MIMO Energy Harvesting in Full-Duplex Multi-User Networks

Tam, H. H. M., Tuan, H. D., Nasir, A. A., Duong, Q., & Poor, H. V. (2017). MIMO Energy Harvesting in Full-
Duplex Multi-User Networks. IEEE Transactions on Wireless Communications, 16(5), 3282-3297. DOI:
10.1109/TWC.2017.2679055

Published in:
IEEE Transactions on Wireless Communications

Document Version:
Peer reviewed version

Queen's University Belfast - Research Portal:

Link to publication record in Queen's University Belfast Research Portal

Publisher rights
Copyright 2017 IEEE.
This work is made available online in accordance with the publisher’s policies. Please refer to any applicable terms of use of the publisher.

General rights
Copyright for the publications made accessible via the Queen's University Belfast Research Portal is retained by the author(s) and / or other
copyright owners and it is a condition of accessing these publications that users recognise and abide by the legal requirements associated
with these rights.

Take down policy

The Research Portal is Queen's institutional repository that provides access to Queen's research output. Every effort has been made to
ensure that content in the Research Portal does not infringe any person's rights, or applicable UK laws. If you discover content in the
Research Portal that you believe breaches copyright or violates any law, please contact [email protected].

Download date:18. Dec. 2017

 1

MIMO Energy Harvesting in Full-Duplex

Multi-user Networks
H. H. M. Tam, H. D. Tuan, A. A. Nasir, T. Q. Duong and H. V. Poor

Abstract—The paper aims at the efficient design of precoding practical applications [1], [2], [5] due to the close BS-UE
matrices for the sum throughput maximization under throughput proximity. In order to transfer both energy and information
QoS constraints and energy harvesting (EH) constraints for by the same communication channel, UEs are equipped with
energy-constrained devices in a full-duplex (FD) multicell multi-
user multiple-input-multiple-output (MU-MIMO) network. Both both information decoding receiver and energy harvesting
time splitting (TS) and power splitting (PS) are considered receiver. Since the received signal cannot be used for energy
to ensure practical EH and information decoding (ID). These harvesting after being decoded, there are two available im-
problems are quite complex due to highly non-concave objec- plementations for wireless energy harvesting and information
tives and nonconvex constraints. Especially, with TS, which is decoding: (i) receive power splitting in which a receiver splits
implementation-wise quite simple, the problem is even more
challenging because time splitting variable is not only coupled the received signal into two streams of different power for
with the downlink (DL) throughput function but also coupled decoding information and harvesting energy separately and
with the self-interference (SI) in the uplink (UL) throughput (ii) transmit time splitting to enable the receiver to decode
function. New path-following algorithms are developed for their information for a portion of a time frame and harvest energy
solutions, which just require a single convex quadratic program for the rest. Beamforming can be applied to focus the RF
for each iteration and ensure fast convergence too. Finally, the
FD EH maximization problem under throughput QoS constraints signal to energy harvesting receiver or enhance throughput at
in TS is also considered. The performance of the proposed information decoding receiver [5].
algorithms is also compared with that of the modified problems Most of the previous works (see e.g. [6], [7] and references
assuming half-duplex systems. In the end, the merit of the therein) only focus on beamforming power optimization sub-
proposed algorithms is shown through extensive simulations. ject to ID throughput and EH constraints with PS in multi-
Index Terms—Full-duplexing transceiver, energy harvesting, input single-output (MISO) networks. The ID throughput
information precoder, energy precoder, path-following algorithm, constraints are equivalent to signal-to-interference-plus-noise
matrix inequality ratio (SINR) constraints, which are indefinite quadratic in
beamforming vectors. The harvested energy constraints are
I. I NTRODUCTION also indefinite quadratic constraints. Thus, [6], [7] used semi-
Recently, wireless energy harvesting (i.e. energy constrained definite relaxation (SDR) to relax such indefinite quadratic
devices scavenge energy from the surrounding RF signals) optimization problems to semi-definite programs (SDP) by
is gaining more and more attraction from both industry and dropping the matrix rank-one constraints on the outer products
academia [1], [2]. Since the amount of energy opportunis- of beamforming vectors. The variable dimension of SDP
tically harvested from the ambient/natural energy sources is is explosively large, and the beamforming vectors that are
uncertain and cannot be controlled, base stations (BSs) in recovered based on the matrix solution of SDR perform poorly
small-cell networks can be configured to become dedicated [8]. Moreover, SDR cannot be applied to throughput or EH
and reliable wireless energy sources [3]. The small cell size maximization as the problems resultant by SDR are still highly
not only gives the benefit of efficient resource reuse across nonconvex. Only recently there was an effective development
a geographic area [4] but also provides an adequate amount to address these problems in [9] and [10].
of RF energy to battery powered user equipments (UEs) for Considering multi-input multi-output (MIMO) interference
channels, information throughput and harvested energy, i.e.,
This work was supported in part by the Australian Research Councils rate-energy (R-E) trade-off, was investigated in [11] and [12],
Discovery Projects under Project DP130104617, in part by the U.K. Royal assuming that any UE either acts as an ID receiver or an EH
Academy of Engineering Research Fellowship under Grant RF1415/14/22
and U.K. Engineering and Physical Sciences Research Council under Grant receiver. In case that UEs can operate both as an ID receiver
EP/P019374/1, and in part by the U.S. National Science Foundation under and EH receiver (namely co-located cases), the R-E region of
Grants CNS-1456793 and ECCS-1647198. point-to-point MIMO channel was studied in [13]. Note that
Ho Huu Minh Tam and Hoang Duong Tuan are with the Faculty of
Engineering and Information Technology, University of Technology Sydney, in MIMO networks, the information throughput function is
Broadway, NSW 2007, Australia (email: [email protected], involved with the determinant operation of a matrix and can
[email protected]). no longer be expressed in the form of SINR. Consequently,
Ali Arshad Nasir is with the Department of Electrical Engineering, King
Fahd University of Petroleum and Minerals (KFUPM), Dhahran, 31261, KSA the throughput constraints are always very challenging in
(Email: [email protected]). precoding signals. [14], [15] used zero-forcing or interference-
Trung Quang Duong is with Queen’s University Belfast, Belfast BT7 1NN, alignment to cancel all interferences, making the throughput
U.K. (Email: [email protected])
H. Vincent Poor is with the Department of Electrical Engineering, Princeton functions concave in the signal covariance. The covariance
University, Princeton, NJ 08544 USA (e-mail: [email protected]). optimization becomes convex but it is still computationally
2

difficult with no available algorithm of polynomial time. function and nonconvex constraints. However, we will see
Moreover, there is no known method to recover the pre- that the PS problem can be efficiently addressed by adapting
coder matrices from the signal covariance. Only recently, the the algorithm of [16]. On the other hand, the TS problem is
MIMO throughput function optimization has been successfully much more challenging because the TS variable α is not only
addressed for non-EH system in our previous work via a coupled with the DL throughput function but also coupled
successive convex quadratic programming [16]. The result of with the SI in the UL throughput function. It is nontrivial to
[16] can be adapted to MIMO networks that employ EH by extend [16] to solve the problem for the TS problem. Toward
PS approach. However, there is almost no serious research for this end, we develop a new inner approximation of the original
the systems employing TS in MIMO networks. Though TS- problem and solve the problem by a path-following algorithm.
based system is practically easier to implement, the related Finally, we also consider the FD EH maximization problem
formulated problem is quite complex because the throughput with throughput QoS constraints with TS. This problem also
function in such case is coupled with the TS variable that has a nonconvex objective function and nonconvex constraints
defines the portion of time slot dedicated to EH and ID. and will be addressed by applying an approach similar to that
This renders the aforementioned precoder design [14]–[16] of proposed for the TS problem.
for PS inapplicable. To the best of our knowledge, both the The rest of this paper is organized as follows: Section
throughput maximization problem and the harvested energy II presents the system model the SCP algorithm of the PS
maximization problem with TS are still very open. problem. The main contribution of the paper is Section III
All aforementioned works only assume that UEs only har- and Section IV, which develop algorithms for the TS problem
vest energy arriving from BSs’ downlink (DL) transmission. and FD EH maximization problem. Section V evaluates the
In reality, UEs can also opportunistically harvest energy from performance of our devised solutions by numerical examples.
other UEs’ signals during their uplink (UL) transmission. Finally, Section VI concludes the paper.
Furthermore, by allowing the BSs to simultaneously transmit Notation. All variables are boldfaced. In denotes the identity
and receive information, both the spectral efficiency and the matrix of size n×n. The notation (.)H stands for the Hermitian
amount of transferred energy will be improved. With the recent transpose. |A| denotes the determinant of a square matrix A
advances in antenna design and RF circuits in reducing self- and hAi denotes the trace of a matrix A. (A)2 is Hermitian
interference (SI) [17]–[20], which is the interference from a symmetric positive definite AAH . The inner product hX, Y i is
BS’s DL transmission to its UL receiver, the full duplex (FD) defined as hX H Y i and therefore the Frobenius squared norm
technology is recently proposed as one of the key transceiving of a matrix X is ||X||2 = h(X)2 i. The notation A  B
techniques for the fifth generation (5G) networks [20]–[24]. (A  B, respectively) means that A − B is a positive
In this paper, we are interested in a network in which each semidefinite (definite, respectively) matrix. E[.] denotes the
FD multi-antenna BS simultaneously serves a group of UL expectation operator and <{.} denotes the real part of a
UEs (ULUs) and a group of DL UEs (DLUs). In the same complex number.
time, the BS also transfers energy to DLUs via TS or PS. The following concept of function approximation [28] plays
FD transmission introduces even more interferences into the an important in our development.
network by adding not only SI but also the interference from Definition. A function f˜ is called a (global) minorant of a
UL users (ULUs) toward downlink users (DLUs) and the function f at a point x̄ in the definition domain dom(f ) of f
interference from DL transmission of other BSs. Consequently, if f˜(x̄) = f (x̄) and f (x) ≥ f˜(x) ∀ x ∈ dom(f ).
the UL and DL precoders are coupled in both DL and The following result [16] is used.
UL throughput functions, respectively, which makes the op- Theorem 1: For function f (V, Y) = ln |In + VH Y−1 V|
timization problems for UL transmission and DL transmission in matrix variable V ∈ Cn×m and positive definite matrix
inseparable. variable Y ∈ Cm×m , the following quadratic function is its
In literature, [14], [25], [26] proposed covariance matrices minorant at (V̄ , Ȳ )
design in (non-EH) FD MU-MIMO networks using D.C.
iterations [27], which are still very computationally demanding f˜(V, Y) = a + 2<{hA, Vi} − hB, VVH + Yi,
as they require log-determinant function optimizations as men- where 0 > a , f (V̄ , Ȳ ) − hV̄ H Ȳ −1 V̄ i, A = Ȳ −1 V̄ and
tioned above. Our previous work [16] has recently proposed a 0  B = Ȳ −1 − (Ȳ + V̄ V̄ H )−1 .
framework to directly find the optimal precoding matrices for
the sum throughput maximization under throughput constraints
in FD MU-MIMO multi-cell networks, which requires only a II. EH- ENABLED FD MU-MIMO N ETWORKS
convex quadratic program of moderate size in each iteration We consider an MU-MIMO EH-enable network consisting
and thus is very computationally efficient. of I cells. In cell i ∈ {1, . . . , I}, a group of D DLUs in the
In this paper, we propose the design of efficient precoding downlink (DL) channel and a group of U ULUs in uplink
matrices for the network sum throughput maximization under (UL) channel are served by a BS i as illustrated in Fig. 1.
QoS in terms of MIMO throughput constraints and EH con- Each BS operates in the FD mode and is equipped with
straints in an FD EH-enabled multicell MU-MIMO network. N , N1 + N2 antennas, where N1 antennas are used to
Both PS and TS are considered for the precoder designs and transmit and the remaining N2 antennas to receive signals.
called by PS problem and TS problem, respectively. They are In cell i, DLU (i, jD ) and ULU (i, jU ) operate in the HD
quite challenging computationally due to nonconcave objective mode and each is equipped with Nr antennas. In the DL,
 3

Assuming that DLUs are equipped by both devices for ID

(1,1U) and EH, the power splitting technique is applied at each DLU
ITI1 (2,1D)
to simultaneously conduct information decoding and energy
SI ITI3 SI harvesting. The power splitter divides the received signal yi,jD
ITI2 ITI1 into two parts in the proportion of α i,jD : (1 − α i,jD ) where
BS 2
ITI3
α i,jD ∈ (0, 1) is termed as the PS ratio for DLU (i, jD ). In
BS 1 ITI3
(1,1D) particular, the signal split to the ID receiver of DLU (i, jD ) is
ITI2 (2,2U) given by
(1,2D) √
(2,1U) c
α i,jD yi,jD + zi,j , (2)
ITI3 D
ITI1 ITI2
c c
(3,1U)
where each r-th element of zi,j D
(i.e. |zi,j D ,r
|2 = σc2 , r =
SI Base station 1, .., Nr ) is additional noise introduced by the ID receiver
DLU circuity.
ITI2 ULU pAn EH receiver processes the second part of the split
BS 3 signal 1 − α i,jD yi,jD for the harvested energy
Intended signal
ITI1 Interference q
ζi,jD (1 − α i,jD )yi,jD ,
(3,1D)
where ζi,jD ∈ (0.4, 0.6) is the efficiency of energy conversion.
It follows from the receive equation (1) and the split
Fig. 1. Interference scenario in an FD multicell network, where SI denotes equation (2) that the downlink information throughput at DLU
the self-interference and ITIi denotes the interference from the BS and ULUs (i, jD ) is
of cell i.
fi,jD (VD , VU , α i,jD ) ,
 
2 −1
let si,jD  ∈ Cd1 be the ln INr + (Li,jD (Vi,jD )) Ψi,jD (VD , VU , α i,jD ), (3)
 
 symbol intended for DLU (i, jD )
where E si,jD (si,jD )H = Id1 , d1 is the number of concurrent
data streams and d1 ≤ min{N1 , Nr }. The vector of symbols where Li,jD (Vi,jD ) , Hi,i,jD Vi,jD and
si,jD is precoded and transmitted to DLU (i, jD ) through the Ψi,jD (VD , VU , α i,jD ) , Ψ̄i,jD (VD , VU ) + (σc2 /α
αi,jD )INr (4)
precoding matrix Vi,jD ∈ CN1 ×d1 . Analogously, in the UL,
si,jU ∈ Cd2 is the information symbols sent by ULU (i, jU ) with the downlink interference covariance mapping
Nr ×d2
and is precoded  precoding matrix Vi,jU ∈ C
by the , X
 H Ψ̄i,jD (VD , VU ) , (Hm,i,jD Vm,`D )2
where E si,jU (si,jU ) = Id2 , d2 is the number of concurrent
(m,`D )∈S1 \(i,jD )
data streams and d2 ≤ min{N2 , Nr }. For notational conve- X
nience, let us define + (Hi,jD ,`U Vi,`U )2 + σD INr . (5)
`U ∈U
I , {1, 2, . . . , I}; D , {1D , 2D , . . . , DD };
U , {1U , 2U , . . . , UU }; S1 , I × D; S2 , I × U; The harvested energy at UE (i, jD ) is
VD = [Vi,jD ](i,jD )∈S1 ; VU = [Vi,jU ](i,jU )∈S2 ; αi,jD )hΦi,jD (VD , VU )i, (6)
Ei,jD (VD , VU , α i,jD ) = ζi,jD (1−α
V , [VD VU ];
with the downlink signal covariance mapping
In the DL channel, the received signal at DLU (i, jD ) is X
expressed as: Φi,jD (VD , VU ) , (Hm,i,jD Vm,`D )2
(m,`D )∈S1
yi,jD , Hi,i,jD Vi,jD si,jD X
| {z } + (Hi,jD ,`U Vi,`U )2 + σD
2
INr . (7)
desired signal
X `U ∈U
+ Hm,i,jD Vm,`D sm,`D
In the UL channel, the received signal at BS i is expressed as
(m,`D )∈S1 \(i,jD )
| {z } X
DL interference yi , Hi,`U ,i Vi,`U si,`U
`U ∈U
X
+ Hi,jD ,`U Vi,`U si,`U +ni,jD , (1) | {z }
`U ∈U desired signal
| {z } X X
UL intracell interference + Hm,`U ,i Vm,`U sm,`U
Nr ×N1 Nr ×Nr m∈I\{i} `U ∈U
where Hm,i,jD ∈ C and Hi,jD ,`U ∈ C are | {z }
the channel matrices from BS m to DLU (i, jD ) and from X X
UL interference
ULU (i, `U ) to DLU (i, jD ), respectively. Also, ni,jD is the B
+ Hm,i Vm,jD sm,jD + nSIi +ni ,(8)
additive white circularly symmetric complex Gaussian noise m∈I\{i} jD ∈D
|{z}
with variance σD 2
. In this work, the UL intercell interference | {z } residual SI
DL intercell interference
is neglected since it is very small compared to the DL intercell
interference due to the much smaller transmit power of ULUs. where Hm,`U ,i ∈ CN2 ×Nr and Hm,i
B
∈ CN2 ×N1 are the
Nevertheless, it can be incorporated easily in our formulation. channel matrices from ULU (m, `U ) to BS i and from BS
4

U,min
m to BS i, respectively; ni is the additive white circularly and (13h) represent QoS guarantee, where emin i,jD , ri and
symmetric complex Gaussian noise with variance σU2 ; nSI i
D,min
ri,jD are the minimum harvested energy required by DLU
is the residual SI (after self-interference cancellation) at BS (i, jD ), the minimum data throughput required by BS i and
i and depends on the transmit power of BS i. Specifically, the minimum data throughput required by DLU (i, jD ). In
nSI
i is modelled as the additive white circularly symmetric comparison to [16] for FD non-EH-enable networks, the UL
2 2
P
complex Gaussian noise with variance σSI `D ∈D ||Vi,`D || throughput function fi (VD , VU ) in (9) is the same, where the
2
[29], where the SI level σSI is the ratio of the average SI DL throughput function fi,jD (VD , VU , α i,jD ) is now addition-
powers after and before the SI cancellation process. ally dependent on the SP variable α i,jD , is decoupled in (5)
Following [14], [16], [26], the optimal minimum mean and thus does not add more difficulty as we will show now.
square error - Successive interference cancellation (MMSE- We also show that the nonconvex EH constraints (13f) can
SIC) decoder is applied at BSs. Therefore, the achievable easily be innerly approximated.
uplink throughput at BS i is given as [30] Under the definitions,
fi (VD , VU ) , ln IN2 + (Li (VUi ))2 Ψ−1
 
i (VD , VU ) , (9)
Mi,jD (VD , VU , α i,jD ) ,(Li,jD (Vi,jD ))2


where
 VUi , [Vi,`U ]`U ∈U and  Li (VUi ) , + Ψi,jD (VD , VU , α i,jD ) (14)
Hi,1U ,i Vi,1U , Hi,2U ,i Vi,2U , . . . , Hi,UU ,i Vi,UU , which means Ψi,jD (VD , VU , α i,jD ), (15)
U
2
Mi (VD , VU ) ,(Li (VUi )) + Ψi (VD , VU )
X
that (Li (VUi ))2 = (Hi,`U ,i Vi,`U )2 , and (16)
`=1 Ψi (VD , VU ), (17)
Ψi (VD , VU ) , Ψ̄U
i (VU ) + Ψ̄SI
i (VD ) (10)
by applying Theorem 1 as in [16], we obtain the follow-
with uplink interference covariance mapping ing concave quadratic minorants of throughput functions
(κ) (κ) (κ) (κ)
X X fi,jD (VD , VU , αi,jD ) and fi (VD , VU ) at (VD , VU , α(κ) )
Ψ̄Ui (VU ) , (Hm,`U ,i Vm,`U )2 (κ) (κ) (κ)
, ([Vi,jD ](i,jD )∈S1 , [Vi,`U ](i,`U )∈S2 , [αi,jD ](i,jD )∈S1 ):
m∈I\{i} `U ∈U
 
(κ)
Θ (VD , VU , α i,jD ) ,
n i,jD
X X
B  B
+ Hm,i (Vm,jD )2  (Hm,i )H + σU2 IN2 (11) (κ) (κ)
o
m∈I\{i} jD ∈D ai,jD + 2< hAi,jD , Li,jD (Vi,jD )i
(κ)
and SI covariance mapping −hBi,jD , Mi,jD (VD , VU , α i,jD )i (18)
X
Ψ̄SI 2
i (VD ) , σSI ||Vi,`D ||2 IN2 . (12) and
`D ∈D
(κ)
We consider the design problem Θi (VD , VU ) ,
n o
(κ) (κ) (κ)
X ai + 2< hAi , Li (VUi )i − hBi , Mi (VD , VU )i, (19)
max P1 (VD , VU , α) , fi (VD , VU )
α
VD ,VU ,α
i∈I
X where
+ fi,jD (VD , VU , α i,jD ) s.t. (13a)
(κ) (κ) (κ) (κ)
(i,jD )∈S1 0 > ai,jD , fi,jD (VD , VU , αi,jD )
0 < α i,jD < 1, (i, jD ) ∈ S1 , (13b) n
(κ) (κ) (κ) (κ)
o
X X −< hΨ−1
i,jD (VD , VU )Li,jD (Vi,jD ), Li,jD (Vi,jD )i ,
||Vi,jD ||2 + ||Vi,jU ||2 ≤ P, (13c)
(κ) (κ) (κ) (κ) (κ)
(i,jD )∈S1 (i,jU )∈S2 Ai,jD = Ψ−1
i,jD (VD , VU , αi,jD )Li,jD (Vi,jD ),
X
||Vi,jD ||2 ≤ Pi , ∀i ∈ I, (13d) (κ)
0  Bi,jD = Ψ−1
(κ)
i,jD (VD , VU , αi,jD )
(κ) (κ)

jD ∈D
(κ) (κ) (κ)
||Vi,jU ||2 ≤ Pi,jU , ∀(i, jU ) ∈ S2 , (13e) −M−1
i,jD (VD , VU , αi,jD ),
(20)
hΦi,jD (VD , VU )i ≥ emin
i,jD /ζi,jD (1 − α i,jD ), and
∀(i, jD ) ∈ S1 , (13f) (κ) (κ) (κ)
0 > ai = fi (VD , VU )
fi (VD , VU ) ≥ riU,min , ∀i ∈I (13g) n o
(κ) (κ) (κ) (κ)
D,min
fi,jD (VD , VU , α i,jD ) ≥ ri,jD
, ∀(i, jD ) ∈ S1 . (13h) −< hΨ−1
i (VD , VU )L (V
i i ), L (V
i i )i ,
(κ) (κ) (κ) (κ)
In the formulation (13), all channel matrices in the downlink Ai = Ψ−1
i (VD , VU )Li (VUi ),
(21)
equation (1) and uplink (8) are assumed to be known by using (κ) (κ) (κ)
the channel reciprocity, feedback and learning mechanisms 0  Bi = Ψ−1
i (VD , VU )
(see e.g. [31]). The convex constraints (13d) and (13e) spec- (κ) (κ)
−M−1
i (VD , VU ).
ify the maximum transmit power available at the BSs and
the ULUs whereas (13c) limits the total transmit power of To handle the nonconvex EH constraints (13f), we define an
(κ)
the whole network. The nonconvex constraints (13f), (13g) affine function φi,jD (VD , VU ) as the first-order approximation
 5

Algorithm 1 Path-following algorithm for PS sum throughput (0) (0)

point (VD , VU , α(0) ) meeting the nonconvex constraints
maximization (13) (13f)-(13h) we consider the following problem:
Initialization: Set κ := 0, and choose a feasible point
(0) (0) max P1,f (VD , VU , α ) ,
(VD , VU , α(0) ) that satisfies (13b)-(13h). α
VD ,VU ,α
emin
n
κ-th iteration: Solve (23) for an optimal solution min Φi,jD (VD , VU ) − ζi,j (1−α i,jD
(κ) (κ) αi,jD ) ,
(VD∗ , VU∗ , α∗ ) and set κ := κ + 1, (VD , VU , α(κ) ) := (i,jD )∈S1 D
o (27)
∗ ∗ ∗ (κ) (κ) (κ) min min
 D , VU , α ) and calculate P1 (VD , VU , α ). Stop
(V  if fi,jD (VD , VU , α ) − ri,iD
, fi (V D , V U ) − ri
(κ) (κ) (κ) (κ−1) (κ−1) (κ−1)
 P1 (VD , VU , α ) − P1 (VD , VU ,α ) s.t. (13b) − (13e).


(κ−1) (κ−1) (κ−1)  (0) (0)
/P1 (VD , VU ,α ) ≤ . Initialized by a (VD , VU , α(0) ) feasible to the
convex constraints (13b)-(13e), an iterative point
(κ+1) (κ+1)
(VD , VU , α(κ+1) ) for κ = 0, 1, . . . , is generated
(κ) (κ) as the optimal solution of the following convex maximin
of the convex function hΦi,jD (VD , VU )i at (VD , VU ):
program:
(κ) (κ) (κ) (κ)
φi,jD (VD , VU ) , −hΦi,jD (VD , VU )i max P1,f (VD , VU , α ) ,
α
VD ,VU ,α
(κ)
X
H
+2<{ hHm,i,jD Vm,`D Vm,`D Hm,i,jD
i} n
(κ) emin
i,jD
(m,`D )∈S1
min φi,jD (VD , VU ) − ,
(i,jD )∈S1 ζi,jD (1 − α i,joD )
(κ)
X
H H 2 (κ) (κ)
+2<{ hHi,iD ,`U Vi,`U Vi,` U
Hi,iD ,`U
i} + 2σD Nr , (22) min
hΘi,jD (VD , VU , α ) − ri,iD
, Θi (VD , VU ) − rimin ,
`U ∈U
s.t (13b) − (13e).
(κ) (κ) (28)
which is an minorant of hΦi,jD (VD , VU )i at (VD , VU ) [28].
We now address the nonconvex problem (13) by succes- which terminates upon reaching
sively solving its following inner approximation: (κ) (κ)
fi,jD (VD , VU , α(κ) ) ≥ ri,i
min (κ) (κ)
, fi (VD , VU ) ≥ rimin ,
D
min
(κ)
X (κ) ei,jD
max P1 (VD , VU , α ) , Θi (VD , VU ) (κ) (κ)
hΦi,jD (VD , VU )i ≥ , ∀(i, jD ) ∈ S1
α
VD ,VU ,α ζi,jD (1 − α i,jD )
i∈I
(κ)
X
+ Θi,jD (VD , VU , α i,jD ) (23a) to satisfy (13b)-(13h).
(i,jD )∈S1 In parallel, we consider the following transmission strategy
s.t. (13b) − (13e) (23b) to configure FD BSs to operate in the HD mode. Here, all
(κ) antennas N = N1 + N2 at each BS are used to serve all the
φi,jD (VD , VU ) ≥ emin
i,jD /ζi,jD (1 − α i,jD ), DLUs in the downlink and all the ULUs in the uplink using
∀(i, jD ) ∈ S1 , (23c) half time slots, where DLUs are allowed to harvest energy
(κ)
Θi (VD , VU ) ≥ riU,min , ∀i ∈ I, (23d) from ULUs. The problem can be formulated as
(κ) D,min 1h X
Θi,jD (VD , VU , α i,jD ) ≥ ri,j D
, ∀(i, jD ) ∈ S1 . (23e) max fi,jD (VD , 0U , α i,jD )
α 2
VD ,VU ,α
(κ) (κ) (i,jD )∈S1
Initializing from (VD , VU , α(κ) ) being feasible point to X i
(13), the optimal solution (VD
(κ+1) (κ+1)
, VU , α(κ+1) ) of convex + fi (0D , VU ) (29a)
i∈I
program (23) is feasible to the nonconvex program (13) and
(κ) (κ) s.t. (13b), (13c), (13d), (13e),
it is better than (VD , VU , α(κ) ):
1
(Ei,jD (VD , 0U , α i,jD ) + Ei,jD (0D , VU , 0)) ≥
(κ+1)
P1 (VD , VU
(κ+1)
, α(κ+1) ) ≥ 2
(κ) (κ+1) (κ+1) emin
i,jD , ∀(i, jD ) ∈ S1 , (29b)
P1 (VD , VU , α(κ+1) ) ≥ (24)
1 U,min
(κ) (κ) (κ)
P1 (VD , VU , α(κ) ) = (25) fi (0D , VU ) ≥ ri , ∀i ∈ I (29c)
2
(κ) (κ)
P1 (VD , VU , α(κ) ), (26) 1 D,min
fi,j (VD , 0U , α i,jD ) ≥ ri,j , ∀(i, jD ) ∈ S1 , (29d)
2 D D

where the inequality (24) and the equality (26) follow from where 0D and 0U are zero quantity of the same dimension
(κ)
the fact that P1 is a minorant of P1 while the inequal- with VD and VU . In (29), DLU (i, jD ) uses (1 −α αi,jD ) of the
(κ+1) (κ+1)
ity (25) follows from the fact that (VD , VU , α(κ+1) ) received signal during DL transmission and the whole received
(κ) (κ) (κ) signal during UL transmission for EH as formulated in (29b).
and (VD , VU , α ) are the optimal solution and feasi-
ble point of (23), respectively. This generates a sequence The main difference between (13) and (29) is in (29b) where
(κ) (κ)
{(VD , VU , α(κ) )} of feasible and improved points which the harvested energy from UL transmission at DLU (i, jD )
converge to a local optimum of (13) after finitely many does not multiply with α i,jD . The constraint (29b) can be recast
iterations [16]. as
The proposed path-following procedure that solves problem hΦi,jD (0D , VU )i 2emin
i,jD
hΦi,jD (VD , 0U )i + ≥ .
(13) is summarized in Algorithm 1. To find a feasible initial (1 − α i,jD ) ζi,jD (1 − α i,jD )
6

Define the following convex function: where sEm,`D is the energy signal sent for (1 − α ) time. With
hΦi,jD (0D , VU )i the definition (6), the harvested energy is
Λi,jD (VU , α i,jD ) ,
(1 − α i,jD )
h `U ∈U (Hi,jD ,`U Vi,`U )2 + σD
2
P
INr i E
Ei,jD (VD E
, VU , α ) = ζi,jD (1 − α )hΦi,jD (VD , VU )i,
= , (30)
1 − α i,jD
with its first-order approximation where the downlink signal covariance mapping Φi,jD (., .) is
(κ) defined from (7).
Λi,jD (VU , 1 − α i,jD ) ,
P (κ) H Similarly to (1), the signal received at DLU (i, jD ) during the
2<{h `U ∈U (Hi,jD ,`U Vi,`U )(Hi,jD ,`U Vi,`U ) i} information transmission in time fraction α is
(κ)
1 − αi,jD
(κ)
h `U ∈U (Hi,jD ,`U Vi,`U )2 + σD2
P
INr i I
yi,j , Hi,i,jD Vi,jD sIi,jD
− (κ) 2
(1 − α i,jD ), (31) D

(1 − αi,jD )
| {z }
desired signal
X
(κ)
which is its minorant at (VD , VU , α(κ) ).
(κ) + Hm,i,jD Vm,`D sIm,`D
Algorithm 1 can be used with the following convex program (m,`D )∈S1 \(i,jD )
| {z }
solved for κ-iteration: DL interference
1h X (κ) +
X
Hi,jD ,`U Vi,`U si,`U +ni,jD , (34)
max Θi,jD (VD , 0U , α i,jD )
α 2
VD ,VU ,α
(i,jD )∈S1 `U ∈U
X (κ) i | {z }
UL intracell interference
+ Θi (0D , VU , 0) (32a)
i∈I
s.t. (13b), (13c), (13d), (13e), where sIm,`D is the information signal intended for DLU
(κ) (κ)
φi,jD (VD , 0U ) + Λi,jD (VU , 1 − α i,jD ) ≥ (m, `D ). The ID throughput at DLU (i, jD ) is then given as
2emin α fi,jD (V), where
i,jD
, ∀(i, jD ) ∈ S1 , (32b)
ζi,jD (1 − α i,jD )
1 (κ)
 
2 −1
Θ (0D , VU ) ≥ riU,min , ∀i ∈ I (32c) I
fi,jD (VD I
, VU ) = ln INr + (Li,jD (Vi,j )) Ψ̄ (V I
, V )
 
i,jD D U
2 i D
,
1 (κ) D,min
(35)
Θ (VD , 0U , α i,jD ) ≥ ri,j , ∀(i, jD ) ∈ S1 , (32d)
2 i,jD D

(κ) (κ)
where φi,jD (VD , 0U ) and Θi,jD (VD , 0U , α i,jD ) are defined by with the downlink interference covariance mapping Ψ̄(., .)
(κ)
(22) and (18) with both VU and VU replaced by 0U , while defined from (5).
(κ) (κ) The uplink throughput at the BS is
Θi (0D , VU ) is defined by (19) with both VD and VD
replaced by 0D .
Problems (23), (28) and (32) involve n = 2(N1 d1 ID +  
fi (VD , VU , α ) , ln IN2 + (Li (VUi ))2 sΨ−1 (V , V , α )
 
Nr d2 IU ) + ID scalar real decision variables and m = i D U ,
5ID+IU +2I +1 quadratic constraints so their computational (36)
complexity is O(n2 m2.5 + m3.5 ).

III. EH- ENABLED FD MU-MIMO BY TS where Li (VUi ) is already defined from (9) but
A much easier implementation is time splitting 0 < α < 1
in downlink transmission where (1−α α) time is used for DL en- Ψi (VD , VU , α ) , Ψ̄U T SI
i (VU ) + Ψ̄i (VD , α ) (37)
α
ergy transfer and time is used for DL information transmis-
I I E
sion. In this section, we define VD , [Vi,j ]
D (i,jD )∈S1
, VD ,
E
[Vi,jD ](i,jD )∈S1 and redefine the notation VD , [VD , VD I E
] with the uplink interference covariance mapping Ψ̄U
i (.) defined
I E by (11) and the time-splitting SI covariance mapping
where Vi,j D
and V i,jD are the information precoding matrix
for ID and energy precoding matrix for EH, respectively. The
received signal at DLU (i, jD ) for EH is X  
E
X
E
Ψ̄Ti SI (VD , α ) , σSI
2 E
(1 − α )||Vi,j ||2 + α ||Vi,j
I
||2 IN2 .
sE
D D
yi,j D
, Hm,i,jD Vm,`D m,`D jD ∈D
(m,`D )∈S1 (38)
X
+ Hi,jD ,`U Vi,`U si,`U +njD , (33)
`U ∈U
| {z } The problem of maximizing the network total throughput
UL intracell interference under throughput QoS and EH constraints is the following:
 7

all nonconvex and cannot be addressed as in (13). In the

following, we will develop the new minorants of the DL
max P2 (VD , VU , α ) , throughput function and UL throughput function.
α
VD ,VU ,α
X 
I
 Firstly, we address a lower approximation for each
α fi,jD (VD , VU ) + fi (VD , VU , α ) (39a) I
fi,jD (VD , VU )/ρρ in (41a) and (41f). Recalling the definition
(i,jD )∈S1 I
(35) of fi,jD (VD , VU ) we introduce
s.t. 0 < α < 1, (39b)
I
||V ||2 ≤ Pi,jU , ∀(i, jU ) ∈ S2 , (39c) Mi,jD (VD , VU ) , (Li,jD (Vi,jD ))2 + Ψ̄i,jD (VD , VU ),
X  i,jU  (κ) (κ)
E
(1 − α )||Vi,j ||2
+ α ||V I
||2 to have its following minorant at (VD , VU ):
D i,j D
n o
(i,jD )∈S1 (κ) I (κ) (κ) I
X Θi,jD (VD , VU ) ,ai,jD + 2< hAi,jD , Li,jD (Vi,j D
)i
+ ||Vi,jU ||2 ≤ P, (39d) (κ) I
(i,jU )∈S2 − hBi,jD , Mi,jD (VD , VU )i, (42)
X  
E
(1 − α )||Vi,j ||2 + α ||Vi,j
I
||2 where similarly to (20)
D D
jD ∈D (κ) (κ) (κ)
0 > ai,jD = fi,jD (VD , VU )
≤ Pi , ∀i ∈ I, (39e) n o
(κ) (κ) (κ) (κ)
fi (VD , VU , α ) ≥ riU,min , ∀i ∈ I, (39f) −< hΨ̄−1
i,jD (VD , VU )Li,jD (Vi,jD ), Li,jD (Vi,jD )i ,
I D,min (κ) (κ) (κ) (κ)
α fi,jD (VD , VU ) ≥ ri,j , ∀(i, jD ) ∈ S1 , (39g) Ai,jD = Ψ̄−1
D i,jD (VD , VU )Li,jD (Vi,jD ),
E min
Ei,jD (VD , VU , α ) ≥ ei,jD , ∀(i, jD ) ∈ S1 . (39h) (κ) (κ) (κ) (κ) (κ)
0  Bi,jD = Ψ̄−1 −1
i,jD (VD , VU ) − Mi,jD (VD , VU ).
Constraints (39c), (39d) and (39e) limits the transmit power (43)
I (κ) I
of each ULU, the whole network and each BS, respectively. A minorant of fi,jD (VD , VU )/ρρ is Θi,jD (VD , VU )/ρρ but it is
I
Constraints (39h) ensures that each DLUs harvest more than still not concave. As fi,jD (VD , VU ) > 0 it is obvious that its
(κ) I I
a threshold whereas constraints (39f) and (39g) guarantee lower bound Θi,jD (VD , VU ) is meaningful for (VD , VU ) such
the throughput QoS at BSs and DLUs, respectively. The that
(κ) I
key difficulty in problem (39) is to handle the time splitting Θi,jD (VD , VU ) ≥ 0, (i, jD ) ∈ S1 (44)
factor α that is coupled with the objective functions and other
which particularly implies
variables. Using the variable change ρ = 1/α α, which satisfies nD Eo
(κ) I
the convex constraint < (Ai,jD , Li,jD (Vi,j D
) ≥ 0, (i, jD ) ∈ S1 . (45)
ρ > 1, (40)
Under (45), we have
problem (39) is equivalent to n
(κ) I
o
< hAi,jD , Li,jD (Vi,jD
)i
P2 (VD , VU , ρ) ,
max ≥
ρ>0
VD ,VU ,ρ ρ
X X r n
I
, VU )/ρρ + fi (VD , VU , 1/ρρ)
o
fi,jD (VD (41a) (κ) (κ) I )i − c(κ) ρ
2bi,jD < hAi,jD , Li,jD (Vi,j (46)
D i,jD
(i,jD )∈S1 i∈I
s.t. (40), (39c), for
X  
E
||2 I
||2 /ρρ
q
||Vi,j D
+ ||Vi,j D
(κ) I,(κ)
hAi,jD , Li,jD (Vi,jD )i
(κ)
(i,jD )∈S1 0 < bi,jD = ,
X
2
X
E
ρ(κ)
+ ||Vi,jU || ≤ P + ||Vi,j ||2 /ρρ, (41b) (κ) (κ)
D
0 < ci,jD = (bi,jD )2 . (47)
(i,jU )∈S2 (i,jD )∈S1
X  Therefore, the following concave function

E
||Vi,j D
||2 + ||Vi,j
I
D
||2 /ρρ ≤
jD ∈D I (κ)
X gi,jD (VD , VU , ρ ) ,
E
Pi + ||Vi,j D
||2 /ρρ, ∀i ∈ I, (41c) (κ)
ai,jD
r n o
(κ) (κ) I )i − 2c(κ) ρ
jD ∈D + 4bi,jD < hAi,jD , Li,jD (Vi,j i,jD
E
ρ D

Ei,jD (VD , VU , 1/ρρ) ≥ emin

i,jD , ∀(i, jD ) ∈ S1 , (41d)
hB (κ) , Mi,jD (VD
I
, VU )i
fi (VD , VU , 1/ρρ) ≥ riU,min , ∀i ∈ I, (41e) − (48)
ρ
I D,min
fi,jD (VD , VU )/ρρ ≥ ri,j , ∀(i, jD ) ∈ S1 . (41f) I I,(κ) (κ)
D
is a minorant of fi,jD (VD , VU )/ρρ at (VD , VU , ρ(κ) ).
Problem (41) is much more difficult computationally than (13). Next, we address a lower approximation of
Firstly, the DL throughput is now the multiplication of data fi (VD , VU , 1/ρρ) in (41a), (41e). Recalling the definition (36)
throughput and the portion of time 1/ρρ. Secondly, the SI in of fi (VD , VU , 1/ρρ) we introduce
UL throughput is also coupled with 1/ρρ. Finally, the power Mi (VD , VU , ρ ) ,(Li (VUi ))2
constraints (41b), (41c) are also coupled with 1/ρρ. Therefore,
the objective function (41a) and constraints (41b)-(41f) are + Ψ̄U T SI
i (VU ) + Ψ̄i (VD , 1/ρρ), (49)
8

for Ψ̄Ti SI (VD , 1/ρρ) defined from (38) as Algorithm 2 Path-following algorithm for TS optimization
X  problem (41)
Ψ̄Ti SI (VD , 1/ρρ) =σSI
2 E
||Vi,j D
||2 Initialization: Set κ := 0, and choose a feasible point
(0) (0)
jD ∈D (VD , VU , α(0) ) that satisfies (39b)-(39g). Set ρ(0) :=
1 1 1/α(0) .

I 2 E 2
+ ||Vi,j || − ||V i,jD || IN2 , (50)
ρ D
ρ κ-th iteration: Solve (55) for an optimal solution
(κ) (κ)
(κ) (κ)
(VD∗ , VU∗ , ρ∗ ) and set κ := κ + 1, (VD , VU , ρ(κ) ) :=
to have its following minorant at (VD , VU , ρ(κ) ): ∗ ∗ ∗ (κ) (κ)
, VU , 1/ρ(κ) ). Stop if
(VD , VU(κ)
P2 (VD(κ−1)
, ρ ) and calculate
(κ−1)
(κ) (κ)
n
(κ)
o  P2 (x ) − P2 (x ) /P2 (x ) ≤ , where x(κ) ,
Θi (VD , VU , ρ ) ,ai + 2< hAi , Li (VUi )i (κ) (κ)
(VD , VU , 1/ρ(κ) )
(κ)
− hBi , Mi (VD , VU , ρ )i, (51)

where similarly to (21) (41) by successively solving its following innerly approxi-
(κ) (κ) (κ) mated convex program at κ-iteration:
0 > ai = fi (VD , VU )
(κ)
n
(κ) (κ) (κ) (κ)
o max P2 (VD , VU , ρ ) (55a)
−< hΨ−1
i (V D , VU )L i (Vi ), Li (V i )i , ρ>0
VD ,VU ,ρ

(κ) (κ) (κ) (κ)

s.t. (40), (39c), (55b)
Ai = Ψ−1
i (VD , VU )Li (VUi ), X  1 
||Vi,jD ||2 + ||Vi,j
E I
||2
+
0  Bi
(κ) (κ)
= Ψi−1 (VD , VU ) − M−1
(κ) (κ) (κ) ρ D
i (VD , VU ). (i,jD )∈S1
(52) X X (κ)
(κ)
Function Θi (VD , VU , ρ ) is not concave due to the term ||Vi,jU ||2 ≤ P + E
γi,jD (Vi,j D
, ρ ), (55c)
(i,jU )∈S2 (i,jD )∈S1
Ψ̄Ti SI (VD , 1/ρρ) defined by (50). However, the following ma-
trix inequality holds true:
X 
E 2 1 I 2

||Vi,j || + ||V i,jD || ≤
D
ρ
1 2 jD ∈D
E E,(κ)
||Vi,j E
||2 IN2 ( (κ) <{hVi,jD , Vi,j i} X (κ)
E
ρ D
ρ D
Pi + γi,jD (Vi,jD
, ρ ), ∀i ∈ I, (55d)
E,(κ) jD ∈D
||Vi,jD ||2
− ρ )IN2 , (53) (κ) E 1
(ρ(κ) )2 φi,jD (VD , VU ) ≥ emin
i,jD (1 + )/ζi,jD ,
ρ−1
which yields the matrix inequality ∀(i, jD ) ∈ S1 , (55e)
(κ)
Mi (VD , VU , ρ )  Θ̃i (VD , VU , ρ) ≥ riU,min , ∀i ∈ I, (55f)
(κ) (κ) I D,min
Mi (VD , VU , ρ ) , gi,jD (VD , VU , ρ) ≥ ri,j D
.∀(i, jD ) ∈ S1 . (55g)

1
(Li (VUi ))2 + Ψ̄U 2 E 2 I
i (VU ) + σSI ||Vi,jD || + ||Vi,jD ||
2 (κ) (κ)
P
where P2 (VD , VU , ρ ) , i∈I Θ̃i (VD , VU , ρ )
ρ (κ) I
P
E,(κ)
||Vi,jD ||2  + (i,jD )∈S1 gi,jD (VD , VU , ρ ).
2 E,(κ) E
− (κ) <{hVi,jD , Vi,j i} + ρ IN2 . A path-following procedure similar to Algorithm 1 can be
ρ D
(ρ(κ) )2 applied to solve (41) as summarized in Algorithm 2. Thanks
(κ) to the following relation, which is similar to (26):
As Bi  0 by (52), we then have
(κ+1) (κ+1) (κ) (κ)
(κ) (κ) (κ)
P2 (VD , VU , ρ(κ+1) ) ≥ P2 (VD , VU , ρ(κ) ), (56)
hBi , Mi (VD , VU , ρ )i ≥ hBi , Mi (VD , VU , ρ )i
Algorithm 2 improves feasible point at each iteration and then
so a concave minorant of both fi (VD , VU , 1/ρρ) and bring a local optimum after finitely many iterations.
Θ(κ) (VD , VU , ρ ) is To find an initial feasible point for Algorithm 2, we consider
n o the following problem:
(κ) (κ) (κ)
Θ̃i (VD , VU , ρ ) ,ai + 2< hAi , Li (VUi )i n
max min fi (VD , VU , 1/ρρ) − rimin ,
(κ) (κ) ρ (i,jD )∈S1
VD ,VU ,ρ
− hBi , Mi (VD , VU , ρ )i. (54) E
Ei,jD (VD ,oVU , 1/ρρ) − emin
i,jD ,
(57)
E
Concerned with ||Vi,j D
||2 /ρρ in the right hand side (RHS) of I
fi,jD (VD , VU )/ρρ − min
ri,i : (41b) − (41c)
D
(41b) and (41c), it follows from (53) that
which can be addressed by successively solving the following
E (κ)
||Vi,j D
||2 /ρρ ≥ E
γi,jD (Vi,j D
, ρ) convex maximin program:
E,(κ) E
, 2<{hVi,jD , Vi,j i}/ρ(κ)
n
(κ) I min
E,(κ) 2
D max min gi,jD (VD , VU , ρ ) − ri,i D
,
−ρρ||Vi,jD || /(ρ )2 .
(κ) ρ (i,jD )∈S1
VD ,VU ,ρ
(κ) E 1
φi,jD (VD , VU ) − emin
i,jD (1 + )/ζi,jD , (58)
We also have
(κ) E
φi,jD (VD , VU )
defined in (22) as a minorant o ρ − 1
(κ)
E
of hΦi,jD (VD , VU )i. We now address the nonconvex problem Θ̃i (VD , VU , ρ) − rimin : (55b) − (55d),
 9

I,(κ) (κ) (κ)

upon reaching fi,jD (VD , VU , α(κ) ) ≥ with both VD and VD replaced by 0D and both ρ and ρ(κ)
(κ) (κ)
min
ri,iD , fi (VD , VU , α(κ) ) ≥ rimin and replaced by 1.
E,(κ) (κ)
Ei,jD (VD , VU , 1/ρ(κ) ) ≥ emin i,jD , ∀(i, jD ) ∈ S1 .
For the system operating in HD mode, we apply the same IV. T HROUGHPUT Q O S CONSTRAINED
transmission strategy as in Section II. Specifically, we consider ENERGY- HARVESTING OPTIMIZATION
the following problem: We will justify numerically that TS is not only easier imple-
1 X 1 mented but performs better than PS for FD EH-enabled MU
max [ fi,jD (VD , 0U )
VD ,VU ,ρρ 2 ρ MIMO networks. This motivates us to consider the following
(i,jD )∈S1
X EH optimization with TS, which has not been previously
+ fi (0D , VU , 1)] s.t. (39b) − (39e) (59a) considered at all:
i∈I X
E
1 max P3 (V, α ) , Ei,jD (VD , VU , α )
(Ei,jD (VD , 0U , 1/ρρ) + Ei,jD (0D , VU , 1)) ≥, α
VD ,VU ,α
(i,jD )∈S1 (62)
2
emin s.t. (39b) − (39g).
i,jD (i, jD ) ∈ S1 , (59b)
1 D,min By defining ρ = 1/αα, we firstly recast Ei,jD (VD E
, VU , 1/ρρ) as
fi,j (VD , 0U ) ≥ ri,j , ∀(i, jD ) ∈ S1 , (59c)
2ρρ D D

E E
1 Ei,jD (VD , VU , 1/ρρ) =ζi,jD hΦi,jD (VD , VU )i
fi (0D , VU , 1) ≥ riU,min , ∀i ∈ I. (59d)
2 E

− Qi,jD (VD , VU , ρ ) , (63)
In (59), DLUs harvest energy for (1 − α ) of 1/2 time slot
during DL transmission and for the whole 1/2 time slot during E 1 E
UL transmission as formulated in (59b). The constraint (59b)
where Qi,jD (VD , VU , ρ ) , ρ hΦi,jD (VD , VU )i is a convex
(κ) E
can be written by function. Recalling that φi,jD (VD , VU ) defined in (22) is
E
a minorant of hΦi,jD (VD , VU )i, we can now address the
2emin
i,jD 1 nonconvex problem (62) by successively solving the following
Ξi,jD (VD , VU , ρ ) ≥ (1 + ), (60)
ζi,jD ρ−1 convex program at κ-iteration:
for X 
(κ) E
max ζi,jD φi,jD (VD , VU )
Ξi,jD (VD , VU , ρ ) , ρ
V,ρ
(i,jD )∈S1
hΦi,jD (0D , VU )i E
 (64)
hΦi,jD (VD , 0U )i + hΦi,jD (0D , VU )i + . −Qi,jD (VD , VU , ρ )
ρ−1
s.t. (39c), (40), (55c), (55d), (55f), (55g).
As Ξi,jD is convex, its minorant is its first-order approximation
(κ) (κ)
at (VD , VU , ρ(κ) ): A path-following procedure similar to Algorithm 2 can be
(κ) (κ) (κ)
applied to solve (62).
Ξi,jD (VD , VU , ρ ) = φi,jD (VD , 0U ) + φi,jD (0D , VU ) For the system operating in HD mode, the same trans-
(κ)
+Λi,jD (VU , ρ − 1), mission strategy as in Section II is applied. Specifically, we
(κ) (κ) consider the following problem:
for Λi,jD (., .) defined by (31) and φi,jD (0D , VU ) defined from
(κ)
X 1
(22) with both VU and VU replaced by 0U . E
max (Ei,jD (VD , 0U , 1/ρρ) + Ei,jD (0D , VU , 0))
The problem (39) thus can be addressed via a path-following
ρ
V,ρ 2
(i,jD )∈S1
procedure similar to Algorithm 2 where the following convex s.t. (39c), (40), (39d), (39e), (59c), (59d).
program is solved for κ-iteration: (65)
1 X The problem (65) can be addressed via a path-following
(κ) I
max [ gi,jD (VD , 0U , ρ ) procedure similar to Algorithm 2 where the following convex
ρ 2
VD ,VU ,ρ
(i,jD )∈S1 program is solved for κ-iteration:
X (κ)
+ Θ̃i (0D , VU , 1)] : (39b) − (39e) (61a) X ζi,j  (κ)
D E E
i∈I
max φi,jD (VD , 0U ) − Qi,jD (VD , 0U , ρ )
ρ
V,ρ 2
(i,jD )∈S1
(κ) 2emin
i,jD 1 (κ)

Ξi,jD (VD , VU , ρ ) ≥ (1 + ), +φi,jD (0D , VU )
ζi,jD ρ−1
(61b) s.t. (39c), (40), (39d), (39e), (61d), (61e),
(66)
(i, jD ) ∈ S1 , (61c) (κ) E (κ)
where φi,jD (VD , 0U ) (φi,jD (0D , VU ), resp.) is defined by (22)
1 (κ) D,min (κ) (κ)
g (VD , 0U , ρ ) ≥ ri,j , ∀(i, jD ) ∈ S1 , (61d) with both VU and VU (both VD and VD , resp.) replaced
2 i,jD D

1 (κ) by 0U (0D , resp.).

Θ̃ (0D , VU , 1) ≥ riU,min , ∀i ∈ I, (61e) Problems (55), (58), (61), (64) and (66) involve n =
2 i
(κ)
2(N1 d1 ID + Nr d2 IU ) + 3 scalar real decision variables
I
where gi,jD (VD , 0U , ρ) is defined by (48) with both VU and and m = ID + IU + 2I + 3 quadratic constraints so its
(κ) (κ)
VU replaced by 0U , while Θ̃i (0D , VU , 0) is defined by (54) computational complexity is O(n2 m2.5 + m3.5 ).
10

as N1 = N2 = 4. The numbers of concurrent downlink data

streams and the numbers of concurrent uplink data streams
r DLU (i,jD) are equal and d1 = d2 = Nr . To arrive at the final figures,
2
we run each simulation 100 times and average out the result.
In all simulations, we set P = 23 dBW, Pi = 16 dBW
BS i r1 r1 ∀i ∈ I, Pi,jU = 10 dBW ∀(i, jU ) ∈ S2 , ∀i ∈ I ,ζ = 0.5,
σc2 = −90 dBW, σ 2 = −90 dBW, ri,j min
D
= rD = 1 bps/Hz and
min U D
ULU (i,jU) ri = r = U r bps/Hz. We further assume that the required
harvested energies of all DLUs are the same and emin i,jD = e
min
,
min
∀(i, jD ). Unless stated otherwise, we set e = −20 dBm as
in [7], [33]. According to the current state-of-the-art-electronic
circuitry, the sensitivity level of a typical energy harvester is
around -20 dBm (0.01mW) [34], which means that we can
r1 activate the EH circuitry with that much amount of received
2
power. The SI level σSI is choosen within the range of
1 2
[−150, −90] dB as in [14], [16], [26] where σSI = −150
dB represents the almost perfect SI cancellation.

A. Single cell network

Fig. 2. A three-cell network with 3 DLUs and 3 ULUs. DLUs are randomly Firstly, we consider the sum throughput maximization prob-
located on the circles with radius of r1 centered at their serving BS. ULUs lem and the total harvested energy in the single cell networks.
are uniformly distributed within the cell of their serving BS. This will facilitate the analysis of the impact of SI to the
network performance since there is no intercell interference.
The network setting in Fig. 2 is used but only one cell is
V. N UMERICAL RESULTS considered.
In this simulation study, we use the example network in Fig. 3 illustrates the comparison of total network throughput
Fig. 2 to study the total network throughput in the presence between the power splitting mechanism and the time splitting
of SI. The HD system is also implemented as a base line for mechanism in both FD and HD systems. Though FD provides
both time splitting mechanism and power splitting mechanism. a substantial improvement in comparison to HD in both power
DLUs are randomly located on the circles with radius of splitting mechanism (25.8%) and time splitting mechanism
2
r1 = 20 m centered at their serving BSs whereas ULUs (26.1%) for Nr = 2 2 at σSI = −150 dB. Note that we
are uniformly distributed within the cell of their serving BSs cannot expect a FD system to achieve twice the throughput
whose radius are r2 = 40 m. There are two DLUs and of that is achieved by a HD system. This is because even
two ULUs within each cell. We set the path loss exponent when the SI cancellation is perfect, DLUs in FD are still
β = 4. For small-scale fading, we generate the channel vulnerable to the intracell interference from the ULUs of the
matrices Hm,i,jD from BS m to UE (i, jD ), matrices Hi,jD ,`U same cell. Moreover, DLUs and ULUs in HD are served with
from ULU (i, `U ) to DLU (i, jD ), matrices Hm,`U ,i from ULU more BS’s antennas, resulting in a larger spatial diversity.
B
(m, `U ) to BS i and matrices Hm,i from BS m to BS i using Consequently, FD cannot double HD’s throughput even with
the Rician fading model as follows: the almost perfect SI cancellation.
r
KR
r
1 When we reduce the number of antennas at UEs from
LOS
H= H + H N LOS , (67) Nr = 2 to Nr = 1, the total network throughput of FD is
1 + KR 1 + KR
significantly reduced by 42% for the time splitting mechanism
where KR = 10 dB is the Rician factor, H LOS is the line- and by 41% for the power splitting mechanism at σSI 2
= −150
of-sight (LOS) deterministic component and each element dB. Notably, since UEs in FD are exposed to more sources
of Rayleigh fading component H N LOS is the circularly- of interference than UEs in HD, reducing the number of
symmetric complex Gaussian random variable CN (0, 1). Here, antennas of UEs degrades the performance of FD more than
we use the far-field uniform linear antenna array model [32] the counterpart of HD. Consequently, the improvement of FD
2
with in comparison to HD reduces to 16% at σSI = −150 dB for
H LOS = [1, ejθr , ej2θr , . . . , ej(Nr −1)θr ] both time splitting mechanism and power splitting mechanism.
(68) Fig. 4 further illustrates how the total throughput is dis-
×[1, ejθt , ej2θt , . . . , ej(N1 −1)θt ]H ,
tributed into the downlink and uplink channels in the time
for θr = 2πd sin(φ
λ
r)
, θt = 2πd sin(φ
λ
t)
, where d = λ/2 is splitting mechanism. The behaviour of the power splitting
the antenna spacing, λ is the carrier wavelength and φr , φt mechanism is similar and omitted here for brevity. With the
is the angle-of-arrival, the angle-of-departure, respectively. In
1 At σ 2 = −90 dB, if a BS transmits at full power (i.e. 16 dBW), the SI
our simulations, φr and φt are randomly generated between SI
power is 16 dB stronger than the background AWGN.
0 and 2π. Unless stated otherwise, the number of transmit 2 N has been defined in the begining of section II as the number of antennas
r
antennas and the number of receive antennas at a BS are set of UEs (DLUs and ULUs).
 11

29 24
FD − TS FD − UL
28 FD − PS 22 FD − DL
HD − TS HD − UL
27 HD − PS 20 HD − DL
Sum throughput (bps/Hz)

Sum throughput (bps/Hz)

26
18
25
16
24
14
23
12
22

21 10

20 8

19 6
−150 −140 −130 −120 −110 −100 −90 −150 −140 −130 −120 −110 −100 −90
σ2SI (dB) σ2SI (dB)

(a) Nr = 1 (a) Nr = 1

50 32
FD − TS FD − UL
FD − PS 30 FD − DL
HD − TS HD − UL
HD − PS HD − DL
Sum throughput (bps/Hz)

Sum throughput (bps/Hz)

28
45
26

24

22
40
20

18

35 16
−150 −140 −130 −120 −110 −100 −90 −150 −140 −130 −120 −110 −100 −90
σ2SI (dB) σ2SI (dB)

(b) Nr = 2 (b) Nr = 2

Fig. 3. Effect of SI on the sum throughput performance in the single-cell Fig. 4. Effect of SI on the UL/DL throughput performance in the single-cell
networks. networks.

2
increase of σSI , the UL throughput consistently decreases. different precoder matrices VI and VE for data transmission
Moreover, since the UL transmission becomes less efficient, and energy transferring, the time splitting scheme can exploit
ULUs reduce their transmission power to reduce the inter- the spatial diversity better than the power splitting scheme
ference toward DLUs. Consequently, a slight increase in FD which only uses one type of precoder matrix for both purposes.
2
DL throughput is observed as σSI increases. Another note is Thus, the time splitting scheme is more efficient than the
that since the distance between ULU-DLU in small cell can power splitting scheme in term of performance.
be quite small due to the random deployment of ULUs and The comparison of maximum harvested energy of time
DLUs, DLUs’ throughput can be severely degraded by the splitting scheme in both FD and HD systems is studied in
interference from ULUs. In fact, the FD DL throughput is Fig. 6. Interestingly, in case of Nr = 1, FD roughly harvests
2
60% less than the counterpart of HD at Nr = 1, σSI = −150 as much as HD. The reason of this is two folds. Firstly, it has
dB. By implementing multiple antenna at UEs (i.e. Nr = 2), been reported in [16], [26], [35] that FD not always harness
DLUs in FD can handle the interference better and the FD performance gain over HD if the distance between ULUs and
2
DL throughput at σSI = −150 dB is only 10% less than the DLUs are not large enough. Since we consider small cell
counterpart of HD. networks with randomly deployed ULUs and DLUs, the ULU-
To analyze the effect of energy harvesting constraint, we fix DLU distance can be very small, which creates significant
2
Nr = 2, σSI = −110 dB and vary emin . Fig. 5 illustrates a interference to DLUs. Secondly, with Nr = 1, DLUs can not
consistent decreasing trend of all schemes as emin increases. exploit the spatial diversity to mitigate the interference from
The time splitting scheme outperforms the power splitting ULUs. Consequently, ULUs must reduce its transmit power
scheme in the considered range of emin for both FD and HD. to ensure the QoS at the DLUs, which lowers the amount of
A similar conclusion can be drawn from Fig. 3. By using two harvested energy at DLUs. In contrast, the results show that
12

45 10

Total harvested energy (dBm)

40
Sum throughput (bps/Hz)

6
FD − TS FD − TS, Nr=2
FD − PS
35 HD − TS 4 FD − TS, Nr=1
HD − PS HD − TS, Nr=2
2 HD − TS, Nr=1
30
0

25 −2

−4
−150 −140 −130 −120 −110 −100 −90 −80 −70
20
−20 −17.5 −15 −12.5 −10 σ2SI (dB)
emin (dBm)
Fig. 6. Effect of SI on the total harvested energy performance in the single-
(a) Pi = 30 dBm cell networks.

50
68
48
66
Sum throughput (bps/Hz)

46
Sum throughput (bps/Hz)
FD − TS 64
FD − PS
HD − TS 62
44
HD − PS
60
42
58 FD − TS
40 FD − PS
56 HD − TS
HD − PS
38
54

36 52
−20 −17.5 −15 −12.5 −10 −150 −140 −130 −120 −110 −100 −90
emin (dBm) σ2SI (dB)

(b) Pi = 46 dBm (a) Nr = 1

Fig. 5. Effect of energy harvesting constraints on the total harvested energy

performance in the single-cell networks. 110
FD − TS
FD − PS
HD − TS
105 HD − PS
2
Sum throughput (bps/Hz)

FD harvests more energy than HD given that σSI ≤ −90 dB

for Nr = 2. All this implies that having multiple antennas at
UEs is important to combat with extra interferences in FD. 100

95
B. Three-cell network
Now, we consider the sum throughput maximization prob-
90
lem and the total harvested energy in the three-cell networks
as depicted in Fig. 2. In this scenario, DLUs and BSs are
exposed to additional intercell interferences. According to Fig. 85
−150 −140 −130 −120 −110 −100 −90
7, FD now only provides a marginal improvement regarding σ2 (dB)
SI
HD in both power splitting scheme (11.7%) and time split-
2 (b) Nr = 2
ting scheme (11.8%) for Nr = 2, σSI = −150 dB. For
2
Nr = 1, σSI = −150 dB, the improvement is even lower with Fig. 7. Effect of SI on the sum throughput performance in the three-cell
4.1% in case of the power splitting scheme and 4.4% in case of networks.
time splitting scheme. Therefore, FD can give marginal gains
compared to HD in the multi-cell networks with high level of
interference. The effect of energy harvesting constraint to the network
 13

100 15

Total harvested energy per cell (dBm)

90
10
Sum throughput (bps/Hz)

80
5
70
FD − TS
FD − PS
60 HD − TS 0
HD − PS
FD − TS, Nr=2
50 FD − TS, Nr=1
−5
HD − TS, Nr=2
40 HD − TS, Nr=1
−10
−150 −140 −130 −120 −110 −100 −90 −80 −70
30
−20 −17.5 −15 −12.5 −10 σ2SI (dB)
emin (dBm)
Fig. 9. Effect of SI on the total harvested energy performance in the three-cell
(a) Pi = 30 dBm networks.

110

108
in FD to mitigate interference. Another note is that given
Nr = 2 the amount of energy harvested per cell in three-cell
106 2
networks (i.e. 10.09 dBm at σSI = −150 dB) is much higher
Sum throughput (bps/Hz)

104 than the harvested energy of single cell in Fig. 6 (i.e. 8.5 dBm
2
102
at σSI = −150 dB), thanks to the extra energy harvested from
FD − TS
FD − PS the intercell interference.
100 HD − TS
HD − PS
98
C. Convergence behaviour
96 Finally, the convergence behavior of the proposed Algo-
94 rithm 1 is illustrated in Fig. 10. For brevity, we only present the
2
case of the three-cell network at σSI = −110 dB and Nr = 2.
92
Fig. 10(a) plots the convergence of the objective functions
90 of the sum throughput maximization problem for the time
−20 −17.5 −15 −12.5 −10
emin (dBm) splitting scheme and the power splitting scheme, whereas Fig.
(b) Pi = 46 dBm
10(b) plots the convergence of the objective function of the EH
maximization problem. As can be seen, the sum throughput
Fig. 8. Effect of energy harvesting constraints on the total harvested energy maximization problem achieve 90% of its final optimal value
performance in the three-cell networks.
within 40 iterations whereas the EH maximization problem
needs 10 iterations. Table I shows the average number of iter-
ations required to solve each program. Note that each iteration
sum throughput is also investigated in Fig. 8 for the three- of the proposed algorithms invokes a convex subproblem to
2
cell networks with Nr = 2, σSI = −110 dB. As in Fig. 5, a generate a new feasible point (VD
(κ+1) (κ+1)
, VU , α(κ+1) ) that is
consistent decreasing trend of all schemes is observed as emin (κ) (κ)
better than the incumbent (VD , VU , α(κ) ). Such a convex
increases. Since DLUs can also harvest energy from the signals
subproblem can be solved efficiently by the available convex
arriving from other BSs in multicell networks, the FD network
solvers of polynomial complexity such as CVX [36]. To save
throughput only decreases by about 3% for both harvesting
the computational time, it is recommended to input the in-
scheme when emin increases from -20 dBm to -10 dBm. The (κ) (κ)
cumbent (VD , VU , α(κ) ) as the initial point for the process
counterpart throughput decrease in single-cell scenarios was
of solving this subproblem. Also, the high dimensionality
about 8%.
and the nonconvexity of the considered problems imply that
Fig. 9 also illustrates the comparison of total harvested
checking the global optimality of the computed solution is
energy per cell of the EH maximization problem in both FD
both theoretically and practically prohibitive. Nevertheless, our
and HD systems in three-cell network. For Nr = 1, FD even
2 recent results in [9] and [10] for the particular multi-input
harvests lesser amount of energy than HD given σSI > −150
single output (MISO) case of the HD optimization problem
dB due to the increasing level of interference when compared
(29) show that both Algorithm 1 and Algorithm 2 are capable
to a single-cell network. Similar to the single-cell network,
2 of delivering the global optimal solutions.
FD outperforms HD for σSI ≤ −90 dB if more antennas
are equipped at UEs (i.e. Nr = 2). This observation again
emphasizes the importance of having multiple antenna at UEs VI. C ONCLUSION
14

110 design in the presence of channel estimation errors.

105 R EFERENCES
100 [1] X. Lu, D. Niyato, P. Wang, and D. I. Kim, “Wireless charger networking
Sum throughput (bps/Hz)

for mobile devices: fundamentals, standards, and applications,” IEEE

95 Wireless Commun. Mag., vol. 22, no. 2, pp. 126–135, Apr. 2015.
[2] X. Lu, P. Wang, D. Niyato, D. I. Kim, and Z. Han, “Wireless charging
90 technologies: Fundamentals, standards, and network applications,” IEEE
Commun. Surveys Tutorials, vol. 18, no. 2, pp. 1413–1452, Second
85 Quarter 2016.
[3] S. Buzzi, C. L. I, T. E. Klein, H. V. Poor, C. Yang, and A. Zappone, “A
80 survey of energy-efficient techniques for 5G networks and challenges
FD−TS ahead,” IEEE J. Sel. Areas Commun., vol. 34, no. 4, pp. 697–709, Apr.
75 FD−PS 2016.
HD−TS [4] J. G. Andrews, S. Buzzi, W. Choi, S. V. Hanly, A. Lozano, A. C. K.
70
HD−PS Soong, and J. C. Zhang, “What will 5G be?” IEEE J. Sel. Areas
65 Commun., vol. 32, no. 6, pp. 1065–1082, Jun. 2014.
0 20 40 60 80 100 [5] Z. Ding, C. Zhong, D. W. K. Ng, M. Peng, H. A. Suraweera, R. Schober,
Number of iterations
and H. V. Poor, “Application of smart antenna technologies in simulta-
neous wireless information and power transfer,” IEEE Commun. Mag.,
(a) Sum throughput maximization vol. 53, no. 4, pp. 86–93, Apr. 2015.
[6] S. Timotheou, I. Krikidis, G. Zheng, and B. Ottersten, “Beamforming
6 for MISO interference channels with QoS and RF energy transfer,” IEEE
Trans. Wireless Commun., vol. 13, no. 5, pp. 2646–2658, May 2014.
FD [7] Q. Shi, W. Xu, T. H. Chang, Y. Wang, and E. Song, “Joint beamforming
5 HD and power splitting for MISO interference channel with SWIPT: An
Total harvested energy (mW)

SOCP relaxation and decentralized algorithm,” IEEE Trans. Signal

Process., vol. 62, no. 23, pp. 6194–6208, Dec. 2014.
4 [8] A. A. Nasir, D. T. Ngo, H. D. Tuan, and S. Durrani, “Iterative
optimization for max-min SINR in dense small-cell multiuser MISO
SWIPT system,” in Proc. IEEE Global Conf. on Signal Processing
3 (GlobalSIP), 2015, pp. 1392–1396.
[9] A. A. Nasir, H. D. Tuan, D. T. Ngo, S. Durrani, and D. I. Kim, “Path-
following algorithms for beamforming and signal splitting in RF energy
2 harvesting networks,” IEEE Commun. Letters, vol. 20, no. 8, pp. 1687–
1690, Aug 2016.
[10] A. A. Nasir, H. D. Tuan, D. T. Ngo, T. Q. Duong, and H. V. Poor,
1
“Beamforming design for wireless information and power transfer
systems: Receive power-splitting versus transmit time-switching,” IEEE
0 Trans. Commun., vol. 65, no. 2, pp. 876–889, Feb. 2017.
0 5 10 15 20 25 30 [11] J. Park and B. Clerckx, “Joint wireless information and energy transfer in
Number of iterations
a K-user MIMO interference channel,” IEEE Trans. Wireless Commun.,
vol. 13, no. 10, pp. 5781–5796, Oct. 2014.
(b) EH maximization [12] ——, “Joint wireless information and energy transfer with reduced
feedback in MIMO interference channels,” IEEE J. Sel. Areas Commun.,
Fig. 10. Convergence of the proposed algorithm for  = 10−4 . vol. 33, no. 8, pp. 1563–1577, Aug. 2015.
[13] R. Zhang and C. K. Ho, “MIMO broadcasting for simultaneous wire-
TABLE I less information and power transfer,” IEEE Trans. Wireless Commun.,
T HE AVERAGE NUMBER OF ITERATIONS REQUIRED BY THE PROPOSED vol. 12, no. 5, pp. 1989–2001, May 2013.
ALGORITHMS [14] D. Nguyen, L.-N. Tran, P. Pirinen, and M. Latva-aho, “Precoding for
full duplex multiuser MIMO systems: Spectral and energy efficiency
Throughput max., Throughput max., maximization,” IEEE Trans. Signal Process., vol. 61, no. 16, pp. 4038
Programs EH max. –4050, Aug. 2013.
TS PS
[15] Z. Zong, H. Feng, F. R. Yu, N. Zhao, T. Yang, and B. Hu, “Optimal
FD 74 65.4 24.1 transceiver design for SWIPT in K-user MIMO interference channels,”
HD 67.5 50.6 20.2 IEEE Trans. Wireless Commun., vol. 15, no. 1, pp. 430–445, Jan. 2016.
[16] H. H. M. Tam, H. D. Tuan, and D. T. Ngo, “Successive convex quadratic
programming for quality-of-service management in full-duplex MU-
MIMO multicell networks,” IEEE Trans. Commun., vol. 64, no. 6, pp.
We have proposed new optimal precoding designs for EH- 2340–2353, Jun. 2016.
enabled FD multicell MU-MIMO networks. Specifically, sum [17] E. Everett, A. Sahai, and A. Sabharwal, “Passive self-interference
suppression for full-duplex infrastructure nodes,” IEEE Trans. Wireless
throughput maximization under throughput QoS constraints Commun., vol. 13, no. 2, pp. 680–694, Feb. 2014.
and EH constraints for energy-constrained devices under either [18] M. Duarte, A. Sabharwal, V. Aggarwal, R. Jana, K. Ramakrishnan,
TS or PS has been considered. The FD EH maximization C. Rice, and N. Shankaranarayanan, “Design and characterization of a
full-duplex multiantenna system for WiFi networks,” IEEE Trans. Veh.
problem under throughput QoS constraints in TS has also Technol., vol. 63, no. 3, pp. 1160–1177, Mar. 2014.
been addressed. Toward this end, we have developed new path- [19] L. Anttila et al, “Modeling and efficient cancelation of nonlinear self-
following algorithms for their solution, which require a convex interference in MIMO full-duplex transceivers,” in Proc. IEEE Global
Conf. on Communication (Globecom), Dec. 2014, pp. 777–783.
quadratic program for each iteration and are guaranteed to [20] M. Heino et al, “Recent advances in antenna design and interference
monotonically converge at least to a local optimum. Finally, cancelllation algorithms for in-band full duplex relays,” IEEE Commun.
we have demonstrated the merits of our proposed algorithms Mag., vol. 53, no. 5, pp. 91–101, May 2015.
[21] M. Duarte, C. Dick, and A. Sabharwal, “Experiment-driven characteri-
through extensive simulations. Note that an interesting topic zation of full-duplex wireless systems,” IEEE Trans. Wireless Commun.,
for further research is this area is robust precoder/beamformer vol. 11, no. 12, pp. 4296–4307, Dec. 2012.
 15

[22] Y.-S. Choi and H. Shirani-Mehr, “Simultaneous transmission and re- Ali Arshad Nasir (S’09-M’13) is an Assistant Pro-
ception: Algorithm, design and system level performance,” IEEE Trans. fessor in the Department of Electrical Engineering,
Wireless Commun., vol. 12, no. 12, pp. 5992–6010, Dec. 2013. King Fahd University of Petroleum and Minerals
[23] A. Sabharwal et al, “In-band full-duplex wireless: challenges and oppor- (KFUPM), Dhahran, KSA. Previously, he held the
tunities,” IEEE J. Select. Areas. Commun., vol. 32, no. 9, pp. 1637–1652, position of Assistant Professor in the School of Elec-
Sep 2014. trical Engineering and Computer Science (SEECS)
[24] “System scenarios and technical requirements for full-duplex concept,” at National University of Sciences & Technology
DUPLO project, Deliverable D1.1. [Online]. Available: athttp: (NUST), Paksitan, from 2015-2016. He received his
//www.fp7-duplo.eu/index.php/deliverables. Ph.D. in telecommunications engineering from the
[25] S. Huberman and T. Le-Ngoc, “MIMO full-duplex precoding: A joint Australian National University (ANU), Australia in
beamforming and self-interference cancellation structure,” IEEE Trans. 2013 and worked there as a Research Fellow from
Wireless Commun., vol. 14, no. 4, pp. 2205–2217, Apr. 2015. 2012-2015. His research interests are in the area of signal processing in
[26] D. Nguyen, L.-N. Tran, P. Pirinen, and M. Latva-aho, “On the spectral wireless communication systems. He is an Associate Editor for IEEE Canadian
efficiency of full-duplex small cell wireless systems,” IEEE Trans. Journal of Electrical and Computer Engineering.
Wireless Commun., vol. 13, no. 9, pp. 4896–4910, Sep. 2014.
[27] H. H. Kha, H. D. Tuan, and H. H. Nguyen, “Fast global optimal power
allocation in wireless networks by local d.c. programming,” IEEE Trans.
Wireless Commun., vol. 11, no. 2, pp. 510–515, Feb. 2012.
[28] H. Tuy, Convex Analysis and Global Optimization. Kluwer Academic,
1998.
[29] B. Day, A. R. Margetts, D. Bliss, and P. Schniter, “Full-duplex bidi-
Trung Q. Duong (S’05, M’12, SM’13) received his
rectional MIMO: Achievable rates under limited dynamic range,” IEEE
Ph.D. degree in Telecommunications Systems from
Trans. Signal Process., vol. 60, no. 7, pp. 3702–3713, Jul. 2012.
Blekinge Institute of Technology (BTH), Sweden in
[30] D. Tse and P. Viswanath, Fundamentals of Wireless Communication.
2012. Since 2013, he has joined Queen’s Univer-
New York, NY, USA: Cambridge University Press, 2005.
sity Belfast, UK as a Lecturer (Assistant Profes-
[31] D. Katselis, E. Kofidis, and S. Theodoridis, “On training optimization
sor). His current research interests include small-cell
for estimation of correlated MIMO channels in the presence of multiuser
networks, physical layer security, energy-harvesting
interference,” IEEE Trans. Signal Process., vol. 56, no. 10, pp. 4892–
communications, cognitive relay networks. He is
4904, Oct. 2008.
the author or co-author of 240 technical papers
[32] Y. S. Cho, J. Kim, W. Y. Yang, and C. G. Kang, MIMO- OFDM Wireless
published in scientific journals (125 articles) and
Commun. with MATLAB. John Wiley & Sons, Ltd, 2010.
presented at international conferences (115 papers).
[33] Q. Shi, W. Xu, J. Wu, E. Song, and Y. Wang, “Secure beamforming for
Dr. Duong currently serves as an Editor for the IEEE T RANSACTIONS
mimo broadcasting with wireless information and power transfer,” IEEE
ON W IRELESS C OMMUNICATIONS , IEEE T RANSACTIONS ON C OMMUNI -
Trans. Wireless Commun., vol. 14, no. 5, pp. 2841–2853, May 2015.
CATIONS , IET C OMMUNICATIONS , and a Senior Editor for IEEE C OMMU -
[34] ——, “Secure beamforming for mimo broadcasting with wireless infor-
NICATIONS L ETTERS. He was awarded the Best Paper Award at the IEEE
mation and power transfer,” IEEE Trans. Wireless Commun., vol. 14,
Vehicular Technology Conference (VTC-Spring) in 2013, IEEE International
no. 5, pp. 2841–2853, May 2015.
Conference on Communications (ICC) 2014, and IEEE Global Communi-
[35] M. Mohammadi, H. A. Suraweera, Y. Cao, I. Krikidis, and C. Tel-
cations Conference (GLOBECOM) 2016. He is the recipient of prestigious
lambura, “Full-duplex radio for uplink/downlink wireless access with
Royal Academy of Engineering Research Fellowship (2016-2021).
spatially random nodes,” IEEE Trans. Commun., vol. 63, no. 12, pp.
5250–5266, Dec. 2015.
[36] M. Grant and S. Boyd, “CVX: Matlab software for disciplined convex
programming, version 2.1,” http://cvxr.com/cvx, Mar. 2014.

H. Vincent Poor (S72, M77, SM82, F87) received

the Ph.D. degree in EECS from Princeton University
in 1977. From 1977 until 1990, he was on the faculty
Ho Huu Minh Tam was born in Ho Chi Minh City, of the University of Illinois at Urbana-Champaign.
Vietnam. He received the B.S. degree in electrical Since 1990 he has been on the faculty at Princeton,
engineering and telecommunications from the Ho where he is currently the Michael Henry Strater
Chi Minh City University of Technology in 2012, University Professor of Electrical Engineering. Dur-
and he is currently pursuing the Ph.D. degree with ing 2006 to 2016, he served as Dean of Princetons
the University of Technology Sydney, NSW, Aus- School of Engineering and Applied Science. His
tralia, under the supervision of Prof. H. D. Tuan. research interests are in the areas of information
His research interest is in optimization techniques theory, statistical signal processing and stochastic
in signal processing for wireless communications. analysis, and their applications in wireless networks and related fields such as
smart grid and social networks. Among his publications in these areas is the
book Mechanisms and Games for Dynamic Spectrum Allocation (Cambridge
University Press, 2014).
Dr. Poor is a member of the National Academy of Engineering, the National
Hoang Duong Tuan received the Diploma (Hons.) Academy of Sciences, and is a foreign member of the Royal Society. He is
and Ph.D. degrees in applied mathematics from also a fellow of the American Academy of Arts and Sciences, the National
Odessa State University, Ukraine, in 1987 and 1991, Academy of Inventors, and other national and international academies. He
respectively. He spent nine academic years in Japan received the Marconi and Armstrong Awards of the IEEE Communications
as an Assistant Professor in the Department of Society in 2007 and 2009, respectively. Recent recognition of his work
Electronic-Mechanical Engineering, Nagoya Univer- includes the 2016 John Fritz Medal, the 2017 IEEE Alexander Graham
sity, from 1994 to 1999, and then as an Associate Bell Medal, Honorary Professorships at Peking University and Tsinghua
Professor in the Department of Electrical and Com- University, both conferred in 2016, and a D.Sc. honoris causa from Syracuse
puter Engineering, Toyota Technological Institute, University awarded in 2017.
Nagoya, from 1999 to 2003. He was a Profes-
sor with the School of Electrical Engineering and
Telecommunications, University of New South Wales, from 2003 to 2011.
He is currently a Professor with the Faculty of Engineering and Information
Technology, University of Technology Sydney. He has been involved in
research with the areas of optimization, control, signal processing, wireless
communication, and biomedical engineering for more than 20 years.