INTEGRATING UAVS INTO 5G AND BEYOND
Fundamental Trade-offs in
Communication and Trajectory Design for
UAV-Enabled Wireless Network
Qingqing Wu, Liang Liu, and Rui Zhang
Abstract
The use of unmanned aerial vehicles (UAVs)
as aerial communication platforms is of high
practical value to future wireless systems such as
5G, especially for swift and on-demand deploy-
ment in temporary events and emergency situ-
ations. Compared to the static terrestrial base
stations (BSs) in cellular networks, UAV-mounted
aerial BSs possess stronger line-of-sight links with
the ground users due to their high altitude as
well as high and flexible mobility in 3D space,
which can be exploited to enhance the commu-
nication performance. On the other hand, unlike
terrestrial BSs that have reliable power supply,
aerial BSs in practice have limited onboard ener-
gy, but require high propulsion power to stay air-
borne and support high mobility. Motivated by
the above new considerations, this article aims
to revisit some fundamental trade-offs in UAV-en-
abled communication and trajectory design.
Specifically, it is shown that the communication
throughput, delay, and (propulsion) energy con-
sumption can be traded off among each other
by adopting different UAV trajectory designs,
which sheds new light on their existing trade-offs
in terrestrial communication. The main design
challenges and promising directions for future
research are also discussed.
Introduction
Due to their prominent features of high mobil-
ity and flexible deployment, unmanned aerial
vehicles (UAVs), also known as drones, will find
many promising uses in the future fifth gener-
ation (5G) and beyond wireless networks, as
shown in Fig. 1. In particular, UAVs can be used
cost-effectively as on-demand aerial platforms to
provide or enhance the communication services
for terrestrial mobiles/devices in a multitude
of applications, including aerial base stations
(BSs)/relays in situations without terrestrial cel-
lular coverage [1–3], aerial helpers for providing
new services such as data backhaul/offloading,
cached-content multicasting, and edge comput-
ing for terrestrial BSs/users, and mobile hubs for
energy-efficient data collection [4] and wireless
power transfer [5] for low-power Internet of
Things (IoT) devices such as sensors and tags.
On the other hand, UAVs in many civilian appli-
Digital Object Identifier:
10.1109/MWC.2018.1800221 The authors are with the National University of Singapore.
36 1536-1284/19/$25.00 © 2019 IEEE
cations such as cargo delivery and aerial video
surveillance may become new aerial users in the
cellular network, which need to have high-per-
formance two-way communications with the
ground BSs to receive control signals and upload
application data in real time [6].
Despite the above promising UAV appli-
cations, their future success critically depends
on the development of new and effective
UAV-ground communication technologies.
Compared to the conventional terrestrial com-
munications, UAV-ground communications
enjoy the following two main advantages that
can be exploited for throughput enhancement,
namely the line-of-sight (LoS)-dominated UAV-
ground channel and the UAV’s controllable
high mobility in 3D space. On one hand, thanks
to the high altitude of UAVs, the probability of
LoS channels between UAVs and the ground
users/BSs is in general pretty high, and thus
UAV-ground communications are significantly
less affected by channel impairments such as
shadowing and fading as compared to terrestrial
communications. On the other hand, thanks to
the high mobility, swift 3D deployment or even
dynamic movement of UAVs becomes feasible
so that they can adjust their locations/trajecto-
ries based on the locations and/or movement
of the ground BSs/users to maintain favorable
LoS channels with them. It is worth noting that
the LoS channels enable UAVs to have their
signal coverage over a much larger number of
ground users or BSs compared to the BSs/users
in terrestrial communications. Consequently, to
achieve optimal communication and trajecto-
ry design, each UAV should not only maintain
strong channels to its served users or connect-
ed BSs via flying in proximity of them, but also
control its interferences to other UAVs as well
as ground users/BSs so as to achieve the maxi-
mum network throughput.
Besides throughput, two important factors
also need to be considered in UAV communi-
cation and trajectory design, namely delay and
energy. First, to maximize throughput, each
UAV should communicate with a ground user/
BS when flying sufficiently close to it so as to
reduce their distance and hence improve the
link capacity. However, this inevitably incurs
more delay in communication due to the UAV
IEEE Wireless Communications • February 2019
 

 
FIGURE 1. Typical UAV applications in 5G and beyond.
movement [7–9]. Thus, there is an interest-
ing throughput-delay trade-off in UAV-ground
communication, as shown in Fig. 2a. Second,
there is also a new trade-off between through-
put and energy in UAV-enabled communication,
as shown in Fig. 2b, since the UAV generally
needs to consume more propulsion energy to
move closer to the ground users/BSs in order
to gain higher throughput [10, 11]. As commer-
cial UAVs usually have limited onboard ener-
gy, more propulsion energy consumption leads
to shorter endurance of UAVs, thus imposing a
critical constraint on their practical applications.
Last, the above two trade-offs naturally imply a
delay-energy trade-off, shown in Fig. 2c, as delay
in UAV-ground communication can be generally
reduced if more propulsion energy is consumed
by the UAV to move faster to the ground users/
BSs it is designated to communicate with.
Motivated by the above new and interesting
trade-offs among the throughput, delay, and (pro-
pulsion) energy consumption in UAV communi-
cation and trajectory design, this article aims to
provide an overview of the state-of-the-art results
on them. In particular, we focus on the use of
UAVs as communication platforms (e.g., aerial
BSs/relays) to serve the terrestrial users, although
such fundamental trade-offs also exist similarly
in the other paradigm with UAVs as new aerial
users served by the ground BSs in the cellular net-
work [6]. Next, we discuss the main differences
between these trade-offs in UAV-enabled commu-
nication and their existing counterparts in tradi-
tional terrestrial communication.
Fundamental Trade-offs in
UAV-Enabled Communication
It is well known that there are fundamen-
tal trade-offs between the throughput, delay,
and energy in wireless communication [12]. In
this section, we first review the classic results
on such trade-offs in terrestrial communica-
tion, and then explain their main differences
in UAV-ground communication arising from its
new considerations discussed in the preceding
IEEE Wireless Communications • February 2019
 


section, such as the LoS channel, the UAV tra-
jectory design, and its high propulsion energy
consumption.
Throughput-Delay Trade-off
The throughput-delay trade-off has been exten-
sively studied for terrestrial wireless commu-
nication. For a basic point-to-point wireless
communication link, the maximum achievable
rate over fading channels, defined as the ergodic
capacity, is achieved by coding over a sufficient-
ly large number of channel coherence intervals
to fully exploit the ergodicity of fading channels
[12]. However, this comes at the cost of long
transmission delay that may not be tolerable for
applications with stringent latency requirement.
On the other hand, channel coding can be per-
formed over each coherence interval to reduce
the delay, resulting in the so-called delay-limited
capacity [12]. However, the delay-limited capaci-
ty is in general smaller than the ergodic capacity
for a given fading channel, and outage is usually
inevitable in deep fading [12]. For the general
multi-user communication, the multi-user diversi-
ty gain can be attained to improve the network
throughput by scheduling the user with the best
channel among all users to communicate in each
coherence interval, whereas this usually leads to
more significant delay for each user as the num-
ber of users increases [12]. The above results
show that there is a general throughput-delay
trade-off for communication over fading chan-
nels. Moreover, it is shown in [13] that there
is another trade-off between the total through-
put of a mobile ad hoc network (MANET) and
the average delay tolerable by the users in the
network due to the random user movement, as
each user needs to wait to communicate with
another one until they become sufficiently close
to each other.
In contrast, in UAV-enabled communication,
channel fading is no longer a key factor contrib-
uting to the throughput-delay trade-off thanks to
the LoS-dominant channels. Instead, the mobility
of UAVs plays the deciding role in such a trade-
off as the UAV-ground LoS channels are solely
37
The throughput-energy
trade-off in traditional
wireless communica-
tion is fundamentally
rooted in the Shannon
capacity formula,
which explicitly sug-
gests that the achiev-
able rate increases
monotonically with
(a)
the transmit power
[12]. One useful per- FIGURE 2. Three fundamental tradeoffs in UAV-enabled wireless communication: a) throughput-delay
formance metric stem- trade-off; b) throughput-energy trade-off; c) delay-energy trade-off.
ming from this trade-off
is “energy efficiency,” determined by the distances between the UAV
which measures the and ground users, which critically depend on the
number of information UAV’s location. However, in sharp contrast to the
bits that can be trans- random user movement in a MANET, where the
mitted by using a Joule delay is random and difficult to predict for the
users [13], the delay in UAV-enabled communi-
of energy. cation can be properly controlled via a joint UAV
trajectory and communication scheduling design
[7–9]. Moreover, another key difference lies in
the timescale of the delay between the terrestrial
communication and UAV-ground communica-
tion: in the former case, the delay is measured in
terms of channel coherence time (e.g., in millisec-
onds), while in the latter case, the delay is mainly
due to the UAV flying time (distance divided by
speed, e.g., in seconds). As a result, in order to
fully exploit the throughput-delay trade-off via tra-
jectory design in UAV-enabled communication,
the application needs to be more delay-tolerant
compared to that in terrestrial communication [8].
Throughput-Energy Trade-off
The throughput-energy trade-off in traditional
wireless communication is fundamentally rooted
in the Shannon capacity formula, which explicitly
suggests that the achievable rate increases mono-
tonically with the transmit power [12]. One useful
performance metric stemming from this trade-off
is “energy efficiency,” which measures the num-
ber of information bits that can be transmitted by
using a Joule of energy. If only the transmit ener-
gy is considered, it is well known that the energy
efficiency monotonically increases with decreas-
ing the transmit rate/power [12]; while if the cir-
cuit power at the transmitter is considered as well,
it is shown in [14] that the energy efficiency first
increases and then decreases with the transmit
rate/power.
In UAV-enabled communication, the propul-
sion energy (usually on the order of a kilowatt)
required to maintain UAVs airborne and support
their high mobility is generally several orders of
magnitude higher than the transmit and circuit
energy for communication (usually on the order
of a watt or even less). As a result, the effect of
propulsion energy on the UAV trajectory is the
dominant factor determining the throughput-ener-
gy trade-off in UAV-enabled communication. For
example, to enhance the throughput, and suppos-
ing that the UAV already transmits at its maximum
power, each UAV needs to fly over a longer dis-
tance at a faster speed so that it can reach each
of its served ground users as closely as possible
38
(b) (c)
and stay near them as long as possible (given a
finite flight duration) in order to exploit better LoS
channels with them. Moreover, each UAV may
also need to adjust its altitude and/or make sharp
turns to avoid blockages in the directions of its
served ground users. All of these will lead to more
substantial propulsion energy consumption. As
a result, for UAV-enabled communication, the
energy efficiency is more appropriately defined in
terms of information bits per Joule of propulsion
energy, rather than that of transmit/circuit energy
in traditional wireless communication. Such a new
metric has high practical significance, as it indi-
cates the maximum number of information bits
that can be communicated with a finite amount
of UAV onboard energy.
Delay-Energy Trade-off
As discussed in the above two subsections, the
throughput-delay and throughput-energy trade-
offs in UAV-enabled communication exhibit
interesting new aspects compared to their coun-
terparts in terrestrial communication. As a result,
their corresponding delay-energy trade-offs are
also drastically different due to the new UAV
trajectory design and the high UAV propulsion
energy consumption. For example, to reduce
the delay in movement and transmission, each
UAV should fly between its served ground users
at its maximum speed, but remain at its mini-
mum speed (e.g., hovering) when serving them
in proximity, both resulting in more propulsion
energy consumption in general as in practice the
UAV achieves its minimum propulsion energy
when its speed is between zero and the maxi-
mum [10].
In the rest of this article, we focus on examin-
ing the throughput-delay and throughput-energy
trade-offs in the next two sections, respectively.
We provide concrete examples to illustrate them
more clearly, provide overviews of their state-
of-the-art results, and also point out promising
directions for future research. As the delay-ener-
gy trade-off becomes intuitive given the first two
trade-offs, it is not discussed in more detail.
Throughput-Delay Trade-off
In this section, we investigate the joint UAV tra-
jectory and communication design to character-
ize the throughput-delay trade-off. Specifically,
we first consider a simple setup with one UAV
serving two ground users (GUs) to draw useful
insights. Then we extend our study to the general
IEEE Wireless Communications • February 2019
 →

(a)
(b)

(c) (d)

FIGURE 3. Throughput-delay trade-off for a single UAV-enabled network with two GUs. The GUs’ nominal locations are marked by ◊
and the UAV trajectories are marked by w. The maximum transmit power and the receiver noise power are set as 20 dBm (0.1 W)
and –110 dBm, respectively, and the channel power gain at the reference distance of 1 m is set as –50 dB. Other required parame-
ters are set as follows: Vmax = 50 m/s, H = 100 m, and D = 2000 m: a) a UAV-enabled two-user wireless system; b) UAV horizontal
trajectory for different T; c) periodic TDMA of GUs; d) common throughput vs. UAV flight period, T.

case with multiple UAVs serving multiple GUs,
followed by further discussions on related/future
work.
Single-UAV-Enabled Wireless Network
As shown in Fig. 3a, we consider a UAV-enabled
downlink communication system where one UAV
is employed to serve two GUs in a finite period
of T s. The UAV is assumed to fly at a constant
altitude of H in meters with the maximum allow-
able speed denoted by Vmax in meters per sec-
ond. The air-ground channels from the UAV to
GUs are assumed to be dominated by the LoS
links. As such, it is preferable to let the UAV fly as
low as possible in order to reduce the signal path
loss with the GUs. However, the minimum value
of H is practically limited for terrain or building
avoidance. The two GUs are assumed to be qua-
si-stationary with a distance of D m between their
nominal locations, as shown in Fig. 3a, where we
assume that their maximum movement distanc-
es from their respective nominal locations within
the given period T are negligible compared to D
and the UAV altitude H; thus, their effects to the
corresponding LoS channel gains are ignored. We
consider that the UAV communicates with GUs
via time-division multiple access (TDMA); that is,
only one GU is scheduled for communication at
any time instant. Other multiple access schemes
are discussed later. To serve GUs continuously
in a periodic manner, we assume that the UAV
needs to return to its initial location by the end of
each flight period T, while the initial location can
be optimized for maximizing the throughput. To
ensure fairness among GUs, we aim to maximize
the common (minimum) throughput among the
GUs via jointly optimizing the UAV trajectory and
communication scheduling.
In Fig. 3b, we show the UAV’s optimal trajec-
tories projected onto the horizontal ground plane
under different flight periods, T. It is observed that
as T increases, the UAV tends to fly closer to the
two GUs, while when T is sufficiently large (e.g.,
T = 100 s), the UAV flies between the two GUs
at its maximum speed to save more time for hov-
ering right above each of them to maintain the
best channel for communication. Furthermore,
at any time instant, to maximize the throughput,
the GU that is closer to the UAV (thus having a
better channel) is scheduled for communication,
while the other GU has to wait until the UAV flies
closer to it again. As such, each GU will experi-

IEEE Wireless Communications • February 2019 39

 1000

800 t = 40 s t = 20 s
600 t=0s
t = 100 s
t = 80 s
400

200 t=0s
t = 60 s

y (m)
0
t = 60 s
−200

−400 t = 20 s
t = 80 s
−600

−800 t = 40 s
t = 100 s
 −1000
−1500 −1000 −500 0 500 1000 1500
 x(m), Rcom = 1.7631 bps/Hz
(a)
(b)
1000
Joint optimization
800 t = 20 s 2.2 Without power control
Orthogonal transmission
600 t = 100 s
t = 40 s 2 Single UAV
t = 80 s

Common throughput (bps/Hz)

400
1.8
200 t = 60 s t=0s
t=0s 1.6
y (m)

0
t = 60 s 1.4
−200
1.2
−400
t = 20 s
t = 80 s 1
−600
t = 40 s t = 100 s 0.8
−800

−1000 0.6
−1500 −1000 −500 0 500 1000 1500 10 30 50 70 90 110 130 150
x (m), Rcom = 2.1535 bps/Hz UAV flight period, T (s)

(c) (d)

FIGURE 4. Throughput-delay trade-off for a multi-UAV enabled wireless system with IUIC. The GUs’ nominal locations are marked by ◊
and the UAV trajectories are marked by w. The simulation parameters are set the same as those in Fig. 3. The user common through-
put is denoted by Rcom in bits per second per Hertz: a) a multi-UAV enabled wireless system with IUIC; b) UAV horizontal trajecto-
ries without power control; c) UAV horizontal trajectories with power control; d) common throughput vs. UAV flight period, T.

ence a waiting time of T/2 for communicating GUs into small-size clusters, each served by one
with the UAV periodically. This is illustrated in of the UAVs. To demonstrate this, we consider
Fig. 3c, where the user scheduling is plotted over a multi-UAV-enabled downlink transmission sys-
time. It is observed that a larger T leads to a lon- tem as shown in Fig. 4a, where two UAVs are
ger waiting time for each GU. Finally, we show employed to serve a group of K GUs in a finite
in Fig. 3d the achievable common throughput period of duration T. To achieve high spectral
in bits per second per Hertz vs. T. Note that the efficiency, we consider a spectrum sharing sys-
throughput upper bound is obtained by ignoring tem where the UAVs share the same frequency
the time spent traveling between the two GUs, band for communication, and each of the UAVs
which holds when T goes to infinity. In addition, serves its associated GUs via periodic TDMA.
the throughput of a static UAV is obtained by fix- As such, each GU suffers from severe interfer-
ing the UAV above the middle location between ence from other non-associated UAVs due to
the two GUs at all times. One can observe that the LoS channels, which needs to be effectively
compared to the case of a static UAV, the com- mitigated by employing the inter-UAV interfer-
mon throughput is significantly improved as T ence coordination (IUIC) via jointly designing
increases with a mobile UAV. However, such a the UAV trajectories, transmit power, and user
throughput gain comes at the cost of increasing associations. We aim to maximize the common
the user delay (or larger T), which thus reveals a throughput of all GUs with optimally designed
new throughput-delay trade-off in a UAV-enabled IUIC. However, this problem is a non-convex
wireless network. optimization problem involving infinite variables
due to the continuous UAV trajectory. To tackle
Multi-UAV-Enabled Wireless Network this problem, we first apply time discretization to
The use of multiple UAVs for cooperatively equally divide the UAV flight period into a finite
serving GUs is an effective solution to improve number of small-interval time slots, each with a
the throughput-delay trade-off over the sin- nominal location of the UAV. Then we apply the
gle-UAV-enabled network by grouping nearby block coordinate descent and successive convex

40 IEEE Wireless Communications • February 2019

optimization techniques to obtain a suboptimal
solution to the IUIC design [7]. As an initial UAV
trajectory is needed for our algorithm, we adopt
a simple but practical circular UAV trajectory for
initialization [7].
For the purpose of illustration, we consider a
setup with K = 6 GUs that are randomly located
in a square area. Specifically, we show the opti-
mized UAV trajectories without and with power
control in Figs. 4b and 4c, respectively, for T
= 120 s. In the former case, both UAVs trans-
mit with their maximum power at all times. It is
observed from Fig. 4b that the optimized UAV
trajectories tend to not only shorten the com-
munication distances between the UAVs and
their associated GUs (e.g., from t = 0 to t = 20
s), but also enlarge separations of the two UAVs
to help alleviate the co-channel interference
(e.g., from t = 40 s to t = 60 s) in the case with-
out power control. However, at certain pairs of
UAV locations, enlarging the UAVs’ separation
is achieved at the cost of compromising direct
link gains, especially when the UAVs fly on their
ways to serve two GUs (e.g., the two nearby
GUs around the center in Fig. 4b) that are close
to each other.
In contrast, for the case with power control, it
is observed from Fig. 4c that the optimized UAV
trajectories do not tend to compromise the direct
link gains in return for large distance separation.
This is because power control can help avoid
strong interference even when the two UAVs
have to be close to each other (e.g., when serv-
ing the two central GUs). As a result, the com-
mon throughput is substantially improved over
the case without power control, as shown in Fig.
4d. In addition, an orthogonal UAV transmission
scheme is adopted for comparison where the two
UAVs take turns transmitting information to serve
GUs over orthogonal time slots, and the system is
then interference-free. One can observe that for
small flying time T, which implies limited UAVs’
flying ranges, the orthogonal transmission even
achieves higher throughput than non-orthogonal
schemes, which is due to the severe interference
between the UAVs. However, as T increases, the
proposed joint design significantly outperforms
the orthogonal transmission, since the UAVs’ tra-
jectories can be more flexibly designed to enlarge
the inter-UAV distance such that the spectrum
can be better reused by the two UAVs with small
interference. Finally, it is also observed that the
user throughput in the multi-UAV network is sig-
nificantly improved over the single-UAV network
under the same delay, thus verifying the improved
throughput-delay trade-off via effective multi-UAV
cooperation with optimized IUIC.
Further Discussion and Future Work
Besides orthogonal multiple access schemes such
as TDMA considered above, non-orthogonal
multiple access schemes based on superposition
coding (SC) or dirty paper coding [12] can be
jointly designed with the UAV trajectory to further
improve the throughput-delay trade-off and achieve
the capacity limits of UAV-enabled wireless net-
works [9]. For example, a two-user broadcast chan-
nel (BC) is studied in [9], where it is shown that a
simple and practical “hover-fly-hover (HFH)” trajec-
tory with SC achieves the capacity region. Howev-
IEEE Wireless Communications • February 2019
er, whether similar results hold for a UAV-enabled To maximize the coop-
BC with more than two users or other multiuser erative beamforming
channel models still remains an open problem that gain in CoMP, it may
is worth investigating in future work. Furthermore, be desirable to let
in our above study, the user delay is roughly mea-
sured in terms of the UAV flight period. However, some UAVs form a fleet
the delay requirements in 5G networks may vary to serve the GUs along
dramatically in timescale from milliseconds (e.g., the same trajectory,
for online gaming/video streaming) to several sec- while this is appar-
onds (e.g., for large file sharing/sensor data col- ently undesired in the
lection). Thus, how to model such heterogeneous
delay requirements and design the joint UAV tra- IUIC case due to the
jectory and communication resource allocation to inter-UAV interference.
efficiently meet them is also an important problem Another important
for future research. issue worthy of further
In practice, the UAV-ground LoS channel investigation is how
model is appropriate for rural or suburban areas
or when the UAV altitude is sufficiently high. to dynamically adjust
However, for other cases such as in urban envi- the UAV trajectories
ronments, other air-ground channel models, such according to the GUs’
as the probabilistic LoS model and Rician fading movement to improve
model, are more suitable. It is worth noting that their throughput and/
such non-LoS channel models may have signif-
icant impacts on the optimal UAV trajectory or delay performance..
design in UAV-enabled wireless networks [3]. For
example, lowering the UAV’s flying altitude under
the probabilistic LoS model generally decreases
the probability of having LoS links with GUs, while
it is always beneficial under the LoS model. As a
result, a more complex 3D trajectory optimization
problem (as compared to the 2D design in our
previous examples under the LoS model) needs to
be investigated. Moreover, although the presence
of LoS links makes the UAVs very suitable for 5G
technologies such as millimeter-wave (mmWave)
and massive multiple-input multiple-output (M-MI-
MO) communications, the severe air-ground inter-
ference issue and 3D mobility-induced Doppler
effect deserve more investigation in the future.
Last, for the multi-UAV-enabled network, we
propose the IUIC as an effective technique to mit-
igate the strong LoS interference by exploiting the
coordinated multi-UAV trajectory design. Alterna-
tively, motivated by the rapid advance of the wire-
less backhaul technologies, the UAVs can share
messages and perform cooperative beamforming
for more efficient interference mitigation, a tech-
nique called coordinated multipoint (CoMP) in the
sky [15. It is worth noting that the methodology for
designing the optimal UAV trajectories for CoMP is
generally different from that for IUIC. For example,
to maximize the cooperative beamforming gain in
CoMP, it may be desirable to let some UAVs form
a fleet to serve the GUs along the same trajectory,
while this is apparently undesired in the IUIC case
due to inter-UAV interference. Another import-
ant issue worthy of further investigation is how to
dynamically adjust the UAV trajectories according
to the GUs’ movement to improve their through-
put and/or delay performance [15].
Throughput-Energy Trade-off
In this section, we further investigate the through-
put-energy trade-off in UAV-enabled communi-
cation and trajectory design. First, we discuss the
energy consumption models of UAVs. Then we
revisit the single-UAV-enabled system from earlier
by taking into account the UAV’s propulsion ener-
gy consumption, followed by discussions on other
related work and future research directions.
41

Required power (W)
Required power (W)
Flying speed (m/s) Flying speed (m/s)
(a) (b)
FIGURE 5. Typical propulsion power consumption vs. the UAV’s flying speed:
a) fixed-wing UAV; b) rotary-wing UAV.
UAV Propulsion Energy Consumption Model:
Fixed-Wing vs. Rotary-Wing
Fixed-wing and rotary-wing UAVs are the two
main types of UAVs that have been widely used
in practice. To investigate the throughput-ener-
gy trade-off in UAV-enabled communication, the
UAV’s propulsion energy consumption needs to
be properly modeled first. Toward this end, two
analytical propulsion power models have been
presented for fixed-wing and rotary-wing UAVs
in [10, 16], respectively. In general, the propul-
sion power required for the UAV depends on
its velocity (including both the flying speed and
direction) as well as the acceleration. In Fig. 5,
the typical propulsion power consumption vs. the
UAV’s flying speed is illustrated for both fixed-
wing and rotary-wing UAVs. In both cases, it is
observed that as the UAV’s flying speed increas-
es, the corresponding propulsion power required
first decreases and then increases, which implies
that flying at too high or too low speeds is not
energy-efficient. Furthermore, flying at a very low
speed is extremely energy-consuming and even
impossible for fixed-wing UAVs in practice, which
renders it very difficult for them to hover over a
small geographical area to serve GUs, while this is
not an issue for rotary-wing UAVs. However, rota-
ry-wing UAVs suffer from consuming excessive
propulsion power when the UAV’s flying speed is
very high, which makes them inefficient for tasks
over a wide geographical area. In practice, fixed-
wing and rotary-wing UAVs can be leveraged
simultaneously to enhance the communication
efficiency. For example, a promising UAV-enabled
networking architecture is to deploy rotary-wing-
UAV-enabled BSs hovering at well selected loca-
tions for establishing signal hotspots and at the
same time dispatch fixed-wing-UAV-enabled BSs
flying around periodically for wider coverage and
higher throughput.
Energy-Constrained Trajectory Optimization
As shown in Fig. 6a, we considered the same
UAV-enabled two-user system as discussed ear-
lier for a given UAV flight period T where the
UAV has limited onboard energy, and thus the
maximum propulsion energy that can be con-
sumed during this period is denoted by E max .
For the purpose of exposition, we consider a
fixed-wing UAV with the minimum speed and
maximum acceleration denoted by V min in
meters per second and a max in square meters
per second , respectively. Similarto another sec-
42
tion, we consider the common throughput max-
imization for the two GUs via jointly optimizing
the UAV trajectory as well as the user schedul-
ing, subject to the new UAV’s total energy con-
straint and the mobility constraints (on its speed
and acceleration).
In Fig. 6b, we plot the UAV’s optimized tra-
jectories under different constraints of propulsion
energy. It is observed that the UAV flies close to
the two GUs by following a smooth trajectory with
relatively large turning radii when Emax = 13,000
J; whereas when E max is increased to 23,000 J,
the UAV’s trajectory tends to approach that with-
out the propulsion energy constraint shown in
Fig. 3b. This is because in the latter case, sharp
turning in the flight direction to quickly shorten
the UAV-GU distance requires more propulsion
energy consumption. Furthermore, the UAV’s fly-
ing speeds over time in the above two cases are
illustrated in Fig. 6c. It is observed that in the first
case, the UAV’s flying speed does not vary much
around 30 m/s during the total period due to the
limited propulsion energy, while in the latter case,
with more available propulsion energy, the UAV
first flies at the maximum speed (50 m/s) to get
close to each of the GUs and then hovers around
the GU at the minimum speed (5 m/s) so as to
maximize the throughput. Finally, the achievable
throughput vs. propulsion energy is plotted in Fig.
6d. The throughput upper bound is obtained by
ignoring the propulsion energy constraint, which
is the same as that in Fig. 3d under the same T.
The throughput lower bound is achieved by the
initial circular trajectory [7] with the UAV’s speed
equal to 30 m/s. One can observe that the com-
mon throughput can be significantly improved at
the cost of more propulsion energy consumption.
In particular, as the propulsion energy increases,
the common throughput first increases rapidly
and then approaches a constant that is strictly
lower than the throughput upper bound. This is
because in addition to the propulsion energy con-
straint, the practically achievable throughput is
also subjected to the UAV’s mobility constraints
on the minimum speed and maximum accelera-
tion.
Further Discussion and Future Work
The throughput-energy trade-off can be further
extended by taking the GUs’ energy consumption
into account (e.g., in the application of UAV-en-
abled data collection in IoT networks) [4, 11].
Since the IoT devices are generally of low power
and limited battery, how to prolong their lifetime
is critical for the sustainability and proliferation of
the future IoT ecosystem. Thanks to the controlla-
ble mobility, a UAV-enabled mobile data collec-
tor can move sufficiently close to the IoT devices
such as sensors or tags to collect their data with
minimum transmit energy. However, this will incur
more propulsion energy consumption of UAVs,
which implies an interesting new perspective on
the throughput-energy trade-off in UAV-enabled
communication [11].
On the other hand, the UAVs’ energy supply
can also be provisioned by means of other tech-
nologies such as solar energy harvesting and
laser-beamed wireless power transfer by ground
chargers. However, these technologies generally
bring new design considerations that need to be
IEEE Wireless Communications • February 2019
 300

 200

 100

y(m)
0
 −100
−200
−300
−1500 −1000 −500 0 500 1000 1500
+ x(m), E = 13000 J
max
30
20

8VHU 8VHU 10

y(m)
0
§  · § ·
¨  ¸ ¨ ¸ −10
©  ¹ ˄˅ © ¹
−20
−30
−1500 −1000 −500 0 500 1000 1500
(a) x(m), E = 23000 J
max

(b)
60
5.2
UAV speed (m/s)

40
5
Upper bound
Proposed trajectory

Common throughput (bps/Hz)

20
4.8 Circular trajectory

0 4.6
0 20 40 60 80 100 120
Time (s), Emax = 13000 J

60 4.4
UAV speed (m/s)

4.2
40

4
20
3.8
0
0 20 40 60 80 100 120 1.4 1.6 1.8 2 2.2 2.4 2.6
Time (s), Emax = 23000 J Emax (Joule) x 10
4

(c) (d)
FIGURE 6. Throughput-energy trade-off for a single-UAV-enabled network with two GUs. The GUs’ nominal locations are marked by
◊ and the UAV trajectories are marked by w. For the propulsion power consumption model in [16], the constants c1 and c2 are set
as 9.26  10–4 and 2250, respectively: a) a UAV-enabled two-user wireless system with finite onboard energy; b) UAV horizontal
trajectory for different propulsion energy; c) UAV’s speed over time; d) common throughput vs. propulsion energy, Emax.

further studied. For example, for solar-powered
UAVs, while increasing the flying altitude will lead
to higher path loss with GUs, it helps harvest
more solar energy to support more flexible trajec-
tory design to adapt to the GUs’ dynamic loca-
tions and communication requirements. As such,
the throughput-energy trade-off in UAV-enabled
communication needs to be revised with carefully
designed altitude control. Furthermore, in the case
of multiple UAVs cooperatively serving the GUs,
besides their communication cooperation through
IUIC or CoMP as discussed earlier, the design
of multi-UAV trajectories also needs to consider
their individual energy availability. For example,
the propulsion energy consumption of different
UAVs should be balanced via cooperative trajecto-
ry design to maximize their endurance from a UAV
network lifetime maximization perspective.
Conclusions
In this article, we have revisited the fundamen-
tal throughput, delay, and energy trade-offs in
UAV-enabled wireless communication. In par-
ticular, we have shown that the communication
throughput and delay can be optimally trad-
ed off between each other as well as with the
UAV’s/GU’s propulsion/transmit energy via
judiciously optimizing the UAV’s trajectory with
communication resource allocation jointly for
both single-UAV and multi-UAV-enabled net-
works. Although we have focused on employing
UAVs as aerial BSs in this article, the discussed
new trade-offs are general and also applicable
to other UAV-mounted platforms [1] or cellu-
lar-connected UAV users [6]. It is worth pointing
out that besides the three trade-offs considered
in this article, there are other important design
considerations in UAV-enabled communication,
which, due to space limitations, have not been
fully explored in this article and thus require
further investigation. These may include, for
example, the deployment cost of mobile UAVs,
their wireless backhaul constraints, as well as
the severe air-ground interference issue due
to the LoS-dominant channels. Nevertheless,
it is hoped that this article has revealed some
new design trade-offs as well as useful insights
that would be helpful to the practical design
of UAV-enabled communication systems in the
future.

IEEE Wireless Communications • February 2019 43

As the propulsion References
energy increases, the [1] Y. Zeng, R. Zhang, and T. J. Lim, “Wireless Communications
with Unmanned Aerial Vehicles: Opportunities and Chal-
common throughput
lenges,” IEEE Commun. Mag., vol. 54, no. 5, May 2016, pp.
first increases rapidly 36–42.
[2] A. Merwaday et al., “Improved Throughput Coverage in
and then approach-
Natural Disasters: Unmanned Aerial Base Stations for Pub-
es a constant that is lic-Safety Communications,” IEEE Vehic. Tech., vol. 11, no. 4,
Dec. 2016, pp. 53–60.
strictly lower than [3] J. Chen and D. Gesbert, “Optimal Positioning of Flying
the throughput upper Relays for Wireless Networks: A LOS Map Approach,” Proc.
IEEE ICC, 2017, pp. 1–6.
bound. This is because [4] M. Mozaffari et al., “Mobile Unmanned Aerial Vehicles
in addition to the UAVs for Energy-Efficient Internet of Things Communica-
tions,” IEEE Trans. Wireless Commun., vol. 16, no. 11, Nov.
propulsion energy con- 2017, pp. 7574–89.
straint, the practically [5] J. Xu, Y. Zeng, and R. Zhang, “UAV-Enabled Wireless Power
Transfer: Trajectory Design and Energy Optimization,” IEEE
achievable throughput Trans. Wireless Commun., vol. 17, no. 8, Aug. 2018, pp.
is also subjected to 5092–5106.
[6] S. Zhang, Y. Zeng, and R. Zhang, “Cellular-Enabled UAV
the UAV’s mobility Communication: Trajectory Optimization Under Connectiv-
constraints on the min- ity Constraint,” Proc. IEEE ICC, 2018.
[7] Q. Wu, Y. Zeng, and R. Zhang, “Joint Trajectory and Com-
imum speed and maxi- munication Design for Multi-UAV Enabled Wireless Net-
mum acceleration. works,” IEEE Trans. Wireless Commun., vol. 17, no. 3, Mar.
2018, pp. 2109–21.
[8] Q. Wu and R. Zhang, “Common Throughput Maximiza­
tion in UAV-Enabled OFDMA Systems with Delay Consider­
ation,” IEEE Trans. Commun., vol. 66, no. 12, Dec. 2018,
pp. 6614–27.
[9] Q. Wu, J. Xu, and R. Zhang, “Capacity Characterization of
UAV-Enabled Two-User Broadcast Channel,” IEEE JSAC, vol.
36, no. 9, Sept. 2018, pp. 1955–71.
[10] Y. Zeng and R. Zhang, “Energy-Efficient UAV Communi-
cation with Trajectory Optimization,” IEEE Trans. Wireless
Commun., vol. 16, no. 6, June 2017, pp. 3747–60.
[11] D. Yang et al., “Energy Trade-Off in Ground-to-UAV Com-
munication Via Trajectory Design,” IEEE Trans. Vehic. Tech.,
vol. 67, no. 7, July 2018, pp. 6721–26.
[12] D. Tse and P. Viswanath, Fundamentals of Wireless Commu-
nication, Cambridge Univ. Press, 2005.
[13] M. Grossglauser and D. Tse, “Mobility Increases the Capac-
ity of Ad-Hoc Wireless Networks,” IEEE/ACM Trans. Net.,
vol. 4, no. 10, Aug. 2002, pp. 477–86.
[14] G. Miao, N. Himayat, and G. Y. Li, “Energy-Efficient Link
Adaptation in Frequency-Selective Channels,” IEEE Trans.
Commun., vol. 58, no. 2, Feb. 2010, pp. 545–54.
44
[15] L. Liu, S. Zhang, and R. Zhang, “CoMP in the Sky: UAV
Placement and Movement Optimization for Multi-User
Communications,” IEEE Trans. Wireless Commun., submitted;
https://arxiv.org/abs/1802.10371.
[16] Y. Zeng, J. Xu, and R. Zhang, “Energy Minimization for
Wireless Communication with Rotary-Wing UAV,” IEEE
Trans. Wireless Commun., submitted; https://arxiv.org/
abs/1804.02238.
Biographies
Qingqing Wu (elewuqq@nus.edu.sg) received his B.Eng. and
Ph.D. degrees in electronic engineering from South China Uni-
versity of Technology and Shanghai Jiao Tong University (SJTU)
in 2012 and 2016, respectively. He is currently a research fellow
in the Department of Electrical and Computer Engineering at
National University of Singapore. He was the recipient of the
Outstanding Ph.D. Thesis Funding from SJTU in 2016 and the
Best Ph.D. Thesis Award of China Institute of Communications
in 2017. He received the IEEE WCSP Best Paper Award in 2015,
the Exemplary Reviewer of IEEE Communications Letters in 2016
and 2017, and IEEE Transactions on Communications and IEEE
Transactions on Wireless Communications in 2017.
Liang Liu (eleliu@nus.edu.sg) received his Ph.D. degree from
National University of Singapore in 2014. He is currently a
research fellow in the Department of Electrical and Computer
Engineering at National University of Singapore. His research
interests include energy harvesting and machine-type communi-
cations in 5G. He was the recipient of the IEEE Signal Processing
Society Young Author Best Paper Award, 2017 and a Best Paper
Award from IEEE WCSP in 2011. He is recognized by Clarivate
Analytics as a Highly Cited Researcher, 2018.
R ui Z hang [F’17] (elezhang@nus.edu.sg) received his Ph.D.
degree from the Electrical Engineering Department of Stanford
University in 2007 and is now a Dean’s Chair Associate Pro-
fessor in the Electrical and Computer Engineering Department
of National University of Singapore. He has been listed as a
Highly Cited Researcher by Thomson Reuters since 2015. His
research interests include wireless communication and wireless
power transfer. He was the co-recipient of the IEEE Marconi
Prize Paper Award in Wireless Communications, the IEEE Signal
Processing Society Best Paper Award, the IEEE Communications
Society Heinrich Hertz Prize Paper Award, and the IEEE Sig-
nal Processing Society Donald G. Fink Overview Paper Award,
among others. He is now an Editor for IEEE Transactions on
Communications and a member of the Steering Committee of
IEEE Wireless Communications Letters.
IEEE Wireless Communications • February 2019