SPECIAL SECTION ON INDEX MODULATION TECHNIQUES

FOR NEXT-GENERATION WIRELESS NETWORKS

Received August 30, 2017, accepted September 25, 2017, date of publication October 12, 2017,
date of current version November 14, 2017.
Digital Object Identifier 10.1109/ACCESS.2017.2762524

On Simultaneous Wireless Information and

Power Transfer for Receive Spatial Modulation
CHIEN-CHUN CHENG1 , (Student Member, IEEE), MARCO DI RENZO 2 , (Senior Member, IEEE),
FABIO GRAZIOSI1 , (Member, IEEE), AND ALESSIO ZAPPONE2 , (Senior Member, IEEE)
1 Department of Information Engineering, Computer Science and Mathematics, University of L’Aquila, 67100 L’Aquila, Italy
2 Laboratoire des Signaux et Systèmes, CNRS, CentraleSupélec, Univ Paris Sud, Université Paris-Saclay, 91192 Gif-sur-Yvette, France
Corresponding author: Marco Di Renzo (marco.direnzo@l2s.centralesupelec.fr)
The work of C.-C. Cheng and F. Graziosi was supported in part by the Italian Government through CIPE under Grant 135 and in part by the
project INnovating CIty Planning through Information and CommunicaTion Technologies. The work of M. Di Renzo was supported by the
Agence Nationale de la Recherche Scientifique through the research project SpatialModulation (Société de l’Information et de la
Communication–Action Plan 2015). The work of A. Zappone was supported by the H2020 MSCA IF BESMART project under Grant
749336.

ABSTRACT In this paper, we study the performance of receive spatial modulation (RSM) with simultaneous
wireless information and power transfer (SWIPT) capabilities. RSM is a multi-antenna modulation scheme,
where the information bits are encoded into complex constellation symbols and the indices of the receive
antennas. We show how RSM can be combined with SWIPT, by allowing the receivers to increase their data
rate and, at the same time, to recharge their batteries. An optimization problem is formulated in order to
optimize the fraction of transmit power to be used for information decoding and energy harvesting, as well
as the covariance matrix of the energy waveform. Efficient numerical algorithms to tackle the associated
optimization problems are proposed. Our analysis shows that RSM-SWIPT is a flexible transmission scheme
that is capable of achieving different rate-energy demands. RSM-SWIPT, in particular, can be configured to
leverage only the energy waveform for transmitting information data and power simultaneously. Compared
with conventional SWIPT-enabled multiple-antenna systems, the proposed RSM-SWIPT scheme increases
the amount of harvested energy for low values of the rate and avoids the need of using energy cancellation
algorithms if information and power are transmitted only through the energy waveform.

INDEX TERMS Receive spatial modulation, simultaneous wireless information and power transfer,
optimization.

I. INTRODUCTION energy efficiency [6]. A modification of SM that simplifies

Multiple-input multiple-output (MIMO) wireless systems are its incorporation in the broadcast channel is receiver spatial
capable of providing rate, diversity and array gains, by using modulation (RSM) [41], [42]. In RSM, by using linear pre-
transmission techniques such as spatial multiplexing (SMX), coding, the information bits are transferred to the receiver
receive combining, and transmit beamforming [1]–[3]. As a by applying the concept of spatial modulation at the receive
result, MIMO systems are widely used for wireless informa- antennas, i.e., by encoding the bits onto the indices of the
tion transfer (MIMO-WIT) [4]–[6], wireless power transfer receive antennas. Recently, RSM has been studied for appli-
(MIMO-WPT) [7]–[18], and simultaneous wireless informa- cation to a wide range of different communication scenarios
tion and power transfer (MIMO-SWIPT) [19]–[27]. [43]–[49]. Stavridis et al. [43] studied RSM in the context of
MIMO-WIT has a long history of development. It trans- virtual MIMO relaying systems. In [44], RSM is combined
mits multiple data streams from all the available antennas. with SMX for increasing the bandwidth efficiency. In [45],
A more recent alternative for increasing the rate at a low differential RSM is analyzed. In [46], energy pattern RSM
complexity is spatial modulation (SM) [5], [28]–[40], which is introduced for information and power transfer. In addition,
takes advantage of the whole antenna array at the transmitter, the benefits of RSM for improving the secrecy of wireless
while using a limited number of radio-frequency (RF) chains. systems are studied in [48] and [49]. Recent research contri-
In SM, additional information bits are mapped onto a spatial butions can be found in [50], [51] and [52].
constellation diagram, where each constellation symbol is Compared with MIMO-WIT, MIMO-WPT has a shorter
made of a subset of indices of the antenna elements. This is history of development. Most research efforts [7]–[9] are con-
shown to reduce the circuitry complexity and to enhance the cerned with the design of beamforming strategies and channel

2169-3536
2017 IEEE. Translations and content mining are permitted for academic research only.
23204 Personal use is also permitted, but republication/redistribution requires IEEE permission. VOLUME 5, 2017
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
C.-C. Cheng et al.: On SWIPT for RSM

state information (CSI) feedback mechanisms for improving

the charging efficiency. The main aim is to maximize the
harvested energy via power allocation and by taking into
account the tradeoff between channel estimation and power
transfer. An important challenge of WPT is to increase the
RF-DC conversion efficiency of the rectifying circuits [10].
Recent works in this field of research have focused their
attention on the performance of the rectifiers by considering
different types of signals, including multi-sine signals [11],
chaotic waveforms [12], and modulated orthogonal frequency
division multiple access (OFDM) signals [13]. Recently, it
was reported that signals with a high peak-to-average power
ratio (PAPR) can improve the efficiency [14]. FIGURE 1. The RSM-SWIPT system for user k.
MIMO-SWIPT is a concept that jointly combines WIT and
WPT by sharing a common circuitry. It offers the possibility II. SYSTEM MODEL
of delivering controllable and on-demand wireless informa- We consider a multi-user MIMO downlink system model as
tion and energy concurrently, which enables one to design depicted in Fig. 1. The transmitter emits an energy signal and
low-cost wireless systems for self-sustainable operation and an information (or RSM signal) with the aid of ZF precoding.
without modifying the hardware at the transmitter [19]. At the receiver, the signal is input to a rectifier for energy
However, it is known that optimizing SWIPT-enabled sys- harvesting and to an information decoder for retrieving the
tems is a challenging task [20]. More specifically, the optimal information bits. The following notation is used: NU is the
rate is achieved by using SMX with a Gaussian codebook and number of users, NR is the number of receive antennas, NT =
the water-filling power allocation scheme [21]. In contrast, NR NU is the number of transmit antennas, P ∈ CNT ×NR NU is
the maximum harvested power is achieved by using transmit the precoding matrix at the transmitter, ρ is the power split-
beamforming, where the transmitted signal depends on the ting ratio at the receiver, and H ∈ CNR NU ×NT is the channel
strongest eigenmode of the transmit covariance matrix [15]. matrix. In addition, I denotes the identity matrix, (·)H and (·)

T
The optimization of SWIPT systems brings about a new ·
denote Hermitian transpose and transpose, respectively, ·
optimization space and tradeoffs.
denotes the binomial coefficient, tr(·) denotes the trace oper-
In the present paper, we study and optimize RSM-SWIPT.
ator, det(·) denotes the determinant of a matrix, E{·} denotes
The main contributions made by this paper are as
the expectation operator, and ⊗ denotes the Kronecker prod-
follows:
uct.
• The first RSM-SWIPT scheme is proposed and ana-
lyzed.
A. INFORMATION DECODER
• A general model for the energy waveform is introduced.
The signal at the input of the information decoder can be
• The rate bounds of the proposed RSM-SWIPT scheme
written, for the NU users, as follows:
are computed. √
y = ρ HP (xI + wP ) + n (1)
• Efficient algorithms for optimizing RSM-SWIPT are
where y ∈ CNR NU ×1 is the received signal, xI ∈ CNR NU ×1 is
proposed.
the information (RSM) signal, wP ∈ CNR NU ×1 is the energy
Compared with conventional MIMO systems that employ
signal, and n ∈ CNR NU ×1 is the additive white Gaussian
SWIPT based on zero-forcing (ZF) precoding, RSM-SWIPT
noise (AWGN). By assuming ZF precoding, P can be formu-
is shown to have a better flexibility for optimizing the rate-
lated as follows:
energy region. In particular, RSM-SWIPT provides one with  −1
an additional option for information and energy transfer by P = HH HHH (2)
exploiting only the energy waveform and, hence, by encoding As for the kth user, the received signal simplifies as
the information bits only onto the indices of the receive anten- follows: √
nas. For high energy demands, this results in a system that yk = ρ (xk + wk ) + nk (3)
is capable of providing a non-zero rate while conventional where xk ∈ CNR ×1 is the RSM signal, wk ∈ CNR ×1 is the
MIMO-SWIPT systems are shown to offer a rate that is close energy signal, and nk ∼ CN (0, σ 2 I) ∈ CNR ×1 is the AWGN
to zero. where each entry of the vector has variance σ 2 .
The rest of this paper is organized as follows. In Section II,
the system model is introduced. In Section III and IV, B. RSM SIGNAL
RSM-SWIPT with receivers that can and cannot remove the Let NS be the number of information streams to be transmit-
energy signal for information decoding are studied, respec- ted, with NS ≤ NR . The RSM signal xk is a sparse vector,
tively. In Section V, setups with a large RSM codebook are which has NS ≤ NR non-zero elements and NR − NS zero
analyzed. In Section VI, numerical illustrations are shown elements. It can be formulated as follows:
and discussed. Finally, Section VII concludes the paper. xk = Bk sk (4)

VOLUME 5, 2017 23205

 C.-C. Cheng et al.: On SWIPT for RSM

where Bk is a NR × NS binary matrix that is referred to  as and are sufficiently powerful. In this section, as an opti-
NR mistic benchmark (best case), we assume that the information
RSM codeword. The number of matrices Bk is NB = .
NS receivers are capable of removing the energy signal.
Also sk is a NS × 1 vector that collects the NS streams that are
transmitted, where sk ∼ N (0, I). For example, if NS = 2 and A. CAPACITY BOUND
NR = 4, the matrix Bk that corresponds to transmitting the By assuming an ideal cancellation of the energy signal, the
two streams from the first received signal in (3) can be re-written as follows:
 and the third antenna is as follows: √ √
yk = ρxk + nk = ρ (Bk sk ) + nk
T
1 0 0 0 (8)
Bk = The channel capacity is obtained by maximizing the
0 0 1 0
mutual information I(xk ; yk ) and can be obtained as follows:
C. ENERGY HARVESTER I(xk ; yk ) = H(yk ) − H(yk |xk )
The rectifier converts the received signal into a direct cur-
= H(yk ) − H(nk )
rent (DC) and then energy. The signal at the input of the
rectifier can be written ≤ H(yk , Bk ) − H(nk )
p as follows: = H(yk |Bk ) + H(Bk ) − H(nk ) , Rk (9)
rk = 1 − ρ (xk + wk ) + n0k (5)
0
where nk is the associated AWGN vector. where H(·) denotes the entropy function, and the inequality
The average harvested is obtained by taking into account that H(yk , Bk ) ≥ H(yk )
 energy can be formulated as [10]:
[53, Lemma 2.3.2] and by using the chain rule of the
Qk = α0 + α2 E{tr rk rH k } joint entropy, i.e., H(yk , Bk ) = H(yk |Bk ) + H(Bk )
    
= α0 + α2 (1 − ρ) E{tr xk xH H in [53, Lemma 5.3.1].
k } + E{tr wk wk }
  If x is a vector of circularly-symmetric and zero-mean com-
+ α2 E{tr n0k n0H
k }
plex random variables with covariance matrix 6, its entropy
NB
! is equal to H(x) = log det(π e6) [54]. Therefore, we obtain
1 X the following expressions of the entropies:
= α2 (1 − ρ)tr H H
Bk,i E{sk sk }Bk,i NB
NB 1 X
i=1 H(yk |Bk ) = H(yk |Bk.i )
+ α2 (1 − ρ)tr (SE ) + α0 + α2 σ 2 NR NB
i=1
NB
= α2 (1 − ρ)tr (6 B + SE ) + %k (6) 1 X
log det π e8k,i


=
where α0 and α2 are constants that depend on the specific rec- NB
i=1
tifier being used, and the definitions 6 B = N1B N H
P B
i=1 Bk,i Bk,i 1
H(Bk ) = − log( ) = log(NB )
and %k = α0 + α2 σ 2 NR are used. 6 B is an Hermitian matrix. NB 
H(nk ) = log det π eσ 2 I
D. ENERGY SIGNAL
The energy signal wk is assumed to be a zero-mean where 8k,i = ρBk,i BHk,i + σ I.
2
Gaussian random variable with covariance matrix SE , i.e., Thus, the channel capacity can be formulated as:
wk ∼ CN (0, SE ). The covariance matrix SE and the power CICRSM ≤ max Rk
ρ
splitting factor ρ are the two parameters to be optimized for
NB
jointly maximizing the harvested energy and rate. 1 X
log det 8k,i + ξk


= max (10)
ρ NB
E. TRANSMIT POWER i=1
The transmitpower can be written as follows:  where ξk = log(NB ) − log det(π eσ 2 I) is a constant.
In this case, ρ is the only parameter of 8k,i that needs to
 
P = E tr P (xI + wP ) (xI + wP )H PH
be optimized.
     
H H H H B. PROBLEM FORMULATION
= E tr PxI xI P + E tr PwP wP P
        Our objective is to maximize the sum rate subject to a given
H H energy harvesting requirement, Q0 , and a transmit power
= tr PE xI xI P + tr PE wP wP PH
H
constraint, P0 . More specifically, the optimization problem
can be formulated as follows:
   
= tr P (I ⊗ 6 B ) PH + tr P (I ⊗ SE ) PH NU
X
  P1 : max Rk
= tr P (I ⊗ (6 B + SE )) PH (7) ρ,SE
k=1
In the sequel, the optimization problems are formulated by s.t. Qk ≥ Q0 , for k = 1, . . . , NU
assuming a power constraint on P.
P ≤ P0 , SE  0, 0 ≤ ρ ≤ 1
III. IDEAL CANCELLATION OF THE ENERGY SIGNAL P1 is a non-convex optimization problem due to the cou-
Since the energy signal wk is deterministic and does not pled parameters SE and ρ in the harvested energy con-
contain any information, it could be completely removed at straint [55]. Nevertheless, it can be reformulated as a convex
the information receivers if they are appropriately designed problem. To this end, let us define the new variable XE =

23206 VOLUME 5, 2017

C.-C. Cheng et al.: On SWIPT for RSM

ρSE . Replacing SE by XE in P1, with the aid of some alge- where the entropy functions are the following:
braic manipulations we obtain the equivalent optimization NB
1 X
log det π e80k,i


problem as follows: H(yk |Bk ) =
NU
NB
X i=1
max Rk (11a) 1
ρ,XE H(Bk ) = − log( ) = log(NB )
k=1 N
B 
ρ 
s.t. α2 tr (ρ6 B + XE ) + (%k − Q0 ) ≥ 0, H(n0k ) = log det π e ρSE + σ 2 I
1−ρ
k = 1, . . . , NU (11b) with 80k,i = ρBk,i BH
k,i + ρSE + σ I.
2

Therefore, the capacity can be formulated as follows:

 
tr P (I ⊗ (ρ6 B + XE )) P H
≤ ρP0 (11c)
CNCRSM ≤ max R0k
XE  0, 0 ≤ ρ ≤ 1 (11d) ρ,SE
The constraints (11c) and (11d) are affine and thus pose no NB
1 X
log det 80k,i


challenge. Instead, the properties of constraint (11b) are more = max
ρ ρ,SE NB
involved to analyze because the function 1−ρ (%k − Q0 ) can i=1
be concave or convex depending on the range of ρ and on the
 
− log det ρSE + σ 2 I + ξk0 (14)
sign of %k − Q0 . To proceed further, we exploit the following
ξk0
where = log(NB ) is a constant.
remark.
In this case, ρ and SE are the optimization parameters.
Remark 1: Without loss of generality, it is possible to
assume %k − Q0 ≤ 0 in (11b). If %k − Q0 > 0, in fact,
the constraint in (11b) would be always fulfilled for any B. PROBLEM FORMULATION
resource allocation setup that fulfils the other constraints. The optimization problem can be formulated as follows:
NU
This is because α2 tr (ρ6 B + XE ) is non-negative for any X
positive semidefinite XE , 6 B , and ρ ≥ 0, α2 ≥ 0. Moreover, P2 : max R0k
ρ,SE
ρ
1−ρ ≥ 0 holds if 0 ≤ ρ ≤ 1. Therefore, if %k − Q0 > 0,
k=1

the constraint (11b) becomes redundant and can be removed s.t. Qk ≥ Q0 , for k = 1, . . . , NU
from the optimization problem. In summary, the case study P ≤ P0 , SE  0, 0 ≤ ρ ≤ 1
of interest is when %k − Q0 ≤ 0. Unlike P1, the optimization problem P2 poses the addi-
Based on Remark 1, we evince that the constraint function tional challenge that the objective function is not jointly
ρ
in (11b) is concave if 1−ρ is convex. It is easy to check that concave in ρ and SE . This is because SE and ρ appear in
ρ the difference of logarithmic functions, and the difference
the function 1−ρ is indeed convex in the feasible set of (11),
i.e., for 0 ≤ ρ ≤ 1. This can be verified by computing of two concave functions is, in general, neither concave
ρ nor convex. This necessitates more advanced algorithms to
the second-order derivative of 1−ρ and checking that it is
always non-negative for 0 ≤ ρ ≤ 1. In conclusion, (11) is a tackle P2. In the next section, we show that the minorization-
convex maximization problem and, thus, it can be optimally maximization (MM) algorithm is a convenient method to use.
and efficiently solved by using standard convex optimization
algorithms. C. MINORIZATION-MAXIMIZATION ALGORITHM
In order to tackle P2, we first re-formulate it as done for
IV. NO CANCELLATION OF THE ENERGY SIGNAL P1, i.e., by introducing the new variable XE = ρSE .
In this section, we consider a pessimistic benchmark (worst By neglecting irrelevant constants and by introducing, for
case) and assume that the energy signal cannot be removed at ease of notation, the definition, e Bk,i = Bk,i BH k,i , P2 can be
the receiver. In particular, the energy signal is processed by re-formulated as follows:
the information receiver without any pre-processing. NU NB
X 1 X  
max log det ρeBk,i + XE + σ 2 I
ρ,XE NB
k=1 i=1
 
A. CAPACITY BOUND
− log det XE + σ 2 I (15a)
Without interference cancellation, the energy signal wk is ρ
treated as noise. The received signal in (3) can be re-written s.t. α2 tr (ρ6 B + XE ) + (%k − Q0 ) ≥ 0,
1−ρ
as follows: k = 1, .. . , NU (15b)
√
yk = ρ (Bk sk ) + n0k

(12) tr P (I ⊗ (ρ6 B + XE )) PH ≤ ρP0 (15c)
√
where n0k = ρwk + nk and n0k ∼ CN (0, ρSE + σ 2 I).
XE  0, 0 ≤ ρ ≤ 1 (15d)
Similar to the case study with ideal cancellation of the
energy signal, the mutual information can be formulated as: The MM algorithm, also known as sequential optimiza-
tion or inner approximation method [56]–[58], is an iterative
I(xk ; yk ) = H(yk ) − H(n0k )
algorithm that solves, at each iteration, an auxiliary problem
≤ H(yk , Bk ) − H(n0k ) whose objective function is a lower-bound of (15a). More-
= H(yk |Bk ) + H(Bk ) − H(n0k ) , R0k (13) over, such a lower bound and its gradient must be equal,

VOLUME 5, 2017 23207

 C.-C. Cheng et al.: On SWIPT for RSM

respectively, to the true function in (15a) and to the gradient bounded over the feasible set by virtue of the Weierstrass
(0)
of the true function in (15a) at a given point XE , which is extreme value theorem. Moreover, denoting the point gen-
updated at the end of each iteration of the algorithm. erated by Algorithm I after iteration n by (ρn , XE,n ), it can
For the case at hand, more precisely, we resort to the be shown, from [58], that any limit point of the sequence
following inequality, which holds
(0)
 for any XE : {(ρn , XE,n )}n fulfills the Karush-Kuhn-Tucker (KKT)
(0)
log det XE + σ I ≤ log det XE + σ I
2 2 conditions of (15).
 
(0) (0)
+ tr (XE − XE )(XE + σ 2 I)−1 V. LARGE RSM CODEBOOK
(16) If the number of sparse RSM matrices is large, a better
which is obtained by recalling that any concave function is rate can be obtained. However, the optimization complex-
upper-bounded by its first-order Taylor expansion around any ity increases. In this section, we discuss how to reduce the
given point. associated computational complexity under the assumption
As a result, the following concave lower-bound for (15a) that the receiver is capable of removing the energy signal.
is obtained: If the energy signal cannot be removed, only approximated
NB
1 X   solutions can be found, which, for brevity, are not discussed
R0k ≥ log det ρe Bk,i + XE + σ 2 I (17) in the present paper.
NB
i=1
  The capacity in (10) needs the computation of:
(0)
− tr XE (XE + σ 2 I)−1 + ξ̄ 1 X
NB  

(0)
 U= log det ρBk,i BH k,i + σ I
2
(20)
, R̄0k XE |XE (18) NB
i=1
(0) (0) (0)
For a large size of the available codebook, e.g.,
where ξ̄ = log det(XE + σ 2 I) + log(NB ) − tr(XE (XE +

16
σ 2 I)−1 ) is a constant term with respect to the optimization NB = = 12870, the log-det function needs to be
8
variables. computed 12870 times for every instance of the optimization
Then, a sub-problem that fulfills the assumptions of the problem. A possible approach to reduce the complexity is to
MM method can be formulated as follows: use the Jensen’s inequality as follows:
NU   NB
!
(0) 1 X
X
P3 : max R̄0k XE |XE (19a) U ≤ log det ρ B̄i + σ I2
ρ,XE NB
k=1
ρ i=1
s.t. α2 tr (ρ6 B + XE ) +
   
(%k − Q0 ) ≥ 0, NS
1−ρ = log det ρ I + σ I , UJ
2
(21)
k = 1, . . . , NU (19b) NR
 H
where B̄i = Bk,i Bk,i .
tr P (I ⊗ (ρ6 B + XE )) P H
≤ ρP0 (19c)
We propose to use the sparse structure of the RSM code-
XE  0, 0 ≤ ρ ≤ 1 (19d) word to get an optimal solution that avoids using the Jensen’s
Since (19a) is jointly concave in (ρ, XE ), and the feasible inequality. Since only NS entries are non-zero in each matrix,
set can be shown to be concave with the aid of similar we have:
arguments as those used in Section III, we conclude that P3 is 1 X
NB
log det 6 k,i


a convex problem and, thus, it can be globally and efficiently U =
NB
solved by convex optimization methods. The resulting MM i=1
algorithm for solving P2 is formulated in Table 1. 1 X
NB  NS  (NR −NS ) 
= log ρ + σ 2 σ2
TABLE 1. MM algorithm to solve P2 NB
i=1
= NS log(ρ + σ 2 ) + (NR − NS ) log(σ 2 ) , UP (22)
where 6 k,i is a diagonal matrix whose diagonal elements are
the eigenvalues of the matrix in the log-det function of (20):
6 k,i = diag(ρ + σ 2 , · · · , ρ + σ 2 , σ 2 , · · · , σ 2 ) (23)
This result follows from the fact that Bk,i BH k,i is a binary
diagonal matrix, whose trace is equal to NS , that has NS
eigenvalues equal to ρ + σ 2 and NR − NS eigenvalues
equal to σ 2 . As a result, an exact closed-form expression
can be obtained without the need of using the Jensen’s
From the properties of the MM method, it can be shown inequality.
that Algorithm I in Table 1 monotonically increases the value
of the true objective function of P2 after each iteration. This VI. NUMERICAL RESULTS
implies the convergence of Algorithm I to the objective value. In this section, we analyze the performance of the proposed
Since the objective function of P2 is continuous and the optimization methods and show the so-called rate-energy
feasible set is compact, then the objective function of P2 is regions. They correspond to setting the energy harvesting

23208 VOLUME 5, 2017

C.-C. Cheng et al.: On SWIPT for RSM

TABLE 2. Simulation parameters

FIGURE 3. Rate-energy region for non-ideal cancellation receivers

(NS = 1, 2, 4).

FIGURE 2. Rate-energy region for ideal-cancellation receivers

(NS = 1, 2, 4).

requirement Q0 to a given value and then to optimizing FIGURE 4. Rate-energy region for ideal-cancellation receivers
the power splitting ratio and the covariance matrix of the (NS = NR /2).
energy signal by solving P1 and P2. Given the solution of the
optimization problem, the optimal rate is obtained and then signal: the information bits can be encoded only onto the
depicted as a function of Q0 . receive antennas, which results in a very efficient system
In Fig. 2, the rate-energy region as a function of NS and design from the point of view of the power transfer. This is the
under the assumption of ideal cancellation receivers is illus- main benefit of RSM-SWIPT compared with conventional
trated. In particular, the setup NS = NR = 4 corresponds to MIMO-SWIPT. It is worth emphasizing that, based on (1),
the conventional MIMO-SWIPT case. Also, we compare the there is no signal at the input of the information decoder if
solution obtained by solving P1 with the proposed algorithm ρ = 0. This ambiguity originates from the fact that (14) is an
and a brute-force search. We observe a good match. The upper-bound. If ρ = 0, in practice, an information receiver is
numerical results highlight that the optimal value of NS to not needed and the information bits encoded onto the indices
use, i.e., the number of information streams, depends on Q0 . of the receive antennas can be retrieved directly from the
Only if Q0 is small, the optimal setup corresponds to the power received at the receive antennas.
conventional MIMO-SWIPT case. If the energy requirements In Fig. 4, we study the rate-energy regions with ideal-
are high, on the other hand, it is more convenient to reduce cancellation receivers by assuming a large number of RSM
the number of information streams and to encode more bits matrices. In particular, we assume NS = NR /2 and
onto the receive antennas. This result originates from the fact NR = 8, 12, 16. The figure confirms that the proposed closed-
that more power is allocated to each stream if NS decreases. form for computing the objective function is accurate and,
In Fig. 3, we show results similar to those reported in Fig. 2. in addition, it can reduce the computational complexity by
!
The difference is that receivers with non-ideal cancellation a factor equal to NB = NS !(NNRR−N S )!
when computing the
capabilities are assumed. As expected, a worse rate-energy expectation with respect to the number of codewords in (20).
region is obtained. In the figure, we highlight the points that As expected, the rate-energy region increases with the num-
correspond to the setups ρ = 1, i.e., all the received power ber of receive antennas.
is input to the information decoder, and ρ = 0, i.e., all the
received power is input to the energy receiver. It is interesting VII. CONCLUSION
to note, with respect to the setup shown in Fig. 2, that the In this paper, we have proposed, studied and optimized the
rate does not approach zero for NS < NR . In fact, an asymp- performance of a new RSM scheme that transmits informa-
tote that is approximately equal to NU log(NB ) is reached, tion and power simultaneously. We have introduced compu-
which can be obtained from (14) by setting ρ = 0. This is tationally efficient algorithms for performance optimization
because the RSM-SWIPT scheme is capable of transmitting and numerically validated their effectiveness. We have shown
information and power without the need of an information that RSM-SWIPT provides the flexibility of obtaining a

VOLUME 5, 2017 23209

 C.-C. Cheng et al.: On SWIPT for RSM

good rate-energy trade-off, by appropriately optimizing the [21] İ. E. Telatar, ‘‘Capacity of multi-antenna Gaussian channels,’’ Eur. Trans.
number of information streams as a function of the energy Telecommun., vol. 10, no. 6, pp. 585–595, 1999.
[22] X. Lu, P. Wang, D. Niyato, D. I. Kim, and Z. Han, ‘‘Wireless charging
harvesting requirement. Thus, it constitutes a more flexible technologies: Fundamentals, standards, and network applications,’’ IEEE
option compared with conventional MIMO-SWIPT schemes Commun. Surveys Tuts., vol. 18, no. 2, pp. 1413–1452, 2nd Quart., 2015.
where the number of information streams is equal to the num- [23] X. Zhou, R. Zhang, and C. K. Ho, ‘‘Wireless information and power trans-
fer: Architecture design and rate-energy tradeoff,’’ IEEE Trans. Commun.,
ber of receive antennas regardless of the energy harvesting vol. 61, no. 11, pp. 4754–4767, Nov. 2013.
requirements. [24] S. Sanayei and A. Nosratinia, ‘‘Antenna selection in MIMO systems,’’
IEEE Commun. Mag., vol. 42, no. 10, pp. 68–73, Oct. 2004.
REFERENCES [25] A. F. Molisch and M. Z. Win, ‘‘MIMO systems with antenna selection,’’
[1] G. J. Foschini and M. J. Gans, ‘‘On limits of wireless communications IEEE Commun. Mag., vol. 5, no. 1, pp. 46–56, Mar. 2004.
in a fading environment when using multiple antennas,’’ Wireless Pers. [26] S. Zhao, Q. Li, Q. Zhang, and J. Qin, ‘‘Antenna selection for simultaneous
Commun., vol. 6, no. 3, pp. 311–335, Mar. 1998. wireless information and power transfer in MIMO systems,’’ IEEE Com-
[2] D. N. C. Tse, P. Viswanath, and L. Zheng, ‘‘Diversity-multiplexing tradeoff mun. Lett., vol. 18, no. 5, pp. 789–792, May 2014.
in multiple-access channels,’’ IEEE Trans. Inf. Theory, vol. 50, no. 9, [27] B. Koo and D. Park, ‘‘Interference alignment and wireless energy transfer
pp. 1859–1874, Sep. 2004. via antenna selection,’’ IEEE Commun. Lett., vol. 18, no. 4, pp. 548–551,
[3] A. J. Paulraj, D. A. Gore, R. U. Nabar, and H. Bolcskei, ‘‘An overview of Apr. 2014.
MIMO communications—A key to gigabit wireless,’’ Proc. IEEE, vol. 92, [28] M. Di Renzo, H. Haas, A. Ghrayeb, S. Sugiura, and L. Hanzo, ‘‘Spatial
no. 2, pp. 198–218, 2004. modulation for generalized MIMO: Challenges, opportunities, and imple-
[4] J. Xu and L. Qiu, ‘‘Energy efficiency optimization for MIMO broadcast mentation,’’ Proc. IEEE, vol. 102, no. 1, pp. 56–103, Jan. 2014.
channels,’’ IEEE Trans. Wireless Commun., vol. 12, no. 2, pp. 690–701, [29] P. Yang, M. Di Renzo, Y. Xiao, S. Li, and L. Hanzo, ‘‘Design guidelines for
Feb. 2013. spatial modulation,’’ IEEE Commun. Surveys Tuts., vol. 17, no. 1, pp. 6–26,
[5] M. Di Renzo, H. Haas, and P. M. Grant, ‘‘Spatial modulation for multiple- 1st Quart., 2015.
antenna wireless systems: A survey,’’ IEEE Commun. Mag., vol. 49, no. 12, [30] P. Yang et al., ‘‘Single-carrier SM-MIMO: A promising design for broad-
pp. 182–191, Dec. 2011. band large-scale antenna systems,’’ IEEE Commun. Surveys Tuts., vol. 18,
[6] A. Stavridis, S. Sinanovic, M. Di Renzo, and H. Haas, ‘‘Energy evaluation no. 3, pp. 1687–1716, 3rd Quart., 2016.
of spatial modulation at a multi-antenna base station,’’ in Proc. IEEE Veh. [31] M. Di Renzo, H. Haas, A. Ghrayeb, S. Sugiura, and L. Hanzo, Spatial
Technol. Conf., Sep. 2013, pp. 1–5. Modulation for Multiple–Antenna Communication. Hoboken, NJ, USA:
[7] G. Yang, C. K. Ho, and Y. L. Guan, ‘‘Dynamic resource allocation for Wiley, 2016.
multiple-antenna wireless power transfer,’’ IEEE Trans. Signal Process., [32] E. Basar, M. Wen, R. Mesleh, M. Di Renzo, Y. Xiao, and H. Haas,
vol. 62, no. 14, pp. 3565–3577, Jul. 2014. ‘‘Index modulation techniques for next-generation wireless networks,’’
[8] Y. Zeng and R. Zhang, ‘‘Optimized training design for wireless IEEE Access, vol. 5, pp. 16693–16746, 2017.
energy transfer,’’ IEEE Trans. Commun., vol. 63, no. 2, pp. 536–550, [33] M. Di Renzo and H. Haas, ‘‘Bit error probability of SM-MIMO over
Feb. 2015. generalized fading channels,’’ IEEE Trans. Veh. Technol., vol. 61, no. 6,
[9] Y. Zeng and R. Zhang, ‘‘Optimized training for net energy maximization in pp. 1124–1144, Mar. 2012.
multi-antenna wireless energy transfer over frequency-selective channel,’’ [34] N. Serafimovski, ‘‘Practical implementation of spatial modulation,’’ IEEE
IEEE Trans. Commun., vol. 63, no. 6, pp. 2360–2373, Jun. 2015. Trans. Veh. Technol., vol. 62, no. 9, pp. 4511–4523, Nov. 2013.
[10] A. Boaventura, D. Belo, R. Fernandes, A. Collado, A. Georgiadis, and [35] A. Younis, S. Sinanovic, M. Di Renzo, R. Mesleh, and H. Haas, ‘‘Gen-
N. B. Carvalho, ‘‘Boosting the efficiency: Unconventional waveform eralised sphere decoding for spatial modulation,’’ IEEE Trans. Commun.,
design for efficient wireless power transfer,’’ IEEE Microw. Mag., vol. 16, vol. 61, no. 7, pp. 2805–2815, Jul. 2013.
[36] M. D. Renzo and H. Haas, ‘‘On transmit diversity for spatial modulation
no. 3, pp. 87–96, Apr. 2015.
[11] A. J. S. Boaventura, A. Collado, A. Georgiadis, and N. B. Carvalho, MIMO: Impact of spatial constellation diagram and shaping filters at the
‘‘Spatial power combining of multi-sine signals for wireless power trans- transmitter,’’ IEEE Trans. Veh. Technol., vol. 62, no. 6, pp. 2507–2531,
mission applications,’’ IEEE Trans. Microw. Theory Techn., vol. 62, no. 4, Jul. 2013.
[37] K. Ntontin, M. Di Renzo, A. I. Perez-Neira, and C. Verikoukis, ‘‘A low-
pp. 1022–1030, Apr. 2014.
[12] A. Collado and A. Georgiadis, ‘‘Improving wireless power transmission complexity method for antenna selection in spatial modulation systems,’’
efficiency using chaotic waveforms,’’ in IEEE Int. Microw. Symp. Dig., IEEE Commun. Lett., vol. 17, no. 12, pp. 2312–2315, Dec. 2013.
[38] M. Le, V. Ngo, H. Mai, X. Tran, and M. Di Renzo, ‘‘A low-complexity
Jun. 2012, pp. 1–3.
[13] A. Collado and A. Georgiadis, ‘‘Optimal waveforms for efficient wireless method for antenna selection in spatial modulation systems,’’ IEEE Trans.
power transmission,’’ IEEE Microw. Wireless Compon. Lett., vol. 24, no. 5, Commun., vol. 62, no. 1, pp. 85–99, Dec. 2014.
[39] M. Di Renzo and H. Haas, ‘‘Transmit-diversity for spatial modulation
pp. 354–356, May 2014.
[14] A. Boaventura, A. Collado, N. B. Carvalho, and A. Georgiadis, ‘‘Optimum (SM): Towards the design of high-rate spatially-modulated space-time
behavior: Wireless power transmission system design through behavioral block codes,’’ in Proc. IEEE Int. Conf. Commun., Jun. 2011, pp. 1–6.
[40] T. Fath, H. Haas, M. Di Renzo, and R. Meslesh, ‘‘Spatial modulation
models and efficient synthesis techniques,’’ IEEE Microw. Mag., vol. 14,
applied to optical wireless communications in indoor LOS environments,’’
no. 2, pp. 26–35, Mar. 2013.
[15] R. Zhang and C. K. Ho, ‘‘MIMO broadcasting for simultaneous wireless in Proc. IEEE Global Commun. Conf., Dec. 2011, pp. 1–5.
[41] R. Zhang, L.-L. Yang, and L. Hanzo, ‘‘Generalised pre-coding aided
information and power transfer,’’ IEEE Trans. Wireless Commun., vol. 12,
spatial modulation,’’ IEEE Trans. Wireless Commun., vol. 12, no. 11,
no. 5, pp. 1989–2001, May 2013.
[16] B. Clerckx, E. Bayguzina, D. Yates, and P. D. Mitcheson, ‘‘Waveform pp. 5434–5443, Nov. 2013.
[42] R. Zhang, L.-L. Yang, and L. Hanzo, ‘‘Error probability and capacity
optimization for wireless power transfer with nonlinear energy harvester
analysis of generalised pre-coding aided spatial modulation,’’ IEEE Trans.
modeling,’’ in Proc. IEEE Int. Symp. Wireless Commun. Syst., Aug. 2015,
Wireless Commun., vol. 14, no. 1, pp. 364–375, Jan. 2015.
pp. 276–280.
[43] A. Stavridis, D. Basnayaka, S. Sinanovic, M. Di Renzo, and H. Haas,
[17] B. Clerckx. (2016). ‘‘Waveform and transceiver design for simulta-
‘‘A virtual MIMO dual-hop architecture based on hybrid spatial modula-
neous wireless information and power transfer.’’ [Online]. Available:
tion,’’ IEEE Trans. Commun., vol. 62, no. 9, pp. 3161–3179, Sep. 2014.
https://arxiv.org/abs/1607.05602
[44] C. Masouros and L. Hanzo, ‘‘Dual-layered MIMO transmission for
[18] Y. Huang and B. Clerckx. (Sep. 2016). ‘‘Large-scale multi-
increased bandwidth efficiency,’’ IEEE Trans. Veh. Technol., vol. 65, no. 5,
antenna multi-sine wireless power transfer.’’ [Online]. Available:
pp. 3139–3149, May 2016.
https://arxiv.org/abs/1609.02440
[45] J. Zheng, ‘‘Fast receive antenna subset selection for pre-coding aided
[19] X. Lu, P. Wang, D. Niyato, D. I. Kim, and Z. Han, ‘‘Wireless networks with
spatial modulation,’’ IEEE Wireless Commun. Lett., vol. 4, no. 3,
RF energy harvesting: A contemporary survey,’’ IEEE Commun. Surveys
pp. 317–320, Jun. 2015.
Tuts., vol. 17, no. 2, pp. 757–789, 2nd Quart., 2015.
[46] M. Zhang, M. Wen, X. Cheng, and L.-L. Yang, ‘‘Pre-coding aided differen-
[20] P. Grover and A. Sahai, ‘‘Shannon meets tesla: Wireless information
tial spatial modulation,’’ in Proc. IEEE Global Commun. Conf., Dec. 2015,
and power transfer,’’ in Proc. IEEE Int. Symp. Inf. Theory, Jun. 2010,
pp. 1–6.
pp. 2363–2367.

23210 VOLUME 5, 2017

C.-C. Cheng et al.: On SWIPT for RSM

[47] R. Zhang, L.-L. Yang, and L. Hanzo, ‘‘Energy pattern aided simultaneous recipient of several awards, including the 2013 IEEE-COMSOC Best Young
wireless information and power transfer,’’ IEEE J. Sel. Areas Commun., Researcher Award for Europe, Middle East, and Africa (EMEA Region), the
vol. 33, no. 8, pp. 1492–1504, Aug. 2015. 2014–2015 Royal Academy of Engineering Distinguished Visiting Fellow-
[48] F. Wu, L.-L. Yang, W. Wang, and Z. Kong, ‘‘Secret precoding-aided ship, the 2015 IEEE Jack Neubauer Memorial Best System Paper Award,
spatial modulation,’’ IEEE Commun. Lett., vol. 19, no. 9, pp. 1544–1547, the 2015–2018 CNRS Award for Excellence in Research and in Advising
Sep. 2015. Doctoral Students and the 2017 SEE-IEEE Alain Glavieux Award. He serves
[49] F. Wu, R. Zhang, L.-L. Yang, and W. Wang, ‘‘Transmitter precoding-
as the Associate Editor-in-Chief of the IEEE COMMUNICATIONS LETTERS, and
aided spatial modulation for secrecy communications,’’ IEEE Trans. Veh.
Technol., vol. 65, no. 1, pp. 467–471, Jan. 2016.
as an Editor of the IEEE TRANSACTIONS ON COMMUNICATIONS and the IEEE
[50] A. Stavridis, M. Di Renzo, P. M. Grant, and H. Haas, ‘‘On the asymptotic TRANSACTIONS ON WIRELESS COMMUNICATIONS.
performance of receive space modulation in the shadowing broadcast
channel,’’ IEEE Commun. Lett., vol. 20, no. 10, pp. 2103–2106, Oct. 2016.
[51] A. Stavridis, M. Di Renzo, and H. Haas, ‘‘Performance analysis of multi-
stream receive spatial modulation in the MIMO broadcast channel,’’ IEEE FABIO GRAZIOSI (S’96–M’97) was born in
Trans. Wireless Commun., vol. 15, no. 3, pp. 1808–1820, Mar. 2016. L’Aquila, Italy, in 1968. He received the Laurea
[52] P. Yang, Y. L. Guan, Y. Xiao, M. Di Renzo, S. Li, and L. Hanzo, ‘‘Trans- (cum laude) and Ph.D. degrees in electronic engi-
mit precoded spatial modulation: Maximizing the minimum Euclidean neering from the University of L’Aquila, L’Aquila,
distance versus minimizing the bit error ratio,’’ IEEE Trans. Wireless Italy, in 1993 and 1997, respectively. Since 1997,
Commun., vol. 15, no. 3, pp. 2054–2068, Mar. 2016. he has been with the Department of Electrical
[53] R. M. Gray, Entropy and Information Theory. New York, NY, USA:
Engineering (now the Department of Information
Springer, 2011.
[54] T. L. Narasimhan and A. Chockalingam, ‘‘On the capacity and perfor- Engineering, Computer Science and Mathemat-
mance of generalized spatial modulation,’’ IEEE Commun. Lett., vol. 20, ics), University of L’Aquila, where he is cur-
no. 2, pp. 252–255, Feb. 2016. rently an Associate Professor. He has authored
[55] Q. Shi, L. Liu, W. Xu, and R. Zhang, ‘‘Joint transmit beamforming and over 200 papers in technical journals and conference proceedings. His
receive power splitting for MISO SWIPT systems,’’ IEEE Trans. Wireless current research interests are mainly focused on wireless communication
Commun., vol. 13, no. 6, pp. 3269–3280, Jun. 2014. systems with emphasis on the Internet of Things, spatial modulation, and
[56] B. R. Marks and G. P. Wright, ‘‘A general inner approximation algo- cooperative communications. He is a member of the Executive Commit-
rithm for non-convex mathematical programs,’’ Oper. Res., vol. 26, no. 4, tee, Center of Excellence Design methodologies for Embedded controllers,
pp. 681–683, 1978. Wireless interconnect and Systemonchip (DEWS), University of L’Aquila,
[57] A. Beck, A. Ben-Tal, and L. Tetruashvili, ‘‘A sequential parametric con- and the Scientific Committee, Consorzio Nazionale Interuniversitario per le
vex approximation method with applications to nonconvex truss topology Telecomunicazioni. He is also the Chairman of the Board of Directors of
design problems,’’ J. Global Optim., vol. 47, no. 1, pp. 29–51, 2010.
WEST Aquila S.r.l., a spinoff research and development company, University
[58] M. Razaviyayn, M. Hong, and Z.-Q. Luo, ‘‘A unified convergence analysis
of block successive minimization methods for nonsmooth optimization,’’
of L’Aquila, and the Center of Excellence DEWS. He is involved in major
SIAM J. Optim., vol. 23, no. 2, pp. 1126–1153, 2013. national and European research programs in the field of wireless systems and
he has been a reviewer of major technical journals and international confer-
ences on communications. He also serves as a technical program committee
member and the session chairman for several international conferences on
CHIEN-CHUN CHENG (S’11) was born in communications.
Taipei, Taiwan, in 1982. He received the B.S.
and M.S. degrees from NCU, Zhongli, Taiwan,
in 2005 and 2007, respectively, and the joint
Ph.D. degree from NCTU, Hsinchu, Taiwan, and ALESSIO ZAPPONE (SM’16) received the M.Sc.
CentraleSupélec, Gif-sur-Yvette, France, in 2016. and Ph.D. degrees from the University of Cassino
He is currently a doing post-doctoral work with and Southern Lazio. He was with the Consorzio
the University of L’Aquila, L’Aquila, Italy. His Nazionale Interuniversitario per le Telecomuni-
research interests include the areas of digital com- cazioni in the framework of the FP7 EU-funded
munication and statistical signal processing. project TREND, which focused on energy effi-
ciency in communication networks. From 2012
to 2016, he was the Principal Investigator of
the CEMRIN project on energy-efficient resource
MARCO DI RENZO (S’05–AM’07–M’09– allocation in wireless networks, which was funded
SM’14) received the Laurea (cum laude) and Ph.D. by the German Research Foundation. Since 2016, he has been an Aggregate
degrees in electrical engineering from the Univer- Professor with the University of Cassino and Southern Lazio, Cassino, Italy.
sity of L’Aquila, Italy, in 2003 and in 2007, respec- He is a Research Associate with CentraleSupelec (Gif-sur-Yvette, France).
tively, and the Doctor of Science degree (HDR) In 2017, he was a recipient of the H2020 MSCA IF BESMART Fellowship.
from the University Paris-Sud, France, in 2013. His research interests include the area of communication theory and signal
Since 2010, he has been a CNRS Associate Profes- processing, with main focus on optimization techniques for resource alloca-
sor (Chargé de Recherche Titulaire CNRS) with tion and energy efficiency maximization. He held several research appoint-
the Laboratory of Signals and Systems, Paris- ments at international institutions. He was a recipient of the FP7-Newcom
Saclay University–CNRS, CentraleSupélec, Univ Grant in 2014. He was selected as an Exemplary Reviewer of the IEEE
Paris Sud, Paris, France. He is an Adjunct Professor with the University of TRANSACTIONS ON COMMUNICATIONS in 2017. He currently serves as an Asso-
Technology Sydney, Australia, a Visiting Professor with the University of ciate Editor for the IEEE SIGNAL PROCESSING LETTERS and a Guest Editor for
L’Aquila, Italy, and a Co-Founder of the university spin-off company WEST the IEEE JOURNAL ON SELECTED AREAS ON COMMUNICATIONS (Special Issue on
Aquila S.r.l., Italy. He is a Distinguished Lecturer of the IEEE Vehicular Energy-Efficient Techniques for 5G Wireless Communication Systems).
Technology Society and the IEEE Communications Society. He was a

VOLUME 5, 2017 23211