You are on page 1of 30

1

Joint Wireless Information and Energy Transfer


in a K-User MIMO Interference Channel
Jaehyun Park, Member, IEEE, Bruno Clerckx, Member, IEEE,
Abstract
Recently, joint wireless information and energy transfer (JWIET) methods have been proposed
to relieve the battery limitation of mobile nodes. However, the JWIET in a general K-user MIMO
interference channel (IFC) has been unexplored so far. In this paper, we rst investigate the JWIET
in K-user MIMO IFC, in which receivers either decode the incoming information data (information
decoding, ID) or harvest the RF energy (energy harvesting, EH). In the K-user IFC, we consider three
different scenarios according to the receiver mode i) multiple EH receivers and a single ID receiver,
ii) multiple IDs and a single EH, and iii) multiple IDs and multiple EHs. For all scenarios, we have
found a common necessary condition of the optimal transmission strategy and, accordingly, developed
the transmission strategy that satises the common necessary condition, in which all the transmitters
transferring energy exploit a rank-one energy beamforming. Furthermore, we have also proposed an
iterative algorithm to optimize the covariance matrices of the transmitters that transfer information and
the powers of the energy beamforming transmitters simultaneously, and identied the corresponding
achievable rate-energy tradeoff region. Finally, we have shown that by selecting EH receivers according
to their signal-to-leakage-and-harvesting energy-ratio (SLER), we can improve the achievable rate-energy
region further.
Index Terms
Joint wireless information and energy transfer, K-user MIMO interference channel, Rank-one beam-
forming
I. INTRODUCTION
One of the main challenges in modern wireless communication system is that wireless devices are
resource-constrained, mainly due to battery limitation. As smart phones which support various heavy-
J. Park and B. Clerckx are with the Department of Electrical and Electronic Engineering, Imperial College London, South
Kensington Campus London SW7 2AZ, United Kingdom (e-mail:{j.park, b.clerckx}@imperial.ac.uk)
October 28, 2013 DRAFT
a
r
X
i
v
:
1
3
1
0
.
6
8
7
0
v
1


[
c
s
.
I
T
]


2
5

O
c
t

2
0
1
3
2
battery-consuming applications are widely distributed, the battery limitation issue is also importantly
discussed even in the 4th generation (4G) and beyond 4G cellular standards organized by 3GPP, con-
sidering device-to-device communication [1]. During the last decade, there has been a lot of interest to
transfer energy wirelessly and recently, radio-frequency (RF) radiation has become a viable source for
energy harvesting. It is nowadays possible to transfer the energy wirelessly with a reasonable efciency
over small distances and, furthermore, wireless sensor networks (WSNs) in which the sensors are capable
of harvesting RF energy to power their own transmissions have been introduced in industry ( [2][5] and
references therein).
Until now, wireless energy transfer has been developed independently from the wireless information
transfer. Interestingly, because RF signals carry information as well as energy, joint wireless information
and energy transfer (JWIET) has attracted attention very recently [6][14]. Previous works have studied
the fundamental performance limits and the optimal transmission strategies of the JWIET in various
communication scenarios such as the cellular downlink system with a single base station (BS) and
multiple mobile stations (MSs) [8] and in the cooperative relay system [13] and in the broadcasting
system [6], [7] with a single energy receiver and a single information receiver when they are separately
located or co-located. Recently, considering multi-user MISO scenario, several transmission strategies and
power allocation methods have been proposed [15][17]. Furthermore, there have been several studies of
JWIET on the interference channel (IFC) models [18][22]. In [18], [19], the optimal power scheduling
at the energy harvesting transmitters are proposed for two-user single-input single-output (SISO) IFC
such that the sum-rate is maximized for given harvested energy constraints. In [20], JWIET in multi-
cell cellular networks is investigated, where all the BSs and MSs have a single antenna. In [21], by
considering two-user single-input multiple-output (SIMO) IFC, the system throughput is maximized
subject to individual energy harvesting constraints and power constraints and extended it to K-user MISO
IFC. Note that because the interference has different impacts on the performances of information decoding
(ID) (negatively) and energy harvesting (EH) (positively) at the receivers, the transmission strategy for
JWIET is a critical issue especially in IFC. To the best of the authors knowledge, JWIET in the general
K-user MIMO IFC (which describes modern advanced communication systems) has not been addressed
so far. Recently, in [22], a JWIET in a two-user MIMO IFC has been studied and a necessary condition
of the optimal transmission strategy for two-user MIMO IFC has been derived. That is, in a two-user
MIMO IFC, the energy transmitter may create a rank-one beam with the aim to either maximize the
energy harvested at the EH receiver or minimize the interference at the ID receiver. Alternatively, it
may generate multi-rank beams allocating its power on both directions. However, in [22], it is proved
October 28, 2013 DRAFT
3
that to achieve the optimal rate-energy (R-E) performance, the energy transmitter should take a rank-one
beamforming strategy with a proper power control.
In this paper, we extend the results obtained in [22] and investigate JWIET in a K-user MIMO IFC, in
which multiple MIMO transceiver pairs coexist and each receiver either decodes the incoming information
data or harvests the RF energy. Throughout the paper, it is assumed that the receivers cannot perform ID
and EH operations simultaneously, because existing circuit systems that harvest energy from the received
RF signal are not yet able to decode the information carried through the same RF signal [6], [7], [23]. In
[6], considering this practical issue, two different JWIET methods for MIMO broadcasting system - time
switching and power splitting methods - have been proposed. In the time switching method, the receiver
switches between ID mode and EH mode over time, while in the power splitting method, the received
signal is split into two signals with different power that are the input of disjoint ID and EH circuits,
respectively. Because the power splitting method requires higher hardware complexity at the receivers
(e.g., RF signal splitter), in this paper, we consider that each receiver switches between ID and EH modes
in time-basis. Accordingly, we have three different scenarios according to the receiver mode i) multiple
EH receivers and a single ID receiver, ii) multiple IDs and a single EH, and iii) multiple IDs and multiple
EHs. For all scenarios, the optimal achievable R-E trade-off region is not easily identied and the optimal
transmission strategy is still unknown. However, in this paper, we have shown that the optimal energy
transmitters strategies for all three scenarios also become optimal for the properly-transformed two-user
MIMO IFC. Therefore, we have found a common necessary condition of the optimal transmission strategy
and developed the transmission strategy that satises the common necessary condition, in which all the
transmitters transferring energy exploit a rank-one energy beamforming. Here, we have modied three
different rank-one beamforming schemes, originally developed for two-user MIMO IFC [22] - maximum
energy beamforming (MEB), minimum leakage beamforming (MLB), and signal-to-leakage-and-energy
ratio (SLER) maximization beamforming, suitable to K-user MIMO IFC. Given the rank-one beamforming
at the energy transmitters, we have formulated the optimization problem for the achievable rate-energy
region. However, because it is non-convex, we have proposed an iterative algorithm to optimize the
covariance matrices of the transmitters that transfer information and the powers of the energy beamforming
transmitters, simultaneously. We have shown that the powers of the energy beamforming transmitters
converges monotonically, which guarantees the convergence of the proposed algorithm. In addition, when
the number of energy transmitters increases, the ID receivers are affected by an increasing number of
interfering beams (directions and power) that affect their information rate performance. This leads us to
develop a new SLER maximizing beamforming with beam tilting. Here, the beam tilting means that we
October 28, 2013 DRAFT
4
Transmitter 1
Transmitter K-1
Receiver 1(EH)
Receiver K-1 (ID)
Energy flow
Interference
Transmitter 2 Receiver 2(EH)
Transmitter K Receiver K (ID)
Information flow
Fig. 1. K-user MIMO IFC.
change the direction of an energy beam without changing its transmit power. Finally, we have proposed
an efcient SLER-based EH transceiver selection method that further improves the achievable R-E region.
The rest of this paper is organized as follows. In Section II, we introduce the system model for K-user
MIMO IFC. In Section III, we discuss the necessary condition for the optimal transmission strategies in
K-user MIMO IFC. In Section IV, we investigate the achievable R-E region for K-user MIMO IFC and,
after formulating the optimization problem, propose an iterative algorithm to solve it. In Section V, we
provide several simulation results and in Section VI we give our conclusions.
Throughout the paper, matrices and vectors are represented by bold capital letters and bold lower-
case letters, respectively. The notations (A)
H
, (A)

, (A)
i
, [A]
i
, tr(A), det(A), and
k
(A) denote the
conjugate transpose, pseudo-inverse, the ith row, the ith column, the trace, the determinant, and the kth
largest singular value of a matrix A, respectively. The matrix norm |A| and |A|
F
denote the 2-norm
and Frobenius norm of a matrix A, respectively, and the vector norm |a| denotes the 2-norm of a vector
a. In addition, (a)
+
max(a, 0) and A _ 0 means that a matrix A is positive semi-denite. The matrix
diagA
1
, ..., A
M
is a block diagonal matrix with block diagonal elements A
m
. Finally, I
M
denotes the
M M identity matrix.
II. SYSTEM MODEL
We consider a K-user MIMO IFC system where K transmitters, each with M
t
antennas, are simul-
taneously transmitting their signals to K receivers, each with M
r
antennas, as shown in Fig. 1. Note
that each receiver can either decode the information or harvest energy from the received signal, but it
cannot execute the ID and EH operations at the same time due to the hardware limitations. That is, each
October 28, 2013 DRAFT
5
receiver can switch between ID mode and EH mode at each frame or time slot. Here, the mode decided
by the receiver is also sent to all the transmitters through the zero-delay and error-free feedback link
at the beginning of the frame. It is assumed that the transmitters have perfect knowledge of the CSI of
their associated links (i.e. the links between a transmitter and all receivers) but do not share those CSI
among the transmitters. Furthermore, M
t
= M
r
= M for simplicity, but it can be extended to general
antenna congurations. Assuming a frequency at fading channel, which is static over several frames,
the received signal y
i
C
M1
for i = 1, ..., K can be written as
y
i
=
K

j=1
H
ij
x
j
+n
i
,
where n
i
C
M1
is a complex white Gaussian noise vector with a covariance matrix
2
n
I
M
and
H
ij
C
MM
is the normalized frequency-at fading channel from the jth transmitter to the ith receiver
such as

M
l,k=1
[h
(l,k)
ij
[
2
=
ij
M [24]. Here, h
(l,k)
ij
is the (l, k)th element of H
ij
and
ij
[0, 1]. We
assume that H
ij
has a full rank. The vector x
j
C
M1
is the transmit signal, in which the independent
messages can be conveyed, at the jth transmitter with a transmit power constraint for j = 1, ..., K as
E[|x
j
|
2
] P for j = 1, ..., K. (1)
In this paper, the SNR measured at the ith receiver is dened as SNR
i
=
E[Hii
2
F
x
2
]
E[n
2
]
=
iiP

2
n
.
Throughout the paper, to ease readability, it is assumed without loss of generality that
2
n
= 1, unless
otherwise stated. General environments, characterized by other values of the channel/noise power, can
be described simply by adjusting P.
When the receiver operates in ID mode, the achievable rate at ith receiver, R
i
, is given by [25]
R
i
= log det(I
M
+H
H
ii
R
1
i
H
ii
Q
i
), (2)
where R
i
indicates the covariance matrix of noise and interference at the ith receiver, i.e.,
R
i
= I
M
+
K

j=i
H
ij
Q
j
H
H
ij
.
Here, Q
j
= E[x
j
x
H
j
] denotes the covariance matrix of the transmit signal at the jth transmitter and,
from (1), tr(Q
j
) P.
For EH mode, it can be assumed that the total harvested power E
i
at the ith receiver (more exactly,
harvested energy normalized by the baseband symbol period) is given by
E
i
=
i
E[|y
i
|
2
]
=
i
tr
_
_
K

j=1
H
ij
Q
j
H
H
ij
+I
M
_
_
, (3)
October 28, 2013 DRAFT
6
where
i
denotes the efciency constant for converting the harvested energy to electrical energy to be
stored [4], [6]. For simplicity, it is assumed that
i
= 1 and the noise power is negligible compared to
the transferred energy from each transmitters.
1
That is,
E
i
tr
_
_
K

j=1
H
ij
Q
j
H
H
ij
_
_
=
K

j=1
tr
_
H
ij
Q
j
H
H
ij
_
, (4)
where E
ij
= tr
_
H
ij
Q
j
H
H
ij
_
denoting the energy transferred from the jth transmitter to the ith receiver.
Note that, when the receiver decodes the information data from the associated transmitter under the
assumption that the signals from the other transmitters are not decodable [26], the signals from the other
transmitters become an interference to be defeated. In contrast, when the receiver harvests the energy,
they become a useful energy-transferring source. In Fig. 1, the interference denoted by the dashed red
line should be reduced for IDs, while the interference by the dashed green line be maximized for EHs.
III. A NECESSARY CONDITION FOR THE OPTIMAL TRANSMISSION STRATEGY
In [22], a necessary condition of the optimal transmission strategy has been found for the two-user
MIMO IFC with one EH and one ID, in which the energy transmitter should take a rank-one energy
beamforming strategy with a proper power control. In this section, we rst review one EH and one ID in
a two-user MIMO IFC, briey. Then, we will look into the cases of one ID/EH and multiple EHs/IDs.
Finally, we consider the case of multiple IDs and multiple EHs.
A. One ID receiver and One EH receiver
In this subsection, without loss of generality, we consider the transceiver pair (Tx
1
, Rx
1
) operates
in EH mode, while (Tx
2
, Rx
2
) in ID mode. Because information decoding is done only at the second
receiver, by letting R = R
2
and E = E
1
= E
11
+E
12
, we can dene the achievable rate-energy region
as:
C
RE
(P)
_
(R, E) : R log det(I
M
+H
H
22
R
1
2
H
22
Q
2
),
E

2
j=1
tr(H
1j
Q
j
H
H
1j
), tr(Q
j
)P, Q
j
_0, j =1,2
_
. (5)
1
In this paper, we assume the system operates in the high signal-to-noise ratio (SNR) regime, which is also consistent with
a practical wireless energy transfer that requires a high-power transmission.
October 28, 2013 DRAFT
7
The following proposition tells about the rank-one optimality in two-user MIMO IFC.
Proposition 1: In the high SNR regime, the optimal Q
1
at the boundary of the achievable rate-energy
region has a rank one at most. That is, rank(Q
1
) 1.
Proof: Detailed proof is given in [22], but here is its brief sketch. If the energy

E at the boundary
point of achievable rate-energy is small enough such that

E tr(H
12
Q
2
H
H
12
), then rank(Q
1
) = 0 (i.e.,
the rst transmitter does not need to transmit any signals causing the interference to the ID receiver).
If

E > tr(H
12
Q
2
H
H
12
), we then assume m = rank(Q
1
) 1. Based on the generalized singular
value decomposition (GSVD) of (H
11
, H
21
) and the interlacing theorem (Theorem 3.1 in [27]), with Q
1
satisfying the required harvesting energy, the achievable rate

R at high SNR regime can be approximated
as [22]

R f(H
22
) log
_
m

i=1
(1 +
2
x,i
)
_
, (6)
where
x,i
is the singular values of an arbitrary matrix X with Q
1
= TXX
H
T
H
. Here, T is an invertible
matrix obtained from GSVD of (H
11
, H
21
) and
x,1
, ...,
x,m
0 such that

m
i=1

2
x,i
=

E
11
with
1
, ...,
m
0 and a xed constant

E
11
. Then, we can easily nd that

R is maximized when
m = rank(Q
1
) = 1.
From Proposition 1, when transferring the energy in the IFC, the transmitters optimal strategy is
either a rank-one beamforming or no transmission according to the energy transferred from the other
transmitter, which increases the harvested energy at the corresponding EH receiver and simultaneously
reduces the interference at the other ID receiver. Intuitively, from the power transfer point of view, Q
1
should be as close to the dominant eigenvector of H
H
11
H
11
as possible, which implies that the rank one
is optimal for power transfer. From the information transfer point of view, when SNR goes to innity,
the rate maximization is equivalent to the DOF maximization. That is, a larger rank for Q
1
means that
more dimensions at the second receiver will be interfered. Therefore, a rank one for Q
1
is optimal for
both information and power transfer. Note that Proposition 1 is based on the high SNR regime, but the
rank-one optimality is also valid in the low SNR regime as discussed in Section VI.A of [22].
B. One ID and multiple EHs in a K-user IFC
Without loss of generality, the transceiver pairs (Tx
k
, Rx
k
), k = 1, ..., K 1 operate in EH mode,
while (Tx
K
, Rx
K
) in ID mode.
Because information decoding is done only at Rx
K
, by letting R = R
K
and E =

K1
i=1
E
i
with
October 28, 2013 DRAFT
8
E
i
=

K
j=1
E
ij
, we can dene the achievable rate-energy region as:
C
RE
(P)
_
(R, E) : R log det(I
M
+H
H
KK
R
1
K
H
KK
Q
K
),
E

K1
i=1

K
j=1
tr(H
ij
Q
j
H
H
ij
), tr(Q
j
)P, Q
j
_0, j =1, ...,K
_
. (7)
The following lemma gives an important insight into the derivation of the optimal boundary for one
EH and multiple IDs.
Lemma 1: Let f(X) as
f(X) = log det(I
M
+A
H
(I
M
+X)
1
A) (8)
for any positive matrix X. Then, f(X) is monotonically decreasing.
Proof: Because of Sylvesters determinant theorem [28] (det(I + AB) = det(I + BA)), we can
have
f(X) = log det(I
M
+AA
H
(I
M
+X)
1
)
= log det((I
M
+X)(I
M
+X)
1
+AA
H
(I
M
+X)
1
)
= log det(I
M
+AA
H
+X) log det(I
M
+X). (9)
Then, the partial derivative of f(X) with respect to X is given by
f(X) = (I
M
+AA
H
+X)
1
(I
M
+X)
1
_ 0. (10)
Therefore, f(X) is monotonically decreasing.
Then, we have the following proposition.
Proposition 2: All the optimal Q
k
at the boundary of the achievable rate-energy region for (7) become
optimal solutions for the boundary of
C
RE,k
(P)
_
(R, E) : R log det(I
M
+

H
H
22
(R
(k)
2
)
1

H
22

Q
2
),
Etr(

H
(k)
11
Q
k
(

H
(k)
11
)
H
) +tr(

H
12
Q
K

H
H
12
), tr(Q
j
)P, Q
j
_0, j =k, K
_
(11)
for all k = 1, ..., K 1, where

H
(k)
11
=
_

_
H
1k
.
.
.
H
K1k
_

_
,

H
12
=
_

_
H
1K
.
.
.
H
K1K
_

_
C
(K1)MM
, (12)
October 28, 2013 DRAFT
9

H
(k)
21
= H
Kk
,

H
22
= H
KK
, (13)
and

Q
2
= Q
K
, R
(k)
2
= I
M
+

H
(k)
21
Q
k
(

H
(k)
21
)
H
. (14)
Proof: Let us consider the boundary point (

R,

E) of the achievable rate-energy for (7) with any

Q
K
,
the covariance matrix of information transmitter on the boundary point. In addition, the corresponding
covariance matrices of energy transmitters at the boundary point are denoted as

Q
j
, j = 1, ..., K 1.
Note that, at the boundary point, the perturbation on

Q
j
for any j 1, ..., K does not increase both

R
and

E values, simultaneously. That is, the increase of

R (

E) due to the variation on

Q
j
always induces
the decrease of

E (

R) at the boundary.
From (7), for arbitrary Q
j
, j = 1, ..., K exhibiting (R, E) C
RE
(P),
K1

i=1
K

j=1
tr(H
ij
Q
j
H
H
ij
) =
K1

j=1
K1

i=1
tr(H
ij
Q
j
H
H
ij
) +
K1

i=1
tr(H
iK
Q
K
H
H
iK
)
=
K1

j=1
tr(

H
(j)
11
Q
j
(

H
(j)
11
)
H
) +tr(

H
12
Q
K

H
H
12
),
= tr(

H
(k)
11
Q
k
(

H
(k)
11
)
H
) +
K1

j=1,j=k
tr(

H
(j)
11
Q
j
(

H
(j)
11
)
H
) +tr(

H
12
Q
K

H
H
12
). (15)
We can dene functions f(Q
k
) and g(Q
k
) with respect to the covariance matrix of the kth transmitter
as
f(Q
k
) = log det(I
M
+H
H
KK
(I
M
+

j=K,k
H
Kj

Q
j
H
H
Kj
+H
Kk
Q
k
H
H
Kk
)
1
H
KK

Q
K
), (16)
and
g(Q
k
) = tr(

H
(k)
11
Q
k
(

H
(k)
11
)
H
) +
K1

j=1,j=k
tr(

H
(j)
11

Q
j
(

H
(j)
11
)
H
) +tr(

H
12

Q
K

H
H
12
). (17)
Then, the boundary point (

R,

E) at any boundary can be rewritten as
(

R,

E) = (f(

Q
k
), g(

Q
k
)). (18)
Note that

j=K,k
H
Kj

Q
j
H
H
Kj
is positive semi-denite and, from Lemma 1, f(Q
k
) is monotonically
decreasing with respect to Q
k
. Now we dene
f

(Q
k
) = log det(I
M
+H
H
KK
(I
M
+H
Kk
Q
k
H
H
Kk
)
1
H
KK

Q
K
),
= log det(I
M
+

H
H
22
(I
M
+

H
(k)
21
Q
k
(

H
(k)
21
)
H
)
1

H
22

Q
K
), (19)
October 28, 2013 DRAFT
10
Transmitter 1
Transmitter K
Receiver 1(EH)
Receiver K (ID)
Energy flow
Information flow
Interference
Transmitter 2 Receiver 2(EH)
Transmitter K
Receiver 1(EH)
Receiver K (ID)
Energy flow
Information flow
Interference
Transmitter k
Receiver K-1(EH)

21
k
H

11
k
H
12
H
22
H


1
11 11
1,
K
j j H
j
j j k
tr

H Q H
R
E
Fig. 2. One ID and multiple EHs in a K-user IFC.
which is also monotonically decreasing with respect to Q
k
. Accordingly, to satisfy f

k
) f

Q
k
)
with

Q
k
of the boundary point (f(

Q
k
), g(

Q
k
)),

Q

k
should be expressed as

k
=

Q
k
Q (20)
with > 0 and Q _ 0. However, we can easily nd that g(

Q
k
) > g(

k
). That is, the variation on

Q
k
that increases f

Q
k
) always incurs a loss in g(

Q
k
). Therefore,

Q
k
at any boundary become solutions
for the boundary of
C

RE,k
(P)
_
(R, E) : R log det(I
M
+

H
H
22
(R
(k)
2
)
1

H
22

Q
2
),
E

K1
i=1

K
j=1
tr(H
ij
Q
j
H
H
ij
), tr(Q
j
)P, Q
j
_0, j =1, ...,K
_
. (21)
Furthermore, in (21),
K1

i=1
K

j=1
tr(H
ij
Q
j
H
H
ij
) = tr(

H
(k)
11
Q
k
(

H
(k)
11
)
H
) +
K1

j=1,j=k
tr(

H
(j)
11
Q
j
(

H
(j)
11
)
H
) +tr(

H
12
Q
K

H
H
12
), (22)
and

K1
j=1,j=k
tr(

H
(j)
11
Q
j
(

H
(j)
11
)
H
) shifts the trade-off curve of (21) along the E-axis as in Fig. 2. Therefore,
all the optimal

Q
k
for all boundary points (

R,

E) of (7) become solutions for the boundary of (11), even
though the boundaries of (7) and (11) are different.
Remark 1: Note that (11) can be regarded as the achievable R-E tradeoff region for the two-user
MIMO IFC with an effective channel set of (

H
(k)
11
,

H
12
,

H
(k)
21
,

H
22
) as in Fig. 2. Accordingly, we can
have the following corollary.
October 28, 2013 DRAFT
11
Corollary 1: (One ID and multiple EHs) In the high SNR regime, for k = 1, ..., K1, the optimal Q
k
at the boundary of the achievable rate-energy region (7) has a rank one at most. That is, rank(Q
k
) 1
for k = 1, ..., K 1.
Proof: Assuming that rank(Q
k
) 2 with some k for (7). From Proposition 2, Q
k
then becomes
a solution for the boundary of (11). However, from Proposition 1, the optimal Q
i
at the boundary of the
achievable R-E region of (11) has a rank one at most, which is a contradiction.
In other words, if a covariance matrix of an energy transmitter in K-user MIMO IFC (multiple EHs and
one ID) has a rank ( 2), then we can always nd a rank-one beamforming for that transmitter exhibiting
either higher information rate or larger harvested energy.
C. One EH and multiple IDs in a K-user IFC
Without loss of generality, the transceiver pair (Tx
1
, Rx
1
) operate in EH mode, while (Tx
k
, Rx
k
),
k = 2, ..., K in ID mode.
Because information decoding is done only at Rx
i
, i = 2, ..., K by letting R =

K
i=2
R
i
and E = E
1
with E
1
=

K
j=1
E
1j
, we can dene the achievable rate-energy region as:
C
RE
(P)
_
(R, E) : R

K
i=2
log det(I
M
+H
H
ii
R
1
i
H
ii
Q
i
),
E

K
j=1
tr(H
1j
Q
j
H
H
1j
), tr(Q
j
)P, Q
j
_0, j =1, ...,K
_
. (23)
Proposition 3: All the optimal Q
1
at the boundary of the achievable rate-energy region for (23) become
optimal solutions for the boundary of
C
RE
(P)
_
(R, E) : R log det(I
(K1)M
+

H
H
22
(

R
(1)
2
)
1

H
22

Q
2
),
Etr(

H
(1)
11
Q
1
(

H
(1)
11
)
H
) +tr(

H
12

Q
2

H
H
12
), tr(Q
j
)P, Q
j
_0, j =1, ..., K
_
, (24)
where

H
(1)
11
= H
11
,

H
12
=
_
H
12
. . . H
1K
_
C
M(K1)M
, (25)

H
(1)
21
=
_

_
H
21
.
.
.
H
K1
_

_
C
(K1)MM
,

H
22
= diagH
22
, H
33
, ..., H
KK
C
(K1)M(K1)M
, (26)

Q
2
= diagQ
2
, Q
3
, ..., Q
K
,

R
(1)
2
= I
(K1)M
+

H
(1)
21
Q
1
(

H
(1)
21
)
H
(27)
October 28, 2013 DRAFT
12
Proof: Let us consider the boundary point (

R,

E) of the achievable rate-energy for (23) for any

Q
2
,...,

Q
K
, the covariance matrices of information transmitters on the boundary point. The corresponding
covariance matrix of energy transmitter (i.e., the 1st transmitter) at the boundary point is denoted as

Q
1
.
Note that, in (23), for arbitrary Q
j
, j = 1, ..., K exhibiting (R, E) C
RE
(P),
log det(I
M
+H
H
ii
R
1
i
H
ii
Q
i
) =
log det(I
M
+H
H
ii
(I
M
+

K
j=1,i
H
ij
Q
j
H
H
ij
+H
i1
Q
1
H
H
i1
)
1
H
ii
Q
i
), (28)
which is monotonically decreasing with respect to Q
1
from Lemma 1. Furthermore, if we remove the
term

K
j=1,i
H
ij
Q
j
H
H
ij
(the interferences from other information transmitters) in (28), the monotonically
decreasing property with respect to Q
1
is still valid. Therefore,

Q
1
for all the boundary point (

R,

E) of
(23) also become the solutions for the boundary of
C

RE
(P)
_
(R, E) : R

K
i=2
log det(I
M
+H
H
ii
(I
M
+H
i1
Q
1
H
H
i1
)
1
H
ii
Q
i
),
Etr(H
11
Q
1
H
H
11
) +

K
j=2
tr(H
1j
Q
j
H
H
1j
), tr(Q
j
)P, Q
j
_0, j =1, ...,K
_
, (29)
even though the boundaries of C
RE
(P) and C

RE
(P) are different. Now let us consider the boundary
point (

R

,

E

) of (29) with any



Q

i
, i = 2, ..., K, the covariance matrices of information transmitters
on the boundary point. In addition, the corresponding covariance matrix of the energy transmitter at the
boundary point are denoted as

Q

1
. Similarly done in the proof of Proposition 2, from (29), we can dene
functions f

(Q
1
) and g

(Q
1
) with respect to the covariance matrix of the rst transmitter as
f

(Q
1
) =
K

i=2
log det(I
M
+H
H
ii
(I
M
+H
i1
Q
1
H
H
i1
)
1
H
ii

Q

i
), (30)
and
g

(Q
1
) = tr(H
11
Q
1
H
H
11
) +
K

j=2
tr(H
1j

Q

j
H
H
1j
). (31)
Then, the boundary point (

R

,

E

) of (29) can be rewritten as (



R

,

E

) = (f

1
), g

1
)). Note that
f

(Q
1
) is monotonically decreasing with respect to Q
1
. Now we dene
f

(Q
1
) = log det(I
(K1)M
+

H
H
22
(I
(K1)M
+

H
(1)
21
Q
1
(

H
(1)
21
)
H
)
1

H
22

2
), (32)
where

2
= diag

2
,

3
, ...,

K
. Note that f

(Q
1
) in (32) is also monotonically decreasing with
respect to Q
1
. Accordingly, to satisfy f

k
) f

k
) with

Q

k
of the boundary point (f

k
), g

k
)),

k
should be expressed as

k
=

Q

k
Q (33)
October 28, 2013 DRAFT
13
Transmitter 1
Transmitter K
Receiver 1(EH)
Receiver K (ID)
Energy flow
Information flow
Interference
Transmitter 2 Receiver 2(ID)
Transmitter2'
Receiver1 EH
Energy flow
Information flow
Interference1
Transmitter1
Receiver2' ID
Interference2
1
21
H
12
H
22
H
1
11
H
(a)
Transmitter 1
Transmitter K-1
Receiver 1(EH)
Receiver K-1 (ID)
Energy flow
Interference
Transmitter 2 Receiver 2(EH)
Transmitter K Receiver K (ID)
Information flow
Transmitter 2'
Receiver 1(EH)
Receiver 2' (ID)
Energy flow
Information flow
Interference
Transmitter k
1
Receiver K (EH)
22
H
12
H

21
k
H

11
k
H
(b)
Fig. 3. (a) One EH and multiple IDs and (b) Multiple EHs and multiple IDs in a K-user IFC.
with > 0 and Q _ 0. However, we can easily nd that g

k
) > g

k
). That is, the variation on

k
that increases f

k
) always incurs a loss in g

k
). Therefore, all the optimal

Q

k
also becomes
the solutions for the boundary of (24). Accordingly, all the optimal

Q
k
for (

R,

E) of (23) become the
solutions for the boundary of (24).
Remark 2: Note that (24) can be regarded as the achievable R-E tradeoff region for two-user MIMO
IFC with an effective channel set of (

H
(1)
11
,

H
12
,

H
(1)
21
,

H
22
) and the additional constraint on the covariance
matrix of x
2
as in Fig. 3 (a). Because Proposition 1 is still valid even with the block-diagonal structure
on the covariance matrix at the information transmitter, i.e.

Q
2
, we can have the following corollary.
Corollary 2: (Multiple IDs and one EH) In the high SNR regime, the optimal Q
1
at the boundary of
the achievable rate-energy region (23) has a rank one at most. That is, rank(Q
1
) 1.
D. multiple EHs and multiple IDs in a K-user IFC
Now let us consider that multiple EHs and multiple IDs coexist. That is, the transceiver pair (Tx
i
, Rx
i
)
for i = 1, ..., K
1
operate in EH mode, while (Tx
i
, Rx
i
), k = K
1
+ 1, ..., K in ID mode. By letting
October 28, 2013 DRAFT
14
R =

K
i=K1+1
R
k
and E =

K1
i=1
E
i
with E
i
=

K
j=1
E
ij
, we can dene the achievable rate-energy
region as:
C
RE
(P)
_
(R, E) : R

K
i=K1+1
log det(I
M
+H
H
ii
R
1
i
H
ii
Q
i
),
E

K1
i=1

K
j=1
tr(H
ij
Q
j
H
H
ij
), tr(Q
j
)P, Q
j
_0, j =1, ...,K
_
. (34)
Then we have the following proposition.
Proposition 4: All the optimal Q
k
at the boundary of the achievable rate-energy region for (34) also
become the optimal solutions for the boundary of
C
RE,k
(P)
_
(R, E) : R log det(I
(KK1)M
+

H
H
22
(

R
(k)
2
)
1

H
22

Q
2
),
Etr(

H
(k)
11
Q
k
(

H
(k)
11
)
H
) +tr(

H
12

Q
2

H
H
12
), tr(Q
j
)P, Q
j
_0, j =k, K
1
+ 1, ..., K
_
, (35)
for all k = 1, ..., K
1
, where

H
(k)
11
=
_

_
H
1k
.
.
.
H
K1k
_

_
,

H
12
=
_

H
(K1+1)
12
, ...,

H
(K)
12
_
=
_

_
H
1K1+1
. . . H
1K
.
.
.
.
.
.
.
.
.
H
K1K1+1
. . . H
K1K
_

_
C
K1M(KK1)M
, (36)

H
(k)
21
=
_

_
H
K1+1k
.
.
.
H
Kk
_

_
,

H
22
= diagH
K1+1K1+1
, H
K1+2K1+2
, ..., H
KK
C
(KK1)M(KK1)M
, (37)
and

Q
2
= diagQ
K1+1
, ..., Q
K
,

R
(k)
2
= I
(KK1)M
+

H
(k)
21
Q
k
(

H
(k)
21
)
H
C
(KK1)M(KK1)M
.
Proof: Let us consider the boundary point (

R,

E) of the achievable rate-energy for (34) for any
given

Q
K1+1
, ...,

Q
K
. In addition, the corresponding covariance matrix of the kth energy transmitter at
the boundary point are denoted as

Q
k
. In (34),
log det(I
M
+H
H
ii
R
1
i
H
ii
Q
i
) =
log det(I
M
+H
H
ii
(I
M
+

K1
j=1
j=k
H
ij
Q
j
H
H
ij
+

K
j=K1+1
j=i
H
ij
Q
j
H
H
ij
+H
ik
Q
k
H
H
ik
)
1
H
ii
Q
i
). (38)
Here,

K1
j=1
j=k
H
ij
Q
j
H
H
ij
and

K
j=K1+1
j=i
H
ij
Q
j
H
H
ij
are the interferences from the energy-transferring
transmitters except the kth transmitter and from other information-transferring transmitters, respectively.
October 28, 2013 DRAFT
15
Because the sum of monotonically decreasing functions is also monotonically decreasing, similarly
done in Proposition 2,

Q
k
for all (

R,

E) also become the solutions for the boundary of
C

RE,k
(P)
_
(R, E) : R

K
i=K1+1
log det(I
M
+H
H
ii
(I
M
+

K
j=K1+1
j=i
H
ij
Q
j
H
H
ij
+H
ik
Q
k
H
H
ik
)
1
H
ii
Q
i
),
E

K1
i=1
(

K
j=K1+1
tr(H
ij
Q
j
H
H
ij
) +tr(H
ik
Q
k
H
H
ik
)), tr(Q
j
)P, Q
j
_0, j =1, ...,K
_
=
_
(R, E) : R

K
i=K1+1
log det(I
M
+H
H
ii
(I
M
+

K
j=K1+1
j=i
H
ij
Q
j
H
H
ij
+H
ik
Q
k
H
H
ik
)
1
H
ii
Q
i
),
Etr(

H
(k)
11
Q
k
(

H
(k)
11
)
H
) +

K
j=K1+1
tr(

H
(j)
12
Q
j
(

H
(j)
12
)
H
), tr(Q
j
)P, Q
j
_0, j =k, K
1
+ 1, ...,K
_
. (39)
Note that the R-E region of (39) is equivalent with the case of one EH and K K
1
IDs in (23). That
is,

Q
k
for all (

R,

E) also become the solutions for the boundary of
C

RE,k
(P)
_
(R, E) : R

K
i=K1+1
log det(I
M
+H
H
ii
(I
M
+H
ik
Q
k
H
H
ik
)
1
H
ii
Q
i
),
Etr(

H
(k)
11
Q
k
(

H
(k)
11
)
H
) +

K
j=K1+1
tr(

H
(j)
12
Q
j
(

H
(j)
12
)
H
), tr(Q
j
)P, Q
j
_0, j =k, K
1
+ 1, ...,K
_
, (40)
which is analogous with (29). From Proposition 3, we can easily derive that the optimal

Q
k
, k = 1, ..., K
1
for (

R,

E) of (34) also yields the boundary of (35). See also Fig. 3 (b).
Corollary 3: (Multiple IDs and multiple EHs) In the high SNR regime, the optimal Q
k
, k = 1, ..., K
1
at the boundary of the achievable rate-energy region (34) has a rank one at most. That is, rank(Q
k
) 1
for k = 1, ..., K
1
.
Remark 3: From Proposition 4, when transferring the energy in K-user MIMO IFC, the transmitters
optimal strategy is a rank-one beamforming with a proper power allocation, which is a generalized version
of Proposition 1 for two-user MIMO IFC [22]. That is, if the covariance matrix of an energy transmitter
in general K-user MIMO IFC has a rank ( 2), a rank-one beamforming for that transmitter exhibiting
either higher information rate or larger harvesting energy can be found.
IV. RANK-ONE BEAMFORMING DESIGN AND ACHIEVABLE R-E REGION
In this section, we propose rank-one beamforming methods based on Proposition 4, and then propose
an iterative algorithm that computes the achievable R-E trade-off curves for the K-user MIMO IFC
with different beamforming schemes. Again, let us consider that the transceiver pair (Tx
i
, Rx
i
) for
i = 1, ..., K
1
operate in EH mode, while (Tx
i
, Rx
i
), k = K
1
+ 1, ..., K operate in ID mode, without
loss of generality.
October 28, 2013 DRAFT
16
A. Rank-one Beamforming Design
Because there exists multiple EH receivers, each energy transferring transmitter steers its signal to
maximize the energy transferred to all EH receivers. Therefore, because Q
k
for k = 1, ..., K
1
has a rank
one from Proposition 4, it can be given by
Q
k
= P
k
[V
(k)
11
]
1
[V
(k)
11
]
H
1
, (41)
where V
(k)
11
is a M M unitary matrix obtained from the SVD of

H
(k)
11
and 0 P
k
P. That is,

H
(k)
11
= U
(k)
11

(k)
11
(V
(k)
11
)
H
, where
(k)
11
= diag
1
(

H
(k)
11
), ...,
M
(

H
(k)
11
). That is, the kth transmitters
beamforming on the effective channel

H
(k)
11
is analogous to the maximum-energy beamforming (MEB) in
two-user MIMO IFC [22]. Here, the energy harvested from the kth transmitter is given by P
k
(
1
(

H
(k)
11
))
2
.
From an ID perspective at the ID receivers, the kth transmitter should steer its signal to minimize the
interference power to all the ID receivers. That is, the corresponding transmit covariance matrix Q
k
is
then given by
Q
k
= P
k
[

V
(k)
21
]
M
[

V
(k)
21
]
H
M
, (42)
where V
(k)
21
is a MM unitary matrix obtained from the SVD of

H
(k)
21
and 0 P
k
P. That is,

H
(k)
21
=
U
(k)
21

(k)
21
(V
(k)
21
)
H
, where
(k)
21
= diag
1
(

H
(k)
21
), ...,
M
(

H
(k)
21
). The kth transmitters beamforming
that minimizes the interference to the effective channel

H
(k)
12
is also analogous to the minimum-leakage
beamforming (MLB) in two-user MIMO IFC [22]. Then, the energy harvested from the rst transmitter
is given by P
k
|

H
11,k
[V
21,k
]
M
|
2
.
Because MEB and MLB strategies are developed according to different aims - either maximizing
transferred energy to EH or minimizing interference (or, leakage) to ID, respectively, they have their
own weakness - causing either large interference to ID receivers or insufcient harvested energy at EH
receivers.
1) Energy-regularized SLER-maximizing beamforming: To maximize transferred energy to EH and
simultaneously minimize the leakage to ID, we dene a new performance metric, signal-to-leakage-and-
harvested energy ratio (SLER) at the kth transmitter as [22]
SLER
k
=
|

H
(k)
11
v
k
|
2
|H
(k)
21
v
k
|
2
+max(

E/K
1
P|H
(k)
11
|
2
, 0)
=
v
H
k
(H
(k)
11
)
H
H
(k)
11
v
k
v
H
k
_
(H
(k)
21
)
H
H
(k)
21
+max(

E/K
1
P |H
(k)
11
|
2
, 0)I
M
_
v
k
. (43)
October 28, 2013 DRAFT
17
The beamforming vector v
k
that maximizes SLER of (43) is then given by
v
k
=
_
P
k
v
k
| v
k
|
, (44)
where v
k
is the generalized eigenvector associated with the largest generalized eigenvalue of the matrix
pair
((H
(k)
11
)
H
H
(k)
11
, (H
(k)
21
)
H
H
(k)
21
+max(

E/K
1
P |H
(k)
11
|
2
, 0)I
M
). (45)
Here, v
k
can be efciently computed by using a GSVD algorithm [29], [30].
Remark 4: Interestingly, from (43), when the required harvested energy at the EH receiver is large, the
matrix in the denominator of (43) approaches an identity matrix multiplied by a scalar. Accordingly, the
SLER maximizing beamforming is equivalent with the MEB in (41). That is, v
k
becomes

P
k
[V
(k)
11
]
1
.
In contrast, as the required harvested energy becomes smaller, v is steered such that less interference is
leaked into the ID receiver to reduce the denominator of (43). That is, v approaches the MLB weight
vector in (42). Therefore, the proposed SLER maximizing beamforming balances both metrics - energy
maximization to EH and leakage minimization to ID, which has been conrmed in [22].
B. Achievable R-E region
Note that the achievable sum-rate is unknown for general K-user MIMO IFC and, accordingly, the
optimal region for (35) is not easily identied. Instead, based on Corollary 3 that each kth transmitter
for k = 1, ..., K
1
transfers its signal with a rank-one beamforming, we propose an iterative algorithm by
optimizing the transmit powers P
k
, k = 1, ..., K
1
and Q
k
, k = K
1
+1, ..., K simultaneously. The energy
transmitters can choose their covariance matrices among (41), (42), (44), or other rank-one covariance
matrices. In this paper, we assume that they adopt the same beamforming among MLB, MEB, and SLER
beamforming and compare their performance by simulation. Given Q
k
as in (41), (42), or (44), the
achievable rate-energy region is then given as:
C
RE
(P)=
_
(R, E) : R

K
i=K1+1
log det(I
M
+H
H
ii
R
1
i
H
ii
Q
i
),
E

K
j=K1+1
tr(

H
(j)
12
Q
j
(

H
(j)
12
)
H
) +E
11
, tr(Q
j
)P, Q
j
_0, j =1, ...,K
_
. (46)
where
E
11
=
K1

j=1
tr(

H
(j)
11
Q
k
(

H
(j)
11
)
H
) =
K1

j=1

j
P
j
, (47)
October 28, 2013 DRAFT
18
with

j
=
_

_
|

H
(j)
11
[V
(j)
11
]
1
|
2
for MEB
|

H
(j)
11
[

V
(j)
21
]
M
|
2
for MLB
|

H
(j)
11
vj
vj
|
2
for SLER beamforming
, (48)
and
R
i
= I
M
+
K1

j=1
P
j

ij
+
K

j=K1+1
j=i
H
ij
Q
j
H
H
ij
, (49)
with

ij
=
_

_
H
ij
[V
(j)
11
]
1
[V
(j)
11
]
H
1
H
H
ij
for MEB
H
ij
[

V
(j)
21
]
M
[

V
(j)
21
]
H
M
H
H
ij
for MLB
H
ij
v
j
v
H
j
H
H
ij
/| v
j
|
2
for SLER beamforming
. (50)
Accordingly, we have the following optimization problem for the rate-energy region of (46)
(P1) maximize
Qk,k=K1+1,...,K
Pi,i=1,...,K1
J

K
i=K1+1
log det(I
M
+H
H
ii
R
1
i
H
ii
Q
i
) (51)
subject to

K
j=K1+1
tr(

H
(j)
12
Q
j
(

H
(j)
12
)
H
) max(

EE
11
, 0) (52)
tr(Q
i
) P, Q
i
_ 0, i = K
1
, ..., K (53)
P
j
P, j = 1, ..., K
1
, (54)
where

E can take any value less than E
max
denoting the maximum energy transferred from all the
transmitters. Here, it can be easily derived that E
max
is given as
E
max
= P
_
_
K1

j=1

j
+
K

j=K1+1

1
(

H
(j)
12
)
_
_
, (55)
which is obtained when all the information transmitters steer their signals such that their beamforming
energy is maximized on the cross link

H
(j)
12
. The optimization problem (P1) is obviously non-convex
due to the coupled variables in the objective function J. That is, because of the interference at each ID
receiver from other information transmitters, Q
k
are coupled in the objective function. Accordingly, here
we develop a sub-optimal iterative algorithm for (P1).
Before we proceed with (P1), let us consider a simplied optimization problem by removing the
interferences from other information transmitters by assuming that the cross-channel gain among the
October 28, 2013 DRAFT
19
information transceivers is very small as
(P1 UP) maximize
Qk,k=K1+1,...,K
Pi,i=1,...,K1
J
UP

K
i=K1+1
log det(I
M
+H
H
ii

R
1
i
H
ii
Q
i
) (56)
subject to

K
j=K1+1
tr(

H
(j)
12
Q
j
(

H
(j)
12
)
H
) max(

EE
11
, 0) (57)
tr(Q
i
) P, Q
i
_ 0, i = K
1
, ..., K (58)
P
j
P, j = 1, ..., K
1
, (59)
where

R
i
= I
M
+

K1
j=1
P
j

ij
. Note that, because the interferences from all other information
transmitters are removed, (P1-UP) can be an upper-bound of the original R-E region. Even though it
is not tight, it gives an insight on developing the iterative algorithm for the original problem.
By letting P = [P
1
, ..., P
K1
]
T
, using that
d
dx
log det(A(x)) = tr
_
A(x)
1
dA(x)
dx
_
[31] and Sylvesters
determinant theorem [28],
P
J
UP
(P, Q
K1+1
, ..., Q
K
) R
K11
is given as

P
J
UP
(P, Q
K1+1
, ..., Q
K
) =
_

_
tr
_

K
i=K1+1
_
(H
ii
Q
i
H
H
ii
+

R
i
)
1
(

R
i
)
1
_

i1
_
.
.
.
tr
_

K
i=K1+1
_
(H
ii
Q
i
H
H
ii
+

R
i
)
1
(

R
i
)
1
_

iK1
_
_

_
. (60)
Because, from (60), the objective function in (P1-UP) is monotonically decreasing with respect to P, we
iteratively optimize their values using the steepest descent method as:
Algorithm 1. Identication of the achievable R-E region for P1-UP:
1) Initialize n = 0, P
[0]
= [P, ..., P]
T
,
E
[0]
11
=
K1

j=1

j
(P
[0]
)
j
,

R
[0]
i
= I
M
+
K1

j=1
(P
[0]
)
j

ij
. (61)
2) For n = 0 : N
max
a) Solve the optimization problem (P1-UP) for Q
[n]
K1+1
, ..., Q
[n]
K
as a function of E
[n]
11
and

R
[n]
i
.
b) If

K
j=K1+1
tr(

H
(j)
12
Q
[n]
j
(

H
(j)
12
)
H
) +E
[n]
11
>

E
P
[n+1]
= max
_
P
[n]
+
P
J
UP
(P
[n]
, Q
[n]
K1+1
, ..., Q
[n]
K
), 0
_
, (62)
where the step size is given by a xed value on [0,
max
] with

max
=

K
j=K1+1
tr(

H
(j)
12
Q
j
(

H
(j)
12
)
H
)
T
P
[n]

P
J
UP
(P
[n]
, Q
[n]
K1+1
, ..., Q
[n]
K
)
, (63)
where = [
1
, ...,
K1
]
T
. Then, update E
[n+1]
11
and R
[n+1]
i
with P
[n+1]
similarly to (61).
October 28, 2013 DRAFT
20
3) Finally, the boundary point of the achievable R-E region is given as
R =
K

i=K1+1
log det(I
M
+H
H
ii
(

R
[Nmax+1]
i
)
1
H
ii
Q
[Nmax+1]
i
),
E = E
[Nmax+1]
11
+
K

j=K1+1
tr(

H
(j)
12
Q
[Nmax+1]
j
(

H
(j)
12
)
H
).
In (63), if the total transferred energy is larger than the required harvested energy

E, the transmitters
transferring the energy reduce their transmit power to lower the interference to the ID receivers. Fur-
thermore, the maximum allowable step size in (63) is computed from (62) and the fact that E
[n+1]
11
=

K
j=K1+1
tr(

H
(j)
12
Q
[n]
j
(

H
(j)
12
)
H
), which leads
E
[n+1]
11
=
T
P
[n+1]
=
T
P
[n]
+
max

T

P
J
UP
(P
[n]
, Q
[n]
K1+1
, ..., Q
[n]
K
). (64)
Note that, if the energy harvested by the EH receivers from the information transmitters (

K
j=K1+1
tr(

H
(j)
12
Q
[n]
j
(

H
(j)
12
)
H
))
is larger than

E, the energy transmitters do not transmit any signal. That is, rank(Q
k
) = 0.
To complete Algorithm 1, we now show how to solve the optimization problem (P1-UP) for Q
[n]
k
, k =
K
1
+1, ..., K in Step 2 of Algorithm 1. For given E
11
and Q
i
, i = 1, ..., K
1
, J
UP
(P
[n]
, Q
[n]
K1+1
, ..., Q
[n]
K
)
can be derived as
J
UP
=
K

i=K1+1
log det(I
M
+H
H
ii
(

R
[n]
i
)
1
H
ii
Q
i
) (65)
=
K

i=K1+1
log det(I
M
+ (

R
[n]
i
)
1/2
H
ii
Q
i
H
H
ii
(

R
[n]
i
)
1/2
). (66)
By letting

H
ii
= (

R
[n]
i
)
1/2
H
ii
, the Lagrangian function of (P1-UP) can then be written as
L(Q
i
, ,
i
, i = K
1
+ 1, ..., K) =

K
i=K1+1
log det(I
M
+

H
ii
Q
i

H
H
ii
)
+(

K
i=K1+1
tr(

H
(i)
12
Q
i
(

H
(i)
12
)
H
) (

EE
11
))

K
i=K1+1

i
(tr(Q
i
) P), (67)
and the corresponding dual function is then given by [6], [32]
g(,
i
, i = K
1
+ 1, ..., K) = max
Qi0
L(Q
i
, ,
i
, i = K
1
+ 1, ..., K). (68)
Here the optimal solution

i
,

, and Q
i
can be found through the iteration of the following steps [32]
Algorithm 2. Optimization algorithm for P1-UP given P:
1) The maximization of L(Q
i
, ,
i
, i = K
1
+ 1, ..., K) over Q
i
for given ,
i
.
2) The minimization of g(,
i
, i = K
1
+ 1, ..., K) over ,
i
for given Q
i
.
October 28, 2013 DRAFT
21
Note that, for given ,
i
, i = K
1
, ..., K, the maximization of L(Q
2
, , ) in Step 1) can be derived as
max
Qi0
L(Q
i
, ,
i
, i = K
1
+ 1, ..., K) =
K

i=K1+1
log det(I
M
+

H
ii
Q
i

H
H
ii
) +(
K

i=K1+1
tr(

H
(i)
12
Q
i
(

H
(i)
12
)
H
))

i=K1+1

i
(tr(Q
i
)),
=
K

i=K1+1
_
log det(I
M
+

H
ii
Q
i

H
H
ii
) tr(A
i
Q
i
)
_
, (69)
where A
i
=
i
I
M
(

H
(i)
12
)
H

H
(i)
12
. Note that, due to the assumption that the interferences from other
information transmitters is nulled out in (P1-UP), (69) can be easily decoupled into the point-to-point
MIMO capacity optimization with a single weighted power constraint and the solution for Q
i
is then
given by [6], [32]
Q
i
= A
1/2
i

ii

i

V
H
ii
A
1/2
i
, (70)
where

V

ii
is obtained from the SVD of the matrix

H
ii
A
1/2
i
, i.e.,

H
ii
A
1/2
i
=

U

ii

ii

V
H
ii
. Here,

ii
=
diag
1
(

H
ii
A
1/2
i
), ...,
M
(

H
ii
A
1/2
i
) and

i
= diag p
i,1
, ..., p
i,M
with p
i,j
= (11/
2
j
(

H
ii
A
1/2
i
))
+
,
j = 1, ..., M.
In step 2), the parameters
i
and minimizing g(,
i
, i = K
1
+ 1, ..., K) can be solved by the
subgradient-based method [6], [33], where the the subgradient of g(,
i
, i = K
1
+1, ..., K) is given by
g(,
i
, i = K
1
+ 1, ..., K) =
_
K

i=K1+1
tr(

H
(i)
12
Q
i
(

H
(i)
12
)
H
) (

EE
11
),
P tr(Q
K1+1
), ..., P tr(Q
K
)
_
. (71)
Remark 5: Due to the fact that each element in (60) always has a negative value and the step size in
(63) has non-negative values, we can nd that the power of energy transmitters converges monotonically.
In addition, because (56) is concave over Q
i
and monotonically decreasing with respect to P, (56)
is quasi-concave. Furthermore, the constraints, (57), (58) and (59), are the convex set of Q
i
and P
j
.
Therefore, the converged solution of Algorithm 1 is globally optimal [34].
Now let us consider the original problem (P1). Because (P1) is non-convex, the optimal solution can
not be easily computed, but motivated by Algorithm 1, we can also develop the sub-optimal iterative
algorithm. Because
P
J(P, Q
K1+1
, ..., Q
K
) R
K11
has the same form as
P
J
UP
(P, Q
K1+1
, ..., Q
K
)
except that

R
i
is replaced by R
i
. Therefore, the objective function in (P1) is also monotonically
October 28, 2013 DRAFT
22
decreasing with respect to P, regardless of R
i
, and P for (P1) can also be optimized using the steepest
descent method as Algorithm 1, in which

R
[n]
i
is replaced by
R
[n]
i
= I
M
+
K1

j=1
(P
[n]
)
j

ij
+
K

j=K1+1
j=i
H
ij
Q
[n]
j
H
H
ij
. (72)
In addition, we formulate the Lagrangian function similarly to (67)
L

(Q
i
, ,
i
, i = K
1
+ 1, ..., K) =

K
i=K1+1
log det(I
M
+H
H
ii
(R
[n]
i
)
1
H
ii
Q
i
)
+(

K
i=K1+1
tr(

H
(i)
12
Q
i
(

H
(i)
12
)
H
) (

EE
11
))

K
i=K1+1

i
(tr(Q
i
) P),
and the corresponding dual function is then given by
g

(,
i
, i = K
1
+ 1, ..., K) = max
Qi0
L

(Q
i
, ,
i
, i = K
1
+ 1, ..., K). (73)
Note that the only difference between L

(Q
i
, ,
i
, i = K
1
+1, ..., K) and L(Q
i
, ,
i
, i = K
1
+1, ..., K)
is the interference at the kth ID receiver due to the other information transmitters, which hinders nding
the optimal solution for the maximization of L

(Q
i
, ,
i
, i = K
1
+ 1, ..., K) over Q
i
for given and

i
. That is, for given ,
i
, i = K
1
, ..., K, the maximization of L

(Q
2
, , ), analogous with Step 1) of
Algorithm 2, can be derived as
max
Qi0
L

(Q
i
, ,
i
, i = K
1
+ 1, ..., K) =
K

i=K1+1
log det(I
M
+H
H
ii
(R
[n]
i
)
1
H
ii
Q
i
)
+(
K

i=K1+1
tr(

H
(i)
12
Q
i
(

H
(i)
12
)
H
))
K

i=K1+1

i
(tr(Q
i
)),
=
K

i=K1+1
_
log det(I
M
+H
H
ii
(R
[n]
i
)
1
H
ii
Q
i
) tr(A
i
Q
i
)
_
. (74)
Letting

Q
i
= A
1/2
i
Q
i
A
1/2
i
similarly to [35], (74) can be rewritten as
max
Qi0
L

(Q
i
, ,
i
, i = K
1
+ 1, ..., K) =
K

i=K1+1
_
log det(I
M
+A
1/2
i
H
H
ii
(R
[n]
i
)
1
H
ii
A
1/2
i

Q
i
) tr(

Q
i
)
_
, (75)
which becomes a conventional rate maximization problem in MIMO IFC subject to individual power
constraints [25], [36]. Accordingly, for (P1), Q
i
, , and
i
can be optimized based on iterative waterlling
algorithm [25] with effective channel H
ii
A
1/2
i
. Even though (75) is not convex, it is shown that the
global optimum exists and the iterative waterlling approaches converged to the global optimum in which
the achievable rate is maximized in a distributed manner [37], [38]. Furthermore, the objective function
in (P1) is also monotonically decreasing with respect to P, regardless of R
i
, and accordingly, we can
October 28, 2013 DRAFT
23
also nd that the power of energy transmitters converges monotonically. Therefore, the convergence of
the proposed iterative algorithm for (P1) is always guaranteed.
Remark 6: In (P1), if the number of information transmitters becomes large, the interference from
energy transmitters is smaller than that from information transmitters. That is, (49) can be approximated
as
R
i
I
M
+
K

j=K1+1
j=i
H
ij
Q
j
H
H
ij
, (76)
and the achievable rate in (46) is independent of the interference from energy transmitters. That is,
the energy transmitter signals can be designed by caring about their own links, not caring about the
interference link to the ID receivers. Accordingly, as the number of information transmitters increases,
the optimal transmission strategy at energy transmitters becomes a rank-one MEB method with a power
control.
C. K
1
-EH Selection algorithm in K-user MIMO IFC
Motivating that the SLER value indicates how suitable the current channel is to either EH mode or ID
mode, we propose K
1
-EH selection method in K-user MIMO IFC. That is, higher SLER implies that the
transmitter can transfer more energy to its associated EH receiver incurring less interference to the ID
receiver. Note that the kth SLER in (43) depends on (K
1
1) EH receivers. Therefore, in our proposed
selection, to choose K
1
transceiver pairs jointly, we evaluate sum of SLERs of K
1
transceiver pairs for
K
C
K1
possible candidates and choose one candidate having the maximum SLER sum. That is, by letting
I
j
= (Tx
j1
, Rx
j1
), ..., (Tx
jK
1
, Rx
jK
1
)[j
k
1, ..., K, for k = 1, ..., K
1
, (77)
the set of K
1
energy harvesting EHs is then selected as:
I
max
= arg
Ij
max

iIj
SLER
i
. (78)
V. SIMULATION RESULTS
Computer simulations have been performed to evaluate the R-E tradeoff of various transmission
strategies in the K-user MIMO IFC. In the simulations, the normalized channel H
ij
is generated such
as H
ij
= 10
3/2

ijM

HijF

H
ij
, where the elements of

H
ij
are independent and identically distributed
(i.i.d.) zero-mean complex Gaussian random variables (RVs) with a unit variance. The term 10
3/2
is
due to the path loss with a power path loss exponent 3 and 10m distance between Tx i and Rx i
October 28, 2013 DRAFT
24
(30dB = 10 log
10
10
3
). The maximum transmit power is set as P = 50mW and the noise power is
1W, unless otherwise stated.
Figs. 4 (a)-(c) show R-E tradeoff curves for three different rank-one beamforming - the MEB, the MLB,
and the SLER maximizing beamforming described in Section IV-A with M = 4 when K = 2, 3, 4
and K K
1
= 1. Here, the path loss is set as
ii
= 1, and
ij
= 0.6 for i, j = 1, 2, ..., K and i ,= j.
That is, while only the Kth transmitter transfers information to its corresponding receivers, the remaining
transmitters transfer the energy to the remaining EH receivers with rank-one beamforming. As expected,
as the number of energy transmitters increases, total harvested energy increases. Interestingly, in the
regions where the energy is less than a certain threshold [40, 100]W, all the energy transmitters do not
transmit any signals to reduce the interference to the ID receiver. That is, the energy transferred from
the information transmitter is sufcient to satisfy the energy constraint at the EH receivers. Note that the
threshold is linearly proportional to the number of EH receivers that can harvest energy from information
transmitters signal. The dashed lines in Figs. 4 (a)-(c) indicate the R-E curves of the time-sharing of the
full-power rank-one beamforming and the no transmission at the energy transmitters. Accordingly, the
information transmitter switches between the beamforming on

H
(K)
12
and the water-lling on H
KK
in the
corresponding time slots. For MLB, water-lling-like approach (70) exhibits higher R-E performance
than the time-sharing scheme. However, for MEB, when the required energy is less than a certain value,
the time-sharing exhibits better performance than the approach (70). This observation is more apparent
as the number of energy transmitters increases. That is, because the energy transmitters with the MEB
cause large interference to the ID receiver, it is desirable that, for the low required harvested energy, the
energy transmitters turn off their power in the time slots where the information transmitter is assigned
to exploit the water-lling method on H
KK
. Instead, in the remaining time slots, the energy transmitters
opt for a MEB with full power and the information transmitter transfers its information to the ID receiver
by steering its beam on EH receivers channel

H
(K)
12
to help the EH operation. In contrast, with SLER
maximizing beamforming, water-lling-like approach (70) exhibits higher R-E performance than the
time-sharing scheme. Furthermore, its R-E region covers most of those of both MEB and MLB.
In Figs. 5 (a) and (b), we have additionally included the R-E tradeoff curves when (K, K
1
) = (3, 1) and
(4, 1). Together with Fig. 1 (a), we can nd the trend of the R-E region when the number of information
transmitters increases, while only the rst transmitter transfers energy. Note that, as the number of
information transmitters increases, the maximum harvested energy is also increased. However, because
the interferences are increasing proportional to the number of information transmitters, the maximum
information rate is not drastically increasing, which implies that the system becomes interference-limited.
October 28, 2013 DRAFT
25
8.5 9 9.5 10 10.5 11 11.5 12 12.5 13
0
50
100
150
200
250
Rate (bits/s/Hz)
E
n
e
r
g
y

(


W
)
RE tradeoff curves when M=4 and (K, K
1
) =(2,1)


Max. energy beamforming
Min. leakage beamforming
SLER maximizing beamforming
Timesharing of full/zero
power transmissions at
Energy Tx.
No transmission at Energy Tx.
rank(Q
1
)=0
rank(Q
i
)=1 at Energy Tx.
(a)
7.5 8 8.5 9 9.5 10 10.5 11 11.5 12 12.5 13
0
50
100
150
200
250
300
350
400
450
500
Rate (bits/s/Hz)
E
n
e
r
g
y

(


W
)
RE tradeoff curves when M=4 and (K, K
1
) =(3,2)


Max. energy beamforming
Min. leakage beamforming
SLER maximizing beamforming
Timesharing of full/zero
power transmissions at
Energy Tx.
rank(Q
i
)=1 at Energy Tx.
No transmission at Energy Tx.
rank(Q
1
)=0
(b)
6 7 8 9 10 11 12 13
0
100
200
300
400
500
600
700
800
Rate (bits/s/Hz)
E
n
e
r
g
y

(


W
)
RE tradeoff curves when M=4 and (K, K
1
) =(4,3)


Max. energy beamforming
Min. leakage beamforming
SLER maximizing beamforming
Timesharing of full/zero
power transmissions at
Energy Tx.
rank(Q
i
)=1 at Energy Tx.
No transmission at Energy Tx.
rank(Q
1
)=0
(c)
Fig. 4. R-E tradeoff curves for MEB, MLB, and SLER maximizing beamforming when (a) (K, K1) = (2, 1), (b) (K, K1) =
(3, 2), and (c) (K, K1) = (4, 3). Here, M = 4 and ij = 0.6 for i = j.
Furthermore, because the interference due to the information transmitters is dominant, MEB at the energy
transmitter becomes a more attractive strategy, resulting in a wider R-E region compared to that for the
MLB. That is, compared to Fig. 4, the R-E region of MEB covers almost that of MLB (see Fig. 5 (b)),
which is consistent with Remark 6.
To see the effect of interference on the information rate, we evaluate the R-E tradeoff curves in Fig.
6 when (K, K
1
) = (4, 2) (multiple EHs and multiple IDs) with different
ij
= 0.6, 0.3 for i ,= j.
Because a smaller
ij
implies less interference at each receiver, we can nd that R-E region for
ij
= 0.6
exhibits a larger harvested energy but a lower information rate.
Fig. 7 shows the R-E tradeoff curves for SLER maximizing beamforming with/without SLER-based
user selection described in Section IV-C when (K, K
E
) = (4, 2) with M = 4. Note that the case with
October 28, 2013 DRAFT
26
12 12.5 13 13.5 14 14.5 15 15.5 16 16.5
0
50
100
150
200
250
300
Rate (bits/s/Hz)
E
n
e
r
g
y

(


W
)
RE tradeoff curves when M=4 and (K, K
1
) =(3,1)


Max. energy beamforming
Min. leakage beamforming
SLER maximizing beamforming
No transmission at Energy Tx.
rank(Q
1
)=0
rank(Q
i
)=1 at Energy Tx.
(a)
12 12.5 13 13.5 14 14.5 15 15.5 16 16.5 17
0
50
100
150
200
250
300
350
Rate (bits/s/Hz)
E
n
e
r
g
y

(


W
)
RE tradeoff curves when M=4 and (K, K
1
) =(4,1)


Max. energy beamforming
Min. leakage beamforming
SLER maximizing beamforming
rank(Q
i
)=1 at Energy
Tx.
No transmission at Energy Tx.
rank(Q
1
)=0
(b)
Fig. 5. R-E tradeoff curves for MEB, MLB, and SLER maximizing beamforming when (a) (K, K1) = (3, 1) and (b)
(K, K1) = (4, 1). Here, M = 4 and ij = 0.6 for i = j.
10 11 12 13 14 15 16 17
0
100
200
300
400
500
600
Rate (bits/s/Hz)
E
n
e
r
g
y

(


W
)
RE tradeoff curves when M=4 and (K, K
1
) =(4,2), = 0.6


Max. energy beamforming
Min. leakage beamforming
SLER maximizing beamforming
No transmission at Energy Tx.
rank(Q
1
)=0
rank(Q
i
)=1 at Energy Tx.
(a)
12 13 14 15 16 17 18
0
50
100
150
200
250
300
350
400
Rate (bits/s/Hz)
E
n
e
r
g
y

(


W
)
RE tradeoff curves when M=4 and (K, K
1
) =(4,2), = 0.3


Max. energy beamforming
Min. leakage beamforming
SLER maximizing beamforming
rank(Q
i
)=1 at Energy Tx.
No transmission at Energy Tx.
rank(Q
1
)=0
(b)
Fig. 6. R-E tradeoff curves for MEB, MLB, and SLER maximizing beamforming when (a) (K, K1) = (4, 2) and ij = 0.6
for i = j and (b) (K, K1) = (4, 2) and ij = 0.3 for i = j. Here, M = 4.

ij
= 0.3 has weaker cross-link channel (inducing less interference) than that with
ij
= 0.6. The SLER-
based user selection extends the achievable R-E region for both
ij
0.3, 0.6, but the improvement
for
ij
= 0.6 is slightly more apparent. That is, the SLER-based scheduling becomes more effective
when strong interference exists in the system. Note that the case with
ij
= 0.6 exhibits a slightly lower
achievable rate than that with
ij
= 0.3, while achieving a larger harvested energy, which is a similar
observation as that found in Fig. 6. That is, a strong interference degrades the information decoding
performance but it can be effectively utilized in the energy-harvesting.
October 28, 2013 DRAFT
27
12 13 14 15 16 17 18
0
50
100
150
200
250
300
350
400
450
500
Rate (bits/s/Hz)
E
n
e
r
g
y

(


W
)
RE tradeoff curves for SLER max. beamforming when (K, K
1
) =(4,2)


Without SLER based scheduling,
ij
=0.3, , i j
With SLER based scheduling,
ij
=0.3, , i j
(a)
10 11 12 13 14 15 16 17
0
100
200
300
400
500
600
Rate (bits/s/Hz)
E
n
e
r
g
y

(


W
)
RE tradeoff curves for SLER max. beamforming when (K, K
1
) =(4,2)


Without SLER based scheduling,
ij
=0.6, i j
With SLER based scheduling,
ij
=0.6, i j
(b)
Fig. 7. R-E tradeoff curves for SLER maximizing beamforming with/without SLER-based user selection when (a) ij = 0.3
and (b) ij = 0.6 for i = j. Here, M = 4.
Remark 7: In Fig. 4 (c), SLER maximizing beamforming is outperformed by MLB at around a
harvested energy of 450W. This comes from the fact that the beam directions of energy transmitters
are determined independently, because they do not share the CSIT. Accordingly, even though the beam
directions are determined to maximize their own SLERs, the aggregate interference may not be optimized.
That is, because the aggregate interference at the ID receivers is determined by both the directions and
the powers of the energy beams, the information rate performance becomes increasingly sensitive to the
beam steering and the power reduction as the number of energy transmitters increases. Therefore, in
Fig. 4 (c), the SLER maximizing beamforming with a small transmit power may be outperformed by
the MLB with a large transmit power under the same harvesting energy (e.g., 450W). Note that, in
our simulations, we rst xed the beam directions of all energy transmitters as in Section IV-A, and
then reduce the powers of the energy transmitters in Algorithm 1. However, as stated in Remark 4, in
the SLER maximizing beamforming, the beam direction softly bridges MEB and MLB depending on
the scalar value multiplied by the identity matrix of the denominator in (43). Therefore, if the harvested
energy is enough in Step 2.b of Algorithm 1, before reducing the power of energy transmitters, we can
tilt the beam to reduce the interference to the ID receivers by updating the energy beamforming vectors.
Here, they can be updated by computing the GSVD of the matrix pair
((H
(k)
11
)
H
H
(k)
11
, (H
(k)
21
)
H
H
(k)
21
+
n
max(

E/K
1
P |H
(k)
11
|
2
, 0)I
M
),
with a decaying factor (0, 1). In Fig. 8, the R-E curve of a new SLER beamforming with beam tilting
October 28, 2013 DRAFT
28
6 7 8 9 10 11 12 13
0
100
200
300
400
500
600
700
800
Rate (bits/s/Hz)
E
n
e
r
g
y

(


W
)
RE tradeoff curves when M=4 and (K, K
1
) =(4,3)


Max. energy beamforming
Min. leakage beamforming
SLER maximizing beamforming
SLER maximizing beamforming with beam tilting
(a)
5 6 7 8 9 10 11 12 13
0
200
400
600
800
1000
1200
Rate (bits/s/Hz)
E
n
e
r
g
y

(


W
)
RE tradeoff curves when M=4 and (K, K
1
) =(5,4)


Max. energy beamforming
Min. leakage beamforming
SLER maximizing beamforming
SLER maximizing beamforming with beam tilting
(b)
Fig. 8. R-E tradeoff curves for SLER maximizing beamforming with/without beam tilting when (a) (K, K1) = (4, 3) (b)
(K, K1) = (5, 4).
and power allocation is compared with that of SLER beamforming with only a power allocation when
(K, K
1
) = (4, 3), (5, 4). In our simulation, is xed as 0.9. We can see that the new SLER beamforming
scheme exhibits better performance than the SLER without beam tilting and the effect of beam tilting
is more apparent for (K, K
1
) = (5, 4). These evaluations show that further beamforming enhancements
are possible by better jointly designing beam directions and power in K-user MIMO IFC.
VI. CONCLUSION
In this paper, we have investigated the joint wireless information and energy transfer in K-user MIMO
IFC. The exact R-E curve for general K-user MIMO IFC is not known, but we have shown that the
optimal energy transmitters strategies for three different scenarios - i) multiple EH receivers and a single
ID receiver, ii) multiple IDs and a single EH, and iii) multiple IDs and multiple EHs - also become
optimal for the properly-transformed two-user MIMO IFC. Accordingly, we have found a common
necessary condition of the optimal transmission strategy, in which all the transmitters transferring energy
exploit a rank-one energy beamforming. Furthermore, given the rank-one beamforming at the energy
transmitters, we have also developed the iterative algorithm for the non-convex optimization problem of
the achievable rate-energy region. By comparing three different rank-one beamforming - MEB, MLB, and
SLER maximizing beamforming, we can nd that MEB and MLB either maximize the harvested energy
or the information rate, but SLER maximizing beamforming scheme pursues both in a well-balanced
way, showing a wider R-E region than that achieved by both MEB and MLB. Interestingly, when the
October 28, 2013 DRAFT
29
number of information transmitters increases (interference-limited information transfer system where the
interference due the information transmitters is dominant), the optimal strategy at the energy transmitters
becomes close to MEB method. In contrast, when the number of energy transmitters increases, the beam
steering as well as the power reduction affects the information rate performance, which lead us to develop
the SLER maximizing beamforming with beam tilting. Finally, we have proposed an efcient SLER-based
EH transceiver selection method which improves the achievable rate-energy region further.
REFERENCES
[1] Study on Enhancements for MTC, 3GPP TR Std. TR 22.888, v.0.4.0, 2011.
[2] A. Kurs, A. Karalis, R. Moffatt, J. D. Joannopoulos, P. Fisher, and M. Soljacic, Wireless power transfer via strongly
coupled magnetic resonances, Science, vol. 137, no. 83, pp. 8386, Sept. 2007.
[3] M. Pi nuela, P. Mitcheson, and S. Lucyszyn, Ambient RF energy harvesting in urban and semi-urban environments, IEEE
Trans. Microwave Theory Tech., vol. 61, no. 7, pp. 27152726, July 2013.
[4] R. J. M. Vullers, R. V. Schaijk, I. Doms, C. V. Hoof, and R. Merterns, Micropower energy harvesting, Solid-State
Electronics, vol. 53, no. 7, pp. 684693, July 2009.
[5] T. Le, K. Mayaram, and T. Fiez, Efcient far-eld radio frequency energy harvesting for passively powered sensor
networks, IEEE J. Solid-State Circuits, vol. 43, no. 5, pp. 12871302, May 2008.
[6] R. Zhang and C. K. Ho, MIMO broadcasting for simultaneous wireless information and power transfer, IEEE Trans.
Wireless Commun., vol. 12, no. 5, pp. 19892001, May 2013.
[7] L. Liu, R. Zhang, and K. Chua, Wireless information transfer with opportunistic energy harvesting, IEEE Trans. Wireless
Commun., vol. 12, no. 1, pp. 288300, Jan. 2013.
[8] K. Huang and E. G. Larsson, Simultaneous information-and-power transfer for broadband wireless systems, IEEE
Transactions on Singla Processing, to be published, 2013.
[9] O. Ozel, K. Tutuncuoglu, J. Yang, S. Ulukus, and A. Yener, Transmission with energy harvesting nodes in fading wireless
channels: Optimal policies, IEEE J. Select. Areas Commun., vol. 29, no. 8, pp. 17321743, Sept. 2011.
[10] R. Rajesh, V. Sharma, and P. Viswanath, Information capacity of energy harvesting sensor nodes, in Proc. IEEE
International Symposium on Information Theory, 2011, July 2011, pp. 23632367.
[11] , Information capacity of an energy harvesting sensor node, submitted to IEEE Transactions on Information Theory,
http://arxiv.org/abs/1212.3177, 2012.
[12] K. Ishibashi, H. Ochiai, and V. Tarokh, Energy harvesting cooperative communications, in Proc. IEEE International
Symposium on Personal, Indoor and Mobile Radio Communications, 2012, Sept. 2012, pp. 18191823.
[13] A. A. Nasir, X. Zhou, S. Durrani, and R. A. Kennedy, Relaying protocols for wireless energy harvesting and information
processing, IEEE Trans. Wireless Commun., vol. 12, no. 7, pp. 36223636, July 2013.
[14] Y. Luo, J. Zhang, and K. B. Letaief, Optimal scheduling and power allocation for two-hop energy harvesting
communication systems, IEEE Trans. Wireless Commun., vol. 12, no. 9, pp. 47294741, Sept. 2013.
[15] J. Xu, L. Liu, and R. Zhang, Multiuser MISO beamforming for simultaneous wireless information and power transfer,
submitted to IEEE Transactions on Signal Processing, http://arxiv.org/abs/1303.1911, 2013.
[16] X. Zhou, R. Zhang, and C. K. Ho, Wireless information and power transfer in multiuser OFDM systems, submitted to
IEEE Transactions on Wireless Communications, http://arxiv.org/abs/1308.2462, 2013.
[17] D. W. K. Ng, E. S. Lo, and R. Schober, Energy-efcient resource allocation in multiuser OFDM systems with wireless
information and power transfer, in Proc. IEEE Wireless Communications and Networking Conference, 2013, Apr. 2013,
pp. 38233828.
[18] K. Tutuncuoglu and A. Yener, Sum-rate optimal power policies for energy harvesting transmitters in an interference
channel, Journal of Communications and Networks, vol. 14, no. 2, pp. 151161, Apr. 2012.
[19] , Transmission policies for asymmetric interference channels with energy harvesting nodes, in Proc. IEEE
International Workshop on Computational Advances in Multi-sensor Adaptive Processing, 2011, Dec. 2011, pp. 197200.
[20] K. Huang and V. K. N. Lau, Enabling wireless power transfer in cellular networks: architecture, modeling and deployment,
submitted to IEEE Transactions on Signal Processing, http://arxiv.org/abs/1207.5640, 2012.
[21] C. Shen, W. Li, and T. Chang, Wireless information and energy transfer in multi-antenna interference channel, submitted
to IEEE Transactions on Signal Processing, http://xxx.tau.ac.il/abs/1308.2838, 2013.
[22] J. Park and B. Clerckx, Joint wireless information and energy transfer in a two-user mimo interference channel, IEEE
Trans. Wireless Commun., vol. 12, no. 8, pp. 42104221, Aug. 2013.
[23] X. Zhou, R. Zhang, and C. K. Ho, Wireless information and power transfer: architecture design and rate-energy tradeoff,
submitted to IEEE Transactions on Communications, http://arxiv.org/abs/1205.0618, 2012.
October 28, 2013 DRAFT
30
[24] S. L. Loyka, Channel capacity of MIMO architecture using the exponential correlation matrix, IEEE Commun. Lett.,
vol. 5, no. 9, pp. 369371, Sept. 2001.
[25] G. Scutari, D. P. Palomar, and S. Barbarossa, The MIMO iterative waterlling algorithm, IEEE Trans. Signal Processing,
vol. 57, no. 5, pp. 19171935, May 2009.
[26] C. Shen, W. Li, and T. Chang, Simultaneous information and energy transfer: A two-user MISO interference channel
case, in Proc. IEEE GLOBECOM, 2012, Dec. 2012.
[27] R. A. Horn, N. H. Rhee, and W. So, Eigenvalue inequalities and equalities, Linear Algebra and Its Applications, vol.
270, no. 1-3, pp. 2944, Feb. 1998.
[28] D. A. Harville, Matrix algebra from a statisticians perspective. Berlin: Springer, 2008.
[29] J. Park, J. Chun, and H. Park, Generalised singular value decomposition-based algorithm for multi-user MIMO linear
precoding and antenna selection, IET Commun., vol. 4, no. 16, pp. 18991907, Nov. 2010.
[30] J. Park, J. Chun, and B. Jeong, Efcient multi-user MIMO precoding based on GSVD and vector perturbation, Signal
Processing, vol. 92, no. 2, pp. 611615, Feb. 2012.
[31] J. R. Magnus and H. Neudecker, Matrix Differential Calculus with Applications in Statistics and Economics, 3rd ed.
Chichester: John Wiley & Sons, 2007.
[32] S. Boyd and L. Vandenberghe, Convex Optimization, 7th ed. New York: Cambridge University Press, 2009.
[33] X. Zhao, P. B. Luh, and J. Wang, Surrogate gradient algorithm for Lagrangian relaxation, Journal of Optimization Theory
and Applications, vol. 100, no. 3, pp. 699712, Mar. 1999.
[34] M. S. Bazaraa, H. D. Sherali, and C. M. Shetty, Nonlinear Programming: Theory and Algorithms, 3rd ed. New York:
John Wiley and Sons, 1993.
[35] R. Zhang, Y. Liang, and S. Cui, Dynamic resource allocation in cognitive radio networks, IEEE Signal Processing Mag.,
vol. 27, no. 3, pp. 102114, May 2010.
[36] F. Negro, S. P. Shenoy, I. Ghauri, and D. T. Slock, Weighted sum rate maximization in the MIMO interference channel,
in Proc. IEEE International Symposium on Personal, Indoor and Mobile Radio Communications, 2010, Sept. 2010, pp.
684689.
[37] H. Al-Shatri and T. Weber, Optimizing power allocation in interference channels using D.C. programming, in Proc. the
8th International Symposium on Modeling and Optimization in Mobile, Ad Hoc and Wireless Networks, May 2010, pp.
360366.
[38] Y. Xu, T. Le-Ngoc, and S. Panigrahi, Global concave minimization for optimal spectrum balancing in multi-user DSL
networks, IEEE Trans. Signal Processing, vol. 56, no. 7, pp. 28752884, July 2008.
October 28, 2013 DRAFT

You might also like