXIMIZATION W ITHOUT
that k∈K λk = 1.

UAV S PEED C ONSTRAINT

Proof: Suppose that k∈K λk > 1 (or k∈K λk < 1).
To overcome the aforementioned fairness issue in the sum- Then by setting E → −∞ (or E → ∞), we have f ({λk }) →
energy maximization, we consider an alternative performance ∞. Therefore, this lemma is proved.
metric, namely the min-energy maximization. Specifically, Based on Lemma 4, the dual problem of (P3.1) is given by
we maximize the minimum received energy among all the K
ERs via optimizing the UAV’s trajectory {x(t), y(t)}, subject (D3.1) : min f ({λk })
{λk }
to the maximum speed constraints in (1). In general, this 
problem is formulated as s.t. λk = 1 (16)
k∈K
(P2) : max min Ek ({x(t), y(t)}) λk ≥ 0, ∀k ∈ K. (17)
{x(t),y(t)} k∈K
s.t. (1). Then, we can solve problem (P3.1) by equivalently solving
Problem (P2) is a non-convex optimization problem, and it its dual problem (D3.1). Let the feasible set of {λk } character-
is more difficult to solve than the sum-energy maximization ized by the constraints in (16) and (17) as X . In the following,
problem (P1). In particular, the single-location-hovering opti- we first solve problem (15) to obtain f ({λk }) under any given
mal solution of (P1) given in Proposition 1 is no longer valid feasible dual variables {λk } ∈ X , then solve (D3.1) to find
for (P2). the optimal {λk } to minimize f ({λk }), and finally construct
To tackle (P2), in this section we first consider an ideal the optimal primal solution to (P3.1).
case by ignoring the UAV speed constraints in (1) and 1) Obtaining f ({λk }) via Solving Problem (15) Under
solve the relaxed problem optimally. In practice, the speed Given {λk } ∈ X : For any given {λk } ∈ X , problem (15)
constraints in (1) can be ignored if the charging duration T can be decomposed into the following subproblems.
and/or the maximum UAV speed V are sufficiently large (see 
max(1 − λk )E (18)
Proposition 8 for a more rigorous argument). For ease of E
k∈K
presentation, we rewrite problem (P2) without the constraint 
{λk }
(1) in the following problem, denoted by (P3). max ψ̃ (x(t), y(t))  λk Qk (x(t), y(t)), ∀t ∈ T
x(t),y(t)
k∈K
(P3) : max min Ek ({x(t), y(t)}) (19)
{x(t),y(t)} k∈K

In Section V, we will consider the general case of (P2) with In the above, (19) consists of an infinite number of subprob-
the UAV speed constraints (1) included, and propose efficient lems, each corresponding to a time instant t. Let the optimal
solutions to (P2) based on the optimal solution obtained solutions to (18) and (19) be denoted by E {λk } , as well as
for (P3). x{λk } (t) and y {λk } (t), ∀t ∈ T , respectively.
XU et al.: UAV-ENABLED WPT: TRAJECTORY DESIGN AND ENERGY OPTIMIZATION 5097

∗ ∗ ∗
As for subproblem (18), since 1 − k∈K λk = 0 holds for when the optimal x{λk } (t)’s, y {λk } (t)’s, and E {λk } to prob-
any given {λk } ∈ X , its objective value is always zero. In this lem (15) are non-unique, they may not be feasible nor optimal
case, we can choose any arbitrary real number as the optimal to problem (P3.1) in general. In the latter case, we need
solution E ({λk }) for the purpose of obtaining the dual function to time-share among these non-unique optimal solutions as
f ({λk }). follows to construct the optimal primal solution {x∗ (t)},
On the other hand, note that the subproblems in (19) are {y ∗ (t)}, and E ∗ to (P3.1).
identical for all time instant t ∈ T . Therefore, we can drop Under the optimal dual solution {λ∗k }, suppose that prob-
the time index t and re-express each problem in (19) as lem (20) under {λ∗k } has a total number of Γ ≥ 1 opti-
∗
mal location solutions to maximize ψ {λk } (x, y), denoted by
max ψ̃ {λk } (x, y). (20) (x∗1 , y1∗ ), . . . , (x∗Γ , yΓ∗ ), which are obtained via a 2D exhaustive
x,y
search over the box region [x, x] × [y, y]. Let Qk (x∗γ , yγ∗ )
Similarly as for problem (6), problem (20) has two optimiza- denote the corresponding received power at each ER k ∈ K
tion variables, and the optimal solution of x{λk } and y {λk } when the UAV stays at the location (x∗γ , yγ∗ , H). Due to the
satisfies x ≤ x{λk } ≤ x and y ≤ y {λk } ≤ y, with x, x, y, zero duality gap between (P3.1) and (D3.1), it is evident
and y given in (7). As a result, we can adopt a 2D exhaustive that for any time t ∈ [0, T ], the optimal primal solution of
search over the box region [x, x] × [y, y] to find the optimal (x∗ (t), y ∗ (t)) must be chosen from the Γ hovering locations.
x{λk } and y {λk } . Accordingly, the optimal solution to problem Notice that when the UAV hovers at the same location at
(19) is given by different time t, the ERs will receive the same amount of
x{λk } (t) = x{λk } , y {λk } (t) = y {λk } , ∀t ∈ T . (21) power. Therefore, we only need to decide the time-sharing
ratio among the Γ solutions for constructing the optimal primal
Note that the optimal solution of x{λk } and y {λk } to (20) is solution to (P3.1). Here, time-sharing means that the UAV
generally non-unique, and we can arbitrarily choose any one of should hover at each of these different locations for a certain
them for obtaining the dual function f ({λk }).4 By substituting portion of the total duration T . Let τγ denote the optimal
E {λk } , x{λk } (t)’s and y {λk } (t)’s into problem (15), the dual hovering duration at (x∗γ , yγ∗ , H). Then, the optimal τγ∗ ’s,
function f ({λk }) is obtained. together with the maximum min-energy E ∗ , can be obtained
2) Finding Optimal Dual Solution to (D3.1): With f ({λk }) by solving the following problem.
obtained, we then solve the dual problem (D3.1) to find
max E
the optimal {λk } to minimize f ({λk }). Note that the {τγ ≥0},E
dual function f ({λk }) is always convex but generally non- Γ

differentiable [23]. As a result, problem (D3.1) can be s.t. τγ Qk (x∗γ , yγ∗ ) ≥ E, ∀k ∈ K
solved by subgradient-based methods such as the ellipsoid γ=1
method [25]. Note that the subgradient of the objective func- Γ

tion f ({λk }) is given by τγ = T. (22)
γ=1
s0 (λ1 , . . . , λK )
  Note that problem (22) is a linear program (LP), which
= T Q1 (x({λk }) , y ({λk }) ), . . . , T QK (x({λk }) , y ({λk }) ) , can be solved by using standard convex optimization tech-
niques [23]. As a result, the optimal solution of E ∗ to (P3.1)
where E ({λk }) = 0 is chosen for simplicity. Furthermore, is found. Finally, we obtain the optimal trajectory solution
the equality constraint

in (16) can be viewed as
two inequality of {x∗ (t), y ∗ (t)} to (P3.1) (and thus (P3)), which is given
constraints 1 − k∈K λk ≤ 0 and −1 + k∈K λk ≤ 0, in the following proposition based on the above time-sharing
whose subgradients are given by s1 (λ1 , . . . , λK ) = −e and property, with the proof omitted for brevity.
s2 (λ1 , . . . , λK ) = e, respectively, where e denotes an all-one Proposition 5: Partition the whole charging period
vector. We denote the obtained optimal dual solution to (D3.1)
as {λ∗k }.
into Γ
portions,
γ−1
γ denoted by T1 , . . . , TΓ , where
Tγ = ( i=1 τi∗ , i=1 τi∗ ] with duration τγ∗ for γ ≥ 1.
3) Constructing Optimal Primal Solution to (P3.1): Based Then, the optimal trajectory solution of {x∗ (t), y ∗ (t)}
on the optimal dual solution {λ∗k } to (D3.1), we need to obtain to (P3.1) or (P3) is given by
the optimal primal solution to (P3.1), denoted by {x∗ (t)},
{y ∗ (t)}, and E ∗ . It is worth noting that when using the x∗ (t) = x∗γ , y ∗ (t) = yγ∗ , ∀t ∈ Tγ , γ ∈ {1, . . . , Γ}, (23)
Lagrange dual method to solve problem (P3.1) via the dual Γ
problem (D3.1), the optimal solution to problem (15) under where Tγ ∩ Tζ = φ, ∀γ
= ζ, and γ=1 Tγ = T .
∗ ∗
the optimal dual solution {λ∗k } (i.e., x{λk } (t)’s, y {λk } (t)’s, In summary, we present the algorithm for
∗
and E {λk } ) is the optimal primal solution to (P3.1), if such a solving (P3.1) or (P3) as Algorithm 1 in Table I.
solution is unique and primal feasible [23]. On the other hand, Remark 6: Note that Proposition 5 implies that to maximize
the min-energy transferred to the K ERs, the UAV should
4 When the optimal solution x{λk } (t)’s, y {λk } (t)’s, and E ({λk }) are not hover above a number of fixed locations during the charging
unique, they may not be optimal for the primal problem (P3.1) after the dual period, and the optimal hovering locations (i.e., x∗γ ’s and yγ∗ ’s)
problem is solved. As a result, an additional step is required to obtain the
optimal primal solution of E, x(t)’s, and y(t)’s to (P3.1), as will be shown are generally different from the locations of the ERs (i.e., xk ’s
later in Section IV-A.3. and yk ’s). We refer to such a design as multi-location hovering.
5098 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 17, NO. 8, AUGUST 2018

TABLE I V. M IN -E NERGY M AXIMIZATION W ITH

A LGORITHM 1: A LGORITHM FOR S OLVING P ROBLEM (P3.1) OR (P3) UAV S PEED C ONSTRAINT
In this section, we consider the general min-energy maxi-
mization problem (P2) by including the practical UAV maxi-
mum speed constraints in (1). This problem is difficult to be
solved globally optimally in general with K > 2. To tackle
this problem, we propose two suboptimal solutions inspired
by the optimal solution obtained previously for problem (P3)
in the ideal case without the UAV maximum speed constraint.

A. Successive Hover-and-Fly Trajectory

Design for Problem (P2)
In this subsection, we propose a successive hover-and-
fly trajectory design to solve problem (P2) based on the
optimal solution obtained for (P3) in the ideal case. Recall
Different from the single-location hovering for sum-energy that the optimal solution to (P3) corresponds to Γ optimal
maximization, the result here shows that the UAV should in hovering locations, i.e., {(x∗γ , yγ∗ , H)}Γγ=1 . In the proposed
general hover over different locations so as to balance the trajectory design with the maximum speed constraint, the UAV
energy transferred to all ERs. sequentially hovers at each of these locations for a certain
duration and flies from one location to another with the
B. Special Case With K = 2 ERs maximum speed V . As a result, to find the optimal successive
hover-and-fly trajectory, we need to first determine the UAV’s
To provide more insights, in this subsection we consider
traveling path to visit all the Γ locations with the minimum
the min-energy maximization in the special case with K = 2
flying distance so as to minimize the total flying time, and
ERs, by assuming x1 = −D/2, x2 = D/2, and y1 = y2 = 0,
then optimize the hovering time at each of these locations for
similarly as in Section III-B. In this case, by setting y(t) =
the remaining time in the charging duration.
0, ∀t ∈ K, problem (P3) can be simplified as
1) Flying Distance Minimization to Visit Γ Hovering
(P3-2ER) : max min(Ê1 ({x(t)}), Ê2 ({x(t)})). (24) Locations: First, we determine the UAV’s traveling path
{x(t)}
to visit all the Γ hovering locations with the mini-
Proposition 7: The optimal solution to problem (P3-2ER) mum
 flying distance. For ease of description, let dγ,ζ 
is given as follows by considering two different cases:
√ (xγ − x∗ζ )2 + (yγ∗ − yζ∗ )2 denote the distance between the
∗

• If D ≤ 2H/ 3, then the UAV should hover at the γth hovering location (x∗γ , yγ∗ , H) and the ζth hovering loca-
fixed location (0, 0, H) during the whole charging period, tion (x∗ζ , yζ∗ , H). We define a binary variable fγ,ζ for any
i.e., x∗ (t) = 0,
√∀t ∈ T . γ, ζ ∈ {1, . . . , Γ}, γ
= ζ, where fγ,ζ = 1 indicates that the
• If D > 2H/ 3, then the UAV should hover over the UAV should fly from the γth hovering location (x∗γ , yγ∗ , H)
two symmetric locations (−ξ, 0, H) and (ξ, 0, H) with to the ζth hovering location (x∗ζ , yζ∗ , H), and fγ,ζ = 0 other-
equal durations, e.g., x∗ (t) = −ξ, ∀t ∈ [0, T /2), and wise. The trajectory design
x∗ (t) = ξ, ∀t ∈ [T /2, T ], with ξ given in (12).
Γ problem

Γ thus becomes determining
{fγ,ζ } to minimize γ=1 ζ=1,ζ=γ fγ,ζ dγ,ζ , provided that
Proof: See Appendix C. each of the Γ locations is visited exactly once.
By comparing Proposition 7 versus Proposition 3, it is The flying distance minimization problem considered here
observed that in the case with K = 2 ERs, each of the optimal is reminiscent of the well-known traveling salesman prob-
hovering locations for the min-energy maximization is also lem (TSP) (see, [26]–[28]), with the following difference.
optimal for the corresponding case in the sum-energy maxi- In the standard TSP, the salesman (or equivalently the UAV
mization. Thus, the achieved sum-energy in problem (P3-2ER) of our interest) needs to return to the origin city (the initial
is identical to that in (P1) when K = 2. This implies that with- hovering location) after visiting all these cities (or hovering
out the maximum UAV speed constraint, the multi-location- locations here); but our flying distance minimization problem
hovering solution to (P3-2ER) can maximize the sum-energy does not have such a requirement since the initial and final
received by the two ERs, while ensuring their energy fairness hovering locations can be optimized. Fortunately, it has been
thanks to the time sharing. Nevertheless, it is worth noting shown in [29] that our flying distance minimization problem
that in the general case with K > 2, the optimal hovering can be transformed to the standard TSP as follows. First,
locations for (P3) are generally different from that for (P1). we add a dummy hovering location, namely the (Γ + 1)-
This is due to the fact that x and y  are the maximizer of the th hovering location, whose distances to all the existing Γ
sum-power ψ(x, y), but x∗γ ’s and yγ∗ ’s are the maximizers of hovering locations are 0, i.e., dΓ+1,γ = dγ,Γ+1 = 0, ∀γ ∈
the weighted sum-power ψ̃ {λk } (x(t), y(t)), which is different {1, . . . , Γ}. Note that this dummy hovering location is a virtual
from ψ(x, y) if any two λk ’s are not identical in the case of node that does not exist physically. Then, we obtain the
K > 2. desirable traveling path by solving the standard TSP problem
XU et al.: UAV-ENABLED WPT: TRAJECTORY DESIGN AND ENERGY OPTIMIZATION 5099

for the Γ + 1 hovering locations,5 and then removing the two Proposition 8: When the charging duration T is sufficiently
edges associated with the dummy location. For the obtained large such that T  Tfly , the successive hover-and-fly trajec-
traveling path, we define the permutation π(·) over the set tory design is asymptotically optimal for problem (P2).
{1, . . . , Γ}, such that the UAV first visits the π(1)-th hovering Proof: When T  Tfly , the flying time is negligible and
location, followed by the π(2)-th, the π(3)-th, until the π(Γ)-th thus the successive hover-and-fly trajectory is equivalent to
hovering location at last. In this case, the resulting flying dis- the optimal multi-location-hovering solution to (P3). In this
tance and flying

Γ−1duration with the maximum speed V are given case, the objective value achieved by the successive hover-
as Dfly = γ=1 dπ(γ),π(γ+1) , and Tfly = Dfly /V , respectively. and-fly trajectory for (P2) is asymptotically approaching the
Tfly optimal value of (P3), which actually serves as the upper
We denote the corresponding trajectory as {x̂(t), ŷ(t)}t=0 .
It is worth noting that the above traveling path is only bound for that of (P2). Therefore, the proposed trajectory
feasible when the charging duration T is no smaller than Tfly , design is asymptotically optimal for (P2) when T  Tfly .
i.e., T ≥ Tfly , since otherwise the charging duration is not suf- 3) Trajectory Redesign When T < Tfly : In this subsection,
ficient for the UAV to visit all the Γ hovering locations. In the we consider the case when T < Tfly . In this case, the UAV
Tfly
following, we first determine the hovering time allocation over traveling path {x̂(t), ŷ(t)}t=0 based on the TSP solution is
different locations in the case with T ≥ Tfly , and then refine no longer feasible since the charging time is not sufficient for
the trajectory design in the case with T < Tfly . the UAV to visit all the Γ hovering locations. To address this
2) Hovering Time Allocation When T ≥ Tfly : First, we con- issue, we first find the solution to (P2) when T is sufficiently
sider the case when T ≥ Tfly . With the above traveling small (i.e., T → 0) such that the UAV can only hover at
Tfly one single location, and then reconstruct a modified successive
path {x̂(t), ŷ(t)}t=0 , the trajectory design problem remains to
allocate the hovering duration T − Tfly among the Γ locations hover-and-fly trajectory for the case of T < Tfly .
to maximize the min-energy transferred to all the K ERs. Note First, when T → 0, the UAV should hover at one single
Tfly fixed location, denoted by (xfix , yfix , H), where xfix and yfix
that based on the traveling path {x̂(t), ŷ(t)}t=0 , we can obtain
the received energy by each ER k ∈ K during the UAV’s can be obtained by solving the following problem via a 2D
T exhaustive search over (x, x) × (y, y).
flying time as Ekfly = 0 fly Qk (x̂(t), ŷ(t))dt, with Qk (·, ·)
given in (2). Also, recall that Qk (x∗γ , yγ∗ ) denotes the received (xfix , yfix ) = arg max min Qk (x, y). (26)
power at ER k ∈ K when the UAV hovers at the location x,y k∈K

(x∗γ , yγ∗ , H). Then the optimal hovering durations, denoted as Next, we reconstruct the trajectory as follows by
τγ∗∗ ’s, together with the corresponding maximum min-energy down-scaling the previously obtained traveling path
Tfly
of the K ERs, denoted by E ∗∗ , can be obtained by solving {(x̂(t), ŷ(t), H)}t=0 for the case of T = Tfly linearly
the following LP. towards the center point (xfix , yfix , H), such that the resulting
total flying distance equals V T .
max E
{τγ ≥0},E x∗∗ (t) = x̂(t/κ) + (1 − κ)(xfix − x̂(t/κ)),
Γ
 y ∗∗ (t) = ŷ(t/κ)+(1 − κ)(yfix − ŷ(t/κ)), ∀t ∈ [0, T ], (27)
s.t. τγ Qk (x∗γ , yγ∗ ) + Ekfly ≥ E, ∀k ∈ K
γ=1 where κ = T /Tfly < 1 denotes the linear scaling factor.
Γ
 Note that when T → 0, we have κ → 0, and the above
τγ = T − Tfly . (25) redesigned trajectory reduces to hovering at one single fixed
γ=1 location (xfix , yfix , H); while T → Tfly , we have κ → 1,
and the above redesigned trajectory becomes identical to the
Tfly
With the optimal permutation π(·) and the optimal hov- TSP-based trajectory {(x̂(t), ŷ(t), H)}t=0 .
ering durations {τγ∗∗ } obtained, the successive hover-and-fly We summarize the overall algorithm for the proposed
trajectory is finalized, which can be summarized as follows. successive hover-and-fly trajectory design as Algorithm 2 in
By dividing the charging period into 2Γ − 1 slots; in the Table II, for both the cases of T ≥ Tfly and T < Tfly .
∗∗
(2γ − 1)-th slot with duration τπ(γ) , γ ∈ {1, . . . , Γ}, the UAV 4) Optimality in the Case With K = 2 ERs: In the
hovers at the π(γ)-th hovering location (x∗π(γ) , yπ(γ) ∗
, H); and special two-ER case with x1 = −D/2, x2 = D/2, and
in the (2γ)-th slot, γ ∈ {1, . . . , Γ − 1}, the UAV flies from the y1 = y2 = 0, the min-energy maximization problem with the
π(γ)-th hovering location (x∗π(γ) , yπ(γ)
∗
, H) to the π(γ + 1)-th maximum speed constraint is simplified as follows by setting
hovering location (x∗π(γ+1) , yπ(γ+1)
∗
, H) with its maximum y(t) = 0, ∀t ∈ T .
speed V . (P2-2ER) : max min(Ê1 ({x(t)}), Ê2 ({x(t)}))
{x(t)}

5 Note that although the TSP is an NP-hard problem in combinatorial opti- s.t. |ẋ(t)| ≤ V, ∀t ∈ T .
mization, various heuristic and approximation algorithms have been proposed
to give efficient high-quality solutions for it (see, [26]–[28]). In particular, As it has been shown in Proposition 7 that for problem
it has been shown in [28] that the TSP problem can be formulated as a binary (P3-2ER)
√ without the maximum speed constraint, if D ≤
integer program, by incorporating a set of constraints to ensure there is only 2H/ 3, then there√ is one optimal hovering location (0, 0, H);
a single tour connecting all visited locations. The binary integer program can
be solved via CVX [30] by using the Mosek solver that supports the integer while if D > 2H/ 3, then there are two symmetric optimal
program (see https://mosek.com/ for details). hovering locations (−ξ, 0, H) and (ξ, 0, H) with equal time
5100 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 17, NO. 8, AUGUST 2018

TABLE II insufficient charging time. On the other hand, if the charging

A LGORITHM 2: S UCCESSIVE H OVER - AND -F LY T RAJECTORY duration is sufficiently long (i.e., T > 2ξ/V ), then the UAV
D ESIGN FOR S OLVING P ROBLEM (P2)
should hover at the two symmetric locations (−ξ, 0, H) and
(ξ, 0, H) with equal time and travel from one location to the
other with the maximum speed. The optimal trajectory in this
case is different from that for sum-energy maximization in
Proposition 3, where the UAV should hover at only one fixed
location (−ξ, 0, H) or (ξ, 0, H) for all the time. Therefore,
the fairness is achieved by the UAV traveling between different
hovering locations, though a certain loss in the total energy
transferred to the two ERs is incurred.

B. SCP-Based Trajectory Design for Problem (P2)

In this subsection, we develop an alternative SCP-based
algorithm to find a locally optimal solution to problem (P2).
allocation. By applying this result in Algorithm 2, the corre- Note that the SCP-based trajectory design has been stud-
sponding successive hover-and-fly trajectory for (P2-2ER) is ied for throughput or energy efficiency maximization for
obtained as follows by considering three different cases: UAV-enabled wireless communication systems [18], [19], but
√
• If D ≤ 2H/ 3, then the UAV should hover at the the results cannot be directly applied for UAV-enabled WPT
fixed location (0, 0, H) above the middle point between systems considered in this paper. With the SCP-based trajec-
the two ERs during the whole charging period, i.e., tory design, we first discretize the whole charging duration
x∗ (t) = 0, ∀t ∈ √
T. into a finite number of N time slots, each with duration
• If D > 2H/ 3 and T ≤ 2ξ/V = Tfly with ξ Δ = T /N . Note that the duration Δ is chosen to be suffi-
given in (12), then the UAV should fly from the loca- ciently small, such that we can assume that the UAV location
tion (−V T /2, 0, H) to (V T /2, 0, H) with the maximum is approximately unchanged during each slot n, which is
speed V , i.e., x∗∗ denoted as (x[n], y[n], H), n ∈ N  {1, . . . , N }. In this case,
√(t) = −V T /2 + V t, ∀t ∈ T .
• If D > 2H/ 3 and T > 2ξ/V , then the UAV the received energy by each ER k ∈ K at slot n is given by
should follow a successive hover-and-fly trajectory: first,
β0 P Δ
the UAV hovers at the location (−ξ, 0, H) for the duration Êk (x[n], y[n]) = . (29)
t ∈ [0, T /2 − ξ/V ]; next, it flies from (−ξ, 0, H) to (x[n] − xk )2 + (y[n] − yk )2 + H 2
(ξ, 0, H) with the maximum speed V during the time Accordingly, the min-energy maximization problem (P2)
interval t ∈ (T /2 − ξ/V, T /2 + ξ/V ); finally, the UAV over the continuous trajectory {x(t), y(t)} can be refor-
hovers at the location (ξ, 0, H) for the remaining time mulated as follows over the discretized trajectory variables
t ∈ [T /2 + ξ/V, T ]. In other words, the optimal UAV {x[n], y[n]}N
n=1 .
trajectory is
N

x∗∗ (t) max min Êk (x[n], y[n]) (30)
⎧ {x[n],y[n]} k∈K
⎪
⎨−ξ, t ∈ [0, T /2 − ξ/V ] n=1

V t − V T /2, t ∈ (T /2 − ξ/V, T /2 + ξ/V ) (28) s.t. (x[n] − x[n − 1])2 + (y[n] − y[n − 1])2 ≤ V 2 Δ2 ,
⎪
⎩
ξ, t ∈ [T /2 + ξ/V, T ]. ∀n ∈ {2, . . . , N }, (31)
Proposition 9: The above successive hover-and-fly trajec- where the constraints in (31) correspond to the discretized
tory solution is optimal for problem (P2-2ER) in the case of version of the maximum speed constraints in (1). Note that
K = 2 ERs. the constraints in (31) are all convex but the objective function
Proof: See Appendix D. in (30) is not concave. Therefore, problem (30) is a non-convex
Remark 10: Proposition 9 provides important insights on optimization problem.
how to maximize the minimum or equal energy transferred For the non-convex optimization problem (30), we obtain a
to the two ERs. First,
√ when the two ERs are close to each locally optimal solution by proposing an SCP-based algorithm,
other with D ≤ 2H/ 3, the UAV hovers at one fixed location which is operated in an iterative manner to successively
(0, 0, H) during the whole charging period, and this solution is maximize a lower bound of the objective function in (30)
also optimal for problem (P1) to maximize the sum-energy of at each iteration. Particularly, let {x(0) [n], y (0) [n]} denote the
the two ERs (see Proposition 3). Next, consider the case
√ when initial trajectory and {x(i) [n], y (i) [n]} the obtained trajectory
the two ERs are located farther apart with D > 2H/ 3. If the after iteration i ≥ 1. We have the following lemma.
charging duration is short (i.e., T ≤ 2ξ/V ), then the UAV Lemma 11: For any given {x(i) [n], y (i) [n]}, i ≥ 0, it fol-
should keep flying at its maximum speed from one ER to the lows that
other by following a symmetric trajectory around the middle
(i)
point (0, 0, H), without hovering over any of them due to the Êk (x[n], y[n]) ≥ Êk (x[n], y[n]), ∀k ∈ K, n ∈ N , (32)
XU et al.: UAV-ENABLED WPT: TRAJECTORY DESIGN AND ENERGY OPTIMIZATION 5101

where
(i)
Êk (x[n], y[n])
2β0 P Δ

(x(i) [n] − xk )2 + (y (i) [n] − yk )2 + H 2
β0 P Δ((x[n] − xk )2 + (y[n] − yk )2 + H 2 )
− . (33)
((x(i) [n] − xk )2 + (y (i) [n] − yk )2 + H 2 )2

The inequalities in (32) are tight for x[n] = x(i) [n] and
y[n] = y (i) [n], i.e.,
(i)
Êk (x(i) [n], y (i) [n]) = Êk (x(i) [n], y (i) [n]), ∀k ∈ K, n ∈ N .
(34)

Proof: See Appendix E. Fig. 3. The average received power with the sum-energy maximization versus
Based on Lemma 11, at each iteration i + 1, we opti- the distance D between the two ERs.
mize over {x[n], y[n]} by replacing Êk (x[n], y[n])’s in prob-
(i) A. Sum-Energy Maximization
lem (30) with their respective lower bounds Êk (x[n], y[n])
in (33). More specifically, the discretized trajectory is This subsection evaluates the performance of our pro-
updated as posed optimal solution for the sum-energy maximization prob-
lem (P1). First, we consider the case with K = 2 ERs, where
{x(i+1) [n], y (i+1) [n]} x1 = −D/2, x2 = D/2, and y1 = y2 = 0, with D denoting
N
 the distance between the two ERs. √ Note that Proposition 3
(i)
= arg max min Êk (x[n], y[n]), indicates that when D > 2H/ 3 = 5.77 m, there are
{x[n],y[n]} k∈K
n=1 two optimal single-location-hovering solutions to (P1). In the
s.t. (31). (35) simulation, we choose the one with x (t) = −ξ, y  (t) = 0,
(i)
∀t ∈ T (with ξ given in (12)), such that the single hovering
Note that the function Êk (x[n], y[n]) in (33) is jointly con- location is closer to ER 1 than ER 2. Fig. 3 shows the√average
cave with respect to x[n] and y[n], and therefore, the objective received power versus D. When 0 ≤ D ≤ 2H/ 3, it is
function in problem (35) is jointly concave with respect to observed that the average received power by ER 1 or ER 2 is
{x[n], y[n]}. As a result, problem (35) is a convex optimiza- identical. This is because in this case, the optimal solution
tion problem, and thus can be optimally solved by standard is obtained by letting the UAV hover above the middle point
convex optimization techniques such as the interior point between the two ERs during the whole charging period √ (see
method [23]. Furthermore, as shown in Lemma 11, the objec- Proposition 3). On the other hand, when D > 2H/ 3, it is
tive function in problem (35) serves as a lower bound for that observed that as D increases, the average received power by
in problem (30). Therefore, after each iteration i, the objec- ER 1 (the ER closer to the UAV) increases, while that by ER 2
tive function of problem (30) achieved by {x(i) [n], y (i) [n]} (the ER farther away from the UAV) decreases. This is due to
monotonically increases [18]. As problem (30) has a finite the fact that as D increases in this case, the optimal hovering
optimal value, the SCP-based algorithm in (35) will con- location (selected in this example) becomes closer to ER 1.
verge to a locally optimal solution to problem (30) in Accordingly, the average sum received power of the two ERs
general. is dominated by the received power of ER 1, and the near-far
It is worth noting that the performance of the SCP-based fairness issue becomes more severe as D becomes larger.
algorithm depends on the choice of the initial trajectory Next, we consider a UAV-enabled WPT system with
{x(0) [n], y (0) [n]}. Here, we choose the discretized version of K = 10 ERs, whose locations are shown in Fig. 4. In this fig-
the proposed successive hover-and-fly trajectory obtained in ure, the green triangle indicates the optimal hovering location
Algorithm 2 as {x(0) [n], y (0) [n]}. In this case, the SCP-based for sum-energy maximization (i.e., (x , y  , H) given in (6)).
trajectory design can always achieve a performance no worse It is observed that this hovering location is close to ERs 7-10,
than the successive hover-and-fly trajectory design, as will be but far away from other ERs, especially ER 1. Fig. 5 shows the
validated by the numerical results later. corresponding average received power by each individual ER.
It is observed that ERs 7-10 receive much higher energy than
VI. N UMERICAL R ESULTS the other ERs, which illustrates the near-far fairness issue in
the sum-energy maximization for this multiuser WPT system
In this section, we provide numerical results to evaluate with more than two ERs.
the performance of our proposed trajectory designs. In the
simulation, we set β0 = −30 dB, H = 5 m, and P = 40 dBm.
For all simulations given below, we consider the average B. Min-Energy Maximization
received power by the ER, which is obtained by normalizing In this subsection, we evaluate the performance of our
the total received energy by the charging duration T . proposed trajectory designs for the min-energy maximization
5102 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 17, NO. 8, AUGUST 2018

Fig. 4. Trajectory designs for a UAV-enabled WPT system with K = 10 ERs. Fig. 7. The max-min average received power at each ER versus the charging
duration T with K = 10 ERs.

V and T . By contrast, when D > 5.77m, the proposed

mobile UAV design achieves higher received power than the
static UAV design (with the UAV fixed at (0, 0, H)), and the
performance gain becomes more pronounced as D increases.
Furthermore, it is observed that as the UAV’s maximum speed
V increases, the average max-min received power by the ERs
increases, as the traveling time between the two ERs becomes
less significant.
Next, we consider the min-energy maximization problem
for the UAV-enabled WPT system in Fig. 4 with K = 10
ERs. Note that besides the single-location-hovering solution
for the sum-energy maximization problem (P1), Fig. 4 also
shows the multi-location-hovering solution for problem (P3)
without the UAV speed constraint, as well as the proposed suc-
Fig. 5. The average received power by different ERs with the optimal sum-
energy maximization trajectory design. cessive hover-and-fly trajectory design and the SCP-based tra-
jectory design for problem (P2) with the UAV speed constraint
considered, by assuming T = 20s. First, it is observed that
there are Γ = 4 optimal hovering locations for the min-energy
maximization problem (P3), which are close to ERs 1-2, ER 3,
ERs 4-6, and ERs 7-10, respectively. This clearly shows that
when the ERs are close to each other (e.g., ERs 7-10), then
the UAV should hover above one single location above them
for charging them more efficiently. It is also observed that the
SCP-based trajectory design and the successive hover-and-fly
trajectory design both visit the Γ optimal hovering locations.
Furthermore, it is observed that the SCP-based trajectory in
general deviates from the successive hover-and-fly trajectory
when flying from one hovering location to another.
Fig. 7 shows the max-min average power received by all
ERs in the 10-ER WPT system in Fig. 4 versus the charging
duration T , for our proposed trajectory designs, as compared
Fig. 6. The average received power at each ER with the min-energy to the following two benchmark schemes.
maximization trajectory design with K = 2 ERs. • Single-location hovering for minimum energy maximiza-
tion: During the whole charging period, the UAV hovers
problem (P2). First, we consider the case with K = 2 at a fixed location (xfix , yfix , H) obtained in (26).
ERs. Fig. 6 shows the average received power by each ER with • Successive hover-and-fly over all ERs: In this successive
the optimal trajectory obtained by solving problem (P2-2ER), hover-and-fly scheme, instead of optimizing the UAV’s
with different values of the maximum UAV speed V and √ the hovering locations, they are simply set as the K locations
charging duration T . It is observed that when D ≤ 2H/ 3 = above the K ERs. This scheme can be implemented
5.77m, the static UAV design (with V = 0) achieves the by using Algorithm 2 via replacing {(x∗γ , yγ∗ , H)}Γγ=1 as
same performance as the mobile UAV design, regardless of {(xk , yk , H)}K
k=1 .
XU et al.: UAV-ENABLED WPT: TRAJECTORY DESIGN AND ENERGY OPTIMIZATION 5103

In Fig. 7, the upper bound corresponds to the optimal loss exponents is an interesting problem worth future
value achieved by (P3) with the UAV’s maximum speed research efforts.
constraints ignored. It is observed that the two proposed • Another interesting direction is to extend the
trajectory designs, namely the successive hover-and-fly and the UAV-enabled WPT to the general multi-UAV scenario.
SCP-based trajectory designs, outperform the single-location- In this case, multiple UAVs should cooperate with
hovering design, and achieve higher average max-min average each other to further improve the WPT performance.
power as T becomes large. When T ≥ 15 s, the proposed On one hand, the UAVs can cooperatively design their
successive hover-and-fly and the SCP-based trajectory designs trajectories subject to individual UAV speed and inter-
also outperform the successive hover-and-fly trajectory over UAV collision avoidance constraints; on the other hand,
all the ERs. Furthermore, it is observed that the SCP-based the UAVs can cooperatively design their transmit energy
trajectory achieves better performance than the successive signals based on the technique of distributed energy
hover-and-fly trajectory, and converges to the upper bound, beamforming [4]. How to jointly design multiple UAVs’
when T becomes large. trajectories and their distributed energy beamforming is
a very interesting but challenging problem to investigate
VII. C ONCLUDING R EMARKS in future work. Furthermore, multi-UAV cooperation can
This paper studies a new UAV-enabled multiuser WPT be exploited to address the limited UAV endurance issue
system. We exploit the mobility of the UAV to maximize the for achieving sustainable WPT. How to jointly control
energy transferred to all ERs over a given charging period multiple UAVs’ trajectories for WPT performance
by optimizing the UAV’s trajectory under its practical speed optimization with finite endurance of each individual
constraint. First, we consider the sum-energy maximization UAV is also an interesting problem worth further
of all ERs and obtain the optimal solution to this problem, investigation.
which shows that the UAV should hover at only one fixed loca-
tion during the whole charging period. However, this single-
A PPENDIX A
location-hovering solution may lead to unfair performance
P ROOF OF P ROPOSITION 1
among ERs due to their different distances from the UAV.
To achieve fairness among all ERs, we consider an alternative First, we consider a relaxed problem of (P1) in the ideal
problem to maximize the minimum energy transferred to case by ignoring the speed constraints in (1). By using (5),
all ERs. We first consider the relaxed problem by ignoring the relaxed problem can be written as
the UAV speed constraints and derive the optimal solution,  T
which shows that the UAV should hover over multiple fixed max ψ(x(t), y(t))dt. (36)
locations with optimal hovering time allocations among them. {x(t),y(t)} 0
Based on this solution, we further propose two new trajectory Due to the relaxation of the constraints (1), the optimal value
designs for solving the min-energy maximization problem in of problem (36) serves as an upper bound for that of (P1). It is
the general case with the UAV speed constraint considered. evident that problem (36) can be decomposed into different
Numerical results show that the proposed UAV-enabled WPT subproblems as follows:
system with optimized UAV trajectory significantly enhances
the WPT performance over the conventional WPT system max ψ(x(t), y(t)), ∀t ∈ T , (37)
with fixed ETs, and yet achieves fair energy delivery to x(t),y(t)

ERs. It is our hope that this paper will pave a new way for where each subproblem corresponds to a time instant t.
future WPT research. Due to the space limitation, there are By comparing (6) and (37), it is evident that x(t) = x and
various interesting issues unaddressed in this paper, which y(t) = y  is the optimal solution to problem (37) for any
are briefly discussed in the following to motivate future t ∈ T . Therefore, {x (t), y  (t)} given in (8) is indeed the
work. optimal solution to problem (36).
• Efficient trajectory design in UAV-enabled WPT systems
Next, it is evident that {x (t), y  (t)} given in (8) is always
critically depends on the accurate modeling of the wire- feasible to problem (P1), since the optimal location is fixed
less channels from UAV to ERs. In this paper, we assume and thus no UAV flying is needed. Furthermore, the objective
that the UAV-to-ERs channels are LoS-dominated, and value achieved by {x (t), y  (t)} for problem (P1) is the same
thus use the free-space path loss model, similarly as in as the optimal value for problem (36). Since the optimal value
prior works [17], [18]. In certain application scenarios of problem (36) is an upper bound of that of (P1), we can
such as in forests, however, there may exist obstacles and conclude that {x (t), y  (t)} is indeed the optimal solution
rich scatterers between the UAV and ERs, and a more to (P1). Therefore, Proposition 1 is proved.
accurate model with path loss exponent larger than 2
should apply. Fortunately, the change in the path loss
A PPENDIX B
exponent does not alter the fundamental structures of
P ROOF OF L EMMA 2
problems (P1), (P2), and (P3), and therefore, the algo-
rithms and methods proposed in this paper are generally The first property of ψ̂(x) can be easily shown and the detail
extendable to handle such cases. How to analytically is thus omitted here. In the following, we prove the second and
reveal the optimal UAV trajectory under different path third properties for ψ̂(x). Note that the first-order derivative
5104 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 17, NO. 8, AUGUST 2018

of ψ̂(x) can be obtained as A PPENDIX D

P ROOF OF P ROPOSITION 9
ψ̂  (x)
For K = 2, the optimality of the successive hover-and-fly√
= −4β0 P x
trajectory can be easily proved in the case of D ≤ 2H/ 3.
x4 + 2(D2 /4 + H 2 )x2 − 3D4 /16 + H 4 − H 2 D2 /2
× 2 . Therefore, in√ the following, we focus on the proof for the case
(x + D2 /4 + H 2 − Dx)2 (x2 + D2 /4 + H 2 + Dx)2 of D > 2H/ 3.
√ First, similarly as Proposition 7 for problem (P3-2ER),
Then we √ consider the two cases with D ≤ 2H/ 3 and
D > 2H/ 3, respectively. the optimal trajectory solution to (P2-2ER) should satisfy the
√ symmetric property in (38). In this case, problem (P2-2ER) is
When D ≤ 2H/ 3, it can be shown that ψ̂  (x) = 0 has
only one single real solution given by x = 0. Furthermore, re-expressed as follows based on (39).
we have ψ̂  (x) > 0 for any x < 0 and ψ̂  (x) < 0 for any  T /2
x > 0. As a result, ψ̂(x) is monotonically increasing and max ψ̂(x(t))dt
{x(t)} 0
decreasing over x ∈ (−∞, 0) and x ∈ (0, ∞), respectively.
s.t. |ẋ(t)| ≤ V, ∀t ∈ [0, T /2],
Therefore, the second property of ψ̂(x) √ is proved.
On the other hand, when D > 2H/ 3, there exist three real x(T /2) = 0. (41)
solutions to the equation ψ̂  (x) = 0, which are given by −ξ, 0, Without loss of optimality, we only need to consider x(t) ≤ 0,
and ξ, respectively. Furthermore, it can be shown that ψ̂  (x) > ∀t ∈ [0, T /2].
0 for x ∈ (−∞, −ξ), ψ̂  (x) < 0 for x ∈ (−ξ, 0), ψ̂  (x) > 0 Next, based on Lemma 2, ψ̂(x) is monotonically increasing
for x ∈ (0, ξ), and ψ̂  (x) < 0 for x ∈ (ξ, ∞). As a result, over x ∈ (−∞, −ξ) and decreasing over (−ξ, 0). As a result,
ψ̂(x) is monotonically increasing, decreasing, increasing, and it can be shown that if ξ ≤ V T /2, the UAV should maximize
decreasing over x ∈ (−∞, −ξ), (−ξ, 0), (0, ξ), and (ξ, ∞), its hovering time at the location (−ξ, 0, H), and fly to the
respectively. Furthermore, as ψ̂(−ξ) = ψ̂(ξ) based on the first middle point (0, 0, H) with its maximum speed V . In other
property of the function, x = −ξ and ξ correspond to two words, we have x∗∗ (t) = −ξ, ∀t ∈ [0, T /2 − ξ/V ], and
equivalent maximizers of ψ̂(x). Therefore, the third property x∗∗ (t) = V t − V T /2, ∀t ∈ (T /2 − ξ/V, T /2]. On the
of ψ̂(x) is proved. other hand, if ξ > V T /2, then the UAV should fly with
By combining the above properties, this lemma is proved. the maximum speed from the location (−V T /2, 0, H) to the
middle point (0, 0, H) to maximize the objective function
A PPENDIX C in (41), i.e., x∗∗ (t) = −V T /2 + V t, ∀t ∈ [0, T /2].
P ROOF OF P ROPOSITION 7 Combining the above result with the optimal symmetric
First, it is evident that at the optimal solution {x∗ (t)} to trajectory in (38), this proposition is thus proved.
(P3-2ER), the received energy at the two ERs must be equal.
Furthermore, if the UAV stays at a location (x̂, 0, H) for a A PPENDIX E
certain duration, then it should stay at the symmetric location P ROOF OF L EMMA 11
(−x̂, 0, H) for the same duration, since otherwise, we can Define a function g1 (z) = βz+H 0P Δ 2
2 with H > 0, which
always increase the min-energy of the two ERs by adjusting is convex with respect to z ≥ 0. As the first-order Taylor
the two locations to be symmetric. As a result, without loss of expansion of a convex function is a global under-estimator of
optimality, we only need to consider the following symmetric the function values [23], for any given z0 ≥ 0, it follows that
trajectory {x(t)} satisfying g1 (z) ≥ g1 (z0 ) + g1 (z0 )(z − z0 ), or equivalently,
x(t) = −x(T − t) ≤ 0, ∀t ∈ [0, T /2). (38) β0 P Δ β0 P Δ β0 P Δ
2
≥ 2
− (z − z0 ). (42)
Based on (38), it follows from (9), (10), and (33) that the z+H z0 + H (z0 + H 2 )2
received energy values at the two ERs are identical, i.e., For any given k ∈ K, n ∈ N , i ≥ 0, by substituting z =
 T /2 (x[n]−xk )2 +(y[n]−yk )2 and z0 = (x(i) [n]−xk )2 +(y (i) [n]−
Ê1 ({x(t)}) = Ê2 ({x(t)}) = ψ̂(x(t))dt, (39) yk )2 into (42), then (32) follows. Furthermore, note that the
0 equality holds for (42) for z = z0 , and therefore, the equality
where ψ̂(x(t)) is given in (11). Accordingly, we can re-express in (34) holds. This lemma is thus proved.
problem (P3-2ER) as
 T /2 R EFERENCES
max ψ̂(x(t))dt. (40) [1] J. Xu, Y. Zeng, and R. Zhang, “UAV-enabled wireless power transfer:
{x(t)} 0 Trajectory design and energy region characterization,” in Proc. IEEE
√ Globecom Workshop, Dec. 2017, pp. 1–7.
Based on Lemma 2, if D ≤ 2H/ 3, then the optimal solution √ [2] J. Xu, Y. Zeng, and R. Zhang, “UAV-enabled multiuser wireless
to (40) is given by x∗ (t) = 0, ∀t ∈ [0, T /2). If D > 2H/ 3, power transfer: Trajectory design and energy optimization,” in Proc.
then the optimal solution is x∗ (t) = −ξ, ∀t ∈ [0, T /2). IEEE APCC Workshop, Dec. 2017, pp. 1–6.
[3] S. Bi, C. K. Ho, and R. Zhang, “Wireless powered communication:
By also choosing x∗ (t) = ξ, ∀t ∈ [T /2, T ) due to symmetry Opportunities and challenges,” IEEE Commun. Mag., vol. 53, no. 4,
and using (38), this proposition is proved. pp. 117–125, Apr. 2015.
XU et al.: UAV-ENABLED WPT: TRAJECTORY DESIGN AND ENERGY OPTIMIZATION 5105

[4] Y. Zeng, B. Clerckx, and R. Zhang, “Communications and signals design [28] C. E. Miller, A. W. Tucker, and R. A. Zemlin, “Integer programming
for wireless power transmission,” IEEE Trans. Commun., vol. 65, no. 5, formulation of traveling salesman problems,” J. ACM, vol. 7, no. 4,
pp. 2264–2290, May 2017. pp. 326–329, Oct. 1960.
[5] S. Bi and R. Zhang, “Placement optimization of energy and infor- [29] E. L. Lawler, J. K. Lenstra, A. H. G. R. Kan, and D. B. Shmoys,
mation access points in wireless powered communication networks,” The Traveling Salesman Problem: A Guided Tour of Combinatorial
IEEE Trans. Wireless Commun., vol. 15, no. 3, pp. 2351–2364, Optimization, 1st ed. Hoboken, NJ, USA: Wiley, 1985.
Mar. 2016. [30] M. Grant and S. Boyd. (Mar. 2017). CVX: Matlab Software for
[6] J. Xu, L. Liu, and R. Zhang, “Multiuser MISO beamforming for Disciplined Convex Programming, Version 2.1. [Online]. Available:
simultaneous wireless information and power transfer,” IEEE Trans. http://cvxr.com/cvx/
Signal Process., vol. 62, no. 18, pp. 4798–4810, Sep. 2014.
[7] J. Xu and R. Zhang, “Energy beamforming with one-bit feed-
back,” IEEE Trans. Signal Process., vol. 62, no. 20, pp. 5370–5381,
Oct. 2014.
[8] J. Xu and R. Zhang, “A general design framework for MIMO wireless
energy transfer with limited feedback,” IEEE Trans. Signal Process.,
vol. 64, no. 10, pp. 2475–2488, May 2016.
[9] Y. Zeng and R. Zhang, “Optimized training design for wireless
energy transfer,” IEEE Trans. Commun., vol. 63, no. 2, pp. 536–550,
Feb. 2015.
[10] E. Boshkovska, A. Koelpin, D. W. K. Ng, N. Zlatanov, and
R. Schober, “Robust beamforming for swipt systems with non-
linear energy harvesting model,” in Proc. IEEE SPAWC, Jul. 2016, Jie Xu (S’12–M’13) received the B.E. and Ph.D.
pp. 1–5. degrees from the University of Science and Technol-
[11] G. Yang, C. K. Ho, R. Zhang, and Y. L. Guan, “Throughput opti- ogy of China in 2007 and 2012, respectively. From
mization for massive MIMO systems powered by wireless energy 2012 to 2014, he was a Research Fellow with the
transfer,” IEEE J. Sel. Areas Commun., vol. 33, no. 8, pp. 1640–1650, Department of Electrical and Computer Engineering,
Aug. 2015. National University of Singapore. From 2015 to
[12] G. Amarasuriya, E. G. Larsson, and H. V. Poor, “Wireless infor- 2016, he was a Post-Doctoral Research Fellow with
mation and power transfer in multiway massive MIMO relay net- the Engineering Systems and Design Pillar, Singa-
works,” IEEE Trans. Wireless Commun., vol. 15, no. 6, pp. 3837–3855, pore University of Technology and Design. He is
Jun. 2016. currently a Professor with the School of Information
[13] S. Bi and R. Zhang, “Distributed charging control in broadband wireless Engineering, Guangdong University of Technology,
power transfer networks,” IEEE J. Sel. Areas Commun., vol. 34, no. 12, China. His research interests include energy efficiency and energy harvesting
pp. 3380–3393, Dec. 2016. in wireless communications, wireless information and power transfer, wireless
[14] M. Xia and S. Aissa, “On the efficiency of far-field wireless power securities, UAV communications, and mobile edge computing. He was a recip-
transfer,” IEEE Trans. Signal Process., vol. 63, no. 11, pp. 2835–2847, ient of the IEEE Signal Processing Society Young Author Best Paper Award
Jun. 2015. in 2017. He is currently an Editor of the IEEE W IRELESS C OMMUNICATIONS
[15] B. Clerckx and E. Bayguzina, “Waveform design for wireless power L ETTERS , an Associate Editor of the IEEE ACCESS , and a Guest Editor of
transfer,” IEEE Trans. Signal Process., vol. 64, no. 23, pp. 6313–6328, the IEEE W IRELESS C OMMUNICATIONS.
Dec. 2016.
[16] M. R. V. Moghadam, Y. Zeng, and R. Zhang, “Waveform optimization
for radio-frequency wireless power transfer: (Invited paper),” in Proc.
IEEE SPAWC, Jul. 2017, pp. 1–6.
[17] Y. Zeng, R. Zhang, and T. J. Lim, “Wireless communications
with unmanned aerial vehicles: Opportunities and challenges,” IEEE
Commun. Mag., vol. 54, no. 5, pp. 36–42, May 2016.
[18] Y. Zeng et al., “Throughput maximization for UAV-enabled
mobile relaying systems,” IEEE Trans. Commun., vol. 64, no. 12,
pp. 4983–4996, Dec. 2016.
[19] Y. Zeng and R. Zhang, “Energy-efficient UAV communication with
trajectory optimization,” IEEE Trans. Wireless Commun., vol. 16, no. 6,
pp. 3747–3760, Jun. 2017. Yong Zeng (S’12–M’14) received the B.E. (Hons.)
[20] L. Xie, Y. Shi, Y. T. Hou, and H. D. Sherali, “Making sensor networks and Ph.D. degrees from Nanyang Technological
immortal: An energy-renewal approach with wireless power transfer,” University, Singapore, in 2009 and 2014, respec-
IEEE/ACM Trans. Netw., vol. 20, no. 6, pp. 1748–1761, Dec. 2012. tively. He was a Research Fellow and a Senior
Research Fellow with the Department of Electrical
[21] Y. Shu et al., “Near-optimal velocity control for mobile charging in
and Computer Engineering, National University of
wireless rechargeable sensor networks,” IEEE Trans. Mobile Comput.,
Singapore. He is currently a Lecturer with the School
vol. 15, no. 7, pp. 1699–1713, Jul. 2016.
of Electrical and Information Engineering, The Uni-
[22] E. Boshkovska, D. W. K. Ng, N. Zlatanov, and R. Schober, versity of Sydney. His research interests include
“Practical non-linear energy harvesting model and resource alloca- UAV communications, wireless power transfer, mas-
tion for SWIPT systems,” IEEE Commun. Lett., vol. 19, no. 12, sive MIMO and millimeter wave communications,
pp. 2082–2085, Dec. 2015. and multi-user MIMO interfering communications. He has published over
[23] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge, U.K.: 60 IEEE journal and conference papers, including one invited paper in IEEE
Cambridge Univ. Press, 2004. T RANSACTIONS ON C OMMUNICATIONS , six ESI highly cited papers, and
[24] W. Yu and R. Lui, “Dual methods for nonconvex spectrum optimiza- three ESI hot papers. He was a recipient of the 2017 IEEE Communications
tion of multicarrier systems,” IEEE Trans. Commun., vol. 54, no. 7, Society Heinrich Hertz Prize Paper Award, the 2017 IEEE Transactions on
pp. 1310–1322, Jul. 2006. Wireless Communications Best Reviewer, the 2015 and 2017 IEEE Wireless
[25] S. Boyd. EE364b Convex Optimization II, Course Notes. Accessed: Communications Letters Exemplary Reviewer, and the Best Paper Award for
Jun. 29, 2017. [Online]. Available: http://www.stanford.edu/class/ the 10th IEEE International Conference on Information, Communications and
ee364b/ Signal Processing. He is currently serving as an Associate Editor for the IEEE
[26] M. Padberg and G. Rinaldi, “A branch-and-cut algorithm for the resolu- A CCESS , a Leading Guest Editor for the IEEE W IRELESS C OMMUNICATIONS
tion of large-scale symmetric traveling salesman problems,” SIAM Rev., on Integrating UAVs into 5G and Beyond and China Communications on
vol. 33, no. 1, pp. 60–100, 1991. Network-Connected UAV Communications. He is the Workshop Co-Chair
[27] Concorde. Concorde TSP Solver. Accessed: Jun. 29, 2017. [Online]. for two workshops in ICC 2018 and the 23rd Asia-Pacific Conference on
Available: http://www.math.uwaterloo.ca/tsp/concorde/ Communications.
5106 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 17, NO. 8, AUGUST 2018

Rui Zhang (S’00–M’07–SM’15–F’17) received IEEE W IRELESS C OMMUNICATIONS L ETTERS . He was the recipient of
the B.Eng. (Hons.) and M.Eng. degrees from the the 6th IEEE Communications Society Asia–Pacific Region Best Young
National University of Singapore, Singapore, and the Researcher Award in 2011 and the Young Researcher Award of National
Ph.D. degree from the Stanford University, Stanford, University of Singapore in 2015. He was a co-recipient of the IEEE Mar-
CA, USA, all in electrical engineering. coni Prize Paper Award in Wireless Communications in 2015, the IEEE
From 2007 to 2010, he was a Research Scientist Communications Society Asia–Pacific Region Best Paper Award in 2016,
with the Institute for Infocomm Research, ASTAR, the IEEE Signal Processing Society Best Paper Award in 2016, the IEEE
Singapore. In 2010, he joined the Department of Communications Society Heinrich Hertz Prize Paper Award in 2017, the IEEE
Electrical and Computer Engineering, National Uni- Signal Processing Society Donald G. Fink Overview Paper Award in 2017,
versity of Singapore, where he is currently the and the IEEE Technical Committee on Green Communications and Computing
Dean’s Chair Associate Professor with the Faculty of Best Journal Paper Award in 2017. His co-authored paper received the IEEE
Engineering. He has authored over 300 papers. His research interests include Signal Processing Society Young Author Best Paper Award in 2017. He has
wireless information and power transfer, drone communication, wireless served as TPC co-chair or as an organizing committee member for over
eavesdropping and spoofing, energy-efficient and energy-harvesting-enabled 30 international conferences, and as the guest editor for three special issues in
wireless communication, multiuser MIMO, cognitive radio, and optimization the IEEE J OURNAL OF S ELECTED T OPICS IN S IGNAL P ROCESSING and the
methods. He has been listed as a Highly Cited Researcher (also known as the IEEE J OURNAL ON S ELECTED A REAS IN C OMMUNICATIONS. He served as
World’s Most Influential Scientific Minds), by Thomson Reuters since 2015. an Editor for the IEEE T RANSACTIONS ON W IRELESS C OMMUNICATIONS
Dr. Zhang was an Elected Member of the IEEE Signal Processing (2012–2016), the IEEE J OURNAL ON S ELECTED A REAS IN C OMMUNICA -
Society SPCOM Technical Committee (2012–2017) and SAM Techni- TIONS Green Communications and Networking Series (2015–2016), and the
cal Committee (2013–2015), and served as the Vice Chair of the IEEE IEEE T RANSACTIONS ON S IGNAL P ROCESSING (2013–2017). He is cur-
Communications Society Asia–Pacific Board Technical Affairs Committee rently an Editor of the IEEE T RANSACTIONS ON C OMMUNICATIONS and the
(2014–2015). He serves as a member of the Steering Committee of the IEEE T RANSACTIONS ON G REEN C OMMUNICATIONS AND N ETWORKING.