IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 21, NO.

5, MAY 2022 3083

Simultaneously Transmitting and Reflecting (STAR)

RIS Aided Wireless Communications
Xidong Mu , Graduate Student Member, IEEE, Yuanwei Liu , Senior Member, IEEE, Li Guo ,
Member, IEEE, Jiaru Lin , Member, IEEE, and Robert Schober, Fellow, IEEE

Abstract— The novel concept of simultaneously transmitting and Index Terms— Operating protocols, reconfigurable intelligent
reflecting (STAR) reconfigurable intelligent surfaces (RISs) is surfaces, simultaneous transmission and reflection, unicast and
investigated, where the incident wireless signal is divided into multicast communication.
transmitted and reflected signals passing into both sides of the space
surrounding the surface, thus facilitating a full-space manipulation of
signal propagation. Based on the introduced basic signal model of I. INTRODUCTION

W
‘STAR’, three practical operating protocols for STAR-RISs are
proposed, namely energy splitting (ES), mode switching (MS), and ITH the worldwide commercialization of fifth
time switching (TS). Moreover, a STAR-RIS aided downlink generation
communication system is considered for both unicast and multicast (5G) wireless communication networks, growing research efforts
transmission, where a multi-antenna base station (BS) sends
information to two users, i.e., one on each side of the STAR-RIS. A are being devoted to the upcoming beyond 5G (B5G) and sixth-
power consumption minimization problem for the joint optimization generation (6G) wireless com munication networks, which are
of the active beamforming at the BS and the passive transmission expected to impose more stringent requirements such as
and reflection beamforming at the STAR-RIS is formulated for each extremely high spectrum- and energy-efficiency, microsecond
of the proposed operating protocols, subject to communication rate latency, and full-dimensional network coverage [2], [3]. To
constraints of the users. For ES, the resulting highly-coupled non-
convex optimization problem is solved by an iterative algorithm, achieve these goals, exten sions to the existing communication
which exploits the penalty method and successive convex technologies [4], which have already been proven effective, have
approximation. Then, the proposed penalty-based iterative algorithm been proposed, including but not limited to ultra-massive
is extended to solve the mixed-integer non-convex optimization multiple-input multiple-output (UM-MIMO), ultra-dense
problem for MS. For TS, the optimization problem is decomposed networks (UDNs), and terahertz (THz) communication. However,
into two subproblems, which can be consecutively solved using state-
of the-art algorithms and convex optimization techniques. Finally, on the one hand, the relentless increase of the number of
our numerical results reveal that: 1) the TS and ES operating antennas/base stations (BSs) and the use of very high carrier
protocols are generally preferable for unicast and multicast trans frequencies will potentially cause high energy consumption and
mission, respectively; and 2) the required power consumption for hardware cost, since more power-hungry and costly radio
both scenarios is significantly reduced by employing the proposed frequency (RF) chains have to be installed for signal conversion.
On the other hand, deploying such a large number of active
Manuscript received April 3, 2021; revised July 30, 2021; accepted September
components operating at very high frequencies in wireless
27, 2021. Date of publication October 13, 2021; date of current version May 10, networks may not always be beneficial since they also introduce
2022. This work was supported in part by Beijing Natural Science Foundation new chal lenges, such as complicated inter-user/cell interference
under Grant L192032, in part by the National Key Research and Development
Program of China under Grant 2019YFB1406500, in part by the Key Project Plan scenar ios, pilot contamination, and severe hardware
of Blockchain in Ministry of Education of the People’s Republic of China under impairments.
Grant 2020KJ010802, and in part by the National Natural Science Foundation of To overcome the above limitations, new cost-effective
China under Grant 61771066. The work of Xidong Mu was supported by China
Scholarship Council. This article is to be presented at the Asilomar Conference on techniques have to be developed for wireless communica tion
Signals, Systems, and Computers, in 2021. The associate editor coordinating the systems. Specifically, motivated by the rapid develop ment of
review of this article and approving it for publication was X. Yuan. metasurfaces and corresponding advanced fabrication
(Corresponding author: Li Guo.)
Xidong Mu, Li Guo, and Jiaru Lin are with the Key Laboratory of Universal technologies, reconfigurable intelligent surfaces (RISs) have
Wireless Communications, Ministry of Education, Beijing University of Posts and emerged as promising solutions [5]–[9]. A RIS is a man made
Telecommunications, Beijing 100876, China, and also with the School of Artificial two-dimensional (2D) surface, which is equipped with a large
Intelligence, Beijing University of Posts and Telecommunications, Beijing 100876,
China (e-mail: [email protected]; [email protected]; [email protected]). number of low-cost passive elements. With the aid of a smart
Yuanwei Liu is with the School of Electronic Engineering and Computer controller attached to the RIS, the propagation of the wireless
Science, Queen Mary University of London, London E1 4NS, U.K. (e-mail:
[email protected]).
signals incident on the RIS can be adjusted through the
Robert Schober is with the Institute for Digital Communications, Friedrich- customized phase response of each element. Therefore, RIS can
Alexander-Universität Erlangen-Nürnberg (FAU), 91054 Erlangen, Germany (e- facilitate “Smart Radio Environments (SREs)” [6]. Compared to
mail: [email protected]).
Color versions of one or more figures in this article are available at
the conventional multi-antenna and relaying concepts, where
https://doi.org/10.1109/TWC.2021.3118225. wireless signals are actively produced using costly RF chains,
Digital Object Identifier 10.1109/TWC.2021.3118225 RISs only passively recycle signals that are already available in
STAR-RIS instead of conventional reflecting/transmitting-only RISs. the network and do not require RF chains. Thus, RISs are more
economical and environmentally
1536-1276 © 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: University of Ulsan. Downloaded on January 07,2024 at 07:58:57 UTC from IEEE Xplore. Restrictions apply.
3084 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 21, NO. 5, MAY 2022

friendly compared to conventional active antenna systems.

Furthermore, RISs can be seamlessly integrated into existing
wireless networks by deploying them on diverse structures,
such as roadside billboards, building facades, windows, and
even human clothes [7].
A. Prior Works
Motivated by the aforementioned favorable characteristics
of RISs, extensive research efforts have been devoted to
exploiting the new degrees-of-freedom (DoFs) introduced by
RISs with the prospect of mitigating a wide range of chal
lenges encountered in wireless networks, such as the reduction
of transmit power consumption [10]–[12], the improvement of
spectrum- and energy-efficiency [13]–[18], and the establish
ment of unobstructed communication links [19], [20]. More
specifically, the authors of [10] proposed a suboptimal alter However, this geographical restriction may not always be met in
nating optimization based algorithm to minimize the transmit practice, and gravely restricts the flexibility and effective ness of
power of the access point (AP) by jointly optimizing the active RISs, as generally users may be located on both sided of a RIS.
beamforming at the AP and the passive beamform ing at the RIS. To overcome this limitation, the novel concept of simultaneously
The transmit power minimization problem was further studied by transmitting and reflecting RISs (STAR-RISs) was proposed in
the authors of [11] for single-user systems, where a globally [24] and [25]. In particular, as shown in Fig. 1, the wireless signal
optimal solution was obtained with a branch-and-bound method. incident on an element of a STAR-RIS from either side of the
Furthermore, the authors of [12] minimized the transmit power of surface is divided into two parts [26]. One part (reflected signal)
RIS-aided multi user systems employing orthogonal multiple is reflected to the same space as the incident signal, i.e., the
access (OMA) and non-orthogonal multiple access (NOMA), reflection space, and the other part (transmitted signal) is
respectively. Considering a practical power consumption model transmitted to the opposite space as the incident signal, i.e., the
for the RIS elements, the authors of [13] formulated a joint active transmission space. As shown in [24], by manipulating both the
and passive beamforming optimization problem for the maximiza electric and magnetic currents of a STAR-RIS element, the
tion of the network energy-efficiency. The joint beamforming transmitted and reflected signals can be reconfigured via two
optimization problem was further studied by the authors of [14] generally independent coefficients, namely the transmission and
for a RIS-assisted downlink multiple-input single output (MISO) the reflection coefficients. There fore, a highly flexible full-space
NOMA system, where the system sum rate was maximized for SRE can be realized. Two prototypes, which resemble STAR-
continuous and discrete RIS phase shifters, respectively. RISs, have been developed using metasurfaces [26], [27]. For the
Furthermore, the communication capacity of a RIS-aided MIMO prototype proposed in [26], each element was composed of a
system and a RIS-aided multi-user system was characterized in parallel resonant LC tank and small metallic loops to provide the
[15] and [16], respectively. Due to the nearly passive nature of required electric and magnetic surface reactance. Despite the
RISs, the accurate acquisition of channel state information (CSI) above advantages, the investigation of how STAR-RISs can be
is a challenging task. Hence, the authors of [17] investigated integrated into wireless communication systems is still in its
robust resource allocation for a RIS-assisted full-duplex cognitive infancy. Based on a similar idea as STAR-RISs, an intelligent
radio system with imperfect CSI. The authors of [18] invoked omni-surface (IOS) was proposed by the authors of [28] to
deep reinforcement learning to optimize the active and passive achieve full space coverage. However, unlike for STAR-RISs, for
beamforming for maximization of the system sum rate based on IOSs, the phase shifts for transmission and reflection are
partial CSI. To overcome signal blockages, the authors of [19] identical, as they are determined by the state of positive intrinsic
proposed to use two RISs to connect a BS to users in a signal negative (PIN) diodes installed in the IOS elements. In another
dead zone, where cooperative passive beamforming among RISs early work [24], the authors focused on the investigation of
was pro posed. A similar problem was further studied for multiple hardware models and channel models for STAR-RISs. However,
RISs in [20]. The significant benefits of deploying RISs in to the best of the authors’ knowledge, efficient operating
wireless networks have also been verified in the context of protocols for STAR-RISs and corresponding joint transmission
physical layer security [21], unmanned aerial vehicle systems and reflection beamforming optimization tech niques for STAR-
[22], and robotic communications systems [23]. RIS aided wireless networks have not been studied, yet, which
provides the main motivation for this work.
B. Motivations and Contributions
To exploit the full potential of STAR-RISs, in this paper, we
Most of the existing research contributions consider the case propose practical protocols for the operation of STAR-RISs and
where the RISs are only able to reflect the incident wireless investigate the joint transmission and reflection beam forming
signal (referred to as conventional reflecting-only RISs). In this design for both unicast and multicast communication.
case, transmitter and receiver have to be located on the same side
of the RIS [10]–[23], thus leading to a half-space SRE.
Fig. 1. The concept of STAR-RISs.

Authorized licensed use limited to: University of Ulsan. Downloaded on January 07,2024 at 07:58:57 UTC from IEEE Xplore. Restrictions apply.
MU et al.: STAR RIS AIDED WIRELESS COMMUNICATIONS 3085

The main contributions of this paper can be summarized as Considering both unicast and multicast communication, we
follows: formulate a joint active and passive beamforming
• We propose three practical operating protocols for STAR- optimization problem for each of the proposed operating
RISs, namely energy splitting (ES), mode switch ing (MS), protocols for minimization of the power consumption of the
and time switching (TS), along with their respective benefits BS, while satisfying the quality-of-service (QoS)
and drawbacks. requirements of the users.
• We consider a STAR-RIS aided downlink communication • For ES, we first transform the resulting highly-coupled non-
system, where a BS sends information to users exploit ing convex optimization problem into a tractable form with
STAR-RIS-enabled transmission and reflection links. decoupled optimization variables. To solve the trans formed
 non-convex problem, we develop an efficient iterative of matrix A, respectively.
algorithm by exploiting the penalty method and successive
convex approximation (SCA). The algorithm is guaranteed
to converge to a stationary point of the orig inal II. BASIC SIGNAL MODEL AND PRACTICAL OPERATING
optimization problem. Furthermore, we extend the proposed PROTOCOLS FOR STAR-RISS
penalty-based iterative algorithm to solve the mixed-integer
In this section, we present the basic signal model, and three
non-convex optimization problem obtained for MS. For TS,
practical operating protocols for STAR-RISs in wireless
we show that the optimization problem can be decomposed
communication systems.
into two subproblems, which can be efficiently solved using
state-of-the-art algorithms and convex optimization
techniques.
• Our numerical results unveil that 1) TS is generally preferable
A. Basic Signal Model of STAR-RISs
for unicast communication, while ES is supe rior for As shown in Fig. 1, the wireless signal incident on a given
multicast communication; 2) the performance gain of element of the STAR-RIS is divided into a transmitted and a
STAR-RISs over conventional RISs increases with the reflected signal. To characterize this STAR feature, let sm denote
number of RIS elements; 3) for ES, the per formance gain of the signal incident on the mth element of the STAR-RIS, where
element-based amplitude control over group-based m ∈ M {1, 2,...,M} and M denotes the total number of elements.
amplitude control is significant for uni cast communication, The signals transmit ted and reflected by the mth element can be
while it is negligible for multicast communication. modelled as
βtmejθtm sm and rm = βrmejθrm sm [24],
C. Organization and Notation tm =
The rest of this paper is organized as follows: Section II t t r
respectively, where β m ∈ [0, 1] , θ m ∈ [0, 2π) and β m ∈ [0,
introduces the basic signal model for STAR-RISs, based on r
which three practical protocols for operating STAR-RISs are 1] , θ m ∈ [0, 2π) denote the amplitude and phase shift response 1
proposed. Section III presents the system model and the joint of the mth element’s transmission and reflection coefficients. 2
beamforming optimization problem formulations for STAR-RIS Note that, for each element, the phase shifts for transmission and
aided communication systems for both unicast and multicast reflection (i.e., θtm and θrm) can be generally chosen independent
transmission. Efficient algorithms are developed for solving the from each other [24]. However, the ampli
problems formulated for each proposed operating protocol in t
tude adjustments for transmission and reflection (i.e., β m
Section IV. Section V provides numerical results to verify the r
effectiveness of the proposed STAR-RIS designs compared to and β m) are coupled by the law of energy conservation. This
baseline schemes. Finally, Section VI concludes the paper. means that, for each element, the sum of the energies of the
Notations: Scalars, vectors, and matrices are denoted by lower- transmitted and reflected signals has to be equal to the energy of
case, bold-face lower-case, and bold-face upper-case let ters, the incident signal, i.e., |tm|2 + |rm|2 = |sm|2, which leads to the
following condition for the transmission and reflection amplitude
respectively. CN×1 denotes the space of N ×1 complex valued
coefficients of each element3 [24]:
vectors. aH and a denote the conjugate transpose and the
Euclidean norm of vector a, respectively. diag (a) denotes
βtm + βrm = 1, ∀m ∈ M. (1)
a diagonal matrix with the elements of vector a on the main
diagonal. The distribution of a circularly symmetric complex 1
In this paper, the amplitude and phase shift coefficients are assumed to be
Gaussian (CSCG) random variable with mean μ and variance continuously adjustable to be able to determine the maximum performance. For
2 practical hardware implementations, the obtained continuous solutions can be
σ2 is denoted by CN μ, σ . 1m×n and 0m×n denote the all-one quantized to discrete values. It has shown that the performance degradation caused
by phase shift quantization is small when the resolution is larger than 3 bits [14].
and all-zero matrices of size m × n, respectively. HN denotes the 2
To investigate the maximum performance gain of STAR-RISs, we assume that
set of all N-dimensional complex Hermitian matrices. Rank (A) the amplitudes and phase shifts of the transmission and reflection coeffi cients can
and Tr (A) denote the rank and the trace of matrix A, be adjusted independently. However, in practical implementations, the amplitude
and phase shift adjustment may be coupled, thus leading to a performance loss.
respectively. Diag (A) denotes a vector whose elements are Investigating this loss is beyond the scope of this work.
extracted from the main diagonal elements of matrix A. A 0 3
In this paper, we assume that the STAR-RIS does not impose a power loss. Our
indicates that A is a positive semidefinite matrix. A ∗, A 2, and proposed solutions are also applicable to the case with power loss, i.e., βtm + βrm =
c, ∀m ∈ M, where 0 <c< 1.
A F denote the nuclear norm, spectral norm, and Frobenius norm

Authorized licensed use limited to: University of Ulsan. Downloaded on January 07,2024 at 07:58:57 UTC from IEEE Xplore. Restrictions apply.
 3086 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 21, NO. 5, MAY 2022

Fig. 2. Three practical protocols for operating STAR-RISs, where Em denotes the energy of the signal incident on the mth element and T denotes the total communication
duration.

B. Three Practical Protocols for Operating STAR-RISs of the STAR-RIS are assumed to operate in the T&R mode,
where the energy of the signal incident on
As can be observed from (1), by properly adjusting the the amplitude coefficients for transmission and reflection are
amplitude coefficients for transmission and reflection, a given restricted to binary values. Therefore, MS generally cannot
element of a STAR-RIS can be operated in the full transmis sion achieve the same full-dimension transmission and reflection
mode (i.e., βtm = 1, βrm = 0, referred to as T mode), the full beamforming gain as ES, since only a subset of the elements are
reflection mode (i.e., βtm = 0, βrm = 1, referred to as R mode), selected for transmission and reflection, respectively. Nev
and the general simultaneous transmission and reflection mode ertheless, MS is still appealing in practice, since such an “on-off”
(βtm, βrm ∈ [0, 1], referred to as T&R mode). Inspired by these type operating protocol is much easier to implement compared to
observations, in this subsection, we propose three practical the ES protocol.
protocols for operating STAR-RISs in wire less communication 3) Time Switching: Different from ES and MS, the TS STAR-
systems, namely energy splitting (ES), mode switching (MS), and RIS exploits the time domain and periodically switches all
time switching (TS), as illustrated in Fig. 2. elements between the T mode and the R mode in different
1) Energy Splitting: For ES, as shown in Fig. 2(a), all elements orthogonal time slots (referred to as T and R period), as
 illustrated in Fig. 2(c). Let 0 ≤ λt ≤ 1 and 0 ≤ λr ≤ 1 denote the R period, respectively, where λt + λr = 1. Therefore, the corre
percentage of commu nication time allocated to the T period and sponding STAR-RIS transmission- and reflection-coefficient
each element is generally split into the t = diag
energies of the transmitted and reflected matrices are given by ΘTS ejθt1 , ejθt2 ,...,ejθtM and
signals with an energy splitting
ratio of βtm : βrm. In this case, the transmission- and reflection- ΘTS
coefficient matrices of the STAR-RIS are given ∀m ∈ M. Different from the ES and MS protocols, due
jθ jθ jθ t r
r= diag e r1 , e r2 ,...,e rM , where θ m, θ m ∈ [0, 2π),
by ΘES βt1ejθt1 , βt2ejθt2 ,..., βtMejθtM and
to the exploitation of the time domain, the
design of the transmission and reflection
t = diag
coefficients for TS is not coupled,
r jθ , respec mode, where Mt + Mr = M. Accordingly, the STAR-RIS
r= diag β 1 e r1 , βr2ejθr2 r
,..., β M ejθrM
transmission- and reflection-coefficient matrices
tively, where βtm, βrm ∈ [0, 1], βtm + βrm = 1, and θtm, θrm ∈
ΘES and thus, easier to handle. However, the periodical switching of
the elements introduces stringent requirements for time syn
[0, 2π), ∀m ∈ M. For ES, since both the transmission and chronization, which entails a higher hardware implementation
reflection coefficients of each element can be optimized, a high complexity.
degree of flexibility for communication system design is enabled. In Table I, we summarize the optimization variables and
However, the large number of design variables also cause a constraints for the ES, MS, and TS operating protocols.
relatively high overhead for configuration information exchange
between the BS and the STAR-RIS. III. SYSTEM MODEL AND PROBLEM FORMULATION Based on
2) Mode Switching: For MS, as shown in Fig. 2(b), all the proposed operating protocols, in this section, we present the
elements of the STAR-RIS are divided into two groups. Specif system model of a STAR-RIS aided downlink communication
ically, one group contains Mt elements that operate in the T mode, system and formulate joint active and passive beamforming
while the other group contains Mr elements operating in the R optimization problems for both unicast and multicast
transmission.

are given by ΘMS βt1ejθt1 , βt2ejθt2 ,. . . , βtMejθtM

A. System Model
t = diag
r jθ , respec
r= diag β 1 e r1 , βr2ejθr2 ,. . . , βrMejθrM
tively, where βtm, βrm ∈ {0, 1}, βtm + βrm = 1, and θtm,
and ΘMS

θrm ∈ [0, 2π), ∀m ∈ M. We note that MS STAR-RISs can be

regarded as a special case of ES STAR-RISs, where

As shown in Fig. 3, we consider a narrow-band STAR-RIS

aided downlink communication system operating over frequency-
flat channels, where an N-antenna BS com municates with
multiple single-antenna users with the aid of
 Authorized licensed use limited to: University of Ulsan. Downloaded on January 07,2024 at 07:58:57 UTC from IEEE Xplore. Restrictions apply.
MU et al.: STAR RIS AIDED WIRELESS COMMUNICATIONS 3087

TABLE I
SUMMARY OF OPTIMIZATION VARIABLES AND CONSTRAINTS FOR THE CONSIDERED OPERATING PROTOCOL

In this paper, we study the joint design of the active

Fig. 3. Illustration of a STAR-RIS aided downlink communication system, where
the direct BS-user links are blocked by obstacles.

a STAR-RIS comprising M STAR elements. As previously

discussed, the STAR-RIS can create full-space coverage by
simultaneously transmitting and reflecting the incident signal. We
refer to users that are located in the transmission space as T users,
while users that are located in the reflection space
beamforming at the BS and the passive transmission and
reflection beamforming at the STAR-RIS for both unicast and
multicast communication. For unicast, the BS sends independent
information to the T and R users. For multicast, the BS sends the
same information to both users. For both scenarios, the three
proposed STAR-RIS operating protocols are considered.

B. Unicast Communication and Problem Formulation

In unicast transmission, let wk and xk denote the active
beamforming vector and the information-bearing symbol for user
k ∈ {t, r} at the BS, respectively.
1) ES and MS: When the ES or MS protocols are employed at
the STAR-RIS, the received signal at user k ∈ {t, r} is given by

yES/MS
H ES/MS
UC,k = v k Θ
k G (wtxt + wrxr) + nk, (2)
are referred to as R users. As depicted in Fig. 2
3, in this paper, we assume that the direct where E |xk| = 1 and nk ∼ CN 0, σ2k denotes the
communication links between the BS
and the users are blocked by obstacles, 4 as this is one of the most additive white Gaussian noise (AWGN) at user k. Therefore, the
challenging scenarios for conventional communication systems. achievable communication rate of user k ∈ {t, r} for ES and MS
The STAR-RIS is deployed to provide communica is given by
tion service for the transmission and revealing fundamental RES/MS ⎜ 2
users in the resulting reflection links. For design ⎝ 1 +
⎞
signal dead zone by simplicity of ⎛ vHk ΘES/MS
establishing additional presentation and k Gwk
⎟
⎠ , (3)
insights, in the remainder of the communication system serves UC,k = log2 ΘES/MS k Gwk
paper, we consider a two-user one T user and one R user. Let G
vHk 2
+ σ2k
setup,5 i.e., the STAR-RIS aided ∈ CM×N , vHt ∈ C1×M,

and vHr ∈ C1×M represent the narrow-band quasi-static fading where k = r, if k = t; and k = t, otherwise.
channels from the BS to the STAR-RIS, the STAR-RIS to T user, 2) TS: For TS, the BS consecutively sends information to the
and the STAR-RIS to R user, respectively. In order to be able to two users in the T and R periods. Thus, the corresponding
unveil the maximum performance gain enabled by STAR-RISs, achievable communication rate of user k ∈ {t, r} can be expressed
the perfect CSI of all channels is assumed to be as
4
available at the BS.6 The algorithms proposed in this RTS k
UC,k = λ log2 k Gwk
2
work are also applicable for the H TS
case where the direct links v kΘ
between the BS and the T and R
users exist. 1+ λkσ2k
 , (4)

5
The proposed algorithms can be extended to systems with multiple T and R
users, see Remark 2 and Remark 3 in Section IV.
6
Because of the nearly-passive mode of operation of RISs, channel estima tion in
RIS-assisted wireless systems is a challenging task. To address this issue,
numerous efficient channel estimation schemes have been proposed for
conventional RISs [29]–[34], which can also be readily employed for STAR-RISs.
For instance, using the TS protocol, the CSI of the T and R users can be
consecutively estimated using existing channel acquisition methods [29]–[34].
However, the development of more efficient CSI esti mation techniques for STAR-
RISs based on the ES and MS protocols to simultaneously acquire the CSI of the T
and R users is an interesting topic for future work.
where λk in (4) is due to the fact that the BS sends information to
user k employing wk in λk fraction of the total communica
k
tion time. The transmit power of user k is increased by 1 λ to
ensure a fair comparison with the ES and MS protocols. For
unicast transmission, we aim to minimize the total power
consumption of the BS by jointly optimizing the active
beamforming at the BS and the passive transmission and
reflection beamforming at the STAR-RIS for a given operating
protocol, while satisfying the QoS requirement of both users.
 Authorized licensed use limited to: University of Ulsan. Downloaded on January 07,2024 at 07:58:57 UTC from IEEE Xplore. Restrictions apply.
3088 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 21, NO. 5, MAY 2022

Then, the optimization problem can be formulated as follows: 2) TS: Since for TS the users are served in different

min wt 2 + wr 2 (5a) employ different active beamforming

wk,ΘXk ,ιk vectors, which are denoted by wc,t and wc,r,
orthogonal time slots, the BS can still
s.t. RXUC,k ≥ Rk, ∀k ∈ {t, r} , (5b) ΘXk ∈ F X, ∀k ∈ {t, to send the same symbol to the T and R users. Therefore, for
r} , (5c) multicast communication with TS, the achievable communi cation
rate at user k ∈ {t, r} is given by
0 ≤ λt ≤ 1, 0 ≤ λr ≤ 1, λt + where X ∈ {ES, MS, TS} RTS 1+
r
λ = 1, (5d) indicates the employed k Gwc,k
2
. (8)
STAR-RIS H TS
k v k Θ
MC,k = λ log2
λkσ2k

operating protocol, F X characterizes the corresponding feasi ble

set for the transmission- and reflection-coefficient matrices, and
Rk denotes the minimum rate requirement of user k. The
k
time allocation variables, λ , and constraint (5d) are only valid
when the TS protocol is employed, i.e., X = TS. Let H = [ht hr] ∈
CN×2, where hHk = vHk ΘES/MS
k G, ∀k ∈
{t, r}. For the ES and MS protocols, assuming that N ≥ 2, a
sufficient condition for the feasibility of problem (5) for any
finite Rk is that Rank (H)=2, where a zero-forcing solution can
be easily constructed for the active beamforming [10].
H
Given the independent STAR-RIS-user channels, v k , and the
different transmission- and reflection-coefficient matrices,
We note that, for TS, the multicast communication rate in (8)
is identical to the unicast communication rate in (4). This is
because, regardless of whether unicast or multicast
communication is considered, the proposed TS protocol can serve
only one user in each time instant.
Since the information has to be delivered to both users, the
system performance of multicast communication is limited by the
user achieving the smaller communication rate. As a result, the
effective system multicasting rate for the considered operating
protocols is given by
RXMC = min RXMC,t, RXMC,r , (9)
 , such a rank condition can be where X ∈ {ES, MS, TS}. Then,
ΘES/MS k satisfied with a for multicast transmission,
high probability. For the TS operating protocol, problem (5) is the joint beamforming optimization problem for ES, MS, and TS
always feasible for any finite Rk since there is no inter-user can be formulated as
interference during the T and R periods, respectively.
Remark 1: Compared to conventional reflecting- can be summarized as wc 2 k ,ι
k

the power consumption only RISs [10], the main min wc,t 2 + wc,r 2 (10a)
min
mini mization problem for challenges for solving (5) wc,ΘES/MS k wc,k,ΘTS

follows. First, STAR-RISs require the optimization of two types s.t. RXMC ≥ Rc, (10b) ΘXk ∈ F X, (10c) 0 ≤ λt ≤ 1, 0 ≤ λr ≤
of passive beamforming (i.e., transmission and reflec tion 1, λt + λr = 1, (10d)
beamforming), while for conventional reflecting-only RISs, only
the reflection beamforming has to be designed. Mathematically, where (10b) denotes the minimum system multicast rate
the optimization problem for conventional reflecting-only RISs is constraint with required communication rate Rc. We note that
a special case of optimization prob lem (5) for STAR-RISs, k
where the transmission function is turned off and only the R user λ and constraint (10d) are only valid for TS. As there is no
is served. Second, for the proposed ES and MS protocols, inter-user interference in multicast, for each operating protocol,
transmission and reflection beamforming is coupled together, problem (10) is always feasible for any finite Rc.
which further complicates resource allocation compared to
conventional reflecting-only RISs. Therefore, the considered D. Discussion
problem (5) for STAR-RISs is much more challenging to solve
than that for conventional reflecting-only RISs. Note that the formulated joint beamforming optimization
problems for multicast transmission can be regarded as special
cases of those for unicast transmission. More specifically, for ES
C. Multicast Communication and Problem Formulation and MS, problem (10) for multicast transmission can be obtained
1) ES and MS: For ES and MS, in multicast transmission, the from problem (5) for unicast transmission by setting wt = wr =
BS employs one active beamforming vector, denoted by wc, to wc, Rt = Rr = Rc and removing the inter-user interference
convey the same symbol, sc, to the T and R users. As a result, the term in the denominator of the unicast communication rate
corresponding received signal at user k ∈ {t, r} is given by expression in (3). For TS, problem (10) for multicast transmission
can be directly obtained from problem (5) for unicast
yES/MS transmission by setting wt = wc,t, wr = wc,r, and Rt = Rr =
H ES/MS
MC,k = v k Θ Rc. Motivated by this, in the following, we mainly focus on joint
k Gwcsc + nk. (6) beamforming optimization problem (5) for unicast transmission
since problem (10) for multicast transmission can be solved in a
Accordingly, the achievable communication rate of user k ∈ similar manner. How ever, problem (5) is a non-convex
{t, r} can be expressed as optimization problem since

RES/MS ⎜ ⎟ X in general non-

⎝1 + ⎠ . (7) concave, as Θ k and convex. Therefore, it
MC,k = log2 σ2k {wk} are highly- is difficult to obtain a
⎛ vHk ΘES/MS the left-hand-side coupled. Moreover, globally optimal
2 X
k Gwc (LHS) of (5b) is not the feasible set, F , is solution for such a
⎞ challenging problem.

Authorized licensed use limited to: University of Ulsan. Downloaded on January 07,2024 at 07:58:57 UTC from IEEE Xplore. Restrictions apply.
MU et al.: STAR RIS AIDED WIRELESS COMMUNICATIONS 3089

In the following, we develop efficient algorithms to find a high To start with, we first transform problem (5) for ES and MS
quality suboptimal solution for the three considered operating into a more tractable form. To facilitate the design, we define the
protocols. transmission- and reflection coefficient vectors for ES and MS as
qES/MS
k=
IV. SOLUTION OF JOINT BEAMFORMING
OPTIMIZATION PROBLEMS k jθ k jθ k jθ H
[ β 1 e k1 , β 2 e k2 ,..., β Me kM ] , ∀k ∈ {t, r}, which
In this section, we first propose a penalty-based iterative Lemma 1: For any two Hermitian matrices A ∈ HM and B ∈
algorithm for solving the joint beamforming optimization HM having the same size, we have the following two equalities:
problem for ES. This algorithm is then further extended to solve
1 2 1 2 1 2
the problem for MS. Then, we show that the problem for TS can Tr (AB) = 2 A + B F − 2 A F − 2 B F , (12a)
be decomposed into two subproblems, namely a 1 2 1 2 1 2
transmission/reflection coefficient design problem and a resource −Tr (AB) = 2 A − B F − 2 A F − 2 B F . (12b)
allocation problem, which can be solved in a rela tively Proof: We first prove equation (12a), whose right-hand side
straightforward manner. (RHS) can be rewritten as
1
2 1 2 1 2
A. Proposed Solution for ES and MS 2 A + B F− 2 A F− 2 B F
 1 H 1 H 1 H Equation (12b) can be proved in a similar manner, and we omit
= 2Tr (A+B) (A+B) − 2Tr A A − 2Tr B B =
1 1 the details for brevity.
H H
2Tr A B + 2Tr B A (a) Based on Lemma 1, the first and second non-convex terms in
= Tr (AB), (11b) can be transformed into the following difference of convex
(DC) functions:
where (a) is due to the fact that A and B are Hermitian matrices. k HkWkH k
H
H 2
k ) Hkwk| , where

γkTr QES
k+ HkW kHHk 2
F

leads to |vHk ΘES/MS q ES/MS k

ES ES k
2 ES/MS ES/MS =γ 2 Q − γ 2 Q
k Gwk| = |(q k (q
Hk = diag(vHk )G. Moreover, we define
ES/MS
Q
H k
k ) ,∀k ∈ {t, r}, which satisfies QES/MS k 0,
2 k
k= F− γ 2 HkWkHHk 2
FΥ
ES
kk , (13a)

Rank(QES/MS H
k HkWkH k
ES/MS
k )=1, and Diag(Q ES
k k −Tr Q
k ) = β , where β H 2
k − HkWkH k F
[βk1 , βk2 ,...,βkM]. Similarly, we define Wk = wkwHk , ∀k ∈ {t, r},
1 ES
which satisfies Wk 0 and Rank(Wk)=1. Then, = 2 Q
optimization problem (5) for ES and MS can 1 ES 2 1 H 2 ES
be reformulated as follows: − 2 Q k F− 2 HkWkH k FΠ

kk . (13b)

Wk,QYk ,βkTr (Wt) + Tr (Wr) (11a) s.t. γkTr QYk HkW kHHk − Tr QYk HkWkHHk
min
Furthermore, the non-convex rank-one constraint (11d) can be
equivalently written as the following equality constraint [35]:
+ γkσ2k ≤ 0, ∀k ∈ {t, r} , (11b) Y k QES k 2= 0, ∀k ∈ {t, r} , (14)
Diag Q k = β , ∀k ∈ {t, r} ,
(11c) ∗− QES k

Rank Q
Y
k = 1, ∀k ∈ {t, where QES k ∗ = i σi QES k and Q k
ES
k
r} , (11d) 2 = σ1 Q
ES

denote the nuclear norm and the spectral norm, respectively,

Rank (Wk)=1, ∀k ∈ {t, r} , (11e) and σi QES k QES k .
is the ith largest singular value of matrix
Note that for any QES
HM and QES
k∈
QYk 0,Wk 0, ∀k ∈ {t, r} , (11f) Q ES
Q ES
k 0, we always have
0 ≤ βtm, βrm ≤ 1, βtm + βrm = 1, 2 ≥ 0, where equality holds if
ES
∀m ∈ M, (11g) ∗− Q k and only if
k

k is a rank-one matrix. Therefore, equality constraint (14)

βkm ∈ {0, 1} , ∀k ∈ {t, r} , ∀m ∈ M, (11h) ES
is only met for rank-one matrices Q k
.
where Y ∈ {ES, MS} specifies the operating protocol and γk = Next, we employ the penalty method [36] to solve prob lem
2 Rk − 1, ∀k ∈ {t, r}. We note that constraint (11h) is only (11). By exploiting (13a), (13b), and (14), we obtain the
present for MS, i.e., Y = MS. Problem (11) is still non following optimization problem:
convex due to the highly-coupled
non-convex constraint (11b) k ,βkTr(Wt)+Tr(Wr)+η k∈{t,r} QES ∗− QES 2

and the non-convex rank-one constraint min s.t. ΥES

(11d) and (11e). Moreover, for MS, the k kk + Π
ES
optimization problem becomes a mixed Wk,QES
integer non-convex problem due to binary k

constraint (11h). (15a)

In the following, we first focus on optimization problem (11) for highly-coupled non-convex constraint (11b), we first have the
ES, when constraint (11h) is absent. 2
kk + γkσ k ≤ 0, ∀k ∈ {t, r} , (15b) (11c), (11e),
1) Penalty-Based Algorithm for ES: Before handling the (11f), (11g), (15c)
 where equality constraint (14) is relaxed to a penalty term added to the objective function, and η > 0 is the penalty
following lemma. ES is not
if Q k
factor which penalizes the objective function

Authorized licensed use limited to: University of Ulsan. Downloaded on January 07,2024 at 07:58:57 UTC from IEEE Xplore. Restrictions apply.
3090 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 21, NO. 5, MAY 2022

rank-one. It can be verified that, when η → +∞, the solu by ignoring (11e). The tightness of the relaxed version of
ES of problem (15) always satisfies equality problem (18) is shown in the following
tion Q k con theorem. Theorem 1: Without loss of
optimality, the solutions {Wk}
straint (14), i.e., problems (11) and (15) are equivalent [37]. obtained for the relaxed version of problem of (18), i.e., after
However, if the initial value of the penalty factor η is chosen too dropping rank-one constraint (11e), always satisfy Rank
large, the objective function of (15) is dominated by the penalty (Wk)=1, ∀k ∈ {t, r}.
term, and the impact of the desired power minimiza tion on the Proof: See Appendix. The relaxed problem (18), which is a
solution becomes negligible. To avoid this, we first initialize η standard convex semi definite program (SDP), can be efficiently
with a small value to find a good starting point, and then, in the solved via standard convex problem solvers such as CVX [39].
course of several iterations, we gradually increase η to a Then, we propose a penalty-based iterative algorithm for solving
sufficiently large value to eventually obtain feasible rank-one problem (11), which comprises two loops. In the outer loop, the
matrices. Note that, for any given penalty factor η > 0, problem penalty factor is gradually increased from one iteration to the next
(15) is still non-convex due to the non-convexity of the objective as follows: η = ωη, where ω > 1. The algorithm terminates when
function and the non-convex constraints (15b) and (11e). In the the penalty term satisfies the following criterion:
following, we employ SCA [38] to obtain a suboptimal solution
of (11) in an iterative
manner. ES of DC functions. For a given
kk , and Π ES
ES ∗− Q k
Note that the penalty term, Υ kk are in the form point Q(n)
ES 2, ∀k ∈ {t, r} ≤ ε1, (19)
max Q k
k in the nth iteration of
the SCA method, using first-order Taylor expansion, a convex
upper bound for the penalty term can be obtained as follows:
where ε1 denotes a predefined maximum violation of equality
constraint (14). Therefore, (14) will be eventually satisfied
k ,Wk

with accuracy ε1 as η increases. In the inner loop, QES

 QES k ∗− QES(n) are jointly optimized by (18) for the given penalty
2≤ QES k iteratively solving the factor. The objective
k , (16)
∗− QES k relaxed version of problem
where QES(n) QES(n) 2+ function value of the relaxed
version of (18) is non-increasing
k
k in each iteration of the inner loop, and the optimal objective

k
Tr u QES
H function value is bounded η approaches
ES(n) ES(n) of the relaxed from below. infinity, the
Q k Q k
k−Q
ES(n)
u and version of (18) Therefore, as developed
u is guaranteed to converge to a Similarly, for given points in the nth iter
Q ES(n)
k
stationary point of the original
problem (11) [38]. The details of the developed algorithm are
denotes the eigenvector QES(n) summarized in Algorithm 1. The
corresponding to the largest eigenvalue of QES(n) k . (n)
k ,W k computational complexity of
penalty-based iterative algorithm Algorithm 1 is analyzed
ation of the SCA method, convex upper bounds of ΥES
kk and

ΠES
kk are respectively given by (17a) and (17b), shown at the
bottom of the page, respectively. As a result, for given points
as follows. The main complexity is caused by solving the relaxed
version of problem (18) in the inner loop. As the relaxed problem
is a standard SDP, the computational com
 interior point method is
plexity for solving this problem employed [40], where K = 2 is
QES(n) , by replacing the non-convex
3.5
(n) terms with 3.5 the
k ,W k is O KN + 2M if the
their convex upper bounds, problem (15) is transformed into the 3.5 3.5
following optimization problem: Algorithm 1 is O IoutIinn KN + 2M , where

number of users. Then, the overall computational complexity of

β Tr(Wt)+Tr(Wr)+η k∈{t,r} QES min Iinn and Iout denote the number of inner and
k, k
∗− QES(n) outer iterations
Wk,QES ub respectively. As can be observed, ub
ES
+ Π the computational complexity of + γkσ2k ≤ 0, ∀k ∈ {t, r} ,
s.t. Υ
ES Algorithm 1 is polynomial (18b)
k (18a)
k k k in M, which facilitates its
kk practical implementation even if
required for convergence,
the
(11c), (11e), (11f), (11g). (18c) number of STAR-RIS elements is large.
2) Extended Penalty-Based Algorithm for MS: Compared to
Now, the remaining non-convexity of (18) is the non-convex ES, for MS, the formulated optimization problem in (11) involves
rank-one constraint (11e). To handle this issue, we employ the additional non-convex binary constraints (11h). Therefore, we
semidefinite relaxation (SDR) and solve the relaxed problem only need to focus on how to tackle this new

k 2 !
kk ≤ γ 2 QES F− γkTr QES(n)

k H
k+ HkWkHHk 2
F+ γ 2 QES(n) QES
k
ΥES k
2
k
F − γkTr HHk HkW(n)
k (n) H H ! ub
+ γ 2 HkW k H k Hk W k ΥES , (17a)
H
kH k
H 2 1
k − HkWkH k F + 2
ES(n)
Q
1 ES
kk ≤ 2 Q

2 ES(n)
F− Tr Q

H
QES
!
kk
 k
ΠES k
k 2 H (n)
F− Tr H k HkW
1 (n) H H ! ub
+ 2 HkW k H k H
k H k Hk Wk ΠES kk . (17b)

Authorized licensed use limited to: University of Ulsan. Downloaded on January 07,2024 at 07:58:57 UTC from IEEE Xplore. Restrictions apply.
MU et al.: STAR RIS AIDED WIRELESS COMMUNICATIONS 3091

Solving Problem (11) for ES

Algorithm 1 Proposed Penalty-Based Iterative Algorithm for structure as problem (15), we can still employ SCA to solve this
new non-convex optimization problem. For given points
1: Initialize feasible points
QES(0) , the penalty in the nth iteration of
(0) the SCA method, by
k ,W
β(n) k using first
k
factor η.
2: repeat: outer loop
order Taylor expansion, an upper bound for the new penalty term
can be obtained as follows:
2 2
βkm− βkm ≤ βkm− βk(n) m −2βk(n) m βkm−βk(n) m
 3: Set iteration index n = 0 for inner loop.
4: repeat: inner loop

QES(n) (n)
k ,W k
1 − 2βk(n) m βkm + βk(n) m
5: For given =
, solve the relaxed version
2
of problem (18). βkm, βk(n) m , ∀k ∈ {t, r} , m ∈ M. (22)

QES(n+1) with the obtained opti Ω

6: Update (n+1)
k ,W k

mal solutions, and n = n + 1. By replacing the non-convex terms with their upper bounds
7: until the fractional decrease of 1 > 0 or the Q(n) k, β
(n)

the objective function value is (n) k

k ,W
below a predefined threshold for given points
, problem (21) can be

maximum number of inner iterations nmax is reached. reformulated as follows:

(0) with the current solutions min
k ,W k
8: Update
QES(0)
k
k ,βkTr(Wt)+Tr(Wr)
Wk,QMS
QES(n) k ,W
(n) ∗− QMS(n)
. + η k∈{t,r} QMS
9: Update η = ωη. violation is below a ε1 > 0. k
+ χ M m=1
10: until the constraint predefined threshold k
k∈{t,r}Ω βkm, βk(n) m (23a)
MS ub ub
s.t. Υ kk MS
+ Π kk + γkσ2k ≤0, ∀k∈ {t, r} , (23b)
obstacle since the other non-convex terms can be handled in a for ES to solve the optimization problem
similar manner as previously described. We first transform the (11c), (11e), (11f), (11g), (23c)
binary constraint (11h) equivalently into the following equality
constraint: where the remaining non-convexity is caused by rank-one
2 constraint (11e). We can employ again SDR and remove the non-
βkm − βkm = 0, ∀k ∈ {t, r} , m ∈ M. (20) convex rank-one constraint. The tightness of the relaxation can be
Recall that because of constraint (11g) the amplitude coef proved similar to Theorem 1. As a result, the relaxed version of
ficients for transmission and reflection are between 0 and 1. problem (23) is convex and can be efficiently solved via standard
2 convex problem solvers such as CVX [39]. Similar to Algorithm
Thus, we always have βkm − βkm ≥ 0, where equality holds if 1 for ES, for MS, a two-loop penalty based iterative algorithm for
and only if βkm is 0 or 1, i.e., a binary variable. As a result, the solving problem (11) can be developed. This algorithm is
equality constraint in (20) is satisfied only for binary variables. In summarized in Algorithm 2. Since for MS there are two penalty
the following, we extend the proposed penalty based algorithm terms, the termination criterion in the outer loop is given by
for MS. In particular, non-
which is highly-coupled in k ,Wk , and the non-convex " QMS k 2,
convex constraint (11b), rank-one #
QMS ∗− Q
MS
k

max 2 ≤ ε2, (24)

constraint (11d) are handled in a similar βkm − βkm , ∀k ∈ {t, r} , m ∈ M
manner as previously

described. Moreover, by further adding equality constraint (20) as min

another penalty term into the objective function, we obtain the
where ε2 > 0 is the predefined accuracy with which equality
following optimization problem for MS: constraints (14) and (20) are met.
k ,β k Tr(W t)+Tr(W r) Remark 2: Although, in this paper, we focus on systems with
one T user and one R user, the proposed Algorithms 1
Wk,QMS
s.t. ΥMS problem can be obtained from
STAR-RIS aided com 2 optimization problem (11) by
MS + χ M m=1
+ η k∈{t,r} Q k replacing the inter-user
munication systems with multiple interference in the first term
MS
∗− Q k
T and R users. Suppose k ∈{t,r}
of (11b) (i.e., γkTr QYk HkW
and 2 can be extended to ES/MS that there are J > 2 users. The
βkm− βkm 2 (21a) corresponding optimization H
kH k ) with the sum of the

kk + ΠMS inter-user interference caused by all J − 1

MS 2 interfering users and minimizing the power
Υ kk + γkσ k ≤ 0, ∀k ∈ {t, r} , (21b) (11c),
(11e), (11f), (11g), (21c) consumption caused by the J
where QES kk , ΠMS kk ,∀k ∈ {t, r} active beamforming vectors. In
are obtained by replacing particular, the J − 1 new
k in (13a) and (13b) by QMS (14) and (20). As problem (21) has a similar
and χ > 0 is a new penalty non-convex inter-user interference terms can be tackled in a
k,
similar manner as previously discussed (i.e., (13a) and (17a)).
k
factor which penalizes the objective function if β m belongs to Therefore, the optimization problem for ES and MS for multiple
(0, 1). Similarly, it can be verified that, when η, χ → +∞, the T and R users can still be solved with the proposed Algorithms 1
solution obtained from problem (21) satisfies equality constraints and 2.

Authorized licensed use limited to: University of Ulsan. Downloaded on January 07,2024 at 07:58:57 UTC from IEEE Xplore. Restrictions apply.
3092 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 21, NO. 5, MAY 2022

Algorithm 3 Procedure for Solving Problem (25) for TS

Algorithm 2 Proposed Penalty-Based Iterative Algorithm for
Solving Problem (11) for MS 1: Obtain the transmission/reflection-coefficient vectors,

(0)
k ,W
1: Initialize
feasible points the penalty factors k , β(0) ,
η and χ. k corresponding effective channel gain
2: repeat: outer loop TS∗ maximization problem (27).
denoted by q k , ∀k ∈ {t, r} , by solving
TS∗
the given q k , ∀k ∈ {t, r} , solve the
2: For
QMS(0) convex resource
3: Set iteration index n = 0 for inner loop. allocation problem (28) to obtain the optimal solutions
4: repeat: inner loop
p∗k, λk∗, ∀k ∈ {t, r} .
(n) , solve the relaxed
k ,W
5: For given (n)
k, β
QMS(n) k

version of problem (23).

(n+1) k transmission/reflection-
k ,W
6: Update with the coefficient vectors for problem
QMS(n+1) k, β(n+1) Since the two users are (26)
alternatingly served, the optimal
obtained optimal solutions, and n = n + 1. maximize the effective channel gain of each user, which yields
7: untilthe fractional decrease of the objective function the following subproblem:
value is below a predefined threshold H
2 > 0 or the maximum number of inner
qTS∗ k = arg max qTS HkHHk qTS
iterations nmax is reached. k, ∀k ∈ {t, r} (27a)
8: Update (0) (0) k
k ,W k, β TS
s.t. q 2
= 1, ∀m ∈ M. (27b)
MS(0)
Q with the current solutions
k
QMS(n) k ,W
(n)
m
k
k
k, β(n)
. The above transmission/reflection-coefficient design prob
9: Update η = ωη, χ = χ. H
10: until the constraint violation is below a predefined threshold ,...,ejθkM , ∀k ∈ {t, r}, denote the STAR-RIS
ε2 > 0.
lem can be efficiently solved using either the suboptimal low
complexity SDR based algorithm proposed in [10] or the globally
B. Proposed Solution for TS optimal branch-and-bound based algorithm proposed in [11]. The
= details are omit here for brevity.
To solve the optimization problem for TS, let qTS k ejθk1 , ejθk2 With the obtained desired solutions for the transmission/
TS∗
reflection-coefficient vectors, q k , problem (26) is reduced to
the following subproblem:
transmission/reflection-coefficient vector for 0 ≤ λk ≤ 1 min
pt + pr (28a)
fraction of the total H TS pk,ιk
effective chan nel power v k Θ
H
⎛ qTS∗ k Hk 2
communication time. The
2= TS ⎜ ⎞
k Gwk q ⎝1 +pk
gain of user k is given by
 H s.t. λklog2 ⎟
Hkwk 2
, where Hk = diag vHk G, ∀k ∈ {t, r}. k ⎠≥ Rk, ∀k ∈ {t, r},
λkσ2k
Then, for TS, optimization problem (5) can be rewritten as follows: (28b)
min wt 2 + wr 2 (25a) (25d). (28c)
wk,qTS
k ,ι
k
⎜ σ2k ⎟ a convex opti and (25d) is an affine
⎝1+ ⎠≥ Rk, ∀k ∈ {t, r} , mization problem, constraint. Therefore,
qTS k ⎞ since the LHS of problem (28) can be
s.t. λklog2 qTS (25b)
H (28b) is a jointly efficiently solved
2
Hkwk The above resource concave function with k
using standard convex
⎛ λk allocation problem is respect to pk and λ , problem solvers such
as CVX [39].
k m (25c) Based on the above two
2
= 1, , ∀k ∈ {t, r} , m ∈ M, subproblems, the procedure for
0 ≤ λt ≤ 1, 0 ≤ λr ≤ 1, λt + λr = 1. (25d) solving problem (25) for TS is summarized in Algorithm 3. The
complexity of Algorithm 3 is analyzed as follows. Suppose that
As, in each time instant, only one user receives data from the the suboptimal SDR based algorithm [10] is employed for
BS via the STAR-RIS, for any given transmission/reflection solving problem (27) for each user. Then,
coefficient vector, the optimal active beamforming vector at the 3.5
BS is the maximum-ratio transmission (MRT) beam the computational complexity is O KM [10], [40], where
K = 2 is the number of users. For the convex
√ , where pk denotes resource allocation problem (28), the
former [10], i.e., w∗k = p kH
H
kq
TS
k
computational complex
(qTS ity is O 3.5
k )HHk 2 (2K) [41]. Therefore, the overall
computational
the allocated transmit power for user k. By complexity of Algorithm 3 is O pk,qTS
substituting {w∗k} into (25), we obtain the pt + pr (26a)
following problem: 3.5
KM3.5 + (2K) . Algo rithm 3 for TS requires a lower
computational complexity than Algorithms
min
1 and 2 for ES and MS, since transmission
k ,ι
k
⎛ qTS k ⎞ and reflection are proposed TS protocol
⎜ H decoupled for TS. severs only
⎝1+pk 2 ⎟
s.t. λklog2 Hk ⎠≥ Rk, ∀k ∈ {t, r}, Remark 3: Since the
λkσ2k proposed Algorithm 3 can be directly applied as follows. First,
solve each effective channel gain maximization
(26b) (25c), (25d). (26c)
one user in each time instant, for systems with J > 2 users, the problem (27) to obtain J transmission/reflection-coefficient

Authorized licensed use limited to: University of Ulsan. Downloaded on January 07,2024 at 07:58:57 UTC from IEEE Xplore. Restrictions apply.
MU et al.: STAR RIS AIDED WIRELESS COMMUNICATIONS 3093

TABLE II
SYSTEM PARAMETERS

ρ0 represents the path loss at a reference distance of 1 meter,

GLoS and vLoS
k are the deterministic line-of-sight (LoS)
components, and GNLoS and vNLoS
are the random non
k line-of-sight (NLoS) components modeled as Rayleigh fading.
The adopted system parameters are presented in Table II.
 Without loss of generality, we assume that the two users
for unicast transmission have the same QoS requirements,
i.e., Rt = Rr = R0 log2 (1 + γ0) and γ0 is the minimum
required signal-to-interference-plus-noise ratio (SINR). The
Fig. 4. The simulated setup. users are modeled as Rician fading channels as follows:
communication requirement for multicast transmission is set to
Rc log2 (1 + γc), where γc is the minimum required SINR.
vectors. Then, solve the resulting convex resource allocation The power consumption results shown (i.e., Figs. 6-8) were
problem (28) with J power allocation and time allocation obtained by averaging over 100 channel realizations. Specifi
variables, respectively. cally, we first randomly generated 100 user distributions, and
then obtained the corresponding channel realizations.
V. NUMERICAL RESULTS
In this section, numerical results are provided to validate the B. Baseline Schemes
effectiveness of the proposed STAR-RIS aided communication To verify the effectiveness of the proposed STAR-RIS con
system. cept and the corresponding operating protocols, we compare with
the following two baseline schemes.
• Baseline scheme 1 (also referred to as conventional RISs):
A. Simulation Setup
In this case, the full-space coverage facilitated by the
Fig. 4 illustrates the considered three-dimensional (3D) STAR-RIS in Fig. 4 is achieved by employing one
simulation setup, where the BS and the STAR-RIS 7 are located at conventional reflecting-only RIS and one transmitting only
(0, 0, 0) meters and (0, 50, 0) meters, respectively. The T and R RIS. The two conventional RISs are deployed adjacent to
users are randomly located on half-circles centered at the STAR- each other at the same location as the STAR-RIS. For a fair
RIS with a radius of dt = dr = 3 m. The STAR-IRS is assumed to comparison, each conventional reflecting/transmitting RIS is
be equipped with a uniform planar array (UPA) composed of M = assumed to have M/2 elements, where M is assumed to be
MhMv elements, where Mh = 5 denotes the number of elements an even number for simplicity. This baseline scheme can be
along the horizontal plane and we increase Mv linearly with M. regarded as a special case of an MS STAR-RIS, where M/2
For our simulations, the narrow-band quasi-static fading channels ele ments operate in the T mode and M/2 elements operate
from the BS to the STAR-RIS and from the STAR-RIS to the two in the R mode. Therefore, the resulting optimization
problem can be solved by applying Algorithm 1 with

G= ele ments of the ES coefficients for

$ α
KBR + 1GNLoS
STAR-RIS employ the KRU + 1vNLoS
ρ0 d BR transmission and
BR
βt = [11×M/2 01×M/2] and βr same amplitude ,
$ = [01×M/2 11×M/2]. % reflection, respectively, k
ρ0 i.e., βkm = βk, βrm = βr∀m
KBR • Baseline scheme 2 (also $ ∈ M, where 0 ≤ βk, βr ≤ 1
KRU
LoS referred to as vk = and βt+ βr = 1. UES
KBR + 1G + uniform energy ,
$ dαRU RU,k can be regarded as a
splitting (UES)): In this 1 special case
$
1 case, we assume that all (29a) KRU + 1vLoS k +
(29b) of the ES STAR-RIS employing a group/surface-wise
amplitude design. The resulting optimization problem can
where dBR and dRU,k denote the distances between the BS and the be solved by applying Algorithm 1 with the above linear
STAR-RIS and between the STAR-RIS and user k ∈ {t, r}, αBR equality constraints.
and αRU denote the corresponding path loss exponents, KBR and
KRU denote the Rician factors, C. Convergence of Algorithms 1 and 2
7
In Fig. 5, we investigate the convergence behavior and the
In practice, both the horizontal location and the height of the STAR-RIS can be violation of the equality constraints, i.e., (19) and (24)
either optimized or selected based on the environmental conditions.

Authorized licensed use limited to: University of Ulsan. Downloaded on January 07,2024 at 07:58:57 UTC from IEEE Xplore. Restrictions apply.
3094 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 21, NO. 5, MAY 2022
Fig. 5. Convergence behaviour of proposed penalty-based iterative algorithms, where “UC” and “MC” represent unicast and multicast, respectively. of the

proposed Algorithm 1 for ES and Algorithm 2 for

MS. Both unicast and multicast transmission are considered. We

set N = 2, M = 10, and γ0 = 0 dB for unicast communication, and
γc = 10 dB for multicast communication. The presented results
were obtained for one random channel realization. As shown in
Fig. 5(a), the required power con sumption obtained for
Algorithms 1 and 2 decreases quickly as the number of outer loop
iterations increases. Specifically, the proposed penalty-based
algorithms converge in 6 iterations and 4 iterations for unicast and
multicast communication, respectively. This is because unicast
communication involves more optimization variables (i.e., two
different active beam forming vectors at the BS) and more complex
constraints (i.e., inter-user interference terms) than multicast
communi cation, and thus more iterations are needed for
convergence. Fig. 5(b) and Fig. 5(c) illustrate the constraint
violation for Algorithms 1 and 2 versus the number of outer loop
iterations for unicast and multicast communication, respectively.
As can be observed, the constraint violation in all setups decreases
quickly as the number of outer loop iterations increases, and
ultimately reaches the predefined accuracy (i.e., ε1 = ε2 = 10−7)
after 8 iterations. This means that feasible rank-one

QES/MS k
transmission/reflection-coefficient matrices, , and
Fig. 6. Power consumption versus M for N = 2.
binary transmission/reflection amplitude coefficients, βkm , are
obtained with Algorithms 1 and 2.

D. Required Power Consumption Versus Number of RIS TS results in interference-free communication for each user, i.e.,
Elements only one user is to be served in each time instant. This
In Fig. 6, we investigate the required power consumption interference-free feature makes TS preferable for unicast
versus the number of RIS elements, M, for both unicast and communication since it prevents the communication rate of the
multicast transmission. Besides the three proposed operating users from being degraded by inter-user interference. In con trast,
protocols, the results obtained for the two baseline schemes are for multicast communication, inter-user interference does not
provided for comparison. We set N = 2, γ0 = 0 dB for unicast exist since the same symbol is sent to all users. Hence, ES
becomes appealing since it can make full use of the entire
communication, and N = 2, γc = 10 dB for multicast available communication time and allows the users to be served
communication. As can be seen from Fig. 6, the required power all the time, which is not possible with TS. Moreover, it can also
consumption for all schemes and scenarios decreases as M
be observed that ES outperforms MS for both unicast and
increases. This is expected since larger M enable a higher
multicast transmission. This is expected since mathematically MS
transmission/reflection beamforming gain, which in turn reduces
is a special case of ES.
the required power consumption of the BS. Regarding the
performance of the three proposed operating protocols for STAR-
RISs, TS achieves the best performance for unicast
communication, whereas ES is preferable for multicast
communication. This can be explained as follows.

Authorized licensed use limited to: University of Ulsan. Downloaded on January 07,2024 at 07:58:57 UTC from IEEE Xplore. Restrictions apply.
MU et al.: STAR RIS AIDED WIRELESS COMMUNICATIONS 3095 Regarding the performance comparison between

STAR-RISs and baseline scheme 1, as can be observed from Fig. communication, we observe that conventional RISs can achieve a
6(a), independent of the adopted operating protocol, STAR-RISs higher performance than TS STAR-RIS for small M. This is
always outperform conventional RISs for unicast communication. because, compared with TS STAR-RIS, conventional RISs (a
The reasons behind this can be explained as follows. Since special case of MS STAR-RISs) can make full use of the entire
conventional RISs employ fixed numbers of transmission and available communication time. This benefit allows conventional
reflection elements, they cannot exploit the same DoFs as RISs to achieve a higher performance than TS when M is small,
available to the STAR-RISs to enhance the desired signal strength i.e., when the available DoFs are limited and using the entire
and mitigate inter-user interference. Therefore, conventional RISs available communication time dominates the achieved
always suffer from the worst performance for unicast performance. However, when M increases, as can be observed
communication. However, in Fig. 6(b), for multicast from Fig. 6(b), conventional RISs become the worst option again
due to the significant loss of DoFs caused by their inflexibility in
choosing between transmission and reflection. This limitation
also causes the performance gap between conventional RISs and
STAR-RISs to become more pronounced as M increases. The
above performance comparison confirms the effectiveness of
employing STAR-RIS in wireless communication systems.
For baseline scheme 2, a special case of ES, it is interesting to
observe that there is a noticeable performance gap between ES
and UES for unicast communication in Fig. 6(a). However, the
performance gap is negligible for multicast communication in
Fig. 6(b). Recall that the difference between ES and UES only
lies in the element-wise and group-wise transmission and
reflection amplitude control. The extra DoFs provided by the
element-wise amplitude control allow ES to achieve an improved
desired signal enhancement and inter-user interfer ence
Fig. 7. Power consumption versus N for M = 10.
mitigation compared to UES for unicast communication, thus
achieving higher performance. For multicast communica tion,
since there is no inter-user interference to be mitigated, the
performance gain caused by the element-wise amplitude control protocols for STAR-RISs and the baselines for both unicast and
vanishes. On the one hand, this result underscores the importance multicast communication. We set M = 10, γ0 = 0 dB for unicast
of element-wise transmission and reflection amplitude control for communication, and M = 10, γc = 10 dB for multicast
mitigating inter-user interference for unicast communication. On communication. Fig. 7 shows that the required power
the other hand, the result also implies that UES is a promising consumption for all schemes decreases as N increases thanks to a
operating protocol for multi cast communication, where it higher active beamforming gain. For unicast commu nication in
achieves similar performance as ES. However, compared with Fig. 7(a), the proposed STAR-RISs outperform conventional
ES, the group-wise amplitude control of UES reduces the RISs since more DoFs for transmission and reflection design can
overhead caused by exchanging configuration information be exploited. As we consider a small M, for multicast
between the BS and the STAR-RIS, which is a major bottleneck communication in Fig. 7(b), the loss in DoFs for employing
for the practical implementation of RISs. conventional RISs is not significant. Thus, in this case,
conventional RISs outperform the TS STAR-RIS because of the
E. Required Power Consumption Versus Number of BS more efficient exploitation of the time resources. As can be
Antennas observed in Fig. 7(a), there is a considerable performance gap
In Fig. 7, we study the required power consumption versus the between UES and ES for unicast communication, while UES and
number of BS antennas, N, for the proposed operating ES achieve a similar performance for multicast communication in
Fig. 7(b). This is consistent with the results in Fig. 6.

F. Required Power Consumption Versus QoS Requirements

In Fig. 8, we study the required power consumption versus the
minimum required SINR, γ0 and γc, for both unicast and
multicast transmission. We set N = 2 and M = 10 for all
considered schemes. For unicast communication in Fig. 8(a), we
observe that TS achieves the best performance for small γ0 (i.e.,
when γ0 < 12 dB) but is outperformed by the other schemes for
large γ0. This is because the

Authorized licensed use limited to: University of Ulsan. Downloaded on January 07,2024 at 07:58:57 UTC from IEEE Xplore. Restrictions apply.
3096 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 21, NO. 5, MAY 2022 problem was solved using state-of-the-art algorithms and

Fig. 8. Power consumption versus γ0 and γc for N = 2 and M = 10.

interference-free communication for TS is achieved at the

expense of an inefficient use of the communication time, which proposed for conventional RISs can be applied for STAR-RIS
becomes a major performance bottleneck for large γ0. In with TS, considerable pilot overhead is required to consecutively
contrast, since for ES, MS, and the two baseline schemes, both estimate the transmission and reflection channels. To tackle this
users are served during the entire communication time, they issue, channel acquisition methods based on the ES protocol have
achieve a higher performance than TS for large γ0 despite the to be developed to simul taneously estimate both channels with
inter-user interference. Similarly, for multicast communication, less pilot overhead. Moreover, the full-space coverage created by
we can observe from Fig. 8(b) that the per formance gap between STAR-RISs also imposes new challenges for deployment.
TS and the other schemes becomes more pronounced as γc Different from conventional reflecting-only RISs, whose optimal
increases due to the inefficient use of the communication time for deployment strategy requires only to assign either the users or the
TS. The obtained results highlight the importance of employing BSs to its local reflection region [42], [43], the deployment
different operating protocols for different communication locations and orientations of STAR-RISs have to be further
objectives and scenarios. optimized to balance the numbers users on both sides of the
STAR-RIS, i.e., the number of T users and the number of R users.
VI. CONCLUSION
In this paper, wireless communication system design for a APPENDIX
novel RIS concept enabling simultaneous transmission and PROOF OF THEOREM 1
reflection was investigated. Based on the basic signal model for The relaxed version of problem (18) without rank one
the individual RIS elements, three practical operating protocols constraint (11e) is jointly concave with respect to
for STAR-RISs were proposed and their respective advan tages Wk, QES
and disadvantages were discussed. For each of these operating and satisfies Slater’s constraint qualifica
protocols, the joint active and passive beamforming optimization k , βk tion [41].
problem was formulated for minimization of the power Therefore, strong duality holds and the Lagrangian function is
consumption of the BS while satisfying the QoS requirements of given by (30), shown at the top of the next page, where τ is the
the users for both unicast and multicast transmission. For the ES collection of all terms which do not depend on {Wk}, and μk and
and MS protocols, the resulting non-convex problems were Yk are the Lagrange multipliers associated with constraints (18b)
efficiently solved by penalty-based iterative algorithms. For the and (11f), respectively. Based on the Karush-Kuhn-Tucker
TS protocol, the optimization (KKT) conditions with respect to Wk, the structure of the optimal
convex optimization techniques. Numerical results showed that W∗k can be characterized as follows:
STAR-RISs can significantly reduce the BS power con sumption
compared to conventional reflecting/transmitting only RISs. μ∗k ≥ 0, Y∗k 0, (31a)
Furthermore, the obtained results also revealed that the TS Y∗kW∗k = 0, (31b)
protocol is preferable for unicast communication and low QoS ∇W∗k L = 0, (31c)
requirements. However, for unicast communication and high QoS ∗ ∗
where μ k and Y k denote the optimal Lagrange multipliers and
requirements, and for multicast communication, the ES protocol
is the best option. These insights provide useful guidelines for the ∇W∗k L is the gradient of L with respect to W∗k. Then, ∇W∗k L = 0
design of STAR-RIS aided wireless communication systems. can be further expressed as
The results obtained in this paper confirm the effectiveness of Y∗k = IN − Ω∗k, (32)
employing STAR-RISs for improving the performance of where Ω∗k is given by (33), shown at the top of the next page. By
wireless networks, which motivates related future research on exploiting the results in [17, Appendix A], it can be proved that
STAR-RISs, such as channel estimation and deployment design. Rank (Y∗k) = N − 1. Furthermore, (31b) implies that Rank
More specifically, although the channel acquisition methods
(Y∗k) + Rank (W∗k) ≤ N. Therefore, Rank (W∗k) ≤

Authorized licensed use limited to: University of Ulsan. Downloaded on January 07,2024 at 07:58:57 UTC from IEEE Xplore. Restrictions apply.

MU et al.: STAR RIS AIDED WIRELESS COMMUNICATIONS 3097 k + HkWkHHk 2

F − γkTr HHk HkW(n)

k H
L = k∈{t,r} Tr (Wk) + k∈{t,r} μk γ 2 QES k − HkWkHHk 2
F− Tr H
k H k Hk W k!
HHk HkW(n)
1 ES Ω∗k = μk HkQES H H !!
+ 2 Q k H k Hk Wk − k∈{t,r} Tr (YkWk) + τ (30)
H
k H k Hk
H
kH k− HkHkWkHHk HHk + HHk HkW(n) kH
H
kH k (33)

H
−μ k γ k kH k+ HkHkWkHHk HHk − HHk HkW(n) HkQES

1 has to hold. Thus, due to the QoS constraint in (18b), Rank [1] X. Mu, Y. Liu, J. Xu, L. Guo, and J. Lin, “Joint beamforming optimization for
simultaneously transmitting and reflecting (STAR) RIS aided
(W∗k)=1 for the optimal solution. This completes the proof. communications,” in Proc. 55th Asilomar Conf. Signals, Syst., Comput., to
be published.
[2] W. Saad, M. Bennis, and M. Chen, “A vision of 6G wireless systems:
REFERENCES Applications, trends, technologies, and open research problems,” IEEE
Netw., vol. 34, no. 3, pp. 134–142, May/Jun. 2020.
 [3] Z. Zhang et al., “6G wireless networks: Vision, requirements, architec ture, [20] W. Mei and R. Zhang, “Cooperative beam routing for multi-IRS aided
and key technologies,” IEEE Veh. Technol. Mag., vol. 14, no. 3, communication,” IEEE Wireless Commun. Lett., vol. 10, no. 2, pp. 426–430,
pp. 28–41, Sep. 2019. Feb. 2021.
[4] F. Boccardi, R. W. Heath, Jr., A. Lozano, T. L. Marzetta, and P. Popovski, [21] X. Yu, D. Xu, Y. Sun, D. W. K. Ng, and R. Schober, “Robust and secure
“Five disruptive technology directions for 5G,” IEEE Commun. Mag., vol. wireless communications via intelligent reflecting sur faces,” IEEE J. Sel.
52, no. 2, pp. 74–80, Feb. 2014. Areas Commun., vol. 38, no. 11, pp. 2637–2652, Nov. 2020.
[5] E. Basar, M. Di Renzo, J. De Rosny, M. Debbah, M. Alouini, and R. Zhang, [22] S. Li, B. Duo, X. Yuan, Y. Liang, and M. Di Renzo, “Reconfigurable
“Wireless communications through reconfigurable intelligent surfaces,” intelligent surface assisted UAV communication: Joint trajectory design and
IEEE Access, vol. 7, pp. 116753–116773, 2019. passive beamforming,” IEEE Wireless Commun. Lett., vol. 9, no. 5, pp. 716–
[6] M. D. Renzo et al., “Smart radio environments empowered by recon figurable 720, May 2020.
intelligent surfaces: How it works, state of research, and road ahead,” IEEE [23] X. Mu, Y. Liu, L. Guo, J. Lin, and R. Schober, “Intelligent reflect ing surface
J. Sel. Areas Commun., vol. 38, no. 11, pp. 2450–2525, Nov. 2020. enhanced indoor robot path planning: A radio map based approach,” IEEE
[7] Y. Liu et al., “Reconfigurable intelligent surfaces: Principles and oppor Trans. Wireless Commun., vol. 20, no. 7, pp. 4732–4747, Jul. 2021.
tunities,” IEEE Commun. Surveys Tuts., vol. 23, no. 3, pp. 1546–1577, 3rd [24] J. Xu, Y. Liu, X. Mu, and O. A. Dobre, “STAR-RISs: Simultaneous
Quart., 2021. transmitting and reflecting reconfigurable intelligent surfaces,” IEEE
[8] Q. Wu and R. Zhang, “Towards smart and reconfigurable environment: Commun. Lett., vol. 25, no. 9, pp. 3134–3138, Sep. 2021.
Intelligent reflecting surface aided wireless network,” IEEE Commun. Mag., [25] Y. Liu et al., “STAR: Simultaneous transmission and reflection for 360 ◦
vol. 58, no. 1, pp. 106–112, Jan. 2020. coverage by intelligent surfaces,” IEEE Wireless Commun., 2021. [Online].
[9] C. Huang et al., “Holographic MIMO surfaces for 6G wireless networks: Available: http://arxiv.org/abs/2103.09104
Opportunities, challenges, and trends,” IEEE Wireless Commun., vol. 27, no. [26] B. O. Zhu et al., “Dynamic control of electromagnetic wave propagation with
5, pp. 118–125, Oct. 2020. the equivalent principle inspired tunable metasurface,” Sci. Rep., vol. 4, no.
[10] Q. Wu and R. Zhang, “Intelligent reflecting surface enhanced wireless 1, May 2014, Art. no. 4971.
network via joint active and passive beamforming,” IEEE Trans. Wireless [27] NTT DOCOMO. Docomo Conducts World’s First Successful Trial of
Commun., vol. 18, no. 11, pp. 5394–5409, Nov. 2019. Transparent Dynamic Metasurface. [Online]. Available: https://www.
[11] X. Yu, D. Xu, and R. Schober, “Optimal beamforming for MISO com nttdocomo.co.jp/english/info/mediacenter/pr/2020/0117_00.html
munications via intelligent reflecting surfaces,” in Proc. IEEE 21st Int. [28] S. Zhang, H. Zhang, B. Di, Y. Tan, Z. Han, and L. Song, “Beyond intelligent
Workshop Signal Process. Adv. Wireless Commun. (SPAWC), May 2020, pp. reflecting surfaces: Reflective-transmissive metasurface aided
1–5. communications for full-dimensional coverage extension,” IEEE Trans. Veh.
[12] B. Zheng, Q. Wu, and R. Zhang, “Intelligent reflecting surface-assisted Technol., vol. 69, no. 11, pp. 13905–13909, Nov. 2020.
multiple access with user pairing: NOMA or OMA?” IEEE Commun. Lett., [29] Z.-Q. He and X. Yuan, “Cascaded channel estimation for large intelligent
vol. 24, no. 4, pp. 753–757, Apr. 2020. metasurface assisted massive MIMO,” IEEE Wireless Commun. Lett., vol. 9,
[13] C. Huang, A. Zappone, G. C. Alexandropoulos, M. Debbah, and C. Yuen, no. 2, pp. 210–214, Feb. 2020.
“Reconfigurable intelligent surfaces for energy efficiency in wireless [30] D. Mishra and H. Johansson, “Channel estimation and low-complexity
communication,” IEEE Trans. Wireless Commun., vol. 18, no. 8, pp. 4157– beamforming design for passive intelligent surface assisted MISO wire less
4170, Aug. 2019. energy transfer,” in Proc. IEEE Int. Conf. Acoust., Speech Signal Process.
[14] X. Mu, Y. Liu, L. Guo, J. Lin, and N. Al-Dhahir, “Exploiting intelligent (ICASSP), May 2019, pp. 4659–4663.
reflecting surfaces in NOMA networks: Joint beamforming optimiza tion,” [31] H. Liu, X. Yuan, and Y.-J.-A. Zhang, “Matrix-calibration-based cas caded
IEEE Trans. Wireless Commun., vol. 19, no. 10, pp. 6884–6898, Oct. 2020. channel estimation for reconfigurable intelligent surface assisted multiuser
[15] S. Zhang and R. Zhang, “Capacity characterization for intelligent reflecting MIMO,” IEEE J. Sel. Areas Commun., vol. 38, no. 11, pp. 2621–2636, Nov.
surface aided MIMO communication,” IEEE J. Sel. Areas Commun., vol. 38, 2020.
no. 8, pp. 1823–1838, Aug. 2020. [32] Z. Wang, L. Liu, and S. Cui, “Channel estimation for intelligent reflecting
[16] X. Mu, Y. Liu, L. Guo, J. Lin, and N. Al-Dhahir, “Capacity and optimal surface assisted multiuser communications: Framework, algo rithms, and
resource allocation for IRS-assisted multi-user communication systems,” analysis,” IEEE Trans. Wireless Commun., vol. 19, no. 10, pp. 6607–6620,
IEEE Trans. Commun., vol. 69, no. 6, pp. 3771–3786, Jun. 2021. Oct. 2020.
[17] D. Xu, X. Yu, Y. Sun, D. W. K. Ng, and R. Schober, “Resource allocation for [33] C. You, B. Zheng, and R. Zhang, “Channel estimation and passive
IRS-assisted full-duplex cognitive radio systems,” IEEE Trans. Commun., beamforming for intelligent reflecting surface: Discrete phase shift and
vol. 68, no. 12, pp. 7376–7394, Dec. 2020. progressive refinement,” IEEE J. Sel. Areas Commun., vol. 38, no. 11, pp.
[18] C. Huang, R. Mo, and Y. Yuen, “Reconfigurable intelligent surface assisted 2604–2620, Nov. 2020.
multiuser MISO systems exploiting deep reinforcement learn ing,” IEEE J. [34] B. Zheng, C. You, and R. Zhang, “Intelligent reflecting surface assisted multi-
Sel. Areas Commun., vol. 38, no. 8, pp. 1839–1850, Aug. 2020. user OFDMA: Channel estimation and training design,” IEEE Trans.
[19] B. Zheng, C. You, and R. Zhang, “Double-IRS assisted multi-user MIMO: Wireless Commun., vol. 19, no. 12, pp. 8315–8329, Dec. 2020.
Cooperative passive beamforming design,” IEEE Trans. Wireless Commun., [35] T. Jiang and Y. Shi, “Over-the-air computation via intelligent reflect ing
vol. 20, no. 7, pp. 4513–4526, Jul. 2021. surfaces,” in Proc. IEEE Global Commun. Conf. (GLOBECOM), Dec. 2019,
pp. 1–6.

Authorized licensed use limited to: University of Ulsan. Downloaded on January 07,2024 at 07:58:57 UTC from IEEE Xplore. Restrictions apply.
3098 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 21, NO. 5, MAY 2022

[36] J. Nocedal and S. Wright, Numerical Optimization. New York, NY, USA: and Computer Science, Queen Mary University
Springer, 2006. of London, U.K. His research interests include
[37] A. Ben-Tal and M. Zibulevsky, “Penalty/barrier multiplier methods for non-orthogonal multiple access, IRSs/RISs aided
convex programming problems,” SIAM J. Optim., vol. 7, no. 2, pp. 347–366, communications, UAV communications, and opti
May 1997. mization theory. He received the Exemplary Reviewer Certificate of the IEEE
[38] Q. T. Dinh and M. Diehl, “Local convergence of sequential convex TRANSACTIONS ON COMMUNICATIONS in 2020.
programming for nonconvex optimization,” in Recent Advances in Opti
mization and its Applications in Engineering. Berlin, Germany: Springer,
2010. Yuanwei Liu (Senior Member, IEEE) received the
[39] M. Grant and S. Boyd. (2014). CVX: MATLAB Software for Disci plined B.S. and M.S. degrees from Beijing University of
Convex Programming, Version 2.1. [Online]. Available: http:// cvxr.com/cvx Posts and Telecommunications in 2011 and 2014,
[40] Z.-Q. Luo, W.-K. Ma, A. M.-C. So, Y. Ye, and S. Zhang, “Semidefinite respectively, and the Ph.D. degree in electrical engi
relaxation of quadratic optimization problems,” IEEE Signal Process. Mag., neering from the Queen Mary University of London,
vol. 27, no. 3, pp. 20–34, May 2010. U.K., in 2016.
[41] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge, U.K.: He was with the Department of Informatics, King’s
Cambridge Univ. Press, 2004. College London, from 2016 to 2017, where he
[42] Q. Wu, S. Zhang, B. Zheng, C. You, and R. Zhang, “Intelligent reflecting was a Post-Doctoral Research Fellow. He has been
surface-aided wireless communications: A tutorial,” IEEE Trans. Commun., a Lecturer (Assistant Professor) with the School
vol. 69, no. 5, pp. 3313–3351, May 2021. of Electronic Engineering and Computer Science,
[43] X. Mu, Y. Liu, L. Guo, J. Lin, and R. Schober, “Joint deployment and Queen Mary University of London, since 2017. His research interests include non-
multiple access design for intelligent reflecting surface assisted networks,” orthogonal multiple access, 5G/6G networks, machine learning, and stochastic
IEEE Trans. Wireless Commun., early access, May 4, 2021, doi: geometry. He has served as a TPC Member of many IEEE conferences, such as
10.1109/TWC.2021.3075885. GLOBECOM and ICC. He received the IEEE ComSoc Outstanding Young
Researcher Award for EMEA in 2020. He received the Exemplary Reviewer
Xidong Mu (Graduate Student Member, IEEE) Certificate of IEEE WIRELESS COMMUNICA TIONS LETTERS in 2015, IEEE
received the B.S. degree in information engineering TRANSACTIONS ON COMMUNICATIONS in 2016 and 2017, and IEEE TRANSACTIONS
from Beijing University of Posts and Telecommuni ON WIRELESS COMMUNI CATIONS in 2017 and 2018. He has served as the Publicity
cations (BUPT), Beijing, China, in 2015, where he Co-Chair for VTC 2019-Fall. He is the leading contributor for “Best readings for
non orthogonal multiple access (NOMA)” and a primary contributor for “Best
is currently pursuing the Ph.D. degree.
readings for reconfigurable intelligent surfaces (RIS).” He serves as the Chair of
Since May 2021, he has been a Visiting Stu Special Interest Group (SIG) in Signal Processing and Computing for
dent with the School of Electronic Engineering
Communications (SPCC) Technical Committee on the topic of signal processing Artificial Intelligence, BUPT. His research interests
techniques for next generation multiple access (NGMA), the Vice Chair of SIG include wireless communication, personal commu
Wireless Communications Technical Committee (WTC) on the topic of nication, and communication networks.
reconfigurable intelligent surfaces for smart radio environments (RISE), and the
Tutorials and Invited Presentations Officer for Reconfigurable Intelligent Surfaces
Emerging Technology Initiative. He is currently an Editor on the Editorial Board of
the IEEE TRANSACTIONS ON WIRELESS COMMU NICATIONS, the IEEE
TRANSACTIONS ON COMMUNICATIONS, and IEEE COMMUNICATIONS LETTERS. He
serves as the leading Guest Editor for IEEE J OURNAL ON SELECTED AREAS IN
COMMUNICATIONS Special Issue on “Next generation multiple access,” a Guest
Editor for IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING Special
Issue on “Signal processing advances for non-orthogonal multiple access in next
generation wireless networks.”
Li Guo (Member, IEEE) is currently a profes
sor with the School of Artificial Intelligence and
the Key Laboratory of Universal Wireless Commu
nications, Ministry of Education, Beijing Univer
sity of Posts and Telecommunications. Her major
research interests include future wireless communi Robert Schober (Fellow, IEEE) received the
cation, advanced signal processing using AI technol Diploma (Univ.) and Ph.D. degrees in electrical
ogy, blockchain technology, and embedded system engineering from Friedrich-Alexander University of
design. Erlangen-Nüremberg (FAU), Germany, in 1997 and
2000, respectively.
From 2002 to 2011, he was a Professor and
a Canada Research Chair with The University of
British Columbia (UBC), Vancouver, Canada. Since
January 2012, he is currently an Alexander von
Humboldt Professor and the Chair for Digital Com
munication at FAU. His research interests fall into
the broad areas of communication theory, wireless communications, and statistical
signal processing. He received several awards for his work, including the 2002
Heinz Maier Leibnitz Award of the German Science Foundation (DFG), the 2004
Innovations Award of the Vodafone Foundation for Research in Mobile
Communications, the 2006 UBC Killam Research Prize, the 2007 Wilhelm
Friedrich Bessel Research Award of the Alexander von Humboldt Foundation, the
2008 Charles McDowell Award for Excellence in Research from UBC, the 2011
Alexander von Humboldt Professorship, the 2012 NSERC E.W.R. Stacie
Fellowship, and the 2017 Wireless Communi cations Recognition Award by the
Jiaru Lin (Member, IEEE) received the B.S. and IEEE Wireless Communications Technical Committee. Since 2017, he has been
Ph.D. degrees from the School of Information Engi listed as a Highly Cited Researcher by the Web of Science. He is a fellow of the
neering, Beijing University of Posts and Telecom Canadian Academy of Engineering, the Engineering Institute of Canada, and a
munications (BUPT), China, in 1987 and 2001, member of German National Academy of Science and Engineering. From 2012 to
respectively. 2015, he served as the Editor-in-Chief for the IEEE T RANSACTIONS ON
From 1991 to 1994, he studied at the Swiss Federal COMMUNICATIONS. He currently serves as a member of the Editorial Board of the
PROCEEDINGS OF THE IEEE and as the VP of Publications for the IEEE
Institute of Technology, Zürich. He is currently a
Communication Society (ComSoc).
Professor and a Ph.D. Supervisor with the School of

Authorized licensed use limited to: University of Ulsan. Downloaded on January 07,2024 at 07:58:57 UTC from IEEE Xplore. Restrictions apply.