You are on page 1of 47

5 Multiple Antennas and Beamforming

for SWIPT Systems


Derrick Wing Kwan Ng, Shiyang Leng, and Robert Schober

5.1 Introduction

The development of wireless communication networks worldwide has triggered an


exponential growth in the number of wireless communication devices and sensors for
applications such as e-health, automated control, environmental monitoring, energy
management, and safety management. It is expected that, by 2020, the number of
inter-connected devices on the planet may reach 50 billion. Recent efforts in next-
generation communication system development aim at providing secure, ubiquitous,
and high-speed communication with guaranteed quality of service (QoS). However, the
related tremendous increase in the number of transmitters and receivers has also led to a
huge demand for energy. A relevant technique for reducing the energy consumption of
wireless devices is multiple-input multiple-output (MIMO), since it offers extra degrees
of freedom for more efficient resource allocation. In particular, multiuser MIMO, where
a transmitter equipped with multiple antennas serves multiple single-antenna receivers,
is considered an effective solution for realizing the potential performance gains offered
by multiple antennas to improve the system spectral efficiency and reduce the transmit
power. On the other hand, battery-powered mobile devices such as wireless sensors
have been widely deployed and have become critical components of many wireless
communication networks over the past decades. However, batteries have limited
energy storage capacity and their replacement can be costly or even impossible, which
creates a performance bottleneck in wireless networks. As a result, energy harvesting
technology is foreseen as a viable solution to remove the last wires of wireless devices.
The integration of energy harvesting (EH) capabilities into communication devices
facilitates self-sustainability of energy limited communication systems. Solar, wind,
hydroelectric, and piezoelectric are the major conventional energy sources for EH. For
instance, energy harvesters for harvesting wind and solar energy have been successfully
integrated into base station transmitters for providing communication services in remote
areas [1]. However, the availability of these natural energy sources is usually limited by
location, climate, and time of day. Besides, the implementation of conventional energy
harvesters may be problematic and renewable energy from natural sources may not be

Wireless-Powered Communication Networks: Architecture, Protocols, and Applications, ed. Dusit Niyato,
Ekram Hossain, Dong In Kim, Vijay Bhargava, and Lotfollah Shafai. Published by Cambridge University
Press. © Cambridge University Press 2017.

170
Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
Multiple Antennas and Beamforming for SWIPT Systems 171

available in indoor environments. Thus, a new form of controllable energy source for
portable wireless devices is needed in order to extend the lifetime of communication
networks.

5.1.1 Background
In 1899, Nikola Tesla proposed the wireless transmission of electrical power via a
magnifying transmitter, an advanced version of the Tesla coil transmitter. The use of
wireless power transfer (WPT) avoids the potentially high costs of planning, installing,
displacing, and maintaining power cables in buildings and infrastructure. Despite its
convenience, one of the major challenges in realizing WPT is its low power transfer effi-
ciency. In practice, wireless power has to be transferred via a carrier signal with a high
carrier frequency such that antennas of reasonable size can be used for harvesting power.
The associated path loss severely attenuates the signal, leading to only a small amount
of power being harvested at the receiver. Besides, the initial efforts on WPT focused on
high-power-consumption applications. This raised serious public health concerns about
strong electromagnetic radiation which prevented the further development of WPT in
the late twentieth century. As a result, this area developed slowly until recent advances
in silicon technology and multiple-antenna technology made WPT attractive once again.
In particular, the breakthrough in silicon technology has significantly reduced the energy
demand of simple wireless devices. Thus, harvesting energy1 from background radio-
frequency (RF) signals originating from ambient transmitters can support the power
needs of low-power-consumption receivers. Besides, multiple-antenna technology has
revolutionized the design of traditional communication systems for a better utilization
of limited system resources. For example, a multiple-antenna transmitter can focus its
transmitted signal into certain locations to improve the signal reception at the receivers.
It has been shown that the use of multiple antennas in communication systems can
significantly reduce the total transmit power and improve the system energy efficiency
for given QoS requirements. Thus, it is envisioned that multiple-antenna technology is
also the key to unlock the potential of WPT.

5.1.2 Literature Survey


Radio-frequency signals are an abundant source of energy for EH [2–6]. Nowadays, EH
circuits are able to harvest microwatts to milliwatts of power over a range of several
meters for a transmit power of 1 W and a carrier frequency of less than 1 GHz [7]. For
instance, Intel has demonstrated the wireless charging of a temperature and humidity
meter as well as a liquid-crystal display by using the signals radiated by a TV station
4 km away [8]. Thus, RF energy can be a viable energy source for devices with low
power consumption, e.g., wireless sensors [4, 5]. There have been some preliminary
applications of wireless energy transfer such as wireless body area networks (WBANs)

1 In this chapter, a normalized energy unit, i.e., Joule-per-second, is considered. Therefore, the terms “power”
and “energy” are used interchangeably.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
172 Derrick Wing Kwan Ng et al.

for biomedical implants, passive radio-frequency identification (RFID) systems, and


wireless sensor networks. Moreover, RF EH provides the possibility of simultaneous
wireless information and power transfer (SWIPT) since RF signals carry both informa-
tion and energy. Yet, this new technology introduces a paradigm shift in system and
resource allocation algorithm design. In particular, the role of multiple-antenna designs
in SWIPT has to be carefully investigated due to the imposed new challenges and QoS
requirements on efficient WPT. In the following, we provide a brief literature survey on
SWIPT systems.
The fundamental tradeoff between wireless power transfer and wireless information
was studied in [2] and [3] for flat fading and frequency selective fading, respectively.
Specifically, a theoretical optimal receiver was assumed in [2, 3] such that EH and infor-
mation decoding can be performed simultaneously on the same received signal, which
is not implementable in practice yet. As a compromise solution, three different types
of receivers, namely power-splitting, separated, and time-switching, were proposed in
[9, 10]. In particular, the power-splitting receiver splits the received power into two
power streams with a certain power-splitting ratio to facilitate simultaneous EH and
information decoding in the same receiver [11–16]. The authors of [9] and [10] inves-
tigated the rate–energy tradeoff regions for two-receiver and point-to-point systems,
respectively. On the other hand, the authors of [11] focused on the power allocation
algorithm design in ergodic fading channels for a point-to-point single-user SWIPT
system with a power-splitting receiver. The authors of [12], by taking into account the
power consumption in electronic circuitries and RF transmission, proposed different
power allocation algorithms and showed that introducing power-splitting receivers can
improve the energy efficiency of a communication system. On the other hand, SWIPT
raises concerns regarding communication security due to the broadcast nature of wire-
less channels and the relatively high signal power for SWIPT. As a result, beamformer
design for multiple-antenna SWIPT systems with consideration of the physical layer
security was studied in [13, 16–18]. The authors of [13] and [16], by taking into account
potential eavesdropping by EH receivers, designed beamformers for minimization of
the total transmit power for the cases of perfect channel state information (CSI) and
imperfect CSI, respectively. A multi-objective framework was adopted in [17] to handle
conflicting system design goals for providing communication security while guarantee-
ing QoS in WPT to EH receivers. In [18], beamforming design was investigated for the
maximization of secrecy rate. In [13, 16–18], artificial noise generation and multiple
antennas were used to ensure secure SWIPT. Specifically, the artificial noise serves as
interference to degrade the channel quality of potential eavesdroppers and acts as an
energy source for expediting energy harvesting at the receivers.
The aforementioned works in the literature suggest that multiple-antenna systems
provide a higher energy transfer efficiency than single-antenna systems. Specifically,
multiple-antenna transmitters utilize the spatial degrees of freedom to create directional
signal transmission, which is crucial for improving the overall system performance.
In the following, we focus on beamforming design for multiple-antenna SWIPT
systems and investigate different practical design problems. The remainder of this
chapter is organized as follows. In Section 5.2, we describe the adopted system model.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
Multiple Antennas and Beamforming for SWIPT Systems 173

In Sections 5.3 and 5.4, we formulate and solve single-objective and multi-objective
optimization problems arising in the design of SWIPT systems, respectively. Section 5.5
studies beamforming design for guaranteeing secure communications in SWIPT
systems. In Section 5.6, we discuss some research challenges in multiple-antenna
SWIPT systems and paths to potential solutions. Section 5.7 concludes this chapter with
a brief summary. Now, we first introduce the main notations adopted in this chapter.

5.1.3 Notation
The key mathematical notations are summarized in Table 5.1. We use boldface capital
and lower-case letters to denote matrices and vectors, respectively. AH , Tr(A), and
Rank(A) represent the Hermitian transpose, the trace, and the rank of matrix A, respec-
tively; A 0 and A  0 indicate that A is a positive definite and a positive semidef-
inite matrix, respectively; vec(A) denotes the vectorization of matrix A by stacking its
columns from left to right to form a column vector; λmax A denotes the maximum
eigenvalue of Hermitian matrix A; IN is the N × N identity matrix; CN×M and RN×M
denote the set of all N × M matrices with complex and real entries, respectively; HN
denotes the set of all N × N Hermitian matrices; diag(x1 , . . . , xK ) denotes a diagonal
matrix with the diagonal elements given by {x1 , . . . , xK }; |·| and · denote the absolute
value of a complex scalar and the l2 -norm of a vector, respectively; the circularly
symmetric complex Gaussian (CSCG) distribution is denoted by CN (μ, σ 2 ) with mean
μ and variance σ 2 ; ∼ stands
 for “distributed as”; 1 denotes a column vector with all
elements equal to one. · a,b returns the (a, b)th element of the input matrix, θ n is the
   
nth unit column vector, i.e., θ n t,1 = 1, if t = n, and θ n t,1 = 0, ∀t = n.

Table 5.1. Nomenclature adopted in this chapter

Notation Description

h Channel vector between the transmitter and the information receiver


gj Channel vector between the transmitter and energy receiver j
w Information beamforming vector
wE Energy signal vector
v Artificial noise vector
2 , σ2
σant Antenna and signal processing noise power
s
ρ Power splitting ratio
req Minimum required signal-to-noise-plus-interference ratio
NT Number of transmit antennas
RTol
ER Maximum tolerable data rate
ξ Power amplifier efficiency
ηIR Information rate energy efficiency
ηEH Energy transfer energy efficiency
Pmax Maximum transmit power of the transmitter
Pmaxn Maximum transmit power of antenna n
Pant Antenna power consumption
Pc Baseband signal processing circuit power consumption
Pminj Minimum required power transfer to receiver j

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
174 Derrick Wing Kwan Ng et al.

Figure 5.1 A multiuser SWIPT downlink model with an information receiver (IR) and J = 2 idle
receivers. The idle receivers work as energy harvesting receivers (ERs) to harvest energy from
the received radio-frequency (RF) signals for extending their lifetimes.

5.2 System Model

We consider a downlink SWIPT system which consists of a transmitter, one information


receiver (IR), and J energy harvesting receivers (ERs). The transmitter is equipped with
NT transmit antennas while the receivers are single-antenna devices, see Figure 5.1.
The transmission is divided into time slots. In each time slot, the transmitter conveys
information to a given receiver and transfers energy to all receivers. The downlink
received signals at the IR and energy harvesting receiver j are given by, respectively,

y = hH x + za , (5.1)
yER
j = j x + zj ,
gH ∀j ∈ {1, . . . , J}, (5.2)

where x ∈ CNT ×1 denotes the transmitted symbol vector. hH ∈ C1×NT is the channel
vector between the transmitter and the desired receiver and gH j ∈ C
1×NT is the channel

vector between the transmitter and idle receiver j. We note that both variables, h and gj ,
include the effects of the multi-path fading and path loss of the associated channels. za
and zj are additive white Gaussian noises (AWGNs) resulting from the receive antenna
2 and σ 2 , respectively.
at the IR and ER j, respectively, with zero mean and variance σant antj
This simple system model is adopted for illustration of the beamforming design in
multiple-antenna SWIPT systems in Sections 5.3 and 5.4 in this chapter.

5.3 Single-Objective Optimization

The design of a single-objective utility-based resource allocation algorithm is the key


for optimal utilization of the physical layer resources. In particular, different resource

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
Multiple Antennas and Beamforming for SWIPT Systems 175

allocation algorithm designs have been studied in the literature for different wire-
less communication systems [19–24]. Specifically, the use of limited physical layer
resources such as bandwidth, energy, time, and space is optimized for achieving
different system design goals such as maximum spectral efficiency and power efficiency.
In practice, by exploiting the extra degrees of freedom offered by multiple antennas,
information/energy beamforming can be effectively performed when the channel
state information (CSI) of the receivers is available at the transmitter. There are
two main approaches for downlink CSI acquisition. In frequency division duplex
(FDD) systems, the CSI can be acquired by feeding back the CSI from the receivers
to the transmitter. On the other hand, in time division duplex (TDD) systems, the
CSI can be obtained by exploiting the channel reciprocity between the uplink and
downlink channels. In particular, the downlink CSI of the transmitter to the receivers
can be estimated by measuring the uplink training sequences embedded in handshaking
signals.
In addition to the conventional QoS requirements such as throughput, energy effi-
ciency, fairness, and delay, the efficient transfer of energy plays an important role as a
new QoS requirement in SWIPT systems. It is expected that new resource allocation
algorithm designs will be needed to satisfy this requirement. For instance, one possible
design goal is the optimization of the covariance matrix of the transmit signal for
achievable rate maximization of an IR subject with respect to a minimum required
amount of energy transferred to the ERs. In the following, we first study different
beamformer designs for achieving different system objectives by assuming that perfect
CSI is available at the transmitter.

5.3.1 Energy Beamforming


In this section, we focus on beamformer design to facilitate efficient wireless power
transfer (WPT) from the transmitter to the receivers. In particular, we adopt random
signals as energy carriers and the transmitter chooses the transmit signal vector x as

x= ws
!"# , (5.3)
random signal

where s ∈ C1×1 and w ∈ CNT ×1 are the pseudo-random transmit signal and the corre-
sponding beamforming vector. We assume without loss of generality that E{|s|2 } = 1.
In practice, random energy signals may be preferable insofar as they spread the sig-
nal power evenly over the operating bandwidth and avoid the power spikes typical of
deterministic sinusoidal carrier signals.
Different hardware circuitries [25, 26] are available for harvesting energy from the
RF. The associated system models and the corresponding energy harvesting efficiencies
can be significantly different. Therefore, we do not assume a particular type of energy
harvesting circuit, and hence the following discussion of beamformer design is isolated
from the specific hardware implementation details. The total harvested RF power at ER
j is modeled as

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
176 Derrick Wing Kwan Ng et al.

Qj = ηj E{|gH
j x| }
2

= ηj |gH
j w|
2

= ηj Tr(gH H
j ww gj ), (5.4)
where 0 ≤ ηj ≤ 1 is a constant that accounts for the RF-to-electrical energy conversion
efficiency.
Now, we examine the optimal transmit covariance matrix, W = wwH , for maximiza-
tion of the total power transfer. The optimal transmit covariance matrix design can be
formulated as the following optimization problem.

Problem 1. Total harvested power maximization:


maximize Tr(WG)
W∈HNT
subject to Tr(W) ≤ Pmax , W  0, (5.5)

where Pmax is the maximum transmit power budget for the transmitter and G =
J
j=1 ηj gj gj is the equivalent channel between the transmitter and the J ERs. Let R =
H

min{NT , J} and let the singular value decomposition of G be given by G = U 1/2 VH ,


where  = diag(g1 , g2 , . . . , gR ) with g1 ≥ g2 ≥ . . . ≥ gR . U ∈ CJ×R and V ∈ CJ×R
are two matrices with orthonormal columns. Besides that, we denote v1 as the first
column of V. Then, we introduce the following proposition for revealing the solution
of Problem 1.
proposition 5.1. [27] The optimal solution for Problem 1 is given by W =
Pmax v1 vH
1.

In other words, the maximum total harvested power can be achieved by beamforming.
In particular, the energy beam, w, aligns with the direction of the strongest eigenmode
of the matrix GGH . This transmission strategy is known as energy beamforming in the
literature [27].
Up to now, we have studied only WPT to the receivers without considering the
possibility of concurrent wireless information transfer. In fact, practical receivers for
information decoding cannot be used to harvest energy from the same RF signals due
to the different nature of the signal processing and receiver sensitivities required [26].
In other words, the received RF signal exploited for energy harvesting cannot be reused
for information decoding, and vice versa. As a result, a special receiver structure is
required in order to enable SWIPT. In the next section, we focus on two commonly used
receiver architectures, namely, “separated receivers” and “power splitting receivers,”
both of which facilitate SWIPT. Also, we develop corresponding beamforming designs
for better utilization of the limited system resources.

5.3.2 SWIPT: Separated Receivers


In a separated receiver architecture, an EH circuit and an ID circuit are implemented
as two separate receivers, which are served by a common multiple-antenna transmitter

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
Multiple Antennas and Beamforming for SWIPT Systems 177

[26–39]. Separated receivers can be easily implemented by using off-the-shelf compo-


nents for the two individual receivers. On the other hand, the transmitter chooses the
transmit signal vector x as

x= ws
!"# , (5.6)
beamformed information signal

where s ∈ C1×1 now represents the information-bearing


  signal. As in the case of WPT,
we assume without loss of generality that E |s| = 1.
2

5.3.2.1 Beamforming Design


In this section, we focus on beamforming design for maximization of the system achiev-
able rate, which can be formulated as the following optimization problem.

Problem 2. Achievable rate maximization:


 
Tr(WH)
maximize log2 1 + 2
W∈HNT σant + σs2
subject to C1 : Tr(WG) ≥ Pmin ,
C2 : Tr(W) ≤ Pmax , (5.7)

where H = hhH , Pmin in constraint C1 is the minimum required total system power
transfer, and σs2 is the signal processing noise at the receiver, which is modeled as
AWGN.
The optimal transmit covariance matrix is given by [27]
 
1 2 + σ2
σant
∗ −1
W =B h − s
hH B−1 , (5.8)
B−1 h2 B−1 h4

where B = μ∗ IT − λ∗ G. Here, λ∗ and μ∗ are the optimal dual variables which can be
found by the gradient method or the ellipsoid method for satisfying constraints C1 and
C2 , respectively. It can be observed from (5.8) that the optimal covariance is a rank-
one matrix, i.e., Rank(W) = 1. In other words, beamforming is the optimal strategy
to maximize the achievable rate while guaranteeing a minimum required total system
power transfer. Specifically, the transmit beamforming direction should align with the
vector B−1 h which represents the equivalent channel spanned jointly by the channels of
the IR and the ERs.

5.3.2.2 Numerical Example


In this section, we present a numerical example to illustrate the benefits of multiple
transmit antennas and multiple ERs for SWIPT. The most important simulation param-
eters are listed in Table 5.2 and the maximum transmit power Pmax is set to 30 dBm. We
assume that all the receivers are located 10 meters away from the transmitter.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
178 Derrick Wing Kwan Ng et al.

Table 5.2. System parameters

Parameter Value

Carrier center frequency 915 MHz


Path-loss exponent 2
Path-loss model TGn Model A [40]
Multi-path fading distribution Rician fading
Rician factor 6 dB
Transmit and receive antenna gain 10 dBi and 0 dBi
Noise variance, σs2 −23 dBm
2
Antenna noise variance, σant −114 dBm
System bandwidth 200 kHz
RF-to-electrical energy conversion efficiency, ηj 0.5

6.6
Harvested power gain
Average achievable rate (bit/s/Hz)

due to extra ERs


6.4

6.2

Achievable rate gain


6 due to extra N T

5.8 N T = 4, J = 2
N T = 4, J = 4
5.6 N T = 8, J = 2
N T = 8, J = 4
5.4
200 300 400 500 600 700 800 800
Average total harvested power (mW)

Figure 5.2 Average achievable rate (bit/s/Hz) versus the average total harvested power (μW) for
one IR, different numbers of ERs, J, and different numbers of transmit antennas, NT .

Figure 5.2 depicts an example of beamforming in SWIPT systems. We show the


average system achievable rate versus the average total harvested energy in a down-
link system for the optimal beamforming scheme. In particular, a transmitter equipped
with NT antennas serves one single-antenna IR and J single-antenna ERs. As can be
observed, with the optimal beamformer, the tradeoff region of the system achievable rate
and the harvested energy increases significantly with NT . This due to the fact that the
extra degrees of freedom offered by multiple transmit antennas help the transmitter to
focus its energy and thus improve the beamforming efficiency. Besides, the average total
system harvested energy improves with the number of ERs as more receivers participate
in energy harvesting.
In the next section, we focus on the general case where the IR wants to decode the
modulated information and harvest energy from the received signal.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
Multiple Antennas and Beamforming for SWIPT Systems 179

Information and power transfer Signal processing noise


Power ss2
amplifier
Antenna noise Power-splitting ratio
=
Transmitter s 2ant r
Receiver signal
signal processing core
Processing Power-splitting ratio Rechargeable
core Battery
= 1–r Energy
harvesting circuit
Power-splitting
unit
Power-splitting receiver

Figure 5.3 The block diagram of the transceiver model for wireless information and power
transfer with a power-splitting receiver.

5.3.3 SWIPT: Power-Splitting Receivers


In this section, we consider power-splitting receivers [11–16], which split the received
signal into two power streams with power-splitting fractions 1 − ρ and ρ, see Fig-
ure 5.3, for harvesting energy and decoding the modulated information, respectively.
Specifically, the power-splitting unit is installed in the analog front-end of the receiver
and is assumed to be a perfect passive analog device; it does not introduce any extra
power gain, i.e., 0 ≤ ρ ≤ 1, and does not add noise to the received signal. Indeed,
the power-splitting receiver architecture is a generalization of traditional information
receivers (IRs) and energy receivers (ERs). In fact, on imposing power splitting fractions
of ρ = 1 and ρ = 0, the power-splitting receiver reduces to a traditional IR and ER,
respectively. Hence, the separated receiver studied in the previous section is a special
case of a power-splitting receiver.

5.3.3.1 System Optimization


In the following, we study the power-efficient resource allocation design for SWIPT
networks. We assume that there is one power-splitting receiver and J pure ERs in the
system. The power-efficient system optimization can be formulated as the following
total transmit power minimization problem.

Problem 3. Total transmit power minimization:


minimize w2
w,ρ

ρ|hH w|2
subject to C1 : 2 + σ2
≥ req ,
ρσant s
C2 : (1 − ρ)η|gH
j w| + (1 − ρ)ησant ≥ Pmin ,
2 2

C3 : ηj |gH
j w| + ηj σant ≥ Pminj , ∀j ∈ {1, . . . , J},
2 2
$ %
C4 : wwH ≤ Pmaxn , ∀n ∈ {1, . . . , NT },
n,n
C5 : 0 ≤ ρ ≤ 1, (5.9)

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
180 Derrick Wing Kwan Ng et al.

where the variable req in C1 specifies the minimum requirement on the signal-
to-noise ratio (SNR) of the receiver for information decoding. η in C2 denotes the
RF-to-electrical energy conversion efficiency for IR. Pmin and Pminj in C2 and C3 are
the minimum required power transfer to the power-splitting IR and ER j, respectively.
In C4 , Pmaxn denotes the maximum transmit power from antenna n, since each antenna
is powered by an individual power amplifier. C5 is the boundary constraint for the
power-splitting variable ρ.
The problem in (5.9) is a non-convex optimization problem. In particular, constraints
C1 –C3 are non-convex, which does not facilitate the design of a computationally effi-
cient beamformer. In order to design a tractable beamfomer, we first rewrite (5.9) in an
equivalent2 form:

minimize Tr(W)
W∈HNT ,ρ
 
σ 2
subject to C1 : Tr(HW) ≥ req σant 2
+ s ,
ρ
Pmin
C2 : Tr(HW) + σant
2
≥ ,
(1 − ρ)η
Pminj
C3 : Tr(Gj W) + σant
2
≥ , ∀j ∈ {1, . . . , J},
ηj
 
C4 : Tr  n W ≤ Pmaxn , ∀n ∈ {1, . . . , NT },
C5 : 0 ≤ ρ ≤ 1,
C6 : W  0,
C7 : Rank(W) ≤ 1, (5.10)

where Gj = gj gH j . We note that the per-antenna transmit power in constraint C4 in


(5.10) can be represented as Tr( n W), where  n = θ n θ H NT ×1 is the nth
n and θ n ∈ R
unit vector of length NT defined in the notation section. Now, the only non-convexity
in problem (5.10) is due to constraint C7 . In particular, constraint C7 is a combinatorial
constraint that requires a brute force search for finding a global optimal solution. To
circumvent the non-convexity, we adopt a semidefinite programming (SDP) relaxation
to (5.10) by relaxing constraint C7 : Rank(W) = 1, i.e., removing it from the problem
formulation, which leads to the following problem.

Problem 4. SDP relaxation of Problem 3:


minimize Tr(W)
W∈HNT ,ρ
subject to C1 –C6 ,
(
C7 : ( ((((
Rank(W) ≤ 1. (5.11)

2 In this chapter, “equivalent” means that two problem formulations share the same optimal solution.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
Multiple Antennas and Beamforming for SWIPT Systems 181

Problem 4 is a convex SDP that can be solved efficiently by standard numerical


solvers for convex programs such as SeDuMi [41] and SDPT3 [42]. From the basic
principles of optimization theory, if the obtained solution W for the relaxed problem
admits a rank-one matrix, then it is the optimal solution of the original problem in (5.10).
Then, the optimal w can be obtained by performing an eigenvalue decomposition of W.
However, in general, the constraint relaxation may not be tight since it is possible that
Rank(W) > 1. In the following, we reveal the tightness of the adopted SDP relaxation
in (5.11) via examination of the dual problem and the Karush–Kuhn–Tucker (KKT)
conditions of (5.11). To this end, we need the Lagrangian function of (5.11), which is
given by

L(W, ρ, α, β, φ, γ , δ, θ , Y)
  
σ2
= Tr(W) − Tr(YW) + α − Tr(HW) + req σant
2
+ s
ρ
 
Pmin
+ β − Tr(HW) − σant
2
+
(1 − ρ)η

J  
Pminj
+ φj − Tr(Gj W) − σant +
2
ηj
j=1


NT    
+ γn Tr  n W − Pmaxn − δρ + θ (ρ − 1). (5.12)
n=1

Here, α ≥ 0 is the dual variable for the minimum required signal-to-interference-


plus-noise ratio (SINR) of the hybrid receiver in C1 . φ is the vector of dual variables
of the requirement for a minimum transferred power in C3 with elements φj ≥ 0,
j ∈ {1, . . . , J}. The dual variable β ≥ 0 corresponds to the minimum required power
transfer to the desired receiver in C2 . γ , with elements γn ≥ 0, n ∈ {1, . . . , NT },
is the dual variable vector associated with the per-antenna maximum transmit power
constraint in C4 . δ, θ ≥ 0 are the dual variables for the boundary constraints of opti-
mization variable ρ in C5 . The matrix Y  0 is the dual variable for the semidefinite-
ness constraint on matrix W. Thus, the dual problem for the SDP relaxed problem is
given by

maximize minimize L(W, ρ, α, β, φ, γ , δ, θ , Y). (5.13)


α,β,φ,γ ,δ,θ≥0, W∈HNT ,ρ
Y0

Now, we reveal the tightness of the SDP relaxation adopted in (5.11) in the following
theorem.

theorem 5.1. Assuming that the channel vectors of the IR, h, and the ERs, gj , j ∈
{1, . . . , J}, can be modeled as statistically independent random variables, the solution
of (5.11) is rank-one, i.e., Rank(W) = 1, with probability one.

Proof. Please refer to Section 5.8.1.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
182 Derrick Wing Kwan Ng et al.

In other words, the SDP relaxation is tight whenever the channels satisfy the condition
stated in Theorem 5.1. Besides, the optimal beamformer for (5.9), w∗ , can be obtained
with probability one by performing an eigenvalue decomposition of the solution W of
(5.11) and selecting the principal eigenvector as the beamformer.

5.3.3.2 Numerical Examples


In this section, we evaluate the system performance of the proposed beamformer design
via simulations. The simulation parameters adopted are listed in Table 5.2. In particular,
we assume that there are one-power splitting IR and J = 3 ERs, and that they are
located 10 m away from the transmitter. The maximum transmit power per antenna
is set to Pmaxn = 30 dBm, ∀n ∈ {1, . . . , NT }, and ηj = η = 0.5. The average total
transmit power of the transmitter is obtained by averaging over different multi-path
fading realizations.

Average Total Transmit Power


Figure 5.4 depicts the average total transmit power versus the minimum required SNR
of the power splitting receiver, req , for different resource allocation schemes. It can
be observed that the average total transmit power of the proposed scheme is a mono-
tonically increasing function of req . This is due to the fact that a higher transmit
power is required for satisfying constraint C1 when the requirement on req becomes
more stringent. Besides, the total transmit power decreases with increasing number
of transmit antennas NT . The extra degrees of freedom offered by the multiple trans-
mit antennas help in focusing energy on the power-splitting receiver, which improves
the power efficiency of the system. Furthermore, we compare our optimal resource

36 N T = 8, optimal scheme Baseline scheme


Average total transmit power (dBm)

34 N T = 8, baseline scheme
N T = 10, optimal scheme
32 N T = 10, baseline scheme
N T = 12, optimal scheme
30
N T = 12, baseline scheme
28

26

24 Opitmal scheme

22

20

18
6 8 10 12 14 16 18 20 22 24
Minimum requried SNR (dB)

Figure 5.4 Average total transmit power (dBm) versus the minimum required SNR of the
power-splitting receiver, req (dB), for different resource allocation schemes and different
numbers of transmit antennas, NT .

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
Multiple Antennas and Beamforming for SWIPT Systems 183

allocation scheme with a baseline scheme. Specifically, a fixed power-splitting ratio


of ρ = 0.5 and maximum ratio transmission (MRT) are adopted in the baseline scheme
for delivering the information signal to the power-splitting receiver. Then, we optimize
the power allocated to the MRT beamforming matrix W for the minimization of the
total transmit power subject to the same constraints as in (5.11). We note that, although
the MRT matrix is a rank-one matrix, the solution obtained is generally a suboptimal
solution with respect to (5.9). It can be observed that the proposed optimal scheme has
a superior performance, i.e., a lower total transmit power, to the baseline scheme. In
particular, a 3 dB power saving is achieved with the optimal scheme compared with
the baseline scheme in the high-SNR regime, because of the proposed optimization
framework.

Average Power-Splitting Ratio


It can be observed from Figure 5.5 that the average power-splitting ratio increases with
the required SNR, req . As the SNR requirement becomes more stringent, the power-
splitting receiver is forced to use more of the received signal power for information
decoding at the power-splitting receiver to improve the receive SNR. In other words,
adopting a fixed power-splitting ratio, e.g. baseline scheme, is strictly suboptimal for
reducing the total transmit power in general. On the other hand, it is expected that
the number of transmit antennas does not have a large impact on the value of the
power-splitting ratio. This is because, for a fixed receive SNR requirement at the power-
splitting receiver, the transmitter has to maintain the same level of receive signal strength
at the power-splitting receiver regardless of the number of transmit antennas. In fact,
the extra degrees of freedom offered by the increased number of transmit antennas are

0.9 NT = 8
NT = 10
Average power-splitting ratio r

0.8 NT = 12
0.7

0.6

0.5

0.4

0.5

0.2

0.1

0
6 9 12 15 18 21 24
Minimum requried SNR (dB)

Figure 5.5 Average power-splitting ratio versus the minimum required SNR of the
power-splitting receiver, req (dB), for the proposed optimal resource allocation scheme.
The groups of bars from left to right represent NT = 8, 10, 12, respectively.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
184 Derrick Wing Kwan Ng et al.

Average total harvested power at power-splitting receiver

600
harvested power (mW)

Optimal scheme, NT = 8
Average total

Baseline scheme, NT = 8
400
Minimum required power transfer

200

0
6 9 12 15 18 21 24
Minimum requried SNR (dB)

Average total harvested power per ER


harvested power (mW)

800
Optimal scheme, NT = 8
Average total

Baseline scheme, NT = 8
600
Minimum required power transfer
400

200

0
6 9 12 15 18 21 24
Minimum requried SNR (dB)

Figure 5.6 Average total harvested power at the power-splitting receiver and per ER, versus the
minimum required SNR of req (dB). Each group of bars from left to right represents the
proposed optimal resource allocation scheme and the baseline scheme, respectively.

exploited to reduce the transmit power, but not to increase the receive SNR at the power-
splitting receiver.

Average Total Harvested Power


Figure 5.6 shows the average total harvested power versus the minimum required SNR
of the power-splitting receiver, req , for different resource allocation schemes and
NT = 8. It can be seen from the upper half of Figure 5.6 that the proposed optimal
resource allocation scheme always satisfies the minimum required power transfer
constraint C2 with equality, despite the increasing required SNR, which confirms the
observation from the KKT conditions in Section 5.8.1. In contrast, for the baseline
scheme, an exceedingly large amount of power is transferred to the power-splitting
receiver which leads to an unnecessarily large transmit power, see Figure 5.4. On the
other hand, it can be observed from the lower half of Figure 5.6 that the total average
power harvested per ER increases with req . In fact, the transmitter has to allocate
more power to the information-bearing signal to achieve a larger req . As a result, more
power is available in the RF and can be harvested by the ERs.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
Multiple Antennas and Beamforming for SWIPT Systems 185

34
Average total transmit power (dBm)

32

30

28

26

24 NT = 8, optimal scheme
NT = 8, baseline scheme
22 NT = 10, optimal scheme
NT = 10, baseline scheme
20
NT = 12, optimal scheme
NT = 12, baseline scheme
18
50 100 150 200 250 300
Minimum required power transfer Pmin (mW)

Figure 5.7 Average total transmit power (dBm) versus the minimum required power transfer per
receiver, Pmin (μW), for different resource allocation schemes and different numbers of transmit
antennas, NT .

Average Total Transmit Power versus Pmin


Figure 5.7 shows the average total transmit power versus the minimum required power
transfer to the receivers, Pmin , for different resource allocation schemes. We assume
that both the IR and the ERs require the same minimum amount of power transfer,
i.e., Pmin = Pminj , ∀j ∈ {1, . . . , J}. The minimum required SNR of the power-splitting
receiver is set to req = 6 dB. It is expected that the total transmit power increases
with the minimum required power transfer. In fact, as the minimum required power
transfer becomes more stringent, the feasible solution set shrinks, which reduces the
flexibility of the optimal beamforming. On the other hand, the average total transmit
power decreases with increasing number of transmit antennas. The extra degrees of
freedom offered by the multiple transmit antennas facilitate a more power-efficient
resource allocation. Besides, the proposed optimal scheme provides a substantial power
saving compared with the baseline scheme owing to the optimization adopted.

5.3.4 Robust Beamforming


In the last section, perfect CSI of all receivers is assumed to be available at the transmit-
ter for beamformer design. In practice, the IR performs handshaking with the transmitter
at the beginning of each scheduling slot. Besides, the IR is required to send positive
acknowledgement (ACK) packets to inform the transmitter of successful reception of
the information packets. Hence, the transmitter is able to track and update the CSI
estimate of the IR frequently, on the basis of training sequences or feedback information
in handshaking signals and ACK packets. Therefore, perfect CSI for the transmitter-to-
IR link, i.e., h, can be assumed to be available during the entire transmission period.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
186 Derrick Wing Kwan Ng et al.

However, the ERs are idle and there is no interaction between them and the transmitter
after handshaking. As a result, the CSI of the ERs becomes outdated during transmis-
sion, which should be taken into account for the design of beamformers.
To capture the impact of the CSI imperfection for the beamformer design, we adopt
a deterministic model [43–46] for modeling the CSI uncertainty. In particular, the CSI
of the link between the transmitter and ER j is modeled as

gj = ĝj + gj , ∀j ∈ {1, . . . , J}, (5.14)


 
j  gj ∈ CNT ×1 : gH j gj ≤ εj , ∀j,
2
(5.15)

where ĝj ∈ CNT ×1 is the CSI estimate available at the transmitter at the beginning
of a scheduling slot and gj represents the unknown channel uncertainty due to the
time-varying nature of the channel during transmission. The continuous set j in (5.15)
defines a Euclidean sphere and contains all possible channel uncertainties. Specifically,
the radius εj represents the size of the sphere and defines the uncertainty region of the
CSI of ER j.

remark 1. In practice, the value of εj2 depends on the coherence time of the
associated channel, the channel estimation method adopted, and the duration of
transmission.

5.3.4.1 Beamforming for Robust and Power-Efficient SWIPT


In this section, we consider beamformer design for separated receivers when the CSI
imperfectness of the ERs is taken into account. The algorithm design is formulated as
the following optimization problem.

Problem 5. Robust beamforming:


minimize w2
w
wH Hw
subject to C1 : 2 + σ2
≥ req ,
σant s
$ %
C2 : wwH ≤ Pmaxn , ∀n ∈ {1, . . . , NT },
n,n
C3 : min ηj |gH
j w| ≥ Pminj , ∀j ∈ {1, . . . , J}.
2
(5.16)
gj ∈j

This problem formulation aims at minimizing the transmit power while guaranteeing
a minimum required power transfer to ER j, assuming that the estimation error gj is
in the set j . Problem 5 is a non-convex optimization problem. In particular, constraint
C1 and constraint C3 span a non-convex feasible solution set, which is an obstacle in
designing computationally efficient beamformers. In order to obtain a tractable result,
we re-write Problem 5 in equivalent form as

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
Multiple Antennas and Beamforming for SWIPT Systems 187

minimize Tr(W)
W∈HNT
Tr(HW)
subject to C1 : 2 + σ2
≥ req ,
σant s
 
C2 : Tr  n W ≤ Pmaxn , ∀n ∈ {1, . . . , NT },
C3 : min ηj Tr(WGj ) ≥ Pminj , ∀j ∈ {1, . . . , J},
gj ∈j
C4: Rank(W) ≤ 1. (5.17)

We note that constraint C4 is non-convex with respect to the optimization variables.


Although constraint C3 is a convex constraint, it represents semi-infinite constraints that
are generally intractable for beamforming design. To facilitate the solution, we trans-
form constraint C3 into linear matrix inequalities (LMIs) using the following lemma.

lemma 1 (S-Procedure [47]). Let a function fm (x), m ∈ {1, 2}, x ∈ CN×1 , be defined
as

fm (x) = xH Am x + 2 Re{bH
m x} + cm , (5.18)

where Am ∈ HN , bm ∈ CN×1 , and cm ∈ R. Then, the implication f1 (x) ≤ 0 ⇒ f2 (x) ≤


0 holds if and only if there exists a δ ≥ 0 such that
& ' & '
A b1 A b2
δ H1 − H2  0, (5.19)
b1 c1 b2 c2

provided that there exists a point x̂ such that fk (x̂) < 0.

Now, we apply Lemma 1 to constraint C3 . In particular, we substitute gj = ĝj + gj


into constraint C3 . Therefore, the implication,
   
gj gj ≤ εj ⇒ C3 :0 ≥ − max gj W gj + 2 Re ĝj W gj + ĝj Wĝj
H 2 H H H
gj ∈j
Pminj
+ , ∀j, (5.20)
ηj
holds if and only if there exists a δj ≥ 0 such that the following LMI constraint holds:
( )
δj INT Wĝj
C3 :SC3j (W, δj ) = H
ĝj W −δj εj2 − Pminj /ηj + ĝH
j Wĝj
( )
δk INT 0
= + UH gj WUgj  0, ∀j, (5.21)
0 −δj εj2 − Pminj /ηj
$ %
for δj ≥ 0, j ∈ {1, . . . , J}, where Ugj = INT ĝj .
Then, we adopt SDP relaxation to handle the non-convexity originating from con-
straint C4 and the SDP relaxation problem is given as follows.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
188 Derrick Wing Kwan Ng et al.

Problem 6. Robust beamforming – SDP relaxation:


minimize Tr(W)
W∈HNT ,δj
Tr(HW)
subject to C1 : 2 + σ2
≥ req ,
σant s
 
C2 : Tr  n W ≤ Pmaxn , ∀n ∈ {1, . . . , NT },
C3 : SC3j (W, δj )  0, ∀j ∈ {1, . . . , J},
((
C4 : ( ((( ≤ 1.
Rank(W) (5.22)

Problem 6 is a convex SDP that can be solved by standard convex numerical


solvers [48] via interior point methods. Furthermore, by following a similar approach
to that in the proof of Theorem 5.1, it can be proved that the SDP relaxation is tight
when h and the ERs, gj , j ∈ {1, . . . , J}, can be modeled as statistically independent
random variables. In other words, the SDP relaxation is tight and the optimal solution
for Problem 5 can be obtained in polynomial time.

5.3.5 Numerical Examples


In this section, we study the performance of the proposed robust beamforming scheme.
The simulation parameters adopted are listed in Table 5.2. Besides that, the normalized
maximum channel estimation error of ER j is given by σER 2 = ε 2 /g 2 = 0.1, ∀j. The
j j j
maximum transmit power per antenna is set to Pmaxn = 30 dBm, ∀n ∈ {1, . . . , NT }.

Average Total Transmit Power


Figure 5.8 illustrates the average total transmit power versus the minimum required
power transfer Pmin for different beamforming schemes and different numbers of trans-
mit antennas. It can be seen that the average total transmit power of the proposed robust
scheme increases monotonically with Pmin . In fact, the transmitter is forced to transmit
a higher power in order to satisfy the more stringent required Pmin whenever some of
the ERs are experiencing bad channel conditions. For comparison, we also show the
performance of a benchmark scheme and a naive scheme. For the benchmark scheme,
we perform optimal beamforming with perfect CSI, which yields the minimum transmit
power required for satisfying constraints C1 –C3 . The naive scheme treats imperfect
CSI as if it were perfect CSI and performs the same optimization as the benchmark
scheme. It can be observed that a higher transmit power is required for the optimal
robust scheme than for the benchmark scheme due to the imperfectness of the CSI.
Besides, although the average total transmit power of the naive scheme is smaller than
that of the proposed robust scheme, the naive scheme is unable to satisfy the mini-
mum required power transfer at all time instants, which is undesirable in practice, see
Figure 5.9.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
Multiple Antennas and Beamforming for SWIPT Systems 189

Robust scheme, NT = 6
30
Average total transmit power (dBm)

Benchmark scheme, NT = 6
Naive scheme, NT = 6 Robust scheme
28 Robust scheme, NT = 8
Benchmark scheme, NT = 8
26 Naive scheme, NT = 8

24

22

20

18

16
5 10 15 20 25 30 35
Minimum required power transfer Pmin (mW)

Figure 5.8 Average total transmit power (dBm) versus the minimum required power transfer
per receiver, Pmin (μW), for different beamforming schemes and different numbers of
transmit antennas, NT .

0.9

0.8

0.7
QoS outage probability

0.6

0.5

0.4

0.3 Naive scheme, N T = 8


0.2 Naive scheme, N T = 6
Benchmark scheme, N T = 6, 8
0.1 Robust scheme, N T = 6, 8

0
5 10 15 20 25 30 35
Minimum required power transfer Pmin (mW)

Figure 5.9 QoS outage probability versus the minimum required power transfer per receiver,
Pmin (μW), for different beamforming schemes and different numbers of transmit antennas, NT .

QoS Outage Probability


Figure 5.9 depicts the QoS outage probability versus the minimum required power trans-
fer per receiver for different beamforming schemes and different numbers of transmit
antennas, NT . A QoS outage event occurs whenever the beamforming scheme cannot
fulfill the QoS requirement for the minimum power transfer in C3 . It can be seen
that both the benchmark scheme and the proposed robust scheme achieve zero outage

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
190 Derrick Wing Kwan Ng et al.

probability. In other words, the proposed robust scheme is able to guarantee the mini-
mum required power transfer in the scenarios considered, despite the imperfectness of
the CSI. However, the outage probability of the naive scheme increases dramatically as
Pmin increases, due to the non-robust beamformer design.

5.4 Multi-Objective SWIPT Optimization

In the above sections, we studied optimization for SWIPT systems by solving different
conventional single-objective optimization problems. However, in practice, multiple
conflicting system design, objectives arise naturally in system design, and applying
the solutions of single-objective optimization to multi-objective optimization problems
(MOOPs) may not lead to satisfactory system performance. Therefore, the concept of
multi-objective optimization (MOO) or vector optimization is discussed in this section
to provide a systematic procedure for handling conflicting objective functions.3 First,
we introduce the MOOP in its standard form as follows.

Problem 7. Multi-objective optimization:


$ %
minimize f(x) = f1 (x), f2 (x), . . . , fK (x)
x
subject to gl (x) ≤ 0, l ∈ {1, . . . , L},
hn (x) = 0, n ∈ {1, . . . , N}, (5.23)

where K, L, and N are the numbers of objective functions, inequality constraints, and
equality constraints, respectively. x is the vector of optimization variables, fk (x), ∀k ∈
{1, . . . , K}, is the kth objective function,4 gl (x) is the lth inequality constraint, and hn (x)
is the nth equality constraint.

5.4.1 Optimization Solution


In contrast to single-objective optimization, a solution to a MOOP is more of an abstract
concept than a fixed point. In general, there is no single global solution that optimizes
all the objective functions simultaneously. Typically, it is necessary to determine a set
of points that fit a predetermined definition of an optimum. To this end, we introduce
the concept of Pareto optimality for MOOP.
definition 5.1 (Pareto optimal). A point, x∗ ∈ F, is Pareto optimal if and only if
(iff) there does not exist another point, x ∈ F, such that f(x) ≤ f(x∗ ) and fk (x) < fk (x∗ )
for at least one function.

3 MOO has been applied extensively in the fields of engineering and economics for handling conflicting
design objectives [17, 32, 49, 50].
4 We note that, in contrast to conventional heuristic approaches where some objectives are converted into
constraints, MOO enables a more rigorous and more flexible system design.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
Multiple Antennas and Beamforming for SWIPT Systems 191

The Pareto optimal set consists of the attainable operating points that cannot be
neglected if the system designer has no preference for a particular system objective. In
particular, none of the objectives can be improved without degrading other objectives.
Evidently, any point that is not in the Pareto optimal set is strictly suboptimal because
there exist other operating points that are better or at least as good with respect to every
objective. The Pareto optimal set is an analogy to global optimality that can be achieved
in multi-objective optimization. We note that single-objective optimization problems
are special cases of MOOPs with K = 1. In other words, if an algorithm can solve
the MOOP, then it can be used to solve the corresponding single-objective optimization
problem.

5.4.1.1 MOOP Scalarization


A common approach to handle MOOPs in practice is the “a-priori method.” This
method allows the system designer to specify preferences that may correspond to
certain design goals or the relative importance of different objectives. In particular, the
system designer adopts a function that produces a scalar describing the preference of
the objectives. The function scalarizes the multi-objective function which describes
a certain subjective tradeoff between the objectives and thus imposes an order on
the objective function vectors in the objective set f(x). There are many scalarization
methods in the literature [17, 49, 50]. Here, we introduce the weighted Chebyshev for-
mulation, also known as the weighted max–min formulation, which plays a key role in
capturing the Pareto optimal set. The weighted Chebyshev objective function is given by
  
fChebyshev (x) = max wk fk (x) − fk∗ (x) , (5.24)
1≤k≤K

where fk∗ (x) is the optimal value of the kth objective function and 0 ≤ wk ≤ 1 is a

constant weight factor on objective function k such that K k=1 wk = 1. In fact, the
multi-objective problem in (5.24) can provide the complete Pareto optimal set [17] on
varying the weights, even if the MOOP is non-convex.

5.4.2 Multi-Objective Optimization for SWIPT


In this section, we study multi-objective optimization in SWIPT systems with sepa-
rated receivers. Recently, driven by environmental concerns, energy efficiency (EE) has
become an important metric for evaluating the performance of wireless communica-
tion systems [12, 22, 51–54]. However, with SWIPT, the EE of WPT becomes just as
important as the EE of information transmission. Thus, in the following, we focus on
three important system design objectives for SWIPT networks, namely, information rate
(IR)-EE maximization, energy transfer (ET)-EE maximization, and total transmit power
minimization.

5.4.3 System Model


In this section, we present a system model for studying MOOP in SWIPT. We assume
that the transmitter chooses the transmit signal vector x as

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
192 Derrick Wing Kwan Ng et al.

x= ws
!"# + wE . (5.25)
!"#
desired information signal energy signal
Here, wE is a deterministic pseudo-random sequence that is used to facilitate efficient
energy transfer and is known to the receivers. Since the energy signal is known at the
legitimate receiver, it can be cancelled at the IR via successive interference cancellation
before attempting to decode the desired information.
As a result, the achievable rate (bit/s) in the SWIPT system is given by
 
wH Hw
C = B log2 1 +
Tr(HWE ) + σant2 + σ2
s
 
(a) wH Hw
≤ B log2 1 + 2 , (5.26)
σant + σs2
where B is the system bandwidth, WE is the covariance matrix of the random energy
signals, and (a) is due to the fact that interference cancellation can be performed at the
IR to remove hH wE before attempting to decode the desired information.
On the other hand, the energy harvested at the ERs is given by

J
HP(w, WE ) = j w| + |gj wE | )
ηj (|gH 2 H 2

j=1
 
= Tr G(wwH + WE ) , (5.27)

where G = Jj=1 ηj gj gH j is the equivalent channel of the J ERs introduced in Section
5.3.1. We note that the contribution of thermal noise to the total harvested power is neg-
ligible compared with the information and energy signals and thus is neglected in (5.27).
On the other hand, we incorporate the total power dissipation of the system as an
optimization objective function. To this end, we model the power dissipation (in Joules
per second) of the system as
w2 + Tr(WE )
TP(w, WE ) = + NT Pant
ξ !" #
!" # Antenna power consumption
Amplifier power consumption
+ Pc , (5.28)
!"#
Constant circuit power consumption
where ξ is the power amplifier efficiency,5 and the first term in (5.28) is the power
consumption of the power amplifiers. NT Pant accounts for the dynamic circuit power
consumption, which is proportional to the number of transmitting antennas NT . Pant
includes the power dissipation of the transmit filter, mixer, frequency synthesizer,
digital-to-analog converter (DAC), etc. Pc denotes the fixed power consumption due to
the baseband signal processing.

5 We assume that Class A power amplifiers with a linear characteristic are implemented in the transceivers.
In practice, the maximum power efficiency of Class A amplifiers is 25%.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
Multiple Antennas and Beamforming for SWIPT Systems 193

We define the IR-EE and the ET-EE as


  
C B log 2 1 + wH Hw/ σ 2 + σ 2
ant s
ηIR = =  (5.29)
TP(w, WE ) w2 + Tr(WE ) /ξ + NT Pant + Pc

and

HP(w, WE ) wH Gw + Tr(GWE )
ηEH = =  , (5.30)
TP(w, WE ) w2 + Tr(WE ) /ξ + NT Pant + Pc

respectively. Now, we first propose three problem formulations for single-objective


system design for SWIPT networks. In particular, each single-objective problem for-
mulation considers one important aspect of the system design. Then, we consider the
three system design objectives jointly under the framework of multi-objective optimiza-
tion. The multi-objective optimization framework adopted enables the Pareto optimal
beamformer design.
The first problem formulation aims at maximizing the IR-EE in the SWIPT network.
The problem formulation is given by the following.

Problem 8. Information rate energy efficiency maximization:


maximize ηIR
w,WE ∈HNT
 
subject to C1 : wwH n,n + Tr( n WE ) ≤ Pmaxn , ∀n ∈ {1, . . . , NT }
C2 : WE  0 . (5.31)

The second system design objective is the maximization of the energy transfer effi-
ciency in the SWIPT network and can be mathematically formulated as follows.

Problem 9. Energy transfer efficiency maximization:


maximize ηEH
w,WE ∈HNT
subject to C1 , C2 . (5.32)

The third system design objective concerns the minimization of the total transmit
power. The problem formulation is given as follows.

Problem 10. Total transmit power minimization:


minimize w2 + Tr(WE )
w,WE ∈HNT
subject to C1 , C2 . (5.33)

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
194 Derrick Wing Kwan Ng et al.

For notational simplicity, we denote the objective functions in Problems 8 and 9 as


F8 (w, WE ) and F9 (w, WE ), respectively. We note that Problem 10 is a trivial problem
with optimal value zero since the transmitter is not required to provide QoS to the
receivers. Yet, Problem 10 plays an important role in the following when we study
the beamforming design under the MOOP framework. To facilitate the presentation
and without loss of generality, Problem 10 is re-written as an equivalent maximization
problem and the corresponding objective function is given by F10 (w, WE ) = −(w2 +
Tr(WE )).
In practice, the above three optimization objectives are all desirable from the system
operator perspective; however, there are non-trivial tradeoffs among these objectives. In
order to optimize these conflicting system design objectives systematically and simulta-
neously, we apply the MOOP framework introduced at the beginning of this section. In
particular, we handle the three optimization functions by using the weighted Chebyshev
method [17]. Hence, our MOOP is formulated as follows.

Problem 11. Multi-objective optimization problem:


 
minimize max ωj (Fp∗ − Fp (w, WE ))
w,WE ∈HNT p=8, 9, 10
subject to C1 , C2 , (5.34)

where Fp∗ is the optimal objective value with respect to Problem p. ωp is a preference

weight imposed on objective function p subject to 0 ≤ ωp ≤ 1 and p ωp = 1, which
reflects the preference of the decision maker for the pth objective function over the
others. In the extreme case, when ωp = 1 and ωt = 0, ∀t = p, Problem 11 is equivalent
to single-objective optimization problem p.

5.4.4 Multi-Objective Optimization Solution


In this section, we solve the multi-objective problem optimally by the Charnes–Cooper
transformation and SDP relaxation. It can be observed that Problems 8–11 are non-
convex with respect to the optimization variables. In order to obtain a tractable solution,
we overcome the non-convexity by recasting the problems as convex problems based
on the proposed transformation and SDP relaxation.
We first reformulate the aforementioned three single-objective optimization problems
by defining a set of new optimization variables as follows:

1
W = wwH , θ = , W = θW, and WE = θ WE . (5.35)
TP(w, WE )

Then, the original problems can be reformulated in terms of the new optimization
variables, i.e., W, WE , and θ , as follows.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
Multiple Antennas and Beamforming for SWIPT Systems 195

Problem 12. Transformed problem 8:


 
Tr(HW)
maximize F 8 = θ B log2 1+ 2 + σ 2)
W,WE ∈HNT ,θ θ (σant s
subject to C1 : Tr( n (W + WE )) ≤ θ Pmaxn , ∀n ∈ {1, . . . , NT },
C2 : W  0, WE  0,
C3 : Rank(W) ≤ 1,
Tr(W + WE )
C4 : + θ (NT Pant + Pc ) ≤ 1,
ξ
C5 : θ ≥ 0. (5.36)

Problem 13. Transformed problem 9:


maximize F 9 = Tr(G(W + WE ))
W,WE ∈HNT ,θ

subject to C1 –C5 . (5.37)

Problem 14. Transformed problem 10:


 
1
maximize F 10 = −ξ − (NT Pant + Pc )
W,WE ∈HNT ,θ θ
subject to C1 –C5 . (5.38)

Constraints W  0, W ∈ HNT , and Rank(W) = 1 are imposed to guarantee that


W = θwwH . Constraints C4 and C5 are introduced due to the proposed transformation.
Then, we further transform the MOOP into its equivalent epigraph representation
[47] as follows.

Problem 15. Transformed multi-objective optimization problem:


minimize τ
W,WE ∈HNT ,θ,τ

subject to C1 –C5 ,
C6 : ωp (Fp∗ − Fp ) ≤ τ , ∀p ∈ {8, 9, 10}, (5.39)

where τ is the auxiliary optimization variable.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
196 Derrick Wing Kwan Ng et al.

proposition 5.2. The transformed Problems 12–15 are equivalent transformations


of the original Problems 8–11, respectively. In particular, we can recover the solution
of the original problems based on (5.35).

Proof. Please refer to Section 5.8.2.

We note that, if Problem 15 can be solved optimally by an algorithm, then the algo-
rithm can also be used to solve Problems 10–12, since Problem 15 is a generalization
of Problems 10–12. Thus, we focus on solving Problem 15 in the following. It can be
verified that Problem 15 is non-convex due to the rank-one beamforming matrix con-
straint in C3 . Now, we apply SDP relaxation by removing constraint C3 : Rank(W) = 1
from Problem 15 in the following. As a result, the SDP relaxed version of Problem 15
is given by the following.

Problem 16. Transformed MOOP – SDP relaxation:


minimize τ
W,WE ∈HNT ,θ,τ

subject to C1 , C2 , C4 –C6 ,
(
C3 : ( ((((
Rank(W) = 1, (5.40)

which is a convex SDP problem and can be solved by numerical convex program solvers
such as CVX [48]. Next, we study the tightness of the adopted SDP relaxation in the
following theorem.

theorem 5.2. The optimal solution of Problem 16 satisfies Rank(W ) ≤ 1. Besides,
this solution can be obtained by construction with a similar approach to that in [49].

Proof. The proof of Theorem 5.2 closely follows the proof of [49, Proposition 1] and is
omitted here for brevity.

Thus, the adopted SDP relaxation is tight. Besides, similarly to Problem 16, Problems
12–14 can be solved using the SDP relaxation.

5.4.5 Numerical examples


In this section, we provide some numerical examples to examine the tradeoff between
the conflicting design objectives considered via the proposed optimal beamforming
schemes. The system parameters in Table 5.2 are adopted. There are two ERs and one
IR located 10 m away from the transmitter. The transmitter is equipped with NT = 12
antennas. The maximum transmit power per antenna is set to Pmaxn = 30 dBm, ∀n ∈
{1, . . . , NT }, the baseband signal processing power consumption is Pc = 1 W, the
per-antenna circuit power consumption is Pant = 150 mW, and the power amplifier
efficiency is ξ = 0.2. The tradeoff region in Figure 5.10 is obtained by solving Problem

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
Multiple Antennas and Beamforming for SWIPT Systems 197

Figure 5.10 Three-dimensional system objective tradeoff regions achieved by the proposed
optimal beamforming scheme.

Figure 5.11 Tradeoff region between IR-EE and ET-EE.

15, where the values of 0 ≤ wp ≤ 1, ∀p ∈ {8, 9, 10}, are uniformly varied for a step size

of 0.05 such that p wp = 1. The average system performance is obtained by averaging
the obtained results over different channel realizations. Besides, for a better illustration,
we have also provided different side-views of the three-dimensional tradeoff region in
Figures 5.11–5.13 to reveal the tradeoffs between different pairs of objective functions,
namely (1) IR-EE and ET-EE; (2) IR-EE and total transmit power; and (3) total transmit
power and ET-EE.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
198 Derrick Wing Kwan Ng et al.

Figure 5.12 Tradeoff region between IR-EE and the total transmit power.

Figure 5.13 Tradeoff region between total transmit power and ET-EE.

It can be observed from Figures 5.10 and 5.11 that the system design objectives of
average ET-EE maximization and average IR-EE maximization are partially aligned
with each other for small transmit powers. In particular, both objective functions
increase rapidly when the transmit power increases from zero. However, the IR-EE
decreases dramatically in the high-transmit-power regime. This is because there is

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
Multiple Antennas and Beamforming for SWIPT Systems 199

a diminishing return in the system achievable rate with respect to the increment in
transmit power. Also, we observe similar increasing and then decreasing trends of
IR-EE from Figures 5.10 and 5.12 for the tradeoff between IR-EE and the total transmit
power, due to the diminishing return in data rate gain achieved by the increment in
transmit power. In contrast, ET-EE increases monotonically with respect to the transmit
power with increasing slope, see Figures 5.10 and 5.13. Thus, a high transmit power
is preferable to maximize the energy transfer efficiency. Furthermore, the tradeoff
region in Figure 5.12 is non-convex. In other words, the proposed optimal beamforming
scheme is able to attain the non-convex tradeoff region, despite the non-convexity
of the MOOP. In fact, the two extremes points in Figure 5.12 correspond to the
maximum/minimum of two of the objective functions considered. Zero transmit power
represents the minimum total transmit power. It is the optimal value of Problem 10,
and it can be obtained by solving Problem 11 with w10 = 1. The second extreme
point occurs in the middle of Figure 5.12, which is the maximum IR-EE, i.e., the
optimal value for Problem 9. Similarly, the maximum ET can be obtained by solving
Problem 11 with w9 = 1.
On the other hand, it can be seen from Figures 5.10, 5.12, and 5.13 that the objective
of total transmit power minimization conflicts with the other two objectives. In particu-
lar, in order to maximize ET-EE, the transmitter has to transmit with almost full power
over each antenna at every time instant. The associated beamformers correspond to the
single point at the right tail in the curve of Figure 5.10 and the rightmost corner point
in Figures 5.12 and 5.13. However, if the transmitter employs a large transmit power, a
low IR-EE will result, see Figure 5.12.

5.5 Secure Communications in SWIPT Systems

Security is a fundamental problem in wireless communication systems due to the


broadcast nature of the wireless medium. Traditionally, cryptographic encryption
technologies have been used to enable communication security in the application
layer. However, the commonly used encryption algorithms are based on the assumption
of limited computational capabilities of the eavesdroppers, which may not hold in
the future due to the development of quantum computers. Besides, these algorithms
assume a perfect secret key management and distribution, which may not be possible in
certain wireless networks such as ad-hoc networks. As an alternative, physical (PHY)
layer security utilizes the physical properties of wireless communication channels,
such as channel fading and interference, to ensure perfectly secure communication
[55–60], regardless of the potentially unlimited computational capabilities of the
potential eavesdroppers.
On the other hand, in SWIPT systems, transmitters can increase the energy of
the information-carrying signal to facilitate energy harvesting at the receivers, see
Sections 5.3 and 5.4. However, this may also increase their susceptibility to

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
200 Derrick Wing Kwan Ng et al.

al Information receiver
sign
ation
orm
Inf
Energy signal
Ar
tifi
cia
ln Energy harvesting receiver 2
ois
e (potential eavesdropper)

Transmitter
Energy harvesting receiver 1
(potential eavesdropper)

Figure 5.14 A multiuser SWIPT downlink model with an active IR and J = 2 ERs. The idle
receivers harvest energy from the received RF signal and are treated as potential eavesdroppers
by the transmitter in terms of providing secure SWIPT services.

eavesdropping6 due to the broadcast nature of wireless channels. Therefore, new


QoS concerns regarding communication and energy security naturally arise in systems
providing SWIPT services [13, 16, 18, 32, 61, 62]. Thus, in this section, we study
beamformer optimization taking into account communication security.

5.5.1 System Model


In this section, we present a system model for separated receivers that has commonly
been adopted in the literature for secure SWIPT [13, 16, 18, 32, 61, 62], see Figure 5.14.
In order to provide secure communication and to facilitate energy harvesting, artificial
noise signals and energy signals are generated at the transmitter. In particular, both
signals are transmitted concurrently with the information-bearing signal. Besides, they
are able to degrade the channels between the transmitter and potential eavesdroppers
and act as an energy source for energy harvesting. The transmitter chooses the transmit
signal vector x as
x= ws
!"# + wE + v
!"# , (5.41)
!"#
desired information signal energy signal artificial noise

where v ∈ CNT ×1 is the artificial noise vector generated by the transmitter to combat
potential eavesdroppers. v is modeled as a complex Gaussian random vector with
v ∼ CN (0, V), (5.42)

6 We note that, in practice, the malicious ERs do not have to decode the eavesdropped information in real time.
They can act as information collectors to sample the received signals and store them for future decoding by
other energy-unlimited and powerful computational devices.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
Multiple Antennas and Beamforming for SWIPT Systems 201

where V ∈ HNT , V  0, denotes the covariance matrix of the artificial noise. The
artificial noise signal interferes with the legitimate receiver and potential eavesdroppers
since v is unknown to both types of receivers. Hence, artificial noise transmission has to
be carefully designed to degrade the channels of potential eavesdroppers while having
a minimal effect on the desired receiver.
On the other hand, wE is a Gaussian pseudo-random sequence7 that is used to facili-
tate efficient energy transfer and is known to the legitimate receiver. wE is modeled as a
complex Gaussian pseudo-random vector with
wE ∼ CN (0, WE ), (5.43)
where WE ∈ HNT , WE  0, denotes the covariance matrix of the pseudo-random energy
signal. Since the energy signal is known at the legitimate receiver it can be cancelled
at the legitimate receiver via successive interference cancellation. Besides, the energy
signal is not known to potential eavesdroppers and can be exploited by the transmitter
to provide communication security.
The above system model advocates the dual use of both artificial noise and energy sig-
nals in providing secure communication and facilitating efficient WPT. In fact, whether
artificial noise or an energy signal is preferable depends on the scenario considered.
In order to exploit the energy signal efficiently, a short secret key is needed at the
desired receiver as seed information for the pseudo-random sequence generator used for
generating the energy signal sequences. Besides, the transmitter is required to regularly
change the seed to prevent the sequence from being cracked by potential eavesdroppers.
The seed information used at the transmitter can be delivered securely to the desired
receivers by exploiting, e.g., the reciprocity of the channels between the transmitter and
the legitimate receiver [63]. However, if the seed information is for some reason also
available to the potential eavesdroppers, allocating all the energy of the energy signal to
the artificial noise may be a better choice for guaranteeing communication security, i.e.,
WE = 0.

5.5.2 SWIPT for Multiple-Antenna Potential Eavesdroppers


In practice, the ERs may be equipped with multiple receive antennas to improve the
energy harvesting efficiency. Thus, in this section, we assume that each ER is equipped
with NR receive antennas. The received signal at ER j ∈ {1, . . . , J} is given by
yERj = GH
j x + nERj , ∀j ∈ {1, . . . , J}, (5.44)
where the channel matrix between the transmitter and ER j is denoted by Gj ∈ CNT ×NR .
The channel matrices capture the joint effects of multi-path fading and path loss.
nERj ∼ CN (0, (σs2 + σant
2 )I ) is the AWGN at ER j.
NR

7 For energy transfer, the energy sequence is not required to be generated by a Gaussian pseudo-
random source. However, a Gaussian pseudo-random energy sequence can be exploited to provide secure
communication.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
202 Derrick Wing Kwan Ng et al.

Then, the total amount of power harvested by ER j is given by


   
EERj = ηj Tr GH j ww + V + WE Gj .
H
(5.45)

5.5.2.1 Achievable Rate and Secrecy Rate


In this section, we assume that NT ≥ NR to study the beamformer design for providing
secure communication. Given perfect CSI at the receiver, the achievable rate (bit/s/Hz)
between the transmitter and the IR is given by
 
wH Hw
R = log2 1 +
Tr(HV) + Tr(HWE ) + σant2 + σ2
s
 
(a) wH Hw
≤ log2 1 + , (5.46)
Tr(HV) + σant2 + σ2
s

where (a) is due to the fact that interference cancellation can be performed at the IR to
remove hH wE before attempting to decode the desired information.
On the other hand, we focus on the worst-case scenario for the decoding capability
of the ERs for providing communication security to the IR. We assume that energy
harvesting receiver j performs interference cancellation to remove all multiuser
interference and eavesdrops on the message intended for the IR. Therefore, the
achievable rate between the transmitter and ER j for decoding the signal of the IR
can be expressed as

RERj = log2 det(INR + Q−1 H H


j Gj ww Gj ), (5.47)

where
 
Qj = GH
j WE + V Gj + (σant + σs )INR 0 .
2 2
(5.48)

Where Qj is the interference-plus-noise covariance matrix for ER j assuming the


worst case for communication secrecy. Thus, the achievable secrecy rate of the IR is
given by
$ %+
Rsec = R − max {RERj } . (5.49)
∀j

5.5.2.2 Problem Formulation


In Problem 1, we studied the beamforming design for the maximization of the total
power transfer to the ERs. However, this problem formulation does not take into account
the fairness in the amount of power harvested per ER. For instance, if there is an
ER located closer to the transmitter than other receivers, then the transmitter has a
tendency to steer the beamforming direction towards that ER for maximization of the
total harvested energy. However, this leads to energy starvation of the other receivers.
In order to take into account the fairness in WPT, we formulate the following optimiza-
tion problem.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
Multiple Antennas and Beamforming for SWIPT Systems 203

Problem 17. Max–min fairness for secure SWIPT:


    
min ηj Tr GH j ww + WE + V Gj
H
maximize
V,WE ∈HNT ,w j∈{1,...,J}
wH Hw
subject to C1 : ≥ req ,
Tr(HV) + σant
2 + σ2
s
C2 : RERj ≤ RTol
ER , ∀j,
$ %
C3 : Tr( n V) + Tr( n WE ) + wwH ≤ Pmaxn ,
n,n
∀n ∈ {1, . . . , NT },
C4 : WE , V  0, (5.50)

where the maximum tolerable data rate RTol ER > 0 in C2 is imposed to restrict the
achievable rate of ER j if it attempts to decode the message of the IR. Constraint C4 and
WE , V ∈ HNT ensure that the covariance matrices V and WE are positive semidefinite
Hermitian matrices.

remark 2. We note that the proposed optimization framework can be extended to


include additional passive eavesdroppers, for which instantaneous CSI is not available
at the transmitter, by introducing probabilistic maximum tolerable SINR constraints for
the passive eavesdroppers following a similar approach to that in [16].

Problem 17 is a non-convex optimization problem. In particular, the non-convexity


arises from constraint C1 and the log det function in C2 . To overcome the non-convexity,
we first advance the following proposition and then recast the problem into a convex
optimization problem using SDP relaxation.

ER > 0, ∀j, the following implication holds for constraint


proposition 5.3. For RTol
C2 :

C2 ⇒ C2 : GH
j WGj  αER Qj , ∀j, (5.51)

Tol
where αER = 2RER − 1 is an auxiliary constant and C2 is an LMI constraint. We note
that constraints C2 and C2 are equivalent, i.e., C2 ⇔ C2 , if Rank(W) ≤ 1.

Proof. Please refer to Section 5.8.3 for the proof.

Now, we apply Proposition 5.3 to Problem 17 by replacing constraint C2 with


constraint C2 . By setting Wk ∈ HNT , ∀k, Wk = wk wHk , and Rank(Wk ) ≤ 1, ∀k, we can
rewrite the optimization problem in its hypograph form.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
204 Derrick Wing Kwan Ng et al.

Problem 18. Equivalent formulation of problem 17:


maximize τ
W,V,WE ∈HNT ,τ
 
subject to C1 : Tr(HW) ≥ req Tr(HV) + σant
2
+ σs2 ,

j WGj  αER Qj , ∀j,


C2 : GH
C3 : Tr( n V) + Tr( n WE ) + Tr( n W) ≤ Pmaxn ,
C4 : WE  0, V  0,
   
j W + V + WE Gj , ∀j,
C5 : τ ≤ ηj Tr GH
C6 : Rank(W) ≤ 1. (5.52)

After the transformation, the problem is non-convex due to the rank constraint
in C6 . As in the cases studied in the previous sections, we adopt SDP relaxation
to obtain a tractable solution. The SDP relaxation of Problem 18 is given by the
following.

Problem 19. SDP relaxation of problem 18:


maximize τ
W,V,WE ∈HNT ,τ

subject to C1 , C2 , C3 –C5 ,
((
C6 : ( ((( ≤ 1 .
Rank(W) (5.53)

It can be verified that the objective function of Problem 18 is an affine


function and the constraints span a convex set. Beside, the problem satisfies Slater’s
constraint qualification. Thus, the SDP relaxed problem can be solved efficiently
in polynomial time via standard numerical solvers for solving convex programs. In
the following, we introduce a theorem that reveals the tightness of the adopted SDP
relaxation.

theorem 5.3. There exists an optimal solution of Problem 18 that satisfies


Rank(W) = 1. Besides, this solution can be obtained by construction via a similar
approach to that in [18]. Furthermore, V = 0 and constraint C1 is active at the optimal
solution.

Proof. The proof of Theorem 5.3 closely follows the proof of [18, Proposition 4.1] and
is omitted here for brevity.

5.5.3 Numerical Examples


In this section, we study the system performance of the proposed beamformer design.
The simulation parameters adopted are listed in Table 5.2. We assume that there are

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
Multiple Antennas and Beamforming for SWIPT Systems 205

always two ERs and one IR in the system. They are all located 10 m away from the
transmitter. The minimum required SNR of the desired IR and the maximum tolerable
SINR of the ERs are set to req = 3 dB and −10 dB, respectively. In other words, the
maximum tolerable data rate at each ER is RTol ER = 0.1375 bit/s/Hz and the minimum
achievable secrecy rate is log2 (1 + req ) − 0.1 = 1.4475 bit/s/Hz.

Average Minimum Harvested Power


Figure 5.15 shows the average minimum harvested power per ER versus the maximum
transmit power per antenna (dBm) for different numbers of transmit antennas, NT , and
different numbers of receive antennas, NR , installed at each ER. It is expected that the
average minimum harvested power per ER increases with the maximum transmit power
per antenna since the transmitter is able to transfer more power to the RF at every time
instant. On the other hand, the minimum harvested power per ER increases both with
the number of transmit antennas, NT , and with the number of receive antennas, NR . By
increasing the number of transmit antennas, the direction of the energy signal beam-
forming matrix WE can be more accurately steered toward the ERs, which improves the
energy transfer efficiency. Besides, the extra receive antennas at each ER act as extra
independent energy collectors to harvest energy from the RF.

800
NT = 8, NR = 1
NT = 8, NR = 2
700
NT = 8, NR = 3
Average minimum harvested power

NT = 10, NR = 1
600
NT = 10, NR = 2
NT = 10
NT = 10, NR = 3
500
per ER (mW)

400 NT = 8

300

200

100

0
13 14 15 16 17 18 19 20
Maximum transmit power per antenna (dBm)

Figure 5.15 Average transmit power allocation (dBm) versus the maximum transmit power
per antenna, Pmaxn for different numbers of transmit antennas, NT , and receive antennas, NR ,
installed at each ER.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
206 Derrick Wing Kwan Ng et al.

Number of transmit antennas NT = 8


Average power allocation (dBm)

30
Tr(W) Tr(WE)

20

10

0
13 14 15 16 17 18 19 20
Maximum transmit power per antenna (dBm)

Number of transmit antennas NT = 10


Average power allocation (dBm)

30
Tr(W) Tr(WE)

20

10

0
13 14 15 16 17 18 19 20
Maximum transmit power per antenna (dBm)

Figure 5.16 Average transmit power allocation (dBm) versus the maximum transmit power
per antenna, Pmaxn , for different numbers of transmit antennas, NT , and receive antennas, NR ,
installed at each ER.

Average Power Allocation


Figure 5.16 depicts the average transmit power allocation to the two components8 of
the transmitted signal, i.e., Tr(W) and Tr(WE ), versus the maximum transmit power
per antenna for the proposed optimal beamforming scheme. It can be observed that
the amount of power allocated to the energy signal is higher than that allocated to the
information signal when the required SNR of the IR is low. Besides, the power allocated
to the information signal and the energy signal increases as the maximum transmit
power per antenna increases because of the higher transmit power budget. In particular,
the portion of the total transmit power allocated to the energy signal increases with
increasing NT . In fact, the number of degrees of freedom for beamforming increases
with the number of transmit antennas. Specifically, with more transmit antennas, the
transmitter is able to apply more power-efficient beamforming to the information signal.
Thus, a small transmit power is sufficient to satisfy the SINR requirement of the IR in
constraint C1 , and more power can be reserved for the energy signal for a more effective
transfer of energy to the ERs.
8 As expected from Theorem 5.3, V = 0 holds for the optimal solution. Thus, the power allocated to the
energy signal is not shown in Figure 5.16.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
Multiple Antennas and Beamforming for SWIPT Systems 207

1.51
NT = 8, NR = 1
NT = 8, NR = 2
NT = 10
1.5 NT = 8, NR = 3
Average achivable secrecy rate (bit/s/Hz)

NT = 10, NR = 1

1.49 NT = 10, NR = 2
NT = 10, NR = 3
Minimum required
1.48 secrecy rate

1.47
NT = 8

1.46

1.45

13 14 15 16 17 18 18 20
Maximum transmit power per antenna (dBm)

Figure 5.17 Average achievable secrecy rate (bit/s/Hz) versus the maximum transmit power
per antenna, Pmaxn for different numbers of transmit antennas, NT , and receive antennas, NR ,
installed at each ER.

Average Achievable Secrecy Rate


Figure 5.17 shows the average achievable secrecy rate versus the maximum transmit
power per antenna for the proposed optimal beamforming scheme. It can be observed
that the average achievable secrecy rate increases with Pmaxn . This is due to the fact
that for larger transmit power budgets the transmitter can allocate more transmit power
to the energy signal to degrade the channel of the potential eavesdroppers. Besides,
the average achievable secrecy rate increases with the number of transmit antennas NT .
The extra degrees of freedom offered by the transmit antennas help in focusing the
power of the energy signal at the potential eavesdroppers, which also facilitates efficient
jamming. On the other hand, the proposed optimal scheme is able to meet the minimum
required secrecy rate even for NR = 3 receive antennas at the potential eavesdroppers.

5.6 Research Challenges

The introduction of SWIPT into traditional communication systems revolutionizes the


design of beamformers, receiver architectures, and network topologies. It is expected
that SWIPT will be a key technology for cutting the last wires of energy-limited
low-power-consumption devices to enable truly mobile communications. As discussed

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
208 Derrick Wing Kwan Ng et al.

in the previous sections, multiple-antenna technology facilitates SWIPT and at the


same time imposes many interesting and challenging new research problems. In
the following, we discuss some other fundamental research challenges and potential
solutions for realizing SWIPT.
Modern communication systems operate in microwave frequency bands for informa-
tion transfer. Over a short distance of 10 m in free space, the attenuation of a wireless
signal can be up to 50 dB for a carrier frequency of 915 MHz in the ISM frequency
band. Hence, employing SWIPT in microwave frequency bands directly may result in
unsatisfactory performance for long-distance transmission. Three possible approaches
to overcome this problem and their potential drawbacks are discussed in the following.
• Massive MIMO. The extra degrees of freedom offered by massive MIMO facil-
itate the transmission of narrow energy and information signal beams [64]. In
particular, the beams can be more accurately steered toward the receivers to
improve system performance. As a result, the combination of massive MIMO and
SWIPT seems to be a viable approach to improve the energy transfer efficiency
[65]. Nevertheless, the implementation of massive MIMO relies heavily on the
availability of cheap and power-efficient radio and base-band hardware. Decreas-
ing the hardware cost and increasing the power amplifier efficiency may have
significant negative side effects on the communication link, such as signal distor-
tion and interference. In particular, the non-linearity of the power amplifier can
cause the appearance of strong harmonics of the energy-carrying signal that inter-
fere with wireless communications. Thus, the potential benefits of implementing
massive MIMO with SWIPT when low-cost non-ideal hardware is adopted at the
transmitters remain unknown.
• Joint beamforming and receiver scheduling. Long-distance transmission is an
obstacle to the realization of WPT since the WPT efficiency decreases with dis-
tance. As a result, receiver scheduling is a key aspect of facilitating SWIPT in
practice. For instance, receivers experiencing high channel gains can be sched-
uled for WPT and the remaining receivers may be scheduled for information
transmission. Also, opportunistic beamforming can be applied to focus the trans-
mitted signals at the targeted receivers. Thus, only a small transmit power is
needed from the transmitters so as to improve the power efficiency of the system.
Nevertheless, the above receiver scheduling scheme may cause energy starvation
in some EH receivers that experience poor channel conditions.
• Distributed antenna system (DAS). In SWIPT-DAS, dedicated power beacons
(stations) and traditional information base stations are distributed across the
network and connected to a central processing unit via backhaul links for joint
transmission. In particular, the spatial diversity provided by the DAS architecture
can effectively combat path loss by reducing the distance between transmitters
and receivers. The implementation of DAS for SWIPT relies on information
exchange via backhaul links among all transmitters for joint beamforming
optimization. Yet, the backhaul capacity can be limited due to the deployment
costs of the backhaul links, and full cooperation is impossible. If any power
beacons are not connected to the other information transmitters, exceedingly
large interference with the information receivers may be created by the power
Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
Multiple Antennas and Beamforming for SWIPT Systems 209

beacons, which jeopardizes the communication links. Furthermore, the optimal


placement of the power beacons and the information base stations is non-trivial
and requires thorough study.

5.7 Summary

WPT is a fundamental technology that enables self-sustainability of low-power energy-


limited mobile devices. More importantly, WPT provides the possibility for SWIPT,
which introduces a paradigm shift in both system and beamforming design. In this chap-
ter, we have presented a survey of beamforming designs for multiple-antenna SWIPT
systems. We first discussed the optimal transmit strategy for maximization of the total
power transfer. Then, we focused on system optimization for separated receivers and
power-splitting receivers. Our discussion has covered different aspects in SWIPT net-
works such as physical layer security, multi-objective optimization, and robust beam-
forming. The beamforming designs for the scenarios considered have been formulated
as non-convex optimization problems and were solved optimally via SDP relaxation.
Our simulation results have unveiled the potential benefits offered by multiple transmit
antennas in improving wireless power transfer efficiency, enhancing spectral/energy
efficiency of information transmission, and guaranteeing communication security in
SWIPT systems. In addition, future research challenges and some potential solutions
for the design of future multiple-antenna SWIPT systems have been discussed.

5.8 Appendix

5.8.1 Proof of Theorem 5.1


Since the relaxed version of problem (5.11) is jointly convex with respect to the opti-
mization variables and satisfies Slater’s constraint qualification, the KKT conditions are
necessary and sufficient conditions [47] for the optimal solution of the relaxed problem.
In the following, we focus on those KKT conditions which are useful for the proof:

Y∗  0, α ∗ , β ∗ , φj∗ , γn∗ , δ ∗ , θ ∗ ≥ 0 (5.54)


∗ ∗
Y W = 0, (5.55)

NT 
J
Y∗ = INT + γn  n − φj Gj − (α ∗ + β ∗ )H
n=1 j=1
∗ ∗
= A − (α + β )H, (5.56)

and
*
σs2 α ∗ req
ρ∗ = * √ , (5.57)
σs2 α ∗ req + β ∗ Pmin /η

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
210 Derrick Wing Kwan Ng et al.

 T J ∗ ∗ ∗ ∗ ∗ ∗ ∗
where A = INT + N n=1 γn  n − j=1 φj Gj and Y , α , β , φj , γn , δ , and θ are the
optimal Lagrange multipliers for (5.13). Equation (5.55) is the complementary slack-
ness condition and is satisfied when the columns of W∗ lie in the null space of Y∗ .
Therefore, if Rank(Y∗ ) = NT − 1, then the optimal W∗ = 0 must be a rank-one matrix
and the optimal w∗ can be obtained by performing eigenvalue decomposition of W∗ . On
the other hand, it can be observed from (5.57) that α ∗ > 0 and β ∗ > 0 at the optimal
solution for req > 0 and Pmin > 0. In other words, constraints C1 and C2 are satisfied
with equality simultaneously at the optimum point.
Now, we prove by contradiction that A is a full-rank matrix with rank NT whenever
the condition stated in Theorem 5.1 is satisfied. Let us focus on the dual problem in
(5.13). For a given set of optimal dual variables, D = {α ∗ , β ∗ , φ ∗ , γ ∗ , δ ∗ , θ ∗ , Y∗ } and
the optimal power-splitting ratio, ρ ∗ , the dual problem in (5.13) can be written as

 
minimize L W, ρ ∗ , α ∗ , β ∗ , φ ∗ , γ ∗ , δ ∗ , θ ∗ , Y∗ . (5.58)
W∈HNT

Supposing that A∗ is not positive definite, i.e., A∗  0, we can choose W = twwH as


a solution of (5.58), where t > 0 is a scaling parameter and w is the eigenvector with
respect to one of the non-positive eigenvalues of A∗ . Next, we substitute W = twwH
into (5.58), which leads to
  
Tr(tA∗ wwH ) − t Tr wwH Y∗ + (α ∗ + β ∗ )H + . (5.59)
!" #
≤0

Here  denotes a collection of the variables that are independent of W. On the other
hand, since channel vectors gj and h are assumed to be statistically independent, it
 ∗ ∗ ∗

follows that, on setting t → ∞, the term −t Tr ww Y + (α + β )H → −∞ and
H

the dual optimal value becomes unbounded from below. However, the optimal value
of the primal problem is strictly positive for req > 0. Thus, strong duality does not
hold, which leads to a contradiction. Therefore, A∗ is a positive definite matrix with
probability one, i.e., Rank(A∗ ) = NT .
By exploiting (5.56) and a basic inequality for the rank of matrices, we have

Rank(Y∗ ) + Rank((α ∗ + β ∗ )H)


≥ Rank(Y∗ + (α ∗ + β ∗ )H)
= Rank(A) = NT ⇒ Rank(Y∗ ) ≥ NT − 1. (5.60)

Thus, Rank(Y∗ ) is either NT − 1 or NT . Furthermore, W∗ = 0 is required to satisfy


the minimum SINR requirement of the power-splitting receiver in C1 for req > 0.
Hence, Rank(Y∗ ) = NT − 1 and Rank(W∗ ) = 1 hold with probability one. In other
words, the optimal beamformer w∗ can be obtained by performing eigenvalue decom-
position of W∗ and selecting the principal eigenvector as the beamformer.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
Multiple Antennas and Beamforming for SWIPT Systems 211

5.8.2 Proof of Proposition 5.2


The proof is based on the Charnes–Cooper transformation [66]. By substituting the new
optimization variables in (5.35) into Problem 8, we can rewrite Problem 8 as

θ B log2 (1 + Tr(HW)/(θ σ 2 ))
maximize F8 =
W,WE ∈HNT ,θ Tr(W + WE )/ρ + θ (NT Pant + Pc )
subject to C1 , C2 , C3 , C5 : θ > 0. (5.61)

Now, we show that the above problem is equivalent to Problem 12. First, it can
be observed that in problem (5.61) θ = 0 is not an optimal solution. Thus, without
loss of generality, the constraint θ > 0 can be replaced by θ ≥ 0. Second, we
prove by contradiction that C4 in Problem 12 is satisfied by equality for the optimal
solution, i.e.,
∗ ∗
Tr(W + WE )
+ θ ∗ (NT Pant + Pc ) = 1. (5.62)
ρ
∗ ∗
We denote the optimal solution of Problem 12 by (W , WE , θ ∗ ). Suppose that C4 is
∗ ∗
satisfied by strict inequality at the optimal solution, i.e., Tr(W + WE )/ρ +θ ∗ (NT Pant +
Pc ) < 1. Then, we construct a new feasible solution by applying a positive scaling to
  ∗ ∗
W and θ. The new solution is given by (W , WE , θ  ) = (cW , WE , cθ ∗ ), where
  
c > 1, such that Tr(W + WE )/ρ + θ (NT Pant + Pc ) = 1. It can be verified that
  ∗ ∗
(W , WE , θ  ) achieves a larger objective value in Problem 12 than does (W , WE , θ ∗ ).
∗ ∗ ∗
Thus, (W , WE , θ ) cannot be the optimal solution, which leads to a contradiction.
Thus, constraint C4 must hold with equality at the optimal solution. The equivalence
of (5.61) and Problem 12 is proved, which implies that Problem 12 is equivalent to
Problem 8. In particular, the solution of the original problem can be recovered from
(5.35). Similarly, the equivalence of Problems 13–15 and Problems 9–11 can be proved
by following the same approach.

5.8.3 Proof of Proposition 5.3


We start the proof by re-writing constraint C2 in the equivalent form

C2 : log2 det(INR + Q−1


j Gj WGj ) ≤ RER
H Tol
(5.63)
−1/2 H −1/2
⇐⇒ det(INR + Qj Gj WGj Qj ) ≤ 1 + αER . (5.64)

Then, we propose a lower bound on the left-hand side of (5.64) by introducing the
following lemma.

lemma 2. For any positive semidefinite square matrix A  0, the following inequality
holds [67]:

det(I + A) ≥ 1 + Tr(A), (5.65)

where the equality holds if and only if Rank(A) ≤ 1.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
212 Derrick Wing Kwan Ng et al.

Exploiting Lemma 2, the left-hand side of (5.64) is bounded from below by


−1/2 −1/2 −1/2 −1/2
det(INR + Qj GH
j WGj Qj ) ≥ 1 + Tr(Qj GH
j WGj Qj ). (5.66)

Subsequently, by combining equations (5.63), (5.64), and (5.66), we have the follow-
ing implications:
−1/2 −1/2
(5.63) ⇐⇒ (5.64) ⇒ Tr(Qj GH
j WGj Qj ) ≤ αER (5.67a)
−1/2 −1/2
⇒ λmax (Qj GH
j WGj Qj ) ≤ αER (5.67b)
−1/2 −1/2
⇐⇒ Qj GH
j WGj Qj  αER INR (5.67c)
⇐⇒ GH
j WGj  αER Qj . (5.67d)

We note that equations (5.63) and (5.67d) are equivalent, i.e., C2 ⇔ C2 , when
Rank(W) = 1.

Acknowledgements
This work was supported in part by the AvH Professorship Program of the Alexander
von Humboldt Foundation and by the Qatar National Research Fund (QNRF) under
project NPRP 5-401-2-161.

References

[1] Huawei, Green Energy Solution by Huawei (available at www.huawei.com/en/solutions/go-


greener/hw-001339-greencommunication-energyefficiency-emissionreduct.htm).
[2] L. Varshney, “Transporting information and energy simultaneously,” in Proc. IEEE Interna-
tional Symposium on Information Theory, July 2008, pp. 1612–1616.
[3] P. Grover and A. Sahai, “Shannon meets Tesla: Wireless information and power transfer,”
in Proc. IEEE International Symposium on Information Theory, June 2010, pp. 2363–2367.
[4] I. Krikidis, S. Timotheou, S. Nikolaou et al., “Simultaneous wireless information and power
transfer in modern communication systems,” IEEE Communications Magazine, vol. 52,
no. 11, pp. 104–110, November 2014.
[5] Z. Ding, C. Zhong, D. W. K. Ng et al., “Application of smart antenna technologies in
simultaneous wireless information and power transfer,” IEEE Communications Magazine,
vol. 53, no. 4, pp. 86–93, April 2015.
[6] X. Chen, Z. Zhang, H.-H. Chen, and H. Zhang, “Enhancing wireless information and power
transfer by exploiting multi-antenna techniques,” IEEE Communications Magazine, no. 4,
pp. 133–141, April 2015.
[7] Powercast Coporation, RF Energy Harvesting and Wireless Power for Low-Power Applica-
tions, 2011 (available at www.mouser.com/pdfdocs/Powercast-Overview-2011-01-25.pdf).
[8] A. Sample and J. Smith, “Experimental results with two wireless power transfer systems,”
in Proc. IEEE Radio and Wireless Symposium, January 2009, pp. 16–18.
[9] R. Zhang and C. K. Ho, “MIMO broadcasting for simultaneous wireless information and
power transfer,” in Proc. IEEE Global Telecommunications Conference, December 2011,
pp. 1–5.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
Multiple Antennas and Beamforming for SWIPT Systems 213

[10] L. Liu, R. Zhang, and K.-C. Chua, “Wireless information transfer with opportunistic energy
harvesting,” IEEE Transactions of Wireless Communications, vol. 12, no. 1, pp. 288–300,
January 2013.
[11] X. Zhou, R. Zhang, and C. K. Ho, “Wireless information and power transfer: Architecture
design and rate–energy tradeoff,” IEEE Transaction on Communication, vol. 61, pp. 4754–
4767, November 2013.
[12] D. W. K. Ng, E. S. Lo, and R. Schober, “Wireless information and power transfer: Energy
efficiency optimization in OFDMA systems,” IEEE Transaction on Wireless Communica-
tions, vol. 12, pp. 6352–6370, December 2013.
[13] D. W. K. Ng and R. Schober, “Resource allocation for secure communication in systems
with wireless information and power transfer,” in Proc. IEEE Global Telecommunications
Conference, December 2013.
[14] D. W. K. Ng and R. Schober, “Spectral efficient optimization in OFDM systems with
wireless information and power transfer,” in 21st European Signal Processing Conference
(EUSIPCO), September 2013, pp. 1–5.
[15] D. W. K. Ng, E. S. Lo, and R. Schober, “Energy-efficient power allocation in OFDM systems
with wireless information and power transfer,” in Proc. IEEE International Communications
Conference, June 2013, pp. 4125–4130.
[16] D. W. K. Ng, E. S. Lo, and R. Schober, “Robust beamforming for secure communication
in systems with wireless information and power transfer,” IEEE Transactions on Wireless
Communications, vol. 13, pp. 4599–4615, August 2014.
[17] R. T. Marler and J. S. Arora, “Survey of multi-objective optimization methods for
engineering,” Structural and Multidisciplinary Optimization, vol. 26, pp. 369–395, April
2004.
[18] L. Liu, R. Zhang, and K.-C. Chua, “Secrecy wireless information and power transfer with
MISO beamforming,” IEEE Transactions on Signal Processing, vol. 62, pp. 1850–1863,
April 2014.
[19] C. Y. Wong, R. S. Cheng, K. B. Letaief, and R. D. Murch, “Multiuser OFDM with adaptive
subcarrier, bit, and power allocation,” IEEE Journal on Selected Areas in Communications,
vol. 17, pp. 1747–1758, October 1999.
[20] V. K. N. Lau and Y. K. Kwok, Channel Adaptation Technologies and Cross Layer Design
for Wireless Systems with Multiple Antennas – Theory and Applications, 1st edn. New York:
Wiley, 2005.
[21] W.-L. Li, Y. J. Zhang, A.-C. So, and M. Win, “Slow adaptive OFDMA systems through
chance constrained programming,” IEEE Transactions on Signal Processing, vol. 58, pp.
3858–3869, July 2010.
[22] D. W. K. Ng, E. S. Lo, and R. Schober, “Energy-efficient resource allocation in multi-
cell OFDMA systems with limited backhaul capacity,” IEEE Transactions on Wireless
Communications, vol. 11, pp. 3618–3631, October 2012.
[23] D. W. K. Ng and R. Schober, “Cross-layer scheduling for OFDMA amplify-and-forward
relay networks,” IEEE Transactions on Vehicular Technology, vol. 59, pp. 1443–1458,
March 2010.
[24] D. W. K. Ng, E. S. Lo, and R. Schober, “Dynamic resource allocation in MIMO-OFDMA
systems with full-duplex and hybrid relaying,” IEEE Transactions on Communication,
vol. 60, May 2012.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
214 Derrick Wing Kwan Ng et al.

[25] H. Jabbar, Y. Song, and T. Jeong, “RF energy harvesting system and circuits for charging
of mobile devices,” IEEE Transactions on Consumer Electronics, vol. 56, pp. 247–253,
February 2010.
[26] X. Zhou, R. Zhang, and C. K. Ho, “Wireless information and power transfer: Architecture
design and rate–energy tradeoff,” in Proc. IEEE Global Telecommunications Conference,
December 2012.
[27] R. Zhang and C. K. Ho, “MIMO broadcasting for simultaneous wireless information and
power transfer,” IEEE Transactions on Wireless Communications, vol. 12, pp. 1989–2001,
May 2013.
[28] E. Boshkovska, D. Ng, N. Zlatanov, and R. Schober, “Practical non-linear energy har-
vesting model and resource allocation for SWIPT systems,” IEEE Communication Letters,
arXiv:1509.02956v1, 2015.
[29] S. Leng, D. W. K. Ng, N. Zlatanov, and R. Schober, “Multi-objective beamforming
for energy-efficient SWIPT systems,” in Proc. International Conference on Computing,
Networking and Communications, February 2016.
[30] D. W. K. Ng, E. S. Lo, and R. Schober, “Energy-efficient resource allocation in multiuser
OFDM systems with wireless information and power transfer,” in Proc. IEEE Wireless
Communications and Networking Conference, 2013.
[31] S. Leng, D. W. K. Ng, and R. Schober, “Power efficient and secure multiuser communication
systems with wireless information and power transfer,” in Proc. IEEE International
Communications Conference, June 2014.
[32] D. W. K. Ng, L. Xiang, and R. Schober, “Multi-objective beamforming for secure
communication in systems with wireless information and power transfer,” in Proc. IEEE
Personal, Indoor and Mobile Radio Communication Symposium, September 2013.
[33] D. W. K. Ng, R. Schober, and H. Alnuweiri, “Secure layered transmission in multicast
systems with wireless information and power transfer,” in Proc. IEEE International
Communications Conference, June 2014, pp. 5389–5395.
[34] D. W. K. Ng and R. Schober, “Resource allocation for coordinated multipoint networks
with wireless information and power transfer,” in Proc. IEEE Global Telecommunications
Conference, December 2014, pp. 4281–4287.
[35] M. Chynonova, R. Morsi, D. W. K. Ng, and R. Schober, “Optimal multiuser scheduling
schemes for simultaneous wireless information and power transfer,” in 23rd European
Signal Processing Conference (EUSIPCO), August 2015.
[36] Q. Wu, M. Tao, D. W. K. Ng, W. Chen, and R. Schober, “Energy-efficient transmission
for wireless powered multiuser communication networks,” in Proc. IEEE International
Communications Conference, June 2015.
[37] D. Ng and R. Schober, “Max–min fair wireless energy transfer for secure multiuser
communication systems,” in IEEE Information Theory Workshop (ITW), November 2014,
pp. 326–330.
[38] X. Chen, C. Yuen, and Z. Zhang, “Wireless energy and information transfer tradeoff for
limited-feedback multiantenna systems with energy beamforming,” IEEE Transactions on
Vehicular Technology, vol. 63, pp. 407–412, January 2014.
[39] H. Wang, W. Wang, X. Chen, and Z. Zhang, “Wireless information and energy transfer in
interference aware massive MIMO systems,” in Proc. IEEE Global Telecommunications
Conference, December 2014, pp. 2556–2561.
[40] IEEE P802.11 Wireless LANs, “TGn channel models”, IEEE 802.11-03/940r4, technical
report, May 2004.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
Multiple Antennas and Beamforming for SWIPT Systems 215

[41] J. F. Sturm, “Using SeDuMi 1.02, a MATLAB toolbox for optimization over symmetric
cones,” Optimization Methods and Software, vol. 11–12, pp. 625–653, September 1999.
[42] K. C. Toh, M. J. Todd, and R. H. Tütüncü, “SDPT3 – A Matlab software package for
semidefinite programming, version 1.3,” Optimization Methods and Software, vol. 11, pp.
545–581, January 1999.
[43] G. Zheng, K. K. Wong, and T. S. Ng, “Robust linear MIMO in the downlink: A worst-case
optimization with ellipsoidal uncertainty regions,” EURASIP Journal on Advances in Signal
Processing, article ID 609028, 2008.
[44] C. Shen, T.-H. Chang, K.-Y. Wang, Z. Qiu, and C.-Y. Chi, “Distributed robust multicell
coordinated beamforming with imperfect CSI: An ADMM approach,” IEEE Transactions
on Signal Processing, vol. 60, pp. 2988–3003, June 2012.
[45] N. Vucic and H. Boche, “Robust QoS-constrained optimization of downlink multiuser
MISO systems,” IEEE Transactions on Signal Processing, vol. 57, pp. 714–725, February
2009.
[46] J. Wang and D. Palomar, “Worst-case robust MIMO transmission with imperfect channel
knowledge,” IEEE Transactions on Signal Processing, vol. 57, pp. 3086–3100, August
2009.
[47] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge: Cambridge University
Press, 2004.
[48] M. Grant and S. Boyd, CVX: Matlab Software for Disciplined Convex Programming,
Version 2.0 Beta, September 2013 (available at https://cvxr.com/cvx).
[49] D. W. K. Ng, E. S. Lo, and R. Schober, “Multi-objective resource allocation for secure
communication in cognitive radio networks with wireless information and power transfer,”
IEEE Transactions on Vehicular Technology, arXiv:1403.0054, May 2015.
[50] E. Björnson and E. Jorswieck, Optimal Resource Allocation in Coordinated Multi-Cell
Systems. Boston, MA: Now Publishers Inc., 2013.
[51] Z. Hasan, G. Bansal, E. Hossain, and V. Bhargava, “Energy-efficient power allocation
in OFDM-based cognitive radio systems: A risk–return model,” IEEE Transactions on
Wireless Communications, vol. 8, pp. 6078–6088, December 2009.
[52] D. W. K. Ng, E. Lo, and R. Schober, “Energy-efficient resource allocation in OFDMA
systems with large numbers of base station antennas,” IEEE Transactions on Wireless
Communications, vol. 11, pp. 3292–3304, September 2012.
[53] D. W. K. Ng, E. S. Lo, and R. Schober, “Energy-efficient resource allocation in OFDMA
systems with hybrid energy harvesting base station,” IEEE Transactions on Wireless
Communications, vol. 12, pp. 3412–3427, July 2013.
[54] X. Chen, X. Wang, and X. Chen, “Energy-efficient optimization for wireless information
and power transfer in large-scale MIMO systems employing energy beamforming,” IEEE
Wireless Communications Letters, vol. 2, pp. 1–4, December 2013.
[55] A. D. Wyner, “The wire-tap channel,” The Bell System Technical Journal, vol. 54, no. 8,
pp. 1355–1387, October 1975.
[56] S. Goel and R. Negi, “Guaranteeing secrecy using artificial noise,” IEEE Transactions on
Wireless Communications, vol. 7, pp. 2180–2189, June 2008.
[57] D. W. K. Ng, E. S. Lo, and R. Schober, “Secure resource allocation and scheduling for
OFDMA decode-and-forward relay networks,” IEEE Transications on Wireless Communi-
ations, vol. 10, pp. 3528–3540, August 2011.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006
216 Derrick Wing Kwan Ng et al.

[58] X. Chen and H.-H. Chen, “Physical layer security in multi-cell MISO downlinks with
incomplete CSI – A unified secrecy performance analysis,” IEEE Transactions on Signal
Processing, vol. 62, pp. 6286–6297, December 2014.
[59] D. W. K. Ng, E. S. Lo, and R. Schober, “Efficient resource allocation for secure OFDMA
systems,” IEEE Transactions on Vehicular Technology, vol. 61, pp. 2572–2585, July 2012.
[60] J. Zhu, V. K. Bhargava, and R. Schober, “Secure downlink transmission in massive MIMO
system with zero-forcing precoding,” in Proc. European Wireless Conference, May 2014,
pp. 1–6.
[61] L. Liu, R. Zhang, and K.-C. Chua, “Secrecy wireless information and power transfer with
MISO beamforming,” in Proc. IEEE Global Telecommunications Conference, December
2013.
[62] D. W. K. Ng and R. Schober, “Secure and green SWIPT in distributed antenna networks
with limited backhaul capacity,” IEEE Transactions on Wireless Communications, vol. 14,
no. 9, pp. 5082–5097, September 2015.
[63] Q. Wang, H. Su, K. Ren, and K. Kim, “Fast and scalable secret key generation exploiting
channel phase randomness in wireless networks,” in Proc. IEEE International Conference
on Computer Communications, April 2011, pp. 1422–1430.
[64] J. Zhu, R. Schober, and V. Bhargava, “Secure transmission in multicell massive MIMO
systems,” IEEE Transactions Wireless Communications, vol. 13, pp. 4766–4781, September
2014.
[65] X. Chen, J. Chen, and T. Liu, “Secure wireless information and power transfer in large-scale
MIMO relaying systems with imperfect CSI,” in Proc. IEEE Global Telecommunications
Conference, December 2014, pp. 4131–4136.
[66] A. Charnes and W. W. Cooper, “Programming with linear fractional functions,” Naval
Research Logistics Quarterly, vol. 9, pp. 181–186, April 1962.
[67] Q. Li and W. K. Ma, “Spatially selective artificial-noise aided transmit optimization for
MISO multi-eves secrecy rate maximization,” IEEE Transactions on Signal Processing,
vol. 61, pp. 2704–2717, May 2013.

Downloaded from https:/www.cambridge.org/core. University of Exeter, on 10 Mar 2017 at 08:07:47, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781316471845.006

You might also like