IEEE VEHICULAR TECHNOLOGY SOCIETY SECTION

Received February 26, 2022, accepted March 6, 2022, date of publication March 10, 2022, date of current version March 21, 2022.
Digital Object Identifier 10.1109/ACCESS.2022.3158486

A Pricing-Based Approach for Energy-Efficiency

Maximization in RIS-Aided Multi-User MIMO
SWIPT-Enabled Wireless Networks
VAIBHAV SHARMA1 , JETTI YASWANTH1 , SANDEEP KUMAR SINGH 1 , (Member, IEEE),
SUDIP BISWAS2 , (Member, IEEE), KESHAV SINGH 1 , (Member, IEEE),
AND FAHEEM KHAN 3 , (Member, IEEE)
1 Instituteof Communications Engineering, National Sun Yat-sen University, Kaohsiung 80424, Taiwan
2 Department of Electronics and Communications Engineering (ECE), Indian Institute of Information Technology Guwahati (IIITG), Guwahati,
Assam 781015, India
3 Electronic Engineering Department, School of Computing and Engineering, University of Huddersfield, Huddersfield HD1 3DH, U.K.

Corresponding author: Keshav Singh ([email protected])

This work was supported by the Ministry of Science and Technology of Taiwan under Grant MOST 110-2224-E-110-001.
The work of Sudip Biswas was supported by the Science and Engineering Research Board (SERB), DST, Government
of India, under Grant SRG/2020/001145.

ABSTRACT In this work, we investigate the performance of a reconfigurable intelligent surface (RIS)-aided
multi-user simultaneous wireless information and power transfer (SWIPT) network, where a multiple-input
multiple-output (MIMO) base station (BS) serves multiple MIMO information receivers (IRs) while ensuring
a minimum harvested power at multiple MIMO energy receivers (ERs). In order to improve the energy
efficiency (EE) of the network, we consider a pricing-based performance metric called network utility.
We then establish an optimization framework to jointly optimize the transmit precoding matrix (TPM) and
phase shift matrix (PSM) to maximize the network utility function with constraints on the available transmit
power at BS, minimum harvested power required at each ER, and unit modulus phase shift condition at
RIS. Due to the non-convex nature of this problem, we divide it into two sub-problems where a sub-optimal
solution of TPM and PSM are obtained separately using successive convex approximation and bisection
search-based algorithms. Further, we propose an EE maximization (EEM) algorithm based on the block
coordinate descent method to achieve the optimal solution of the master problem by iteratively obtaining
the sub-optimal TPM, PSM, and network price using their respective algorithms. Moreover, we also prove
that the solution obtained for each problem using their respective algorithm converges to the Karush-Kuhn-
Tucker (KKT) optimum point of that problem. We also show the efficacy of the proposed algorithm using
simulation results. In particular, we highlight the importance of using RIS in a multi-user MIMO SWIPT
network and demonstrate the effect of various parameters on the network’s EE performance.

INDEX TERMS Reconfigurable intelligent surfaces (RIS), multiple-input and multiple-output (MIMO),
multi-user, simultaneous wireless information and power transfer (SWIPT), energy efficiency (EE).

I. INTRODUCTION consumption, very high energy efficiency (EE), better spec-

The impending deployment of fifth-generation (5G) mobile tral efficiency (SE), etc., researchers have already started to
communications around the world will be a major factor explore the evolution of 5G, commonly referred to as 5G and
in driving productivity and will be the key enabler for beyond (5GB) and sixth generation (6G) communications [1].
long-envisaged verticals including personalised healthcare, In this context, to enhance the experience of relaying in wire-
manufacturing, smart energy grids, smart cities, finance, and less communication networks, the idea of multiple passive
transportation. However, realizing the ever-growing demand reflecting surfaces/elements made of meta-materials has been
for a better communication network with improved qual- floated to assist the communication between multiple devices
ity of service (QoS) requirements such as lower power such as base station (BS), users etc., [2]–[4]. This set of
discrete reflecting elements is termed as reconfigurable intel-
The associate editor coordinating the review of this manuscript and ligent surface (RIS) or intelligent reflecting surface (IRS),
approving it for publication was Jie Gao . which is neither a part of the transmitter radio nor the receiver

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
29132 VOLUME 10, 2022
V. Sharma et al.: Pricing-Based Approach for EEM in RIS-Aided Multi-User MIMO SWIPT-Enabled Wireless Networks

radio. It is a novel and low-cost/maintenance technique to and optimized phase shift matrix (PSM), which lead to a
control the wireless propagation medium, which until now significant reduction in training overhead and design com-
had been deemed as uncontrollable. Accordingly, the funda- plexity. Similarly, in [11], joint optimization of transmitting
mental role of a RIS is to affect the dispersion of wireless precoder and PSM was performed to maximize the received
signals transmitted by other devices, without producing its power for a multiple RIS aided single user wireless net-
signals [5]. The phase/reflection angle of each element in a work. Specifically, a semi-definite relaxation (SDR)-based
RIS can be independently controlled or reconfigured through algorithm was proposed to find a sub-optimal solution to the
software. Due to their passive nature, RISs do not impose any optimization problem. Next with regards to power efficiency,
thermal noise as they simply reflect the signals incident upon in [9], the authors minimized the total dynamic power by
them. Therefore, RISs consume less power as compared to the joint design of active and passive beamformers for a
conventional decode-and-forward and amplify-and-forward RIS-aided MISO network. Similarly, joint symbol level pre-
relays, thus enhancing the EE of the communication network. coding and PSM were optimized to minimize the total power
Further, RIS uses smart reflection characteristics to create usage in a RIS-aided multi-user network in [17]. In [18], joint
a virtual line of sight (LoS) link that helps in eliminating optimization of the transmit power and PSM was performed
the fading caused by obstacles between the transmitter and to maximize the EE of a RIS-aided multi-user MISO network
receiver. In addition to this, RISs also provide multi-path for Terahertz communication using covariance matrix adap-
propagation that leads to improvement in the rank of the tation evolution strategy and Dinkelbach’s method.
channel along with an increase in the achievable diversity. While the works mentioned above are seminal for a
Hence, a wireless network‘s performance can be significantly RIS-aided network, the consideration of a MISO network or
improved by mere adjustments of the angles of the reflective a network without EH capabilities or without an emphasis
elements in a RIS [6], [7], which make it more remunerative on EE, might not be ideal considering the requirements and
than conventional relays for network operators. specifications of future 6G networks. Accordingly, a multi-
While several key performance indices such as throughput, user MIMO network with multiple antennas at each node
coverage, EE, and SE play pivotal roles in the design and insinuates a more practical consideration. Similarly, the
deployment of RISs in a wireless network [8]–[12], this consideration of EH is also of paramount importance for
paper particularly focuses on the EE of the network1 [13]. future energy-efficient networks. Only a handful of works
To achieve this, we not only consider the problem of max- investigate networks with such considerations, where the
imizing the EE of the network but also implement simulta- communication between devices is assisted by RISs. For
neous information and power transfer (SWIPT) [14], [15]. example, in [19] the trade-off between EE and SE was inves-
Accordingly, certain devices in the network have abilities tigated in a RIS-aided multi-user MIMO uplink network and
to extract either only-power or only-symbol/information or the resource efficiency was maximized by jointly optimiz-
some amount of power along with the desired symbol from ing the transmit precoding and RIS reflective beamforming.
the received signal using the energy harvesting (EH) tech- A similar RIS-aided multi-user MIMO uplink network was
nique. Now, both EE and EH are strictly related to the total investigated in [20], where the passive beamforming and
power consumed by a network in the form of static and on-off reflecting modulation based information transfer were
dynamic power to attain the desired level of QoS [16]. Static optimized to maximize the sum-rate. Next with regards to EH,
power consumption is the constant/fixed power required the authors in [21] investigated the performance of SWIPT in
for non-communication-related tasks such as network sig- a RIS-aided wireless multi-user network. Similarly, in [22],
nal processing, hardware maintenance, cooling, etc. On the the authors investigated a RIS-assisted SWIPT multi-user
other hand, dynamic power is the total transmit power used network and maximized its sum rate by jointly optimizing
by the network to complete an end to end communication the active and passive beamforming, albeit considering a
among several devices wirelessly. Due to continuous varia- single antenna at each user. However, to the best of the
tions in wireless channel conditions, the total transmit power authors’ knowledge no work to date has investigated the EE
required by the network changes dynamically. Therefore, of a RIS-aided SWIPT enabled Multiuser MIMO Networks.
judicious selection of transmit power is required to signifi- Accordingly, the goal of this paper is to maximize the EE
cantly improve the network’s performance. of a RIS-aided multi-user MIMO SWIPT network by jointly
designing the optimal PSM and optimal precoding matrix.
A. RELATED LITERATURE AND MOTIVATION
A RIS-aided multiple-input single-output (MISO) network B. CONTRIBUTIONS
was considered in [10] and a two-timescale (TTS) transmis- In particular, we consider a multi-user MIMO SWIPT
sion protocol was proposed to maximize the sum rate of network having a multi-antenna BS serving multiple
the network. In particular, the transmit precoding vector was multi-antenna information receivers (IRs) while ensuring a
designed using instantaneous channel state information (CSI) minimum harvested power at multiple multi-antenna energy
receivers (ERs). Unlike previous works, which have only
1 EE is evaluated as the ratio of the achievable sum-rate of a network to the considered sum rate maximization as a performance metric,
total power consumed. we use a pricing-based approach and adopt a performance

VOLUME 10, 2022 29133

 V. Sharma et al.: Pricing-Based Approach for EEM in RIS-Aided Multi-User MIMO SWIPT-Enabled Wireless Networks

metric called network utility, which provides a striking bal- |X| represent the trace, Frobenius norm and determinant,
ance between the sum rate and the total dissipated power, respectively. Re{x} denotes the real part of the complex value
thus indirectly controlling the achievable EE of the network. x. Ca×b denote a complex vector of size a×b. The expectation
We then provide a framework to jointly optimize the transmit operation is denoted by E{·}. (·) (·), (·)H and (·)T repre-
precoding matrix (TPM) and PSM to maximize the network sent Hadamard product, Hermitian and transpose operations,
utility function. The primary distinctions of this work are respectively. arg{·} and diag(·) represents extraction of phase
summarized below. information and diagonalization operation, respectively. (·)†
• We formulate a network utility function maximization and (·)−1 denotes the matrix pseudo-inverse and inverse oper-
problem and jointly optimize TPM and PSM consid- ations, respectively. I denotes the identity matrix. Ofx (x) is
ering three different constraints: 1) maximum available used for denoting the gradient of f w.r.t. the vector x and
transmit power available at BS, 2) minimum harvested CN (0, σ 2 I) denote a random vector with zero mean and σ 2
power required at each ER, and 3) unit modulus phase variance.
shift at RIS. This problem is non-convex and extremely
difficult to solve due to the involvement of phase and II. SYSTEM MODEL
power constraints. We consider a RIS-aided multi-user MIMO downlink wire-
• We establish an optimization framework and reformu- less communication network which consists of a BS, NI
late the master problem into a simpler and more tractable information receivers (IRs) and NE energy receivers (ERs).
form using the mean squared error (MSE) minimization We further assume that the BS is equipped with AB anten-
approach. Then we use the block coordinate descent nas, each IR and ER are fitted with AI and AE antennas,
method to solve this reformulated problem by dividing respectively. With the aid of NR reflective elements of RIS,
it into two sub-problems and solve them separately. BS transmits information to all IRs while providing sufficient
In particular, first, we optimize TPM for fixed PSM and energy to all ERs simultaneously using SWIPT protocol,
then optimize PSM for fixed TPM. as shown in the Fig. 1.
• To solve the TPM optimization problem, we pro-
pose a successive convex approximation (SCA) based A. INFORMATION AND ENERGY TRANSFER
algorithm. It uses the bisection search method to pro- As mentioned earlier, BS is fitted with AB antennas to aid the
vide a near-optimal solution that converges at the signal transmission and it need to have channel state informa-
Karush-Kuhn-Tucker (KKT) optimum point of this tion (CSI) for applying transmit beamforming. Thus, for sim-
sub-problem. Similarly, we simplify the PSM opti- plicity, we assume BS has perfect CSI using some standard
mization problem using the majorization-minimization estimation method2 Therefore, using this CSI, BS designs an
method and propose a SCA based algorithm to find a appropriate TPM and combines the symbols of all user to
near-optimal PSM that converges at the KKT optimum form a superimposed symbol given by
point.
NI
• Further, we propose an EE maximization (EEM) algo- X
s= Tn an , (1)
rithm based on the block coordinate descent method to
n=1
achieve the optimal solution of the master problem by
iteratively obtaining the sub-optimal TPM, PSM, and where an ∈ Cc×1 represents the unit energy data symbol
network price using their respective algorithms. vector of the nth IR with c ∈ min(AB , AI ), AI represents the
• Finally, the efficacy of the proposed algorithm is number of antennas at each IR. Tn ∈ CAB ×c denotes the
demonstrated using simulation results. In particular, linear TPM corresponding to nth IR with ||Tn ||2F = Pn and
PNI
we highlight the importance of using RIS in a multi-user n=1 Pn ≤ Pmax . Pmax is the maximum transmission power
MIMO SWIPT network and demonstrate the effect of available at BS. Next, BS transmits this superimposed symbol
various parameters on the network’s EE performance. to all users simultaneously. Thus, the signal received at the nth
IR can be expressed as
1) ORGANIZATION
The remainder part of the paper is organized as follows: bI ,n = (Ln,B + Ln,R 8PR,B )s + dI ,n , (2)
Section II discusses the considered RIS-aided multiuser wire-
where 8 = diag{φ1 , φ2 , . . . , φNR } is the phase shift
less network in detail and formulated optimization problem.
matrix (PSM) with φi = ejθ i and θi ∈ [0, 2π] denoting the
Sections III and IV present the problem formulation and
phase shift of the ith reflective element of the RIS, Ln,B ∈
its detailed solution, respectively. The simulation results are
CAI ×AB and Ln,R ∈ CAI ×NR represent the channel gain from
discussed in Section VI whereas Section VII concludes the
work. 2 CSI estimation in RIS-aided networks can be performed using some
standard algorithms such as parallel factor decomposition (PARAFAC)
2) NOTATIONS [23]–[25]. However, similar to [26], [27], we assume perfect CSI in this
paper for simplicity and analytical tractability. Moreover, on the reviewers
X? and X∗ denote the converged solution and conjugate suggestion and for better insights, we highlight the impact of CSI estimation
operator, respectively, of a given matrix X. tr(X), ||X||F and error on the performance of the considered network in Fig. 11.

29134 VOLUME 10, 2022

V. Sharma et al.: Pricing-Based Approach for EEM in RIS-Aided Multi-User MIMO SWIPT-Enabled Wireless Networks

FIGURE 1. A RIS-assisted multi-user MIMO SWIPT-enabled wireless network.

BS and IRs, respectively, to the nth IR, PR,B ∈ CNR ×AB harvest sufficient power from received signal using SWIPT
represents the channel gain for the link between BS and RIS, based EH [29]. Thus, using (1), the received signal at the jth
and dI ,n ∼ CN (0, σI2 IAI ) is the additive white Gaussian ER can be expressed as
noise (AWGN). Substituting s from (1) into (2), bI ,n can be bE,j = (Mj,B + Mj,R 8PR,B )s + dE,j , (6)
expressed as:
where Mj,B ∈ CAE ×AB and Mj,R ∈ CAE ×NR are the channel
NI
gain from BS and RIS to the jth ER, respectively, and dE,j ∼
X
bI ,n = L̄n Tn an + L̄n Ti ai + dI ,n , (3)
i=1,6=n CN (0, σE2 IAE ) is the AWGN. Each ER applies EH protocol
to harvest power from the received signal [14]. The total
1
where L̄n = Ln,B + Ln,R 8PR,B . The corresponding achiev- harvested power by the jth ER can be expressed as
able data rate of the nth IR is given by NI
X
!
Hj = κj tr M̄j Tn Tn M̄j ,
H H
(7)
Rn = log|I + L̄n Tn TH H −1
n L̄n Kn |, (4)
n=1
where Kn isPinterference plus-noise covariance matrix given 1
NI where M̄j = Mj,B + Mj,R 8PR,B , 0 < κj ≤
f =1,f 6=n L̄n Tf Tf L̄n + σI I [28]. Further, the
as Kn = H H 2

sum rate (SR) can be expressed as 1, ∀j ∈ {1, 2, · · · , NE } denotes the EH efficiency.3

The
weighted sum of the power harvested by all the ERs is given
NI
X by
R= Rn . (5)
NE NI
n=1 X X  
H= δj Hj = tr TH
n MTn , (8)
As mentioned earlier, this network also consists of low j=1 n=1
power ERs/sensor nodes which require certain amount of
power (very low) to complete their designated task without 3 Note that the EH efficiency depends on the harvesting circuit used at

any interruption. Therefore, considering this requirement, each user. Thus, considering the fact that efficiency is independent of these
constraints and randomness of the channel, we have assumed this efficiency
BS designs the beamforming/TPM so that alongwith satisfy- to be constant at each user, which is a common practice in MIMO-aided
ing the desired performance at each IRs, these ERs can also EH [30]–[32].

VOLUME 10, 2022 29135

 V. Sharma et al.: Pricing-Based Approach for EEM in RIS-Aided Multi-User MIMO SWIPT-Enabled Wireless Networks

PNE NI
j=1 δj κj M̄j M̄j , δj is the energy weighting
where M = H X  
th s.t. n ≤ Pmax ,
tr Tn TH (13b)
factor for the j ER.
n=1
NI
!
III. NETWORK UTILITY FUNCTION AND X
tr TH
n MTn ≥ H̄ , (13c)
PROBLEM FORMULATION
n=1
Let us consider that BS consumes a fixed static power of |φi | = 1, i ∈ {1, 2, . . . , NR }, (13d)
PSB for performing non-communication task such as network
signal processing, hardware maintenance, cooling, etc. Sim- where H̄ is the minimum power required by each ER. Clearly,
ilarly, the static power consumed at the RIS is assumed to due to coupling of variables T and 8, and involvement of EH
be PSI . Thus, total static power consumed by the considered constraint, this problem becomes non-convex in nature. Thus,
network is given by it is extremely difficult to find its solution using standard
methods.
Pstat = PSB + PSI . (9)
Further, the total dynamic power consumed by the network is IV. BLOCK COORDINATE DESCENT BASED ALGORITHM
equal to the total transmit power used by BS. Therefore, from As discussed above, problem (13) is extremely difficult
(1), the dynamic power can be evaluated as to solve. Therefore, using minimum mean square error
(MMSE) [33], problem (13) is simplified to a much simpler
NI NI
X X   and tractable form. Thus, using (3) and approach similar
Pdyna = Pn = n .
tr Tn TH (10)
to [33], the estimated signal vector at nth IR can be evaluated
n=1 n=1
as
From (9) and (10), the total power consumed by the network
is obtained as n bI ,n ,
ân = JH (14)

Ptot (T) = Pdyna + Pstat , where Jn ∈ CAI ×c represents the decoding matrix. The mean
NI square error (MSE) corresponding to (14) can be expressed as
X  
n + PB + PI .
tr Tn TH S S h
H i
= (11)

On = E ân − an ân − an ,
n=1
  H
A. NETWORK UTILITY FUNCTION = JH n L̄n Tn − I JH
n L̄n Tn − I
In this paper, we study the network’s performance by inves- NI
X
n L̄n Tf Tf L̄n Jn + σ Jn Jn .
JH H H 2 H
tigating a pricing based EE maximization problem. Using + (15)
(5) and (11), we define a new performance metric termed as f =1,f 6 =n
network utility function (NUF) and expressed as Using (4), (12) and (15), problem (13) can be reformulated as
Uq (T, 8) = R(T, 8) − qPtot (T), NI
X
NI NI max ln (V, J, T, 8)
!
X X  
= Rn (T, 8)−q n +PB + PI ,
tr Tn TH S S V,J,T,8
n=1
NI
n=1 n=1
!!
X  
(12) −q tr Tn TH S S
n + PB + PI (16)
n=1
where q ≥ 0 denotes the network price. Note that the power
s.t. (13b), (13c), (13d)
allocation problem will be equivalent to sum rate maximiza-
tion problem when q → 0 and the power resource utilization where V is the set of auxiliary matrices that denotes V =
cost becomes negligible. However, with increase in q, the {Vn ≥ 0, ∀n ∈ NI } and J is the set of Jn for all IRs, and
optimal design of TPM and PSM becomes very important.
ln (V, J, T, 8) = log |Vn | − Tr (Vn On ) + c. (17)
B. PROBLEM FORMULATION To find optimal J, we take first order partial derivative of
As evident from (12), TPM T and PSM 8 play a vital role ln (V, J, T, 8) w.r.t. Jn and equate it to zero. Thus, the
in obtaining a desired performance. Therefore, it is extremely optimal value of Jn can be expressed as
important to jointly optimize and design these parameters and  −1
maximize NUF. Thus, using (10), (8) and (12), we formulate a J?n = Kn + L̄n Tn TH H
n L̄n L̄n Tn . (18)
joint TPM and PSM optimization based NUF maximization
problem considering the constraints of maximum available Similarly, the optimal value of Vn can be expressed as
transmit at BS alongwith minimum power requirement of   −1 −1
N
ERs and unit modulus of phase shift of RIS elements given I
V?n = Ic − TH
X
f L̄n + σI I L̄n Tn  .
H
n L̄n L̄n Tf TH H 2 
 
by
f =1
max Uq (T, 8)


(13a)
T,8 (19)

29136 VOLUME 10, 2022

V. Sharma et al.: Pricing-Based Approach for EEM in RIS-Aided Multi-User MIMO SWIPT-Enabled Wireless Networks

After obtaining optimal decoding matrix Jn and auxiliary this problem and propose an algorithm using Lagrangian dual
matrix Vn , we use these values to find sub-optimal TPM decomposition method that provides a near-optimal solution
and PSM in next subsections. In particular, we use block with much lower complexity [37]. Owing to the fact that it sat-
coordinate descent (BCD) method to solve problem (16) isfies Slater’s condition, we use the dual problem approach to
and divide it into two sub problems where one problem will find the solution as its dual gap is zero. The partial Lagrangian
find sub-optimal TPM where as the other problem provides function of problem (23) can be expressed as
sub-optimal PSM. NI   XNI  
X
L (T, λ) = H
tr Tn DTn − tr Vn JH
n L̄n Tn
A. SUB-OPTIMAL TPM n=1 n=1
This section presents the optimization of TPM for fixed value NI NI
of other variables (J, V and 8). Using (15) and neglecting
X   X
− tr Vn TH L̄H
n n n J + q tr(TH
n Tn )
the constant terms, the maximization problem (16) can be n=1 n=1
reformulated in a minimization problem w.r.t T give as XNI  
NI
X   XNI   +λ n Tn − λPmax ,
tr TH (24)
min tr TH
n DTn − tr Vn JH
n L̄n Tn
n=1
T
n=1 n=1 where λ denotes the Lagrange multiplier. Next, using (24),
XNI   NI
X   we solve the following problem
H H
− tr Vn Tn L̄n Jn + q tr TH
n Tn 1
n=1 n=1 g (λ) = min L(T, λ) s.t. (22). (25)
T
s.t. (13b), (13c), (20)
The Lagrangian function of problem (25) can be expressed as
where D = N H H
P I
f =1 L̄f Jf Vf Jf L̄f . Note that similar to previous NI
X  
problems, problem (20) is also non-convex. However, unlike L (T, ρ) = tr TH
n (D + (q + λ)I) Tn
previous problem, it is equivalent to a difference of convex n=1
(d.c.) program. Therefore, we can use successive convex NI
X  
approximation (SCA) method to solve problem (20) [34]. − tr Vn JH
n L̄n Tn
In particular, using first-order Taylor series expansion and n=1
Jensen’ inequality, similar to [35], we obtain: XNI  
NI
! NI
! − tr Vn TH L̄
n n
H
Jn + ρ H̃
X X
tr THn MTn ≥ −tr T(i)H
n MTn
(i) n=1
NI
" !#
n=1 n=1 X
" NI
!# − 2ρRe tr T(i)H
n MTn − λPmax , (26)
X
+ 2Re tr T(i)H
n MTn , (21) n=1
n=1 where ρ ≥ 0 denotes the dual variable. Differentiating
(i) L(T, ρ) w.r.t. Tn and solving it for zero, we obtain
where Tn is the previous value. After some algebra,  
we obtain Tn (λ, ρ) = (D + (q + λ)I)† L̄H n Jn Vn + ρMTn .
(i)
(27)
NI NI
" !# !
Further, considering the complementary slackness condition
X X
2Re tr (i)H
Tn MTn ≥ H̄ +tr Tn MTn . (22)
(i)H (i)

n=1 n=1
of constraint (22), we obtain
NI
" !#
Using (22), we reformulate the problem (20) as X
2Re tr Tn MTn (λi , 0)
(i)H
≥ H̃ , (28)
NI NI
X   X   n=1
min tr TH DTn − tr Vn JH
L̄n Tn
T
n n
where λi is the previous value of λ. Note that if inequality (28)
n=1 n=1
NI NI
hold true, then Tn (λi , 0) is the solution to the problem (25).
   
−
X
tr Vn TH L̄H
J + q
X
tr TH
T (23a) On the other hand, if inequality (28) does not satisfy, then the
n n n n n
solution is given by Tn (λi , ρo ) with
n=1 n=1
h P i
s.t. (13b), NI (i)H
n=1 Tn M(D+(q+λi )I) L̄n Jn Vn
H̃ −2Re tr −1 H
NI ρo = .
" !#
X P 
2Re tr T(i)H
n MTn ≥ H̃ , (23b) 2tr NI
n=1 T
(i)H
n M(D+(q+λ i )I)−1 MT(i)
n
n=1
(29)
(i)H (i)
where H̃ = H̄ +tr( N
P I
n=1 Tn MTn ). Clearly, the OF in (23) Now, the dual problem is corresponding to g (λ) can be
is convex and it can be solved using standard optimization
evaluated as
tool such as CVX [36]. However, the computational com-
plexity of the CVX tool is very high. So, we further simply max g (λ) s.t. λ ≥ 0. (30)
λ
VOLUME 10, 2022 29137
 V. Sharma et al.: Pricing-Based Approach for EEM in RIS-Aided Multi-User MIMO SWIPT-Enabled Wireless Networks

Now, considering the complementary slackness condition of Algorithm 1 Sub-Optimal Solution to Problem (23)
constraint (13b), we obtain 1: Initialize Pmax , ε, λin , λfi .
NI
! 2: repeat
X 3: λ = (λin + λfi )/2;
tr n (0, ρo )Tn (0, ρo ) ≤ Pmax .
TH (31)
n=1
4: Evaluateh {T
P n (λ, 0), ∀n ∈ {1, 2 . . . ,i
NI }} using (27)
NI (i)H
If the inequality (31) holds true, then Tn (0, ρo ) is the solution
5: if 2Re tr n=1 Tn MTn (λi , 0) ≥ H̃ (i) then,
6: ρo = 0.
to the problem (25). However, if inequality (31) does not
7: else
satisfy, then the solution is given by Tn (λ, ρo ) [38], where
8: Evaluate ρo using (29).
λ is obtained by solving
9: end if
NI Evaluate {Tn (λ, ρo ), ∀n ∈ {1, 2 . . . , NI }} using (27)
!
1 X 10:
P(λ) = tr Tn (λ, ρo )Tn (λ, ρo ) = Pmax
H
(32) 11: Evaluate P(λ) using (32)
n=1 12: if P(λ) ≥ Pmax then
Due to involvement of ρo in (29), obtaining a closed form 13: λin = λ.
solution of (32) is extremely difficult. So, in order to solve 14: else
(32), we first obtain the behavior of P(λ) w.r.t. λ in the 15: λfi = λ.
following lemma. 16: end if.
Lemma 1: P(λ) decreases monotonically w.r.t. λ. 17: until |λin − λfi | ≤ ε
Proof: Refer to Appendix A. 18: Tn (i+1) = {Tn (λ, ρo ), ∀n ∈ {1, 2 . . . , NI }} using (27)
Further considering the nature of P(λ), we propose Algo-
rithm 1 that uses bisection search method to solve (23). Next Algorithm 2 Sub-Optimal TPM
using Algorithm 1 and SCA method, we propose Algorithm 2
to solve problem (20). 1: Initialize ε, T(1) , i = 0, imax
Theorem 1: Algorithm 2 provides a solution that con- 2: repeat
verges at the KKT optimum point of (20). 3: i←i+1
Proof: Proof is similar to [39], so for paucity of space 4: Evaluate z(T(i) ) using (20)
Calculate H̃ (i) = H̄ + tr( N (i)H
P I
we have omitted the proof. 5: n=1 Tn );
6: Obtain {Tn (i+1)
, ∀n ∈ {1, 2 . . . , NI }} using Algo-
1) COMPUTATIONAL COMPLEXITY rithm 1;
The calculation of T in (27) requires approximately O NI A3

7: Evaluate z(T(i+1) ) using (23)
computations with A = max(AB , AI , c). Considering 8: until i ≥ imax or |z(T(i+1) ) − z(T(i) )|/|z(T(i+1) )| ≤ ε
the fact that  Algorithm 1 converges at approximately
λfi −λin
log2 ε iterations, the complexity of Algorithm 1 to 1 1 1

λfi −λin

 where Yn = LH H H
n,R Jn Vn Jn Ln,R , U = PR,B T̃PR,B , and Wn =
solve the problem (23) is given by Olog2 ε NI A3 .
PR,B T̃H LH H
n,B Jn Vn Jn Ln,R . C1 is summation of entities that are
Thus,
 Algorithm 2 solves the (20) with a complexity of independent of 8. Similarly, we obtain
λ −λ

O imax log2 fi ε in NI A3 .  
tr Vn JH n n n = tr (8Bn ) + C2 ,
L̄ T (35)
B. SUB-OPTIMAL PSM 8
This section presents the optimization of PSM for fixed value 1
where Bn = PR,B Tn Vn JH n Ln,R and C2 constitutes to the
of other variables (J, V and T). Using (15) and neglecting
entities independent of 8. Further, using (8), constraint (13c)
the constant terms, the maximization problem (16) can be
can be improvised as
reformulated in a minimization problem w.r.t 8 given as    
NI
X   XNI   tr 8H MR 8U + tr 8H MH BR
min tr Vn JH
n L̄n T̃L̄H
n Jn − tr Vn J H
n L̄n Tn  
8
n=1 n=1 + tr (8MBR ) + tr MB T̃ ≥ H̄ , (36)
XNI  
− tr Vn TH H
n L̄n Jn (33) 1
NE
1
NE
δj κj MH δj κj MH
P P
where MB = j,B Mj,B , MR = j,R Mj,R , and
n=1 j=1 j=1
s.t. (13c), (13d), NE
1
δj κj MH
P
NI
MBR = PR,B T̃ j,B Mj,R . Substituting (34) and (35)
j=1
Tf TH
P
where T̃ = f . From (3), we can write in (33), we simplify the problem as
f =1
         
tr Vn JH H
n L̄n T̃L̄n Jn = tr 8H Yn 8U + tr 8H WH
n
min tr 8H Y8U + tr 8H XH + tr (8X)
8
+ tr (8Wn ) + C1 , (34) s.t. (13d), (36), (37)

29138 VOLUME 10, 2022

V. Sharma et al.: Pricing-Based Approach for EEM in RIS-Aided Multi-User MIMO SWIPT-Enabled Wireless Networks

with Y = N
PNI PNI
We formulate a new sub-problem considering ξ (φ|φ (n) ) as
P I
n=1 Yn and X = n=1 Wn − n=1 Bn . Note that
C1 and C2 are constant and have no impact on analysis, so we the new OF which can be obtained using the following
have neglected these terms. Using the identity from [40], inequality [42]
we can write n o
    φ H 4φ ≤ φ H Zφ − 2Re φ H (Z4) φ (n)
tr 8H Y8U = φ H Y UT φ, (38)  H
    + φ (n) (Z − 4) φ (n) , (50)
tr 8H MR 8U = φ H MR UT φ, (39)
where Z = λmax INR and λmax denotes the maximum eigen
where φ = [φ1 , . . . , φNR ]T . Similarly, we have value of with 4. After some simplification, we obtain
  n o
tr (8X) = xT φ, tr 8H XH = φ H x∗ , (40) ξ (φ|φ (n) ) = y(φ|φ (n) ) + 2Re φ H x∗ , (51)
 
tr (8MBR ) = mT φ, tr 8H MH BR = φ m ,
H ∗
where y(φ|φ (n) ) = φ H Zφ − 2Re φ H (Z4) φ (n) +

(41)

H
φ (n) (Z − 4) φ (n) , Using (51), problem (46) is reformu-
where x = [[X]1,1 , . . . , [X]NR ,NR ]T and m = [[MBR ]1,1 , . . . , lated as
[MBR ]NR ,NR ]T . Further, from (36), we obtain
o _ min ξ (φ|φ (n) ), s.t. (13d), (45). (52)
φ
n
φ H ϒφ + 2Re φ H m∗ ≥H , (42)
Since, φ H φ = NR , thus φ H Zφ = NR λmax . Therefore, remov-
_
where H = H̄ − Tr(MB T̃) and ϒ = MR UT is a semidef- ing constant terms from (52), it is converted to a maximization
inite matrix, as MR and UT are non-negative semidefinite problem given by
matrices [40]. From (36) and (42), (37) is reformulated as n o
max 2Re φ H w(n) s.t. (13d), (45), (53)
φ
n o
min φ H 4φ + 2Re φ H x∗ (43)
φ
where = (λmax INR − 4)φ (n) − x∗ . Due to involvement
w(n)
s.t. (13d), (42),
of constraint (13d), problem (53) becomes a non-convex
where 4 = Y UT . Similar to ϒ, 4 is also a non-negative optimization problem. Also, unlike problem (23), its dual gap
semidefinite matrix. Due to (42), problem (43) is a non- is not zero. Therefore, we use pricing based approach and
convex problem. Therefore, similar to (20), we use SCA reformulate the problem (53) as
method to solve problem (43). Considering the fact that φϒφ n o h  i
max 2Re φ H w(n) + 2pRe φ H m∗ + ϒφ (n) ,
is a convex function of φ, we have φ
h i s.t. (13d), (54)
φ H ϒφ ≥ −φ (n)H ϒφ (n) + 2Re φ H ϒφ (n) , (44)
where p ≥ 0 denotes the price factor. The solution of this
where φ (n)
is the previous value. From (42) and (44), problem can be obtained as
we obtain w(n) +p m∗ +ϒφ (n)



h  i _ φ(p) = ej arg . (55)
2 Re φ H m∗ + ϒφ (n) ≥H +φ (n)H ϒφ (n) . (45)
Considering the complementary slackness condition p
 
Thus using (45), the problem (43) can be transformed as K (p) − Ĥ = 0 of (45) with
n o
min φ H 4φ + 2Re φ H x∗ (46a)
φ K (p) = 2Re[φ(p)H (m∗ + ϒφ (n) )]. (56)
s.t. (13d),
h  i If (45) is satisfied, then p = 0 provide the optimal solution as
2 Re φ H m∗ + ϒφ (n) ≥ Ĥ , (46b) φ(0). If (45) is not satisfied, then we need to find p for which
the (55) gives the optimal solution. Therefore, using approach
_
where Ĥ , H +φ (n)H ϒφ (n) . We use the majorization- similar to (32), we propose Algorithm 3 that provides the
minimization algorithm to solve the problem (46) by subdi- optimal solution to problem (53).
viding it into simple and tractable sub-problems [41]. First Theorem 2: Algorithm 3 finds the optimal solution of
we obtain the upper bound (ξ (φ|φ (n) )) of OF (χ (φ)) in (46) problem (53) and problem (52).
that satisfies the following conditions: Proof: For the proof refer to Appendix B.
Using Algorithm 3, we propose Algorithm 4 that uses SCA
ξ (φ (n) |φ (n) ) = χ(φ (n) ), (47) method to solve the Problem (46).
∇φ ∗ ξ (φ|φ ) (n)
= ∇φ ∗ χ(φ) , (48) Theorem 3: Algorithm 4 provides a solution that con-
φ=φ (n) φ=φ (n)
verges at the KKT optimum point of (33).
ξ (φ|φ ) ≥ χ(φ).
(n)
(49) Proof: Refer to Appendix C

VOLUME 10, 2022 29139

 V. Sharma et al.: Pricing-Based Approach for EEM in RIS-Aided Multi-User MIMO SWIPT-Enabled Wireless Networks

Algorithm 3 Sub-Optimal Solution to Problem (53) Algorithm 4 Sub-Optimal PSM

1: Initialize the ε, pj and pρ 1: Initialize ε, φ (1) , n = 0, nmax
2: Evaluate K (0) using (56) 2: repeat
3: if K (0) ≥ Ĥ (n) then 3: n←n+1
4: repeat 4: Evaluate f (φ (n) ) using (45)
5: p = (pj + pρ )/2; _
5: Ĥ (n) =H +φ (n)H ϒφ (n) ;
6: Evaluate φ(p) using (55) 6: w(n) = (λmax INR − 4)φ (n) − x∗ ;
7: Evaluate K (p) using (56) 7: Obtain φ (n+1) using Algorithm 3
8: if K (p) ≥ Ĥ (n) then (n+1) )−f (φ (n) )|
8: until n ≥ nmax or |f (φ (n+1) ≤ε
9: pρ = p. f (φ )
10: else
11: pj = p.
12: end if Theorem 5: Algorithm 5 provides a solution that con-
13: until |pj − pρ | ≤ ε verges at the KKT optimum point of (13).
14: Evaluate φ (n+1) = φ(p) using (55) Proof: Refer to Appendix E.
15: else
16: Evaluate φ (n+1) = φ(0) using (55) 1) COMPUTATIONAL COMPLEXITY
17: end if Algorithm 5 uses Algorithm 2 and Algorithm 4 in step (5)
and step (8), respectively. Therefore, these steps consumes the
major portion of the computations with complexity discussed
1) COMPUTATIONAL COMPLEXITY earlier in their respective section. Thus, the computational
Similar to Algorithm 2, the computational
  complexity
  of complexity of Algorithm 5 is given by
pρ −pj
Algorithm 4 is approximately O nmax log2 ε NR3 .
λρ − λj
   
O n5max nmax log 2 NI A3B
C. OPTIMAL NETWORK PRICE (q? ) FOR OPTIMAL EE
1
ε
  
pρ − pj
Using (5) and (11), the EE of the considered network can be + nmax log2 NR3 .
expressed as
2
ε
PNI H H −1
n=1 log|I + L̄n Tn Tn L̄n K̄n | V. FEASIBILITY CHECK FOR PROBLEM (13)
ηEE (T, 8) = P NI We formulate the following optimization problem to check
S S


H
n=1 tr Tn Tn + PB + PI
PNI the feasibility constraints (13b) and (13d) as follows:
n=1 Rn (T, 8)
= . (57) NI
!
Ptot (T) X
H
max tr Tn MTn
The optimal price q? that gives the maximum EE is obtained T,8
n=1
using following theorem. s.t. (13b), (13d). (60)
Theorem 4: The price q? is the optimal price, if and only if
the optimal precoder (T? ) and optimal phase (8? ) in problem Problem (13) is feasible if and only if maximum OF of the
(13) with respect to q? satisfies the balance equation given by: above problem value is larger than H̄ . Due to coupling of
TPM and PSM, obtaining an exact solution to the above prob-
R(T? , 8? ) − q? Ptot (T? ) = 0. (58) lem is extremely difficult. Therefore, we obtain sub-optimal
Proof: Refer to Appendix D. TPM and PSM by optimizing them alternately. Thus, first we
Similar to [13], we use local maximizer of the problem (13) obtain a sub-optimal TPM by solving following optimization
to obtain the local optimum value of q in the (i + 1)th iteration problem
as NI
!
Rn (T?(i) , 8?(i) )
X
PNI H
max tr Tn MTn
q (i+1)
= n=1 , (59) T
Ptot (T?(i) ) n=1
s.t. (13b). (61)
where i is the previous iteration. Using (59), we find the
optimal price q? that can be obtained using iterative method. The optimal solution to the above problem is obtained as
The proposed EEM algorithm is summarized in Algorithm 5. √ NI
Tn =[ pn y, 0AB ×(c−1) ], ∀n = 1, . . . , NI , where
P
pn =
n=1
D. SOLUTION TO PROBLEM (13) Pmax , χ and y denote the maximum eigenvalue and the
Using above discussion, we propose an iterative Algorithm 5 corresponding eigenvector of M. Further assuming pi =
to solve (13). It uses Algorithm 2 and Algorithm 4 to find sub- Pmax /NI , ∀ i ∈ {1, 2 . . . , NI }, we have maximum value of
optimal T and 8, respectively, iteratively until convergence is OF in above problem as χPmax , which represents the optimal
achieved. energy beamforming [43].

29140 VOLUME 10, 2022

V. Sharma et al.: Pricing-Based Approach for EEM in RIS-Aided Multi-User MIMO SWIPT-Enabled Wireless Networks

Algorithm 5 EEM Algorithm where ϑriser,j

AoD ∈ [0, 2π] denotes the angle of departure and

1: Initialize n = 0, nmax , T(1) , φ (1) ,

ε, q(1) ϑriser,j ∈ [0, 2π] denotes the angle of arrival. γ and d
AoA

2: repeat represents the wavelength and antenna separation, respec-

3: n←n+1 tively. Similar to (63), the link between BS to ERs and BS
4: Evaluate R(n) using (5) to RIS also follows Rician fading and are represented by
5: Evaluate J(n) and V(n) using (18) and (19), Mj,B ∈ CAE ×AB , ∀j ∈ {1, 2, . . . , NE } and PR,B ∈ CNR ×AB ,
respectively respectively. Further, the link between BS to IRs and RIS
6: Obtain T(n+1) using Algorithm 2 to IRs are denoted by Ln,B ∈ CAI ×AB and Ln,R ∈ CAI ×NR ,
7: Obtain φ (n+1) using Algorithm 4 with J(n) , V(n) respectively, for all n ∈ {1, 2, . . . , NI }, and follow Rayleigh
and T(n+1) fading. Further, we consider a common large scale-path loss
8: Evaluate J(n+1) using (18) and V(n+1) using (19) model for each link given by
9: Evaluate R(T(n+1) , φ (n+1) ) using (5)  δj
D0
10: Evaluate q(n+1) using (59) PL = PL0 , (66)
11: ε̂ = |q(n+1) − q(n) |/q(n+1) Dj
12: until n ≥ nmax or ε̂ < ε, where PL0 denotes the path loss at a reference distance D0
and Dj , ∀j ∈ {BSRIS, RISER, RISIR, BSIR, BSER} denotes
the length for the links BS-RIS, RIS-ER, RIS-IR, BS-IR and
Next, sub optimal PSM for a given TPM can be obtained BS-ER. δj denotes the path loss exponent of the jth link. Also,
by solving the following problem (κj , ∀j ∈ {1, 2, · · · , NE } = κ). Unless stated otherwise, the
n o considered parameter values used for simulations are given
max φ H ϒφ + 2Re φ H m∗ in Table 1. Note that we consider without-RIS scenario (i.e.,
φ
direct link only) as the benchmark to compare the proposed
s.t. (13d), (62) EEM algorithm.
Note that OF of the above problem is convex w.r.t φ. There- Fig. 2 represents the convergence behavior of the
fore, after applying first order Taylor expansion, it can be Algorithm 5 obtained for different number of RIS elements
solved using SCA method. Finally, problem (61) and (62) are NR with Pmax = 10dBm. It can be observed that EE increases
solved alternatively until the OF of the problem (60) becomes with increase in NR , which is evident from the fact that sum
greater than H̄ . rate is directly proportional to NR (see (5)). Also, the obtained
solution converges after a fixed number of iterations, which
VI. RESULTS AND DISCUSSION verifies Theorem 5. Also, it can be seen that the change
In this section, the performance of the proposed algorithms in performance is around 150 − 200% for initial iterations.
and optimization framework are verified using exhaustive However, this change is around 0 − 2%, which is negligible
numerical simulations. Since ERs are energy constrained low compared to earlier change. This verifies the convergence of
power sensor nodes, thus, they are assumed to be placed in the algorithm graphically.
LoS of the BS and RIS. So, small scale fading of the link Fig. 3 depicts the plot of average EE w.r.t. number of RIS
between RIS and ERs follows Rician distribution which is a elements NR obtained for three different cases: 1) Optimal
combination of LoS and non-LoS (NLoS) components given PSM using Algorithm 5, 2) Random PSM, and 3) without
by RIS (wo-RIS, NR = 0). It can be seen that the performance
s is very low and constant when NR = 0. The main reason for
αriser
r
LoS 1 NLoS
M̃j,R = M̃j,R + M̃j,R , (63) this behavior is that with NR = 0, only direct link is used to
αriser + 1 αriser + 1 decode information. Further, one can observe that there is a
significant improvement in performance when NR increases.
where Mj,R ∈ CAE ×NR , j ∈ {1, 2, . . . , NE }, αriser
NLoS However, the achievable EE obtained using EEM algorithm
denotes the Rician factor, and M̃j,R denote the NLoS is higher as compared to random PSM case. This highlights
LoS
component
 whichfollows Rayleigh distribution. M̃j,R = the importance of the resource optimization in the considered
µAE ϑriser,j µH network in order to achieve maximum performance.
NR ϑriser,j represents the LoS component of
AoA AoD
In Fig. 4, we plot the average EE versus the minimum
the link with
  harvested power H̄ for the three different cases with Pmax =
µAE ϑriser,j
AoA
10dBm. As expected, the achievable EE reduces with increase
h
j 2π d sin ϑriser,j
AoA AoA T
j 2πd (A −1) sin ϑriser,j
i in H̄ . The reason for this behavior is that major portion of
= 1, e γ ,··· ,e γ E , (64) Pmax is used by ERs to achieve higher H̄ . Therefore, net
power received at each IR and hence the performance reduces
and
  with increase in H̄ . It can also be seen that among three
µNR ϑriser,j
AoD
cases, EE is maximum for the optimal TPM and PSM case.
iT For better understanding, we highlight the impact of energy
j 2πd sin ϑriser,j
AoD
γ (NR −1) sin ϑriser,j
j 2πd AoD
h
= 1, e γ ,··· ,e , (65) harvesting efficiency (κ) on the EE of the considered network.

VOLUME 10, 2022 29141

 V. Sharma et al.: Pricing-Based Approach for EEM in RIS-Aided Multi-User MIMO SWIPT-Enabled Wireless Networks

TABLE 1. Simulation parameters.

FIGURE 2. Convergence behavior of the EEM algorithm. FIGURE 4. Average EE versus minimum harvested power at each ER (H̄).

FIGURE 3. Average EE versus number of phase shifters. FIGURE 5. Average EE versus minimum harvested power (H̄).

We plot average EE versus minimum harvested power at average EE versus δRIS in Fig. 6 considering δRIS , δBSRIS =
each ER (H̄ ) in Fig. 5 for different values of harvesting δRISER = δRISIR . It can be observed that because of the
efficiency at each user (KE = 4). In particular, we compare reduction in the strength of the reflected signal from RIS,
different/random efficiency (κ1 = 0.3, κ2 = 0.4, κ3 = the average EE for optimal and random case decreases with
0.5, κ4 = 0.6) with equal efficiency (κ1 = κ2 = κ3 = κ4 = increase with δRIS . Additionally, the performance is worst for
κ) at each user. As expected, performance increases with the wo-RIS case as compared to other cases. However, for
increase in efficiency because of increase in net harvested δRIS ≈ 3, the performance of all the case are approximately
power. same. Thus, using RIS in such scenarios has no significant
All the above figures are obtained for fixed pathloss expo- impact on the network performance.
nent of each link. However, depending on the environmen- Fig. 7 compares the average EE obtained after considering
tal disturbances/obstacles such as buildings, trees, etc., the only direct link and only RIS links. Here, δBSIR = δBSER =
value of these parameters may change drastically. Therefore, δD . As expected, the EE increases for each link with decrease
to understand the impact of these pathloss exponent, we plot in pathloss exponent. In addition to this, it can be seen that,

29142 VOLUME 10, 2022

V. Sharma et al.: Pricing-Based Approach for EEM in RIS-Aided Multi-User MIMO SWIPT-Enabled Wireless Networks

FIGURE 6. Average EE versus RIS-related path loss exponent (δRIS ). FIGURE 8. Average EE versus location of the ER (x ER ).

FIGURE 7. Average EE versus the phase shifters for RIS link and direct FIGURE 9. Average EE versus location of the IR circle center (x IR ).
link.

In similar context, we show the impact of location of the

depending on δD and δRIS , the performance of RIS dominates IRs (xIR ) on the EE performance of all cases and plot average
the performance of direct link after a certain NR . For example, EE versus xIR in Fig. 9. Similar to Fig. 8, the performance for
EE for RIS link is greater than that of the direct link when each case decreases with increase in distance between BS and
δD = 3.6, δRIS = 2 with NR > 100, and δRIS = 2.1 with IR, and optimal case provides the best performance. However,
NR > 140. Similarly, EE for RIS link is greater than that one can observe that the difference between the performance
of the direct link when δD = 3.7, δRIS = 2 ∀ NR , δRIS = of optimal and random case is very low.
2.1 ∀ NR , and δRIS = 2.2 with NR > 100. Thus, however, In Fig. 10, we study the impact of maximum transmit
increase in number of RIS elements can over come the double power (Pmax ) available at BS and observe the average EE gain
fading effect and provide better performance. for each scheme. It is obtained for three different values of
Fig. 8 presents the variation of obtained EE w.r.t. location number of elements at RIS NR . As expected, the average EE
of ER (xER ) with Pmax = 10dBm and BS at the origin. gain increases with increase in maximum transmit power for
Similar to pathloss exponent, the performance for each case each NR . Also, NR = 50 provides much better performance
decreases with increase in distance between BS and ER. than NR = 20 and NR = 10. Moreover, it can also be seen that
The main reason for this behavior is the reduction in effec- after certain Pmax , there is no significant improvement in the
tive power/strength of the received signal due to increase achievable EE. Thus, one can choose Pmax carefully to avoid
in distance. Further, there is a significant improvement in the wastage of power. Moreover, for better insights, we high-
EE obtained using EEM algorithm as compared to other light the impact of CSI estimation error on the performance of
two cases. It can also seen that for high value of xER , the the considered network in Fig. 11. It can be seen that average
performance of optimal and random cases are similar but still EE decreases due to increase in CSI estimation error.
better than wo-RIS case. Therefore, in such case RIS provide Next, in Fig. 12, we plot the average EE versus number
a significant improvement over wo-RIS case. of antennas at BS (AB ) with Pmax = 10dBm and compare

VOLUME 10, 2022 29143

 V. Sharma et al.: Pricing-Based Approach for EEM in RIS-Aided Multi-User MIMO SWIPT-Enabled Wireless Networks

FIGURE 10. Average EE versus maximum transmit power. FIGURE 13. Average EE versus number of energy receivers.

FIGURE 11. Average EE versus perfection and imperfection CSI scenarios. FIGURE 14. Average EE versus number of information receivers.

observe that the difference between RIS-aided and wo-RIS

cases is very high. Thus, using RIS in a multi antenna net-
works significantly improves the network performance.
Finally, we demonstrate the impact of the number of ERs
(NE ) and IRs (NI ) on the network performance. Thus, we plot
EE versus NE and EE versus NI in Fig. 13 and Fig. 14, respec-
tively, for each case. It can be observed that the achievable
EE increases with increase in the number of users in both
the figures. However, observing both the graphs carefully for
NE = NI , one can see that the EE obtained in Fig. 13 is lower
than that of Fig. 14. The reason behind this difference is the
minimum power requirement (EH limit) of ERs considered
in this work. As the number of ER increases, their EH limit
also increases. On the other hand, IR has no minimum power
requirement.
FIGURE 12. Average EE versus number of antennas at BS.

VII. CONCLUSION
the performance obtained using three cases discussed earlier. We investigated the performance of a RIS-adied multi-user
It can be observed that with increase in AB there a significant MIMO SWIPT downlink network where a BS served infor-
improvement in achievable EE. This improvement is obtained mation to multiple information receivers while ensuring
because of the increase in diversity of system which increases a minimum harvested power at multiple energy receivers.
linearly with increase in AB . In addition to this, one can also A joint optimization of transmit precoding matrices and phase

29144 VOLUME 10, 2022

V. Sharma et al.: Pricing-Based Approach for EEM in RIS-Aided Multi-User MIMO SWIPT-Enabled Wireless Networks

shift matrices was formulated for maximizing the network APPENDIX C

utility function considering constraints related to available PROOF OF THEOREM 3
transmit power available at BS, minimum harvested power For the case of exposition, we define
required at each ER, and unit modulus phase shift condi- 1
n o  
tion at RIS. Due to non-convex nature of this problem, we ξ̂ (φ) = φ H ϒφ + 2Re φ H m∗ + tr Mb T̃ , (74)
divided it into two sub-problems and solve them separately. 1 H
ξ̄ (φ|φ (n) ) = −φ (n) ϒφ (n)
In particular, we optimized TPM and PSM separately using h  i  
successive convex approximation and bisection search based + 2Re φ H m∗ + ϒφ (n) +tr Mb T̃ , (75)
algorithms. Further, we proposed an EEM algorithm based on
with ξ̂ (φ (n) ) = ξ̄ (φ (n) |φ (n) ). Since, from (53), φ (n+1) sat-
block coordinate descent method that provided the optimal
isfies (44) and ξ̄ (φ (n+1) |φ (n) ) ≥ H̄ . Also, ξ̂ (φ (n+1) ) ≥
solution of the main problem by iteratively obtaining sub-
ξ̄ (φ (n+1) |φ (n) . So, ξ̂ (φ (n+1) ) ≥ H̄ . Therefore φ (n+1) satisfies
optimal TPM, PSM, and network price using their respective
(13c). Hence {φ (n) , n = 1, 2 . . .} is a
feasible solution of
algorithms. Furthermore, we demonstrated the convergence
(33). Further,
From (47), ξ φ (n+1)
|φ
(n)
= χ(φ (n+1) ). Also,
behavior of the proposed algorithm using simulation results.
ξ φ (n+1) (n)
≤ ξ φ |φ . Thus, ξ φ
(n) (n) (n+1) (n)


|φ |φ ≤
Additionally, we also highlighted the impact of RIS in mul-
χ(φ (n) ). Similarly, from (47), ξ φ (n+1) |φ (n) ≥ χ(φ (n+1) ).


tiuser MIMO SWIPT networks. The importance of use of
multiple antennas was also highlighted. Moreover, in order Therefore, χ(φ (n) ) ≥ χ(φ (n+1) ). Hence, sequences
to avoid the wastage of power, the importance of judicious {χ (φ (n+1) ), n = 1, 2, · · · } is monotonically decreasing and
choice of available transmit power was also discussed. converges after certain n.
Using KKT condition, the Lagrange function of problem
APPENDIX A (52) can be expressed as
 h
i
PROOF OF LEMMA 1 L(φ, ν, τ ) = ξ (φ|φ ? ) + ν Ĥ − 2Re φ H m∗ + ϒφ ?
From (24), we can write
NR
L[T(λ1 ), λ1 ] ≤ L[T(λ2 ), λ1 ].
X
(67) + τi (|φi | − 1), (76)
where T(λ1 ) and T(λ2 ) are the solutions to the problem (25) i=1
corresponding to λ1 and λ2 , respectively, with λ1 > λ2 . where {φ ? } denotes global optimum solution, ν and τ =
Similarly, we obtain {τ1 , · · · , τm } denotes the dual variables. These variables must
satisfy the following conditions
L[T(λ2 ), λ2 ] ≤ L[T(λ1 ), λ2 ]. (68)
∇φ ∗ L(φ, ν, τ )|φ=φ ? = ∇φ ∗ ξ (φ|φ ? )|φ=φ ? −ν ? m∗ +ϒφ ?


Adding (67) and (68), and after some algebra, we obtain
NR
P (λ1 ) ≤ P (λ2 ) . τi? ∇φ ∗ φ i |φ=φ ? = 0,
X

(69) + (77)
i=1
APPENDIX B
 h
i
ν ? Ĥ − 2Re φ ? H m∗ + ϒφ ? = 0, (78)
PROOF OF THEOREM 2
Suppose φ ? represents the global optimum solution of (53) τi? φ ?i − 1 = 0, ∀i.


(79)
and φ(p? ) is the global optimum solution of (54). Now, From (47), we obtain
assuming φ ? 6 = φ(p? ), we obtain
n o n o ∇φ ∗ ξ (φ|φ ? )|φ=φ ? = ∇φ ∗ χ(φ)|φ=φ ? . (80)
2Re φ(p? )H w(n) < 2Re φ ?H w(n) . (70)
Further, ∇φ ∗ χ(φ)|φ=φ ? = ∇φ ∗ ϕ(φ)|φ=φ ? , where ϕ(φ)
However, φ(p? )
is also the global optimum. Thus, we obtain denotes the OF of problem (33). From (80), ∇φ ∗ g(φ|φ ? )|φ=φ ?
n
? H (n)
o h  i = ∇φ ∗ ϕ(φ)|φ=φ ? . Substituting this into (78), we have
2Re φ(p ) w + 2p? Re φ(p? )H m∗ + ϒφ (n)
∇φ ∗ ϕ(φ)|φ=φ ? − ν ? m∗ + ϒφ ?


n o h  i
≥ 2Re φ ?H w(n) + 2p? Re φ ?H m∗ + ϒφ (n) . NR
τi? ∇φ ∗ |φi | |φ=φ ? = 0.
X

(71) + (81)
i=1
According to [37], φ ? and φ(p? ) should satisfy
 h  i  (78), (79) and (81) satisfies the KKT conditions.
2Re φ̂ H m∗ + ϒφ (n) − Ĥ = 0, (72)
APPENDIX D
where φ̂ ∈ {φ ? , φ(p? )}. Therefore, substituting (72) in (71), PROOF OF THEOREM 4
we obtain Sufficiency Part: Let Since T? is the optimal precoder with
n o n o respect to the optimal price q? , it implies that
2Re φ(p? )H w(n) ≥ 2Re φ ?H w(n) . (73)
? ?
 
R T(n) , 8(n)
Contradiction between (73) and (70) falsify our assumption. q̃ , ?

Hence, φ ? = φ(p? ). This proves the Theorem 2. Ptot T(n)

VOLUME 10, 2022 29145

 V. Sharma et al.: Pricing-Based Approach for EEM in RIS-Aided Multi-User MIMO SWIPT-Enabled Wireless Networks

R T(n) , 8(n)

  
= −tr V?n ∇T∗n On J? , T, 8? T =T?
 

≥
, ∀ T(n) , 8(n) ∈ S ∩ B, (82) (92)
n
Ptot T(n)   n
? ?
∇T∗n On (J , T, 8 ) T =T?
where S depicts the feasible set of the problem (13) and B = −tr 

n n  (93)
? ? ? ?
On J , T , 8 ?

represents the norm ball set centered at T(n) , 8(n) with
= ∇T∗n log On J? , T, 8?
? ?
 



the radius r > 0. Thus, T(n) , 8(n) is the local maximizer T =T? n n
(94)
for the price q̃ in the problem (13) because = ∇T∗n Rn (T, 8? ) Tn =T?n
. (95)
   
R T(n) , 8(n) − q̃Ptot T(n) ≤ 0 Here (92) corresponds to the chain rule and (95) is obtained
 ? ?
  ?
 after simplifying (19) using Woodbury matrix identity.
= R T(n) , 8(n) − q̃Ptot T(n) . (83)
From (95) and (90), we obtain
From (82) and (83), it can be concluded that q̃ is theopti-
? ?
∇T∗n z(T, 8? ) Tn =T?n
= ∇T∗n Rn (T, 8? ) Tn =T?n
. (96)
mal price such that the obtained solution T(n) , 8(n) can
Similarly, from (96) and (90), we have
achieve the local maximum ofthe EE, and the? balance equa-
? ?
∇T∗n Rn (T, 8? ) + λ? T?n − µ? MT?n = 0,

tion is given as R T , 8
(n) (n) − q̃Ptot T(n) = 0. Tn =T?n
(97)
Necessity Part: Using the balance equation, we have where n ∈ {1, 2, . . . , NI }. Further, we can write
   
R T(n) , 8(n) − q? Ptot T(n) ∇φ ∗ l V? , J? , T? , 8 |φ=φ ? = ∇φ ∗ ϕ(φ)|φ=φ ? .


(98)
? ? ?
   
≤ R T(n) , 8(n) −q? Ptot T(n) = 0. (84) Similar to (91)-(95), we have
∇φ ∗ l V? , J? , T? , 8 |φ=φ ? = ∇φ ∗ Rn (φ, T? ) . (99)


In other words, we can write φ=φ ?
? ?
 
R T(n) , 8(n) Also, we can write
R T(n) , 8(n)



≤ q? = ?
. (85) ∇φ ∗ ϕ(φ)|φ=φ ? = ∇φ ∗ Rn (φ, T? ) . (100)
Ptot T(n) Ptot T(n) φ=φ ?
Using (100) into (81), we obtain
APPENDIX E
∇φ ∗ Rn (φ, T? ) − ν ? m∗ + ϒφ ?


PROOF OF THEOREM 5 φ=φ ?
Firstly, the monotonic property of Algorithm 5 can be proved NR
τi? ∇φ ∗ |φi | |φ=φ ? = 0.
X
similar to [28]. Next, using KKT conditions corresponding to


+ (101)
problem (20), its Lagrange function can be expressed as i=1
NI
!
Hence, (97), (88), (89) and (101) satisfies the KKT
?
X
L(T, λ, ρ) = z(T, 8 ) + λ 2
kTn kF − PB conditions.
n=1
NI
!!
X REFERENCES
+ ρ H̄ − tr TH
n MTn . (86) [1] G. Gui, M. Liu, F. Tang, N. Kato, and F. Adachi, ‘‘6G: Opening new hori-
n=1 zons for integration of comfort, security, and intelligence,’’ IEEE Wireless
Commun., vol. 27, no. 5, pp. 126–132, Oct. 2020.
Similar to (76), dual variables λ and ρ must satisfy follow- [2] Q. Wu, S. Zhang, B. Zheng, C. You, and R. Zhang, ‘‘Intelligent reflecting
ing conditions surface-aided wireless communications: A tutorial,’’ IEEE Trans. Com-
mun., vol. 69, no. 5, pp. 3313–3351, May 2021.
∇T∗n L(T, λ, ρ) Tn =T?n
[3] A. K. Iyer, A. Alu, and A. Epstein, ‘‘Metamaterials and metasurfaces—
Historical context, recent advances, and future directions,’’ IEEE Trans.
?
= ∇T∗n z(T, 8 ) Tn =T?n
+ λ? T?n − ρ ? MT?n = 0, (87) Antennas Propag., vol. 68, no. 3, pp. 1223–1231, Mar. 2020.
[4] W. Tang, M. Z. Chen, X. Chen, J. Y. Dai, Y. Han, M. Di Renzo, Y. Zeng,
NI
!
S. Jin, Q. Cheng, and T. J. Cui, ‘‘Wireless communications with recon-
λ? T?n
2
X
F
− PB = 0, (88) figurable intelligent surface: Path loss modeling and experimental mea-
surement,’’ IEEE Trans. Wireless Commun., vol. 20, no. 1, pp. 421–439,
n=1
Jan. 2021.
NI
!!
[5] L. Yang, J. Yang, W. Xie, M. O. Hasna, T. Tsiftsis, and M. D. Renzo,
ρ ? H̄ − tr T?H ?
X
n MTn = 0. (89) ‘‘Secrecy performance analysis of RIS-aided wireless communication
systems,’’ IEEE Trans. Veh. Technol., vol. 69, no. 10, pp. 12296–12300,
n=1
Oct. 2020.
Also, we have [6] D. Song, W. Shin, and J. Lee, ‘‘A maximum throughput design for wire-
less powered communication networks with IRS-NOMA,’’ IEEE Wireless
∇T∗n l V? , J? , T, 8? , = ∇T∗n z(T, 8? ) , (90)


Commun. Lett., vol. 10, no. 4, pp. 849–853, Apr. 2021.
Tn =T?n Tn =T?n
[7] O. Özdogan, E. Björnson, and E. G. Larsson, ‘‘Intelligent reflecting
1 PNI surfaces: Physics, propagation, and pathloss modeling,’’ IEEE Wireless
where n ∈ NI and l (V, J, T, 8) = n=1 ln (V, J, T, 8). Commun. Lett., vol. 9, no. 5, pp. 581–585, May 2020.
Further, we use following inequalities [8] H. Guo, Y.-C. Liang, J. Chen, and E. G. Larsson, ‘‘Weighted sum-rate max-
imization for reconfigurable intelligent surface aided wireless networks,’’
∇T∗n ln V? , J? , T, 8? T =T?


(91) IEEE Trans. Wireless Commun., vol. 19, no. 5, pp. 3064–3076, May 2020.
n n

29146 VOLUME 10, 2022

V. Sharma et al.: Pricing-Based Approach for EEM in RIS-Aided Multi-User MIMO SWIPT-Enabled Wireless Networks

[9] S. Abeywickrama, R. Zhang, Q. Wu, and C. Yuen, ‘‘Intelligent reflecting [31] K. Xiong, B. Wang, and K. J. R. Liu, ‘‘Rate-energy region of SWIPT
surface: Practical phase shift model and beamforming optimization,’’ IEEE for MIMO broadcasting under nonlinear energy harvesting model,’’ IEEE
Trans. Commun., vol. 68, no. 9, pp. 5849–5863, Sep. 2020. Trans. Wireless Commun., vol. 16, no. 8, pp. 5147–5161, Aug. 2017.
[10] M.-M. Zhao, Q. Wu, M.-J. Zhao, and R. Zhang, ‘‘Intelligent reflecting sur- [32] B. K. Chalise, W.-K. Ma, Y. D. Zhang, H. A. Suraweera, and M. G. Amin,
face enhanced wireless networks: Two-timescale beamforming optimiza- ‘‘Optimum performance boundaries of OSTBC based AF-MIMO relay
tion,’’ IEEE Trans. Wireless Commun., vol. 20, no. 1, pp. 2–17, Jan. 2021. system with energy harvesting receiver,’’ IEEE Trans. Signal Process.,
[11] P. Wang, J. Fang, X. Yuan, Z. Chen, and H. Li, ‘‘Intelligent reflect- vol. 61, no. 17, pp. 4199–4213, Sep. 2013.
ing surface-assisted millimeter wave communications: Joint active and [33] Q. Shi, M. Razaviyayn, Z.-Q. Luo, and C. He, ‘‘An iteratively weighted
passive precoding design,’’ IEEE Trans. Veh. Technol., vol. 69, no. 12, MMSE approach to distributed sum-utility maximization for a MIMO
pp. 14960–14973, Oct. 2020. interfering broadcast channel,’’ IEEE Trans. Signal Process., vol. 59, no. 9,
[12] B. Zheng, C. You, and R. Zhang, ‘‘Efficient channel estimation for double- pp. 4331–4340, Sep. 2011.
IRS aided multi-user MIMO system,’’ IEEE Trans. Commun., vol. 69, [34] C. Pan, H. Ren, M. Elkashlan, A. Nallanathan, and L. Hanzo, ‘‘The non-
no. 6, pp. 3818–3832, Jun. 2021. coherent ultra-dense C-RAN is capable of outperforming its coherent
counterpart at a limited fronthaul capacity,’’ IEEE J. Sel. Areas Commun.,
[13] K. Singh and M. L. Ku, ‘‘Toward green power allocation in relay-assisted
vol. 36, no. 11, pp. 2549–2560, Nov. 2018.
multiuser networks: A pricing-based approach,’’ IEEE Trans. Wireless
[35] C. Pan, H. Ren, M. Elkashlan, A. Nallanathan, and L. Hanzo, ‘‘Robust
Commun., vol. 14, no. 5, pp. 2470–2486, May 2015.
beamforming design for ultra-dense user-centric C-RAN in the face of
[14] R. Zhang and C. K. Ho, ‘‘MIMO broadcasting for simultaneous wireless
realistic pilot contamination and limited feedback,’’ IEEE Trans. Wireless
information and power transfer,’’ IEEE Trans. Wireless Commun., vol. 12,
Commun., vol. 18, no. 2, pp. 780–795, Feb. 2019.
no. 5, pp. 1989–2001, May 2013.
[36] M. Grant and S. Boyd. (Mar. 2014). CVX: MATLAB Software for
[15] G. Dong, H. Zhang, and D. Yuan, ‘‘Downlink achievable rate of massive Disciplined Convex Programming, Version 2.1. [Online]. Available:
MIMO enabled SWIPT systems over Rician channels,’’ IEEE Commun. http://cvxr.com/cvx
Lett., vol. 22, no. 3, pp. 578–581, Mar. 2018. [37] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge, U.K.:
[16] G. Miao, N. Himayat, and G. Y. Li, ‘‘Energy-efficient link adaptation Cambridge Univ. Press, Aug. 2004.
in frequency-selective channels,’’ IEEE Trans. Commun., vol. 58, no. 2, [38] C. Pan, H. Ren, K. Wang, W. Xu, M. Elkashlan, A. Nallanathan,
pp. 545–554, Feb. 2010. and L. Hanzo, ‘‘Multicell MIMO communications relying on intelligent
[17] R. Liu, M. Li, Q. Liu, and A. L. Swindlehurst, ‘‘Joint symbol-level precod- reflecting surfaces,’’ IEEE Trans. Wireless Commun., vol. 19, no. 8,
ing and reflecting designs for IRS-enhanced MU-MISO systems,’’ IEEE pp. 5218–5233, Aug. 2020.
Trans. Wireless Commun., vol. 20, no. 2, pp. 798–811, Feb. 2021. [39] C. Pan, W. Xu, W. Zhang, J. Wang, H. Ren, and M. Chen, ‘‘Weighted
[18] Q. Wu, Y. Zhang, C. Huang, Y. Chau, Z. Yang, and M. Shikh-Bahaei, sum energy efficiency maximization in ad hoc networks,’’ IEEE Wireless
‘‘Energy efficient intelligent reflecting surface assisted terahertz communi- Commun. Lett., vol. 4, no. 3, pp. 233–236, Jun. 2015.
cations,’’ in Proc. IEEE Int. Conf. Commun. Workshops (ICC Workshops), [40] X.-D. Zhang, Matrix Analysis and Applications. Cambridge, U.K.:
Jun. 2021, pp. 1–6. Cambridge Univ. Press, Oct. 2017.
[19] L. You, J. Xiong, D. W. K. Ng, C. Yuen, W. Wang, and X. Gao, [41] Y. Sun, P. Babu, and D. P. Palomar, ‘‘Majorization-minimization algo-
‘‘Energy efficiency and spectral efficiency tradeoff in RIS-aided mul- rithms in signal processing, communications, and machine learning,’’ IEEE
tiuser MIMO uplink transmission,’’ IEEE Trans. Signal Process., vol. 69, Trans. Signal Process., vol. 65, no. 3, pp. 794–816, Feb. 2017.
pp. 1407–1421, 2021. [42] J. Song, P. Babu, and D. P. Palomar, ‘‘Sequence design to minimize
[20] W. Yan, X. Yuan, Z.-Q. He, and X. Kuai, ‘‘Passive beamforming and the weighted integrated and peak sidelobe levels,’’ IEEE Trans. Signal
information transfer design for reconfigurable intelligent surfaces aided Process., vol. 64, no. 8, pp. 2051–2064, Apr. 2016.
multiuser MIMO systems,’’ IEEE J. Sel. Areas Commun., vol. 38, no. 8, [43] J. Xu, L. Liu, and R. Zhang, ‘‘Multiuser MISO beamforming for simultane-
pp. 1793–1808, Aug. 2020. ous wireless information and power transfer,’’ IEEE Trans. Signal Process.,
[21] N. Hehao and L. Ni, ‘‘Intelligent reflect surface aided secure transmission vol. 62, no. 18, pp. 4798–4810, Sep. 2014.
in MIMO channel with SWIPT,’’ IEEE Access, vol. 8, pp. 192132–192140,
2020.
[22] Q. Wu and R. Zhang, ‘‘Weighted sum power maximization for intelligent VAIBHAV SHARMA received the Bachelor of
reflecting surface aided SWIPT,’’ IEEE Wireless Commun. Lett., vol. 9, Technology degree in electrical engineering from
no. 5, pp. 586–590, May 2020. Dr. A. P. J. Abdul Kalam Technical Univer-
[23] G. Zhou, C. Pan, H. Ren, K. Wang, and A. Nallanathan, ‘‘A frame- sity, Lucknow, Uttar Pradesh, India, in 2020.
work of robust transmission design for IRS-aided MISO communications He is currently pursuing the Master of Science
with imperfect cascaded channels,’’ IEEE Trans. Signal Process., vol. 68, degree with the International Master’s Program
pp. 5092–5106, 2020. in Telecommunication Engineering, Institute
[24] C. Zhang, W. Yi, Y. Liu, Z. Ding, and L. Song, ‘‘STAR-IOS aided of Communications Engineering, National Sun
NOMA networks: Channel model approximation and performance anal- Yat-sen University (NSYSU), Kaohsiung, Taiwan.
ysis,’’ 2021, arXiv:2107.01543. His research interests include MIMO, reconfig-
[25] L. Wei, C. Huang, G. C. Alexandropoulos, C. Yuen, Z. Zhang, and urable intelligent surfaces (RIS), simultaneous wireless information and
M. Debbah, ‘‘Channel estimation for RIS-empowered multi-user MISO
power transfer (SWIPT), and non-orthogonal multiple access (NOMA).
wireless communications,’’ IEEE Trans. Commun., vol. 69, no. 6,
pp. 4144–4157, Mar. 2021.
[26] S. Dhok, P. Raut, P. K. Sharma, K. Singh, and C.-P. Li, ‘‘Non-linear energy
harvesting in RIS-assisted URLLC networks for industry automation,’’ JETTI YASWANTH received the bachelor’s
IEEE Trans. Commun., vol. 69, no. 11, pp. 7761–7774, Nov. 2021. degree in electronics and communication engi-
[27] P. K. Sharma and P. Garg, ‘‘Intelligent reflecting surfaces to achieve the neering from the Swarnandhra College of
full-duplex wireless communication,’’ IEEE Commun. Lett., vol. 25, no. 2, Engineering and Technology, Andhra Pradesh,
pp. 622–626, Feb. 2021. India, in 2020. He is currently pursuing the
[28] C. Pan, H. Zhu, N. J. Gomes, and J. Wang, ‘‘Joint precoding and RRH Master of Science degree with the International
selection for user-centric green MIMO C-RAN,’’ IEEE Trans. Wireless Master’s Program in Telecommunication Engi-
Commun., vol. 16, no. 5, pp. 2891–2906, May 2017. neering, Institute of Communications Engineer-
[29] S. N. Deepa and B. S. S. Rani, ‘‘RF energy harvesting using 900 MHz of ing, National Sun Yat-sen University (NSYSU),
mobile signal frequency to charging the mobile battery,’’ in Proc. Int. Conf. Kaohsiung, Taiwan. His current interests include
Innov. Green Energy Healthcare Technol. (IGEHT), Mar. 2017, pp. 1–5. the multi-input multi-output (MIMO), reconfigurable intelligent surface
[30] H. H. M. Tam, H. D. Tuan, A. A. Nasir, T. Q. Duong, and H. V. Poor, (RIS), energy efficiency (EE), non-orthogonal multiple access (NOMA),
‘‘MIMO energy harvesting in full-duplex multi-user networks,’’ IEEE simultaneous wireless information and power transfer (SWIPT), and energy
Trans. Wireless Commun., vol. 16, no. 5, pp. 3282–3297, May 2017. harvesting (EH).

VOLUME 10, 2022 29147

 V. Sharma et al.: Pricing-Based Approach for EEM in RIS-Aided Multi-User MIMO SWIPT-Enabled Wireless Networks

SANDEEP KUMAR SINGH (Member, IEEE) KESHAV SINGH (Member, IEEE) received the
received the B.E. degree in electronics and com- M.Sc. degree in information and telecommuni-
munication engineering from RGPV University, cations technologies from Athens Information
Bhopal, India, in 2010, the M.-Tech. degree in Technology, Greece, in 2009, and the Ph.D.
ME and VLSI design from MNNIT Allahabad, degree in communication engineering from the
Prayagraj, India, in 2013, and Ph.D. degree in National Central University, Taiwan, in 2015.
electronics and communication engineering from From 2016 to 2019, he was a Research Associate
VNIT Nagpur, India, in 2020. He is currently a with the Institute of Digital Communications, The
Postdoctoral Researcher at the Institute of Com- University of Edinburgh, U.K. From 2019 to 2020,
munications Engineering, National Sun Yat-sen he was associated as a Research Fellow with
University (NSYSU), Taiwan. His current research interests include the University College Dublin, Ireland. He currently works as an Assis-
unmanned aerial vehicles, green communications, resource allocation, full- tant Professor with the Institute of Communications Engineering, National
duplex radio, ultra-reliable low-latency communication, non-orthogonal Sun Yat-sen University (NSYSU), Taiwan. His research interests include
multiple access (NOMA), rate splitting multiple access (RSMA), machine green communications, resource allocation, full-duplex radio, ultra-reliable
learning for communications, and large intelligent surface-assisted commu- low-latency communication, non-orthogonal multiple access, (NOMA), rate
nications, massive MIMO, and OFDM. splitting multiple access (RSMA), wireless edge caching, machine learning
for communications, and large intelligent surface-assisted communications.

FAHEEM KHAN (Member, IEEE) received the

bachelor’s degree in electronics engineering from
Aligarh Muslim University, India, the master’s
SUDIP BISWAS (Member, IEEE) received the degree in communication and radar engineer-
Ph.D. degree in digital communications from The ing from the Indian Institute of Technology
University of Edinburgh (UEDIN), Edinburgh, Delhi, India, and the Ph.D. degree in electri-
U.K., in 2017. He leads research on signal process- cal and electronic engineering from the Institute
ing for wireless communications, with particular of Electronics, Communications and Information
focus on 5G’s long-term evolution, including Technology, Queen’s University Belfast, U.K.,
transceiver design for full-duplex radios, wire- in 2012. His Ph.D. research focused on ‘‘Spectrum
less edge caching, and comms-radar co-existence sharing in cognitive radio wireless networks.’’
and intelligent reflector surface assisted commu- He serves as a Course Leader of M.Sc. in electronic and automotive
nication. From 2017 to 2019, he was a Research engineering. He has more than 15 years of teaching and research experience
Associate with the Institute of Digital Communications, UEDIN. He also has at academic and research institutions in the U.K., Sultanate of Oman, United
industrial experience with Tata Consultancy Services, India (Lucknow and Arab Emirates, and India. He is currently a Senior Lecturer in electronic
Kolkata), where he was an Assistant Systems Engineer, from 2010 to 2012. engineering at the School of Computing and Engineering, University of
He currently works as an Assistant Professor with the Department of Elec- Huddersfield. Prior to joining as Lecturer at the University of Huddersfield,
tronics and Communications Engineering, Indian Institute of Information in 2016, he worked as a Research Associate in wireless communications and
Technology Guwahati (IIITG). He was an Organizer of the IEEE Inter- signal processing at the Institute for Digital Communications, The University
national Workshop on Signal Processing Advances in Wireless Commu- of Edinburgh, where he contributed to research and project management in
nications (SPAWC), Edinburgh, in 2016, and has been involved in EU several EU/EPSRC research projects. During this period, he was also actively
FP7 projects remote radio heads and parasitic antenna arrays (HARP) and involved in several EU/U.K. research projects. He is a fellow of Higher
dynamic licensed shared access (ADEL), a DST UKIERI project on wireless Education Academy (FHEA).
edge caching, and an EPSRC project on NoMA.

29148 VOLUME 10, 2022