2548 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 38, NO.

11, NOVEMBER 2020

Reflecting Modulation
Shuaishuai Guo, Member, IEEE, Shuheng Lv, Student Member, IEEE, Haixia Zhang , Senior Member, IEEE,
Jia Ye , Student Member, IEEE, and Peng Zhang , Member, IEEE

Abstract— Reconfigurable intelligent surface (RIS) has realized with minimal hardware complexity and cost based
emerged as a promising technique for future wireless commu- on low-power and low-complexity electronic circuits [2].
nication networks. How to reliably transmit information in a Recently, there has been a large body of literature investi-
RIS-based communication system arouses much interest. This
paper proposes a reflecting modulation (RM) scheme for RIS- gating RIS on the channel estimation [3]–[6], joint precoding
based communications, where both the reflecting patterns and and reflecting designs [7]–[10], information modulation tech-
transmit signals can carry information. Depending on that the niques [11]–[14], performance analysis [15]–[20], hardware
transmitter and RIS jointly or independently deliver information, implementation and experimental work [21]–[24], etc. Among
RM is further classified into two categories: jointly mapped RM them, the investigation on RIS-based information transfer
(JRM) and separately mapped RM (SRM). JRM and SRM are
naturally superior to existing schemes, because the transmit sig- schemes arouses our special interest. Specifically, we are
nal vectors, reflecting patterns, and bit mapping methods of JRM interested in how to reliably convey information at a fixed
and SRM are more flexibly designed. To enhance transmission rate of r bits per channel use (bpcu) in an Nt × Nr MIMO
reliability, this paper proposes a discrete optimization-based joint communication system assisted by an N -unit RIS.
signal mapping, shaping, and reflecting (DJMSR) design for JRM
and SRM to minimize the bit error rate (BER) with a given trans-
mit signal candidate set and a given reflecting pattern candidate A. Prior Work on RIS-Based Information Transfer
set. To further improve the performance, this paper optimizes
multiple reflecting patterns and their associated transmit signal Various RIS-based information transfer schemes in
sets in continuous fields for JRM and SRM. Numerical results literature can fulfill the transmission. According to the roles
show that JRM and SRM with the proposed system optimization of RIS in various schemes, we classify prior work into four
methods considerably outperform existing schemes in BER.
categories: RIS-aided communications (RIS-C), RIS-based
Index Terms— Reconfigurable intelligent surface, reflecting backscatter communications (RIS-BC), RIS-based spatial
modulation, system optimization, bit error rate. modulation (RIS-SM), passive beamforming and information
transfer (PBIT).
I. I NTRODUCTION
1) RIS-C: In RIS-C, RIS only reflects signals. RIS-C sys-

R ECONFIGURABLE intelligent surface (RIS), also

referred as intelligent reflecting surface (IRS), has newly
emerged as a promising technique for wireless communica-
tems attracted the most research attention in literature includ-
ing the investigation for spectral efficiency (SE)/signal power
maximization, capacity/data rate optimization, security/reliable
tions to against unfavorable wireless environment [1]. RIS transmission analysis, channel estimation, etc. Huang et al. [8]
consists of a massive number of passive reflecting units, which made a valuable contribution on maximizing energy effi-
neither introduce too much additional noise nor need signal ciency by jointly designing the RIS phase rotating matrix
processing circuits. Compared to relays, passive RIS can be and power allocation at the base station. The authors of [9]
Manuscript received December 18, 2019; revised April 22, 2020; accepted and [10] improved the power efficiency through optimizing
May 9, 2020. Date of publication July 3, 2020; date of current version the beamformer at the transmitter and the phase shift matrix
October 16, 2020. The work of Shuaishuai Guo, Shuheng Lv, Jia Ye, and Peng at RIS. The primary and extended works on channel capacity
Zhang was supported in part by the National Natural Science Foundation of
China under Grant 61801266, and in part by the Shandong Natural Science were provided by Hu et al. [15], [16], who established the
Foundation under Grant ZR2018BF033. The work of Haixia Zhang was relationship between capacity per square meter surface area
supported by the National Natural Science Foundation of China under Grant and the average transmit power. Later, Hu et al. further
61860206005 and Grant 61671278. (Corresponding author: Haixia Zhang.)
Shuaishuai Guo, Shuheng Lv, and Haixia Zhang are with the Shan- examined the degradations in capacity assuming RIS has
dong Provincial Key Laboratory of Wireless Communication Technolo- hardware impairments in [17]. Jung et al. [18] derived the
gies, Shandong University, Jinan 250061, China, and also with the School asymptotic results of uplink data rate in a RIS-C system
of Control Science and Engineering, Shandong University, Jinan 250061,
China (e-mail: shuaishuai_guo@[Link]; [Link]@[Link]; considering channel estimation errors and spatially correlated
[Link]@[Link]). Rician fading with channel hardening effects. Zhang and
Jia Ye is with Computer, Electrical and Mathematical Science and Engineer- Zhang [19] analyzed the RIS-aided multiple-input multiple-
ing Division, King Abdullah University of Science and Technology (KAUST),
Thuwal 23955-6900, Saudi Arabia (e-mail: [Link]@[Link]). output (MIMO) capacity by using an alternative optimiza-
Peng Zhang is with the School of Computer Engineering, Weifang Univer- tion approach. The physical layer security of RIS-C systems
sity, Weifang 261061, China (e-mail: sduzhangp@[Link]). also gained the attention of researchers. The recent
Color versions of one or more of the figures in this article are available
online at [Link] papers [25]–[27] have made contributions on improving the
Digital Object Identifier 10.1109/JSAC.2020.3007060 difference between the data rate at the legitimate receiver
0733-8716 © 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See [Link] for more information.

Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on May 18,2026 at 05:32:05 UTC from IEEE Xplore. Restrictions apply.
 GUO et al.: REFLECTING MODULATION 2549

and the one at an eavesdropper. It is noteworthy that these contributions in this field appeared in [11], who investigated
investigations were based on Gaussian input assumption to the effect of modulation orders and blind phases on the
provide the insights into communications performance bounds, error performance of the RIS-based communication system.
but neglected the fact that finite constellation signals are The phenomenon that the RIS-based scheme experiences
the most common input to RIS-C systems. Against the degradation of the error performance as the modulation order
background, Ye et al. investigated the joint reflecting and increases was found in [13]. However, it also showed that
precoding designs to minimize the symbol error rate (SER) RIS could take advantage of large numbers of reflecting
for RIS-C in [7]. More recent attention has focused on the elements to counteract the detrimental effect of increasing
provision of channel estimation in RIS-C systems, which the modulation order. Considering maximum energy-based
becomes challenging because of its massive number of passive suboptimal and exhaustive search-based optimal detectors, the-
elements without any signal processing capability. The pilot oretical analysis, and computer simulation results on average
training signals were first introduced by [28] to obtain the bit error probability were provided to validate the potential of
channel state information (CSI). They found the optimal pilot RIS-assisted index modulation schemes on improving the SE
training length, which could maximize the asymptotic SE. and data rates with remarkably low error rate. Reference [20]
Nadeem et al. [3] designed channel estimation protocol based showed that jointly encoding RIS and the transmitter signals
on minimum mean squared error, while Taha et al. [4] solved outperforms the RIS-C transmission. It is worth mentioning
this problem by using the compressive sensing and deep only the reflecting patterns that steer the beam to a sin-
learning methods. Three-stage mechanisms regarding channel gle receive antenna were adopted for data communication
estimation including sparse matrix factorization, ambiguity in RIS-SM [12]. Using such reflecting patterns cannot benefit
elimination, and matrix completion were proposed by [5]. from the receive diversity gain.
2) RIS-BC: In RIS-BC, RIS plays the role of information 4) PBIT: In PBIT, RIS helps the information delivery from
modulator. Backscatter communication is another promising the transmitter to the receiver and has its own information
communication paradigm that enables the backscatter devices to be transmitted. Most recently, PBIT was firstly proposed
to modulate the information over the ambient radio frequency and investigated by Yan et al in [33]. They maximized the
(RF) signals without requiring active energy-emitting compo- average received signal-to-noise ratio (SNR) assuming RIS
nents. Specifically, a transmitter tag switches its antenna to data adopting SM. In detail, the RIS information is carried by
non-reflecting or reflecting mode based on external energy the ON/OFF states of the reflecting elements, while passive
sources in the ambient environment, such as WiFi, public beamforming is achieved by adjusting the phase shifts of the
radio, and cellular transmit power. The differences and similar- activated reflecting elements. The main difference between
ities between RIS and backscatter leading to the occurrence RIS-SM and PBIT is that the transmitter and RIS in RIS-SM
of RIS-based backscatter communication (RIS-BC) systems can jointly encode the information while the transmitter and
appeared in recent works [12], [22]–[24] by Tang et al. RIS in PBIT cannot because they does not share the informa-
They have established several novel wireless communication tion to be transmitted. It should be noted that the reflecting
systems by designing the hardware structure of the transmit- units can only be ON or OFF to carry information in PBIT,
ter based on the concept of the programmable metasurface. which limits the feasible reflecting patterns.
Reflection coefficient controllable metasurface-based transmit- To conclude, the literature identifies that RIS-based informa-
ter occurred in [22], which could process phase modulation of tion transfer can rely on the radiated signals at the transmitter,
the reflected electromagnetic (EM) wave directly. RF chain- the reflecting patterns at the RIS or both for data delivery. The
free transmitter and space-down-conversion receiver were pro- transmitter and RIS can either independently or jointly deliver
posed in [23] based on the superior EM waves manipulation information. The adopted reflecting patterns can be the patterns
capability of programmable metasurfaces. In [24], they have steering the beam to a single receive antenna or just the
designed a novel transmitter without filter, wideband mixer ON/OFF states of each RIS unit. The dimension of the radiated
and power amplifier on the concept of the new programmable signals can be one-dimensional and multi-dimensional. All
metasurface architecture. From the perspective of design archi- these schemes call a unified transmission model for a fair
tectures and preliminary experimental results, all proposed comparison. Besides, there has been no detailed investigation
wireless programmable metasurfaces networks were verified to of system optimization involving signal mapping, shaping and
achieve low hardware complexity, low cost, and high energy reflecting pattern design. All these motivated our work. It is
efficiency. Another example of RIS-BC is RIS-based space hoped that our work will contribute to a deeper understanding
shift keying (RIS-SSK) proposed by Basar in [12], where the of RIS-based information transfer.
transmitter only radiates a carrier signal.
3) RIS-SM: In RIS-SM, RIS both reflects and carries the B. Our Work and Contributions
information. During the past few years, there has been a • In this paper, we propose a reflecting modulation (RM)
growing interest in using the channel index for informa- scheme for RIS-based communications. Depending on
tion modulation. A notable example is a spatial modula- that the transmitter and RIS jointly or independently
tion (SM) [29]–[32], which could simplify the transceiver deliver information, RM is further classified into two
architecture and increase the energy efficiency. The intention categories: jointly mapped RM (JRM) and separately
propagation environment controllability of RIS boosted the mapped RM (SRM). JRM and SRM (JRM&SRM) can
research on RIS-SM systems [11]–[13], [20]. The preliminary cover the existing schemes, i.e., RIS-C, RIS-BC, RIS-SM

Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on May 18,2026 at 05:32:05 UTC from IEEE Xplore. Restrictions apply.
 2550 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 38, NO. 11, NOVEMBER 2020

and PBIT. Since the transmit signal vectors, reflecting

patterns, and bit mapping methods in JRM&SRM are
more flexibly designed, JRM&SRM are naturally superior
to existing schemes. BER analysis of JRM&SRM is
included, which will generate fresh insight into how
the signal mapping, shaping, and reflecting affect the
system BER.
• To enhance transmission reliability, this paper proposes
a discrete optimization-based joint signal mapping, shap-
ing, and reflecting (DJMSR) design for JRM&SRM to
minimize the system BER with a given transmit signal
candidate set and a given reflecting pattern candidate set.
We compare DJMSR with an exhaustive search method in Fig. 1. A RIS-assisted (Nt , Nr , N , r) MIMO communication system.
complexity and performance to validate its effectiveness.
• To further improve the performance, this paper proposes
a continuous optimization-based joint signal mapping, Natural logarithm. C and R stand for the complex and the real
shaping, and reflecting (CJMSR) design for JRM&SRM. domains. IN denotes an N ×N identity matrix. · denotes the
Using an alternative optimization approach, we iteratively floor operation. Q(·) stands for the tail distribution function of
optimize the signal shaping and reflecting in continuous the standard normal distribution. X is a set and
 n |X | represents
fields. In the reflecting design with given transmit sig- the size of set X . ∅ refers to an empty set. m is a binomial
nal sets, multiple reflecting patterns for reflecting and coefficient. (x) and (x) denote the functions to take the
carrying information are jointly optimized. In the signal real and imaginary part of x.
shaping design with given reflecting patterns, the transmit
signal sets for all reflecting patterns are jointly optimized. II. S YSTEM M ODEL
• Comprehensive numerical results are presented. The pro- In this paper, we consider a RIS-based (Nt , Nr , N , r)
posed JRM&SRM are compared to RIC-C, RIS-BC, MIMO communication system as illustrated in Fig. 1, where
RIS-SM, and PBIT to validate their superiority. The Nt , Nr represent the numbers of transmit and receive antennas
effectiveness of DJMSR&CJMSR is validated in various at the transceivers; N stands for the number of RIS units; and
system setups. The impact of channel estimation errors r is the target transmit rate in bit per channel use (bpcu). Let
on the performance of the JRM&SRM with DJMSR L = 2r and there are L possible bit sequences of length r in a
and CJMSR (DJMSR&CJMSR) is studied. Moreover, channel use. In the proposed RM scheme, similar to [11]–[14],
implementation challenges and future directions regard- a sequence bl of r bits is conveyed per channel use not only
ing JRM&SRM are also discussed. by the index of the transmit signal vector xki ∈ CNt ×1 but also
by the index of the reflecting patterns Φk ∈ CN ×N , i.e., the
C. Paper Organization
tuple (i, k). In the following, Xk = {xki } denotes the transmit
The rest of this paper proceeds as follows. Section II depicts signal set when Φk is activated; X denotes the transmit signal
the system model, including the mapping types, the signal candidate set with size |X | = M ; and Ψ represents the
model and the BER analysis. Section III introduces the system reflection pattern candidate set with size |Ψ| = K. Based on
optimization involving signal mapping, shaping and reflecting above denotations, we have xki ∈ Xk ⊆ X and Φk ∈ Ψ.
in use of a given transmit signal candidate set and a given Moreover, due to the specific nature of RIS units, Φk ∈ Ψ is
reflecting pattern candidate set. Section IV extends the opti- a diagonal matrix with each diagonal entry having modulus
mization in continuous fields. Section V presents the numerical 1 or 0, where 0 represents the OFF state. Mathematically,
results. Section VI points out the implementation challenges (Φk )n,n ∈ {0} ∪ {ejθ , θ ∈ R}. Since xki is a general multi-
and future research directions. Conclusions are drawn in the dimensional signal, Φk is a general reflecting pattern, and the
last section. adopted transmit signal sets corresponding to {Φk } can be the
same or not in the proposed transmission model, it is obvious
D. Notations
that the existing RIS-SM and PBIT [11]–[14] can be regarded
Throughout this paper, the term x refers to a scalar; as special realizations of the proposed RM. Moreover, RIS-C
x represents a vector; and X denotes a matrix. The terms x0 , and RIS-BC can also be regarded as special realizations of
x2 , xp , x∞ will be used in this paper to refer to 0 , 2 , RM by setting |Ψ| = 1 and |X | = 1, respectively.
p and ∞ norms. (X)m,n stands for the entry of X located
at m-th row and n-th column. diag (x) means a diagonal
matrix with diagonal elements being x. diag{X} represents a A. Mapping Types
vector drawn from the diagonal elements from X. (·)T , (·)C , The proposed RM is suitable for two application scenarios.
(·)H refer to the transpose, conjugate and conjugate-transpose In the first scenario, the RIS and the transmitter do not share
operators, respectively.  and ⊗ stands for the Hadamard the information bits and they have to independently transmit
and Kronecker products, respectively. 1m×n represents a all- their information similarly to the PBIT in [33]. In this scenario,
one matrix with m rows and n columns. ln represents the the total r bits can only be separately mapped to i and k. Thus,

Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on May 18,2026 at 05:32:05 UTC from IEEE Xplore. Restrictions apply.
 GUO et al.: REFLECTING MODULATION 2551

TABLE I preferred, since the optimized JRM is surely superior to the

S EPARATELY M APPED RM (SRM) optimized SRM. Also, we remark that given the same X and
Ψ, the optimized JRM is naturally superior to RIS-C, RIS-BC,
RIS-SM and PBIT. This is because RIS-C, RIS-BC, RIS-SM
and PBIT are special realizations of JRM and the globally
optimized solution will surely achieve better performance than
the optimized solutions with additional constraints.

B. Signal Model
Let l be the index of the activated tuple (i, k) corresponding
TABLE II
to the bit sequence bl . For either SRM or JRM, when Φk
J OINTLY M APPED RM (JRM)
and xki are activated to deliver bl , the receiver signal vector
y ∈ CNr ×1 can be expressed by

y = (Hd + H2 Φk H1 )xki + n, (1)

as shown in Fig. 1, where Hd ∈ CNr ×Nt denotes channel

matrix of the direct link; H2 ∈ CNr ×N represents the channel
matrix between the RIS and the receive antennas; H1 ∈
CN ×Nt represents the channel matrix between the transmit
and the RIS; and n ∈ CNr ×1 stands for the additive complex
we refer to such a RIS-based information transfer as SRM.
Gaussian noise with zero mean and variance σ 2 INr . In the
In SRM, the number of information bits mapped to i and k
paper, it is assumed that all channels are perfectly known by
are denoted by r1 and r2 , respectively. Since the number of
the transceivers as well as the RIS.
bits has to be integer, consequently the numbers of adopted
With Hd , H2 and H1 being known by the receiver, maxi-
transmit signals and reflecting patterns are limited to be a
mum likelihood (ML) detection can be performed by
power of two.1 In our work, we use Mc = 2r1 and Kc = 2r2
to represent the numbers of transmit signals and reflecting (î, k̂) = arg max pY (y|xki , Φk , Hd , H2 , H1 )
patterns that can be activated. And we have L = Mc Kc . all legitimate(i,k)
Moreover, since the transmitter does not know the information = arg min ||y − (Hd + H2 Φk H1 )xki ||22 , (2)
all legitimate(i,k)
to be transmitted at the RIS, thus the employed transmit signal
sets for different reflecting pattern activations are the same. which is deduced from
SRM is much similar to the concept of PBIT, but with more
flexible reflecting patterns. pY (y|xki , Φk , Hd , H2 , H1 )
In the second scenario, the RIS and the transmitter jointly ∝ exp(−||y − (Hd + H2 Φk H1 )xki ||22 ). (3)
transmit the information bits. In other words, the information
bits can be jointly mapped to the tuple (i, k), similarly to
the jointly mapped SM in [35], [36]. In this paper, we refer C. BER Analysis
to such a RIS-based information transfer as JRM. To show According to [37], the BER of JRM&SRM can be analyzed
the differences between SRM and JRM, an example is pro- to be upper bounded by
vided, where the candidate set for reflecting patterns Ψ =
1  
L L
{Φ1 , Φ2 , Φ3 } and the candidate set for the transmitted signals
Pe ≤ Pe = DHD (bl , bl̂ )PPEP (l → ˆ
l), (4)
X = {x1 , x2 , x3 , x4 }. The bit mapping methods of SRM and Lr
l=1 l̂=1,l̂=l
JRM are given in Tables I and II, respectively. From the tables,
it can be observed that SRM can only use a power of two where DHD (bl , bl̂ ) represents the Hamming distances between
reflecting patterns, while JRM can use a more flexible (i, k) bl and bl̂ ; PPEP (l → ˆl) represents the pairwise error proba-
tuple set with size L. In JRM, the transmit signal sets for each bility. In (4), PPEP (l → l̂) can be calculated by
reflecting pattern activation can be the same or not. To show ⎛ ⎞
the differences, we denote xki as the i-th transmit signal when D (l, l̂)2
ED
activating Φk . In SRM, {xki } are the same for any Φk that P (l → ˆl) = Q ⎝ ⎠, (5)
2σ 2
can be activated. In the given example, X1 = X2 = X and
X3 = ∅ in SRM. From this point, SRM can be regarded
where
as a special case of JRM. Therefore, if the transmitter and
the RIS do not independently deliver the information, JRM is DED (l, l̂)2 = ||(Hd + H2 Φk H1 )xki − (Hd +H2 Φk̂ H1 )xk̂î ||22 ,
1 It should be mentioned that even though there are some methods such as (6)
fractional bit mapping method [34] that can release the number constraint,
such methods typically suffered from severe error propagation and high and DED (l, l̂) represents the Euclidean distances between the
detection complexity. two noise-free received signal vectors.

Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on May 18,2026 at 05:32:05 UTC from IEEE Xplore. Restrictions apply.
 2552 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 38, NO. 11, NOVEMBER 2020

III. D ISCRETE O PTIMIZATION -BASED J OINT S IGNAL positions in n. Based on these denotations, the DJMSR opti-
M APPING , S HAPING AND R EFLECTING mization in SRM can be formulated to be
D ESIGN (DJMSR)
(OP2) : Given : Hd , H1 , H2 , σ, X , Ψ, Kc , r1
In this paper, one key concern is how to design the sys- Find : Γ1 , Γ2 , X1 , X2 , · · · , XK
tem to minimize the system BER. In the section, we will
Minimize : Pe
present the joint signal mapping, shaping, reflecting design
to minimize the BER upper bound Pe with a given transmit Subject to : Γ1 , Γ2 are bijective, (8a)
signal candidate set X and a given reflecting pattern candidate ||n||0 = Kc , (8b)
set Ψ. Xn1 = Xn2 = · · · = XnKc ⊆ X ,
(8c)
A. Problem Formulation |Xn1 | = Mc , (8d)
For JRM, we refer to the bijective mapping rule of 1 
||xi ||22 = 1, (8e)
sequences {bl } to tuples {(i, k)} as Γ. Observing the expres- Mc
xi ∈Xn1
sion of Pe in (4), we find that the BER upper bound is
jointly determined by the Hamming distances {DHD (bl , bl̂ )} where (8a) is the constraint for the mapping rules; (8b) limits
and the Euclidean distances {DED (l, l̂)}. Among them, the number of the reflecting patterns that can be activated;
the Hamming distances {DHD (bl , bl̂ )} are only affected by (8c) limits the transmit signal sets chosen from the candidate
the mapping rule, i.e., Γ, while the Euclidean distances set for different reflecting patterns are the same; (8d) is the size
{DED (l, l̂)} are jointly affected by the adopted reflecting constraint for the transmit signal set; (8e) is the normalized
patterns Φ1 , Φ2 , · · · , ΦK and the corresponding transmit sig- power constraint for all transmit signals.
nal sets X1 , X2 , · · · , XK . Based on these denotations, given
a transmit signal candidate set X and a reflecting pattern B. Design Procedure
candidate set Ψ, the DJMSR optimization problem for JRM In this subsection, we will analyze the formulated prob-
can be formulated to be lems (OP1) and (OP2) to optimize the signal map-
ping, shaping, and reflecting for JRM&SRM. According to
(OP1) : Given : Hd , H1 , H2 , σ, X , Ψ, L [35, Lemma 1], it is known that the BER is more sensitive
Find : Γ, X1 , X2 , · · · , XK to the Euclidean distances {DED (l, l̂)} than the Hamming
Minimize : Pe distances {DHD (bl , bl̂ )} especially in the high SNR regime.
Based on the fact, one can firstly optimize the signal shaping
Subject to : Γ is bijective, (7a)
and reflecting that affects {DED (l, l̂)} with omitting the impact
Xk ⊆ X , k = 1, 2, · · · , K, (7b) of signal mapping by setting DHD (bl , bl̂ ) = 1, ∀l = l̂, and
|Xk | ≥ 0, k = 1, 2, · · · , K, (7c) then optimize the signal mapping that affects {DHD (bl , bl̂ )}.

K In JRM, the joint signal shaping and reflecting optimization
|Xk | = L, (7d) is indeed a subset selection problem. The equivalence can
k=1 be explained by an understanding of JRM that there are
1 K  KM feasible (i, k) tuples and L out of them are chosen for
||xki ||22 = 1, (7e) data transmission. Thus, one can perform exhaustive search
L
k=1 xk
i ∈Xk with complexity O KM
CPe , where CPe represents the
L
where (7a) is the constraint for the mapping rule; (7b) limits all computational complexity required for computing the objec-
the transmitted signal sets corresponding to different reflecting tive function Pe . The computational complexity is sometimes
patterns to be subsets of X ; (7c) and (7d) represents the set prohibitive, for instance, when K = 10, M = 10, L = 64,
 
size constraints for the transmit signal sets; (7e) stands for
KM
L = 100L ≈ 2.0 × 1027 . To reduce the complexity,
the normalized power constraint for all transmit signals. It is we propose to use a stepwise depletion procedure, which can
noteworthy that even though the optimization variables does be described as follows. The stepwise depletion is started with
not include {Φk }, the reflecting patterns can be optimized, initializing the legitimate tuple set to have all feasible tuples.
because |Xk | being nonzero or not will determine Φk being Pick up a tuple and compute the Pe of the rest tuple set with
used or not. the average transmit power being normalized. Then drop one
For SRM, we denote the bits-to-transmit-signal mapping of the tuples whose rest tuple set has the minimum P e . Then
rule as Γ1 and the bits-to-reflecting-pattern mapping rule as Γ2 . its rest tuple with the minimum P e become the legitimate tuple
Since the adopted transmit signal sets for each reflecting set. In each iteration, the number of tuples in the legitimate
pattern activation have to be the same, the number of feasible tuple set is reduced by 1. Repeat the process step by step until
solutions of X1 , X2 , · · · , XK is greatly reduced. Recall that the legitimate tuple set size being L.
Mc , Kc are used to denote the numbers of the transmit signals In SRM, the joint signal shaping and reflecting optimization
and reflecting patterns that can be activated. It means that there is to solve two subset selection problems, which are selecting
are Kc nonzeros in a vector n = [|X1 |, |X2 |, · · · , |XK |]T , Kc = 2r2 reflecting patterns that can be activated from Ψ
i.e., ||m||0 = Kc and let n1 , n2 , · · · , nKc be the nonzero and selecting Mc = 2r1 legitimate transmit signals from X .

Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on May 18,2026 at 05:32:05 UTC from IEEE Xplore. Restrictions apply.
 GUO et al.: REFLECTING MODULATION 2553

Algorithm 1 DJMSR Design for JRM C. Computational Complexity and Performance Analysis
1: Input: Hd , H1 , H2 , σ, X , Ψ, L. In the DJMSR for JRM, the computational complexity
2: Select the legitimate tuple set by performing the stepwise of the joint shaping and reflecting as well as the signal
depletion procedure, which is equivalent to optimizing mapping in use of BSA [39] are all dominated by the
X1 , X2 , · · · , XK directly. computation of the objective function Pe . In the stepwise
3: Perform BSA [39] to obtain Pseudo Gray mapping for Γ. depletion procedure for joint shaping and reflecting, finding
4: Output Γ and X1 , X2 , · · · , XK . the tuple to be dropped needs to compute the Pe regarding
t tuples by t times, where t denotes of the number of
tuples in the legitimate tuple set. Computing Pe once
Algorithm 2 DJMSR Design for SRM needs to compute the t noise-free received signal vectors
and their mutual Euclidean distances, which requires
1: Input: Hd , H1 , H2 , σ, X , Ψ, Kc , Mc .
O tN 2 (Nr + Nt ) + t2 Nr multiplications. Multiplying it by
2: Select the legitimate transmit signal set by performing a KM 2 2 3
zero-embedding stepwise depletion procedure. t times yields O t=L+1 t N (Nr + Nt ) + t Nr . Adding
3: Select the legitimate reflecting pattern set by performing a the complexity of BSA [39], which requires around L2 times
stepwise depletion procedure. of Pe with L tuples that needs O L3 N 2 (Nr + Nt ) + L4 Nr .
4: Perform BSA [39] to generate Pseudo Gray mapping for Jointly considering all above and the relation KM ≥ L,
Γ1 and Γ2 . the complexity order of DJMSR for JRM is
5: Output Γ1 , Γ2 , and X1 , X2 , · · · , XK .
CDJMSR
JRM
= O[K 3 M 3 N 2 (Nr + Nt ) + K 4 M 4 Nr ]. (9)
Using exhaustive search (ES)-based signal shaping and
reflecting for JRM with Pseudo Gray mapping induces a
The exhaustive search will induce the computational complex- complexity of O KM
CPe . Based on above analysis,
L
ity of O K
Kc Mc CPe . To facilitate the low-complexity
M
the aggregate computational complexity of the ES method in
implementation, we propose a low-complexity two-step opti- the number of multiplications can be written as
mization approach. In the first step, we select Mc legitimate    
KM  
signal vectors in X . In the selection, we propose to use a zero- CES
JRM
=O + L2 LN 2 (Nr + Nt ) + L2 Nr .
embedding stepwise depletion procedure. In detail, we first L
embed a zero vector 0 ∈ CNt ×1 in X to generate Xˆc = (10)
{0} ∪ X and then perform a stepwise depletion procedure Similarly, we can analyze the computation orders of DJMSR
to repeatedly discard the signal vector expect 0 that leads to and ES for SRM to be
unfavorable mutual Euclidean distances until the legitimate
signal set size being Mc + 1. Then by removing 0 from CDJMSR
SRM
= O[K 3 Mc N 2 (Nr + Nt ) + K 4 Mc2 Nr ], (11)
the set, we can obtain the final legitimate transmit signal set
of size Mc , denoted by Xc . The zero embedding approach and
    
is to avoid selecting the signal vector that is close to the K M 2
 2 2

CES
SRM
=O +L LN (Nr +Nt )+L Nr .
all-zero vector 0. Even though a vector close to 0 in Xc Kc Mc
may not affect the mutual Euclidean distances in Xc , it will (12)
severely damage the detection of the activated reflecting pat-
terns. This selection of legitimate transmit signal set can be To quantify the performance improvement brought by the
conducted off-line, since the selection does not involve the proposed DJMSR compared to a benchmark system with given
CSI. In the second step, we select the Kc legitimate reflecting signal vector sets, reflecting pattern set and with a natural
patterns from Ψ. An original stepwise depletion procedure mapping scheme, the SNR gain for achieving a same BER in
can be adopted. That is, repeatedly discarding one of the the high SNR regime can be computed as [39]
reflecting patterns that lead to unfavorable Pe until the set size 10 log10
PeD
being Kc . γG = − lim
PeB
(dB), (13)
After the signal set and the reflecting pattern set are settled, SNR→∞ Nr
Pseudo Gray mapping [38] can be adopted for JRM&SRM to where PeD represents the BER of the system with the proposed
further reduce the BER. Pseudo Gray mapping is a general DJMSR and PeB represents the BER of the benchmark system.
mapping method for any signal constellation, which can be In the high SNR, the BER can be approximated to be
obtained by a low-complexity binary switching algorithm min
  min
 2
min min
(BSA) [39], [40]. BSA starts with an initial mapping. Then DHD DED (l, l̂) DHD DED (l, l̂)
Pe ≈ λ Q √ ≈λ exp − ,
exchanging the bit labels of any two tuples generates a new r 2σ 2r 4σ 2
mapping. Comparing them in terms of Pe , we choose the (14)
one of two mappings with better performance. Repeating the
min
process until all the labels of all tuples are exchanged, then where λ represents the number of closest pairs; DHD stands
the Pseudo-Gray mapping is obtained. To summary, we list for the average Hamming distance between the bit sequences
min
the DJMSR for JRM and SRM in Algorithms 1 and 2. mapped to the closest pairs; and DED refers to the Euclidean

Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on May 18,2026 at 05:32:05 UTC from IEEE Xplore. Restrictions apply.
 2554 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 38, NO. 11, NOVEMBER 2020

min
distance of the closest pairs, i.e., DED = minl=l̂ DED (l, l̂). To solve the problem, we introduce four new matrices
Substituting (14) into (13), we have K̃
  
r2 r−1
Ĥd = [Hd , Hd , · · · , Hd ] ∈ CNr ×K̃Nt , (17)
γG =
Nr (2r − 1) K̃
    
10 log10 ιB min
exp DED (l, l̂)2D − DED
min
(l, l̂)2B Ĥ2 = [H2 , H2 , · · · , H2 ] ∈ CNr ×K̃N , (18)
ιD ⎡ ⎤
+
Nr
(dB), H1 0 ··· 0
⎢ 0 H1 · · · 0 ⎥
(15) ⎢ ⎥ K̃N ×K̃Nt
Ĥ1 = ⎢ . .. . .. ⎥ ∈ C , (19)
⎣ . . . . . . ⎦
where the first term is the mapping gain [39]; the second term
0 0 ··· H1
stands for the gain brought by the proposed signal shaping and ⎡ ⎤
reflecting methods; λB , λD represent the numbers of closest Φ1 0 ··· 0
⎢ 0 Φ2 ··· 0 ⎥
pairs in the system with DJMSR and the benchmark system; ⎢ ⎥
min
DED (l, l̂)2D , DED
min
(l, l̂)2B refer to the Euclidean distance of the
Dq = ⎢ .. .. .. .. ⎥ ∈ CK̃N ×K̃N , (20)
⎣ . . . . ⎦
closest pairs in the system with DJMSR and the benchmark 0 0 ··· ΦK̃
system, respectively.
and L new vectors

IV. C ONTINUOUS O PTIMIZATION -BASED J OINT S IGNAL x̂l = gk ⊗ xki ∈ CK̃Nt ×1 , l = 1, 2, · · · , L, (21)
M APPING , S HAPING AND R EFLECTING where gk ∈ CK̃×1 is the kth K̃-dimensional vector basis
D ESIGN (CJMSR) with all zeros except the kth entry being one. Moreover,
It is worth noting that the aforementioned phase-shift we define q = diag{Dq }. With these new matrices and vec-
reflecting design in DJMSR are based on a given Ψ, which tors, the square of the Euclidean distances can be transformed
corresponds to the RIS equipped with discrete phase-shift to be
reflecting units or ON/OFF reflecting units. With a large DED (l, l̂)2 = ||(Hd + H2 Φk H1 )xki − (Hd + H2 Φk̂ H1 )xk̂î ||22
number of continuous phase shift units equipped on the RIS,
the set size of the feasible reflecting patterns Ψ goes to infinity. = ||(Ĥd + Ĥ2 Dq Ĥ1 )(x̂l − x̂l̂ )||22
! !2
The discrete optimization-based reflecting design in DJMSR ! ! H
= !Ĥ2 Dq Ĥ1 x̂l,l̂ ! + x̂H Ĥd Ĥ2 Dq Ĥ1 x̂l,l̂
will fail. In addition, the sets X1 , X2 , · · · , XK in DJMSR are 2 l, l̂
! !2
optimized with a given transmit signal set X . Consequently, H H H H ! !
+ x̂l,l̂ Ĥ1 Dq Ĥ2 Ĥd x̂l,l̂ + !Ĥd x̂l,l̂ !
the performance is highly depending on X . How to optimize ! !2
2
! !2
X1 , X2 , · · · , XK in the generalized complex field instead of ! ! ! !
= !Ĥ2 Dq Ĥ1 x̂l,l̂ ! + 2 q al,l̂ + !Ĥd x̂l,l̂ ! ,
T
X remains unanswered. To address these issues, we will 2 2
investigate the joint signal shaping and reflecting design in (22)
the continuous fields. In our design, the optimization is firstly where x̂l,l̂ = x̂l − x̂l̂ ∈ CK̃Nt ×1 and al,l̂ ∈ CK̃N ×1 is a vector
initialized with the output of DJMSR for JRM&SRM. Then, T
with n-th element (al,l̂ )n = xH ĥ ĥ x , where ĥd,2,n is
l,l̂ d,2,n 1,n l,l̂
we alternatively optimize the reflecting patterns and transmit
H
signal sets to minimize the system BER. the n-th column of Ĥd Ĥ2 and ĥ1,n denotes the n-th column
T
of Ĥ1 . The first term of (22) can be written as
! !2
A. Alternative Optimization ! ! H
!Ĥ2 Dq Ĥ1 x̂l,l̂ ! = mH DH Ĥ2 Ĥ2 Dq ml,l̂
l,l̂ q
1) Continuous Optimization-Based Reflecting (COR) 2

Design: For either JRM or SRM, given the non-empty q RĤ2 Dq ΔMl,l̂ ,
= tr DH (23)
transmit signal vector sets X1 , X2 , · · · , XK̃ for K̃ reflecting H
patterns having been selected from a candidate set satisfying where ml,l̂ = Ĥ1 x̂l,l̂ ∈ CK̃N ×1 , RĤ2 = Ĥ2 Ĥ2 ∈ CK̃N ×K̃N
K̃
k=1 |Xk | = L by DJMSR, optimizing multiple reflecting and ΔMl,l̂ = ml,l̂ mH ∈ CK̃N ×K̃N . Based on the equation
patterns is our new target. That is, we next aim to  H  l,l̂
tr Du ADv B T
= uH (A  B) v with Du = diag {u} and
jointly optimize Φ1 , · · · , ΦK̃ , to reflect the signals and ! !2
! !
simultaneously carry information. The corresponding Dv = diag {v} [41], the term !Ĥ2 Dq Ĥ1 x̂l,l̂ ! can be trans-
2
optimization problem can be formulated to be formed to be
! !2
! !
(OP3) : Given : Hd , H1 , H2 , σ, X1 , X2 , · · · , XK̃ !Ĥ2 Dq Ĥ1 x̂l,l̂ ! = qH RH2  ΔMTl,l̂ q
2
Find : Φ1 , Φ2 , · · · , ΦK̃ = qH Cl,l̂ q, (24)
Minimize : Pe
where Cl,l̂ = RĤ2  ΔMTl,l̂ ∈ CK̃N ×K̃N . Submitting the term
Subject to : |(Φk )nn | = 1, (Φk )mn = 0, into (22) yields
∀m = n, ! !2
! !
DED (l, l̂)2 = qH Cl,l̂ q + 2 qT al,l̂ + !Ĥd x̂l,l̂ ! . (25)
k = 1, 2, · · · , K̃. (16) 2

Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on May 18,2026 at 05:32:05 UTC from IEEE Xplore. Restrictions apply.
 GUO et al.: REFLECTING MODULATION 2555

The objective function of (OP3) becomes Algorithm 3 Continuous-Optimization Based Reflecting

Design
1  
L L
Pe (q) = DHD (bl , bl̂ ) 1: Initialize a feasible solution q0 , a halting criterion ε1 > 0,
Lr a penalty parameter t, and the iteration number iter = 0.
l=1 l̂=1,l̂=l
⎛# $ ! !2 ⎞ 2: Compute the gradient according to (30).
$ qH C q + 2 qT a ! ! 3: Project the negative gradient to obtain Δq according to
⎜% l,l̂ l,l̂ + !Ĥd l,l̂ ! ⎟
x̂
×Q⎝ ⎜ 2⎟
⎠.
(31).
2σ 2 4: Update qiter+1 = qiter + αiter Δqiter , where αiter is computed
by a line search procedure to satisfy the Wolfe conditions
(26) [42] and iter ← iter + 1.
5: Repeat steps 2-4 until the halting criterion [g(qiter ) −
Then, the problem (OP3) is transformed to be
g(qiter+1 )]/g(qiter ) ≤ ε1 is met.
∗
(OP4) : Given : Cl,l̂ , l = ˆ
l = 1, 2, · · · , L 6: Output qn = |qqnn | for all n = 1, 2, · · · , K̃N and Φk =
∗ ∗
+1 , · · · , qkN +1 ] ), k = 1, 2, · · · , K̃.
T
Find : q diag([q(k−1)N
Minimize : Pe
Subject to : |qn | = 1, n = 1, 2, · · · , K̃N.
qiter+1 can be expressed as
(27)

Similarly g(qiter+1 ) − g(qiter ) = g(qiter + αiter Δqiter ) − g(qiter )

  to [7], we relax the unit modulus constraint to be
tr qqH = K̃N and q∞ ≤ 1. To deal with constraint ≤ uαiter [∇q g(q)]H Δqiter
q∞ ≤ 1, we use a large p norm to replace the infinity = −uαiter [∇q g(q)]H P1 ∇q g(q) (32)
norm and rebuild a new objective function as
where u is a small real positive number in the Wolfe condi-
g(q) = Pe (q) + B(1 − ||q||p ), (28) tions [42] and αiter > 0. Substituting P1 = PH
1 P1 , we have

where B(·) is a penalty function defined by g(qiter+1 ) − g(qiter ) ≤ −cαiter ||P1 ∇q g(q)||22 < 0. (33)
⎧
⎨ − 1 ln (u) , u > 0 Since g(qiter+1 ) in (28) is lower bounded, Algorithm 3 will
B (u) = t (29) converge.
⎩ ∞, u ≤ 0, 2) Continuous Optimization-Based Signal Shaping (COS)
and t is the penalty parameter. Taking the first derivation Design: For JRM, X1 , , · · · , XK̃ may be different. Based on
of the objective function g(q) with respect to q yields Φ1 , · · · , ΦK̃ and the set sizes |X1 |, , · · · , |XK̃ |, set entries of
(30), as shown at the bottom of the page, where pq = X1 , · · · , XK̃ can be further optimized in the complex field by
* +T solving the following problem:
p−2 p−2 p−2
q1 · |q1 | , q2 · |q2 | , . . . , qK̃N · |qK̃N | .
 H (OP5) : Given : Hd , H1 , H2 , σ, Φ1 , Φ2 , · · · , ΦK̃ ,
To satisfy the constraint tr qq = K̃N , we perform a
projection as |X1 |, |X2 |, · · · , |XK̃ |, L
Find : X1 , X2 , · · · , XK̃
Δq = −P1 ∇q g(q) (31)
Minimize : Pe
to ensure qH Δq = 0 in the update process, where P1 is a 1 
K

projection matrix given by P1 = IK̃N − K̃N qqH

and P1 = PH Subject to : ||xki ||22 = 1. (34)
1 = L
2 H k
k=1 xi ∈Xk
P1 = P1 P1 . Based on the obtained Δq, we can search the
solution iteratively as listed in Algorithm 3. Writing Gk = H + H2 Φk H1 , we re-express the square of the
Proposition 1: Algorithm 3 converges. Euclidean distances as
Proof: During each iteration in Algorithm 3, the variance
of the objective function in the update process from qiter to DED (l, l̂)2 = ||Gk xki − Gk̂ xk̂î ||22 . (35)

#
1  
L L $
$ 
1
∇q g(q) = −
Lr
DHD (bl , bl̂ )$
% ! !2 
! !
l=1 l̂=1,l̂=l πσ 2 qH C
l,l̂ q + 2 l,l̂ + !Ĥd x̂l,l̂ !
qT a
2
⎛  ! !2  ⎞
! !
⎜ q Cl,l̂ q + 2 q al,l̂ + !Ĥd x̂l,l̂ !2 ⎟
H T
1−p
qp pq
× exp ⎜
⎝−
⎟ C q + aC +
⎠ l,l̂ (30)
4σ 2 l,l̂
2t 1 − qp

Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on May 18,2026 at 05:32:05 UTC from IEEE Xplore. Restrictions apply.
 2556 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 38, NO. 11, NOVEMBER 2020

For neatness, we introduce several new matrices Algorithm 4 Continuous Optimization-Based Signal Shaping
⎡ ⎤ for JRM
|X1 | |X2 | |XK̃ |
⎢        ⎥ 1: Initialize a feasible solution z0 , a halting criterion ε2 > 0,
W = ⎣G1 , · · · , G1 , G2 , · · · , G2 ,, · · · , GK̃ , · · · , GK̃ ⎦ and the iteration number iter = 0.
2: Compute the gradient according to (42).
∈ CNr ×LNt , (36) 3: Project the negative gradient to obtain Δz according to
⎡ 1 ⎤ (43).
X1 0 0 ··· 0 0 0
⎢ .. ⎥ 4: Update ziter+1 = ziter + βiter Δziter , where βiter is computed
⎢ 0 . 0 ··· · 0 0 ⎥
⎢ ⎥ by a line search procedure to satisfy the Wolfe conditions
⎢ 0 X1|X1 | ··· · ⎥ [42] and iter ← iter + 1.
⎢ 0 0 0 ⎥
⎢ . .. .. .. .. .. ⎥ 5: Repeat steps 2-4 until the halting criterion [Pe (ziter ) −
Dz  ⎢
⎢ .. . .
..
. . . .
⎥
⎥
⎢ ⎥ Pe (ziter+1 )]/Pe (ziter ) ≤ ε2 is met.
⎢ 0 · 0 ··· XK1 0 0 ⎥ 6: Output xk
⎢ ⎥ i = ŴDz ol , where is defined by Ŵ =
⎢ .. ⎥ L
⎣ 0 0 · ··· 0 . 0 ⎦   
0 0 0 ··· 0 0 XK
|XK̃ |
[INt , · · · , INt ] ∈ CNt ×LNt .
LNt ×LNt
∈C , (37)
 
where To satisfy the constraint tr zzH = L, we perform a projection

Xki = diag (xki ) ∈ CNt ×Nt . (38) Δz = −P2 ∇z Pe (z) (43)

We define z = diag{Dz } ∈ CLNt ×1 and ol = el ⊗ 1Nt ×1 ∈ to ensure zH Δz = 0, where P2 is a projection matrix given
H
2
CLNt ×1 , l = 1, 2, · · · , L, where el is the lth L-dimensional by P2 = IL − zzL satisfying P2 = PH H
2 = P2 = P2 P2 . Based
vector basis with all zeros except the lth entry being one. on the derived search direction, the continuous optimization
Based on these new matrices and vectors, the square of the algorithm can be conducted as listed in Algorithm 4.
pairwise Euclidean distances can be expressed as For SRM, K̃ = Kc . Owing to X1 = X2 = · · · = XK̃ ,
the optimization problem becomes
DED (l, l̂)2 = ||Gk xki − Gk̂ xk̂î ||22
(OP7) : Given : G1 , G2 , · · · , GK̃ , σ, Mc
= ||WDz ol − WDz ol̂ ||22 Find : X1 , X2 , · · · , XK̃
= (ol − ol̂ )H DH
z W WDz (ol − ol̂ )
H
Minimize : Pe
 H 
= Tr Dz RW Dz ΔOll̂ 1 
Subject to : ||xi ||22 = 1,
(a) Mc
= zH Zl,l̂ z, (39) xi ∈X1
X1 = X2 = · · · = XK̃ ,
where RW = WH W ∈ CLNt ×LNt , ΔOll̂ = (ol − ol̂ )(ol −
|X1 | = |X2 | = · · · = |XK̃ | = Mc .
ol̂ )H ∈ CLNt ×LNt and Zl,l̂ = RHG ΔOHll̂
∈ CLNt ×LNt . The
(44)
equality (a) holds due to Tr(Du UDv VH ) = uH (U  V)v for
any two diagonal matrices Dv = diag(v) and Du = diag(u). By introducing c = [xT1 , xT2 , · · · , xTMc ]T ∈ CMc Nt ×1 , f =
Based on the reformulation, (OP5) becomes [(c)T , (c)T ]T ∈ R2Mc Nt ×1 , we transform (OP7) to be
(OP6) : Given : Zl,l̂ , l = ˆ
l = 1, 2, · · · , L (OP8) : Given : G1 , G2 , · · · , GK̃ , σ, Mc
Find : z Find : f
Minimize : Pe (z) Minimize : Pe (f)
   
Subject to : tr zzH = L. (40) Subject to : tr ffH = Mc . (45)

where To solve it, we compute the first derivative of Pe (f) with

⎛ ⎞ respect to f to be
1  
L L
zH Zl,l̂ z
Pe (z) = DHD (bl , bl̂ )Q ⎝ ⎠. (41) ∇f Pe (f) = [∇f1 Pe (f), ∇f2 Pe (f), · · · , ∇f2Mc Nt Pe (f2Mc Nt )]T ,
Lr 2σ 2
l=1 l̂=1,l̂=l (46)
Taking the first derivation of the objective function Pe (z) with where ∇fi Pe (f) is approximately computed by
respect to z yields Pe (f + δfi ) − Pe (f)
 ∇fi Pe (f) ≈ , (47)
1  
L L
1 δ
∇z Pe (z) = − DHD (bl , bl̂ ) 2 with δ being a small number and fi denoting the i-th 2Mc Nt ×
Lr πσ z Zl,l̂ z
H
l=1 l̂=1,l̂=l
 H  1 base vector. To satisfy the constraint tr ffH = Mc ,
z Zl,l̂ z we perform a projection as
× exp − Zl,l̂ z, (42)
4σ 2 Δz = −P3 ∇f Pe (f), (48)

Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on May 18,2026 at 05:32:05 UTC from IEEE Xplore. Restrictions apply.
 GUO et al.: REFLECTING MODULATION 2557

Algorithm 5 Continuous Optimization-Based Signal Shaping where Niter5 denotes for the number of iterations in
for SRM Algorithm 5.
1: Initialize a feasible solution f0 , a halting criterion ε3 > 0, Thus, the aggregate computational complexity of CJMSR
and the iteration number iter = 0. for JRM&SRM can be computed by
2: Compute the gradient according to (46).  
CCJMSR
JRM
= CDJMSR
JRM
+ Niter6
J
CCOR + CCOS
JRM
, (52)
3: Project the negative gradient to obtain Δz according to
(48). and
4: Update fiter+1 = fiter + ηiter Δfiter , where ηiter is computed  
CCJMSR
SRM
= CDJMSR
SRM
+ Niter6
S
CCOR + CCOS
SRM
, (53)
by a line search procedure to satisfy the Wolfe conditions
[42] and iter ← iter + 1. where Niter6
J
and Niter6
S
refers to the numbers of iterations in
5: Repeat steps 2-4 until the halting criterion [Pe (fiter ) − use of Algorithm 6 for CJMSR for JRM&SRM, respectively.
Pe (fiter+1 )]/Pe (fiter ) ≤ ε3 is met. Similarly to Section III-B, the performance improvement
6: Compute c = [f1 , · · · fMc Nt ]T + j[fMc Nt +1 , · · · f2Mc Nt ]T . brought by the proposed CJMSR method can be quantified by
7: Output xi = [c(i−1)Nt +1 , c2 , · · · , ciNt ]T . the SNR gain. Moreover, we remark that the proposed signal
shaping can be regarded as a joint optimization of the transmit
beamforming and transmit symbol vectors, which inherently
Algorithm 6 CJMSR Design for JRM&SRM
outperforms the transmit beamforming optimization with the
1: Input: Hd , H1 , H2 , σ, X , Ψ, L, Kc , Mc .
given transmit symbol vectors, e.g., the work in [7].
2: Perform DJMSR to obtain X1 , X2 , · · · , XK̃ .
3: Update Φ1 , Φ2 , · · · , ΦK̃ by performing COR design in use
V. N UMERICAL R ESULTS
of Algorithm 3.
4: Update X1 , X2 , · · · , XK̃ by performing the COS design in In this section, numerical results are presented to investi-
use of Algorithm 4 or 5. gate the performance of the proposed JRM&SRM in various
5: Repeat step 2 and 3 until the halting criterion is met. (Nt , Nr , N, r) RIS-based MIMO communication systems. It is
6: Update the mapping rules by performing the BSA [39]. divided by three subsections. Subsection V-A investigates the
performance of the proposed JRM&SRM with DJMSR based
on given X and Ψ. The comparison among JRM, SRM,
to ensure fH Δf = 0, where P3 is the projection matrix given RIS-C, RIS-BC (e.g, RIS-SSK), RIS-SM and PBIT is also
H
2 presented. Subsection V-B studies the performance of the
by P3 = IMc − qq H H
Mc satisfying P3 = P3 = P3 = P3 P3 . Based
on the derived search direction, the continuous optimization proposed JRM&SRM with CJMSR. Subsection V-C shows
algorithm can be conducted as listed in Algorithm 5. the system performance in the presence of channel estimation
Proposition 2: Both Algorithms 4 and 5 converge. errors. All simulation results are evaluated over 1000 Rayleigh
Proof: This proposition can be proved similarly to channel realizations.
Proposition 1.
Based on the proposed COR and COS designs, the reflecting A. Performance of JRM&SRM With DJMSR
and signal shaping can be alternatively optimized. We sum- Firstly, we compare the proposed DJMSR for JRM&SRM
marize the CJMSR in Algorithm 6. Propositions 1 and 2 can with the exhaustive search (ES) algorithm in both perfor-
guarantee the convergence of Algorithm 6. mance and complexity in (1, 3, 4, 3) RIS-based communication
systems. The numerical results regarding system BER and
B. Computational Complexity and Performance Analysis computational complexity are illustrated in Figs. 2 and 3,
In COR, the gradient calculation dominates the computa- where a randomly-generated X with |X | = 5 and a randomly-
tional complexity, involving calculating Cl,l̂ and qH Cl,l̂ q for generated Ψ with |Ψ| = 3 are adopted. Results in Figs. 2 and 3
all l = l . Based on the fact, the complexity order of COR can demonstrate that the proposed DJMSR for JRM can achieve
be analyzed to be almost the same performance as the ES solution, while the
* + computational complexity is reduced by an order of magnitude
CCOR = O L2 K̃ 2 N 2 (Nr + Nt )Niter3 , (49) in the number of multiplications. It also demonstrated that the
where Niter3 represents the number of iterations in proposed DJMSR for SRM has a certain level of performance
Algorithm 3. loss compared with the ES solution, but enjoys a considerably
In COS for JRM, the computation of Zl,l̂ and zH Zl,l̂ z are reduced complexity.
needed, which introduces the complexity: As aforementioned in Section II, JRM&SRM are more gen-
* + eralized and naturally superior to existing schemes. To verify
CCOS
JRM
= O L2 (L2 Nt2 + K̃N )(Nr + Nt )Niter4 , (50) this, we compare the proposed JRM&SRM with DJMSR with
where Niter4 stands for the number of iterations in Algorithm 4. RIS-C, RIS-BC, RIS-SM and PBIT in a (1, 4, 5, 2) RIS-based
In COS for SRM, the computation of the Pe have to be communication system, as illustrated in Fig. 4. To show the
computed 2Mc Nt + 1 times, which introduces the complexity: superiority of the proposed designs, we use a union set X =
XRIS-C ∪ XRIS-BC ∪ XRIS-SM ∪ XPBIT as the signal candidate set
CCOS
SRM
= O Mc Nt LN 2 (Nr + Nt )Niter5 + L2 Nr Niter5 , and a union set Ψ = ΨRIS-C ∪ ΨRIS-BC ∪ ΨRIS-SM ∪ ΨPBIT as the
(51) reflecting pattern candidate set. Simulation results demonstrate

Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on May 18,2026 at 05:32:05 UTC from IEEE Xplore. Restrictions apply.
 2558 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 38, NO. 11, NOVEMBER 2020

Fig. 2. BER comparison between the proposed DJMSR and exhaustive search Fig. 5. BER performance of the proposed JRM&SRM with CJMSR in
algorithms in (1, 3, 4, 3) RIS-based communication systems. (a):(2, 3, 4, 3), (b): (1, 3, 4, 3) and (c): (2, 4, 4, 3) RIS-based MIMO systems.

PBIT have additional constraints compared to RM, they are

naturally less comparable.

B. Performance of the Proposed JRM&SRM With CJMSR

Secondly, we investigate the performance of CJMSR
for JRM&SRM in (a):(2, 3, 4, 3), (b): (1, 3, 4, 3) and (c):
(2, 4, 4, 3) RIS-based MIMO systems, as illustrated in Fig. 5.
In the simulations, we include the sole COR and COS for
Fig. 3. Computational complexity comparison of DJMSR with exhaustive comparison. It is demonstrated that the sole COR brings
search algorithm (1, 3, 4, 3) RIS-based communication systems in the number around 1 dB gain, the sole COS bring about 4 dB gain, while
of multiplications.
CJMSR brings around 6 dB in JRM systems. In SRM systems,
the gains brought by COR, COS and CJMSR are 2 dB, 2 dB
and 6 dB, respectively. From these observations, we conclude
that both COR and COS can considerably improve the perfor-
mance and iteratively performing both in CJMSR enlarges the
gain. Another significant insight gained from the results is that
COS for JRM can provide more gain than COS for SRM. This
is because that COS in JRM can optimize more optimization
variables (i.e., LNt complex variables) than COS in SRM (i.e.,
Mc Nt complex variables) and more optimization dimensions
lead to a more considerable performance gain. By comparing
the performance of (2, 3, 4, 3) systems with that of (1, 3, 4, 3)
and (2, 4, 4, 3) systems, as expected, increasing the number of
antennas either at the transmitter side or at the receiver side
brings better performance.

C. Performance of the Proposed JRM With the Increase of

the Number of RIS Units
Fig. 4. BER comparison among the proposed JRM&SRM, existing RIS-C,
RIS-BC (e.g., RIS-SSK), RIS-SM and PBIT in (1, 4, 5, 2) RIS-based com- Thirdly, we investigate the performance of the proposed
munication systems. schemes with the increase of the number of RIS units in
systems with Nt = 2, Nr = 3 and r = 3, as illustrated
that the optimized JRM&SRM considerably outperform exist- in Fig. 6. With the increase of the number of RIS units from
ing designs. In details, JRM with DJMSR outperforms RIS-C 4 to 10 and 20, it is found that the system reliability is
and RIS-SM by around 4-5 dB in the depicted high SNR increased substantially by around 6 dB and 11 dB, respectively.
regime. Compared to RIS-BC and PBIT, the performance gain This is because increasing the number of RIS units increases
becomes more obvious. The performance differences mainly the number of reflected paths to the receiver. For comparison,
result from the bit mapping method, transmit signal shaping the performance of RIS-C with the optimized fixed reflect-
and reflecting patterns. Since RIS-C, RIS-BC, RIS-SM, and ing pattern is included. Simulation results demonstrate that

Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on May 18,2026 at 05:32:05 UTC from IEEE Xplore. Restrictions apply.
 GUO et al.: REFLECTING MODULATION 2559

Fig. 7. BER performance in the presence of channel estimation errors in a

(3, 3, 6, 4) RIS-based MIMO communication system.

Fig. 6. BER performance of the proposed JRM with the increase of the
number of the RIS units with Nt = 2, Nr = 3 and r = 3. patterns in a system time. It is challenging for high-data-rate
short-symbol-time communication, since the fast modification
may be limited by the hardware. But, it is worth mentioning
JRM outperforms RIS-C under all system setups, which indi-
that RIS-BC also needs to modify the reflecting patterns per
cates that employing RIS as additional information modulators
symbol time and the data rates reported in current experi-
has performance advantages.
mental work [12], [22]–[24] are growing higher. We believe
that the problem can be addressed with the development of
D. Performance in Presence of Channel Estimation Errors hardware. Third and most importantly, CSI acquisition and
In the paper, a key assumption is that all CSI are globally sharing are problematic. The design in this paper assuming
and perfectly known by the transceivers and the RIS. However, each channels are perfectly known. However, due to the lack
obtaining CSI is costly and the perfect CSI is typically not of signal processing capacity, the separate channels cannot be
available. To investigate the applicability of the proposed known. Thus, how to design the JRM&SRM systems based
designs to the scenarios with imperfect CSI. An channel on cascaded channel estimation and CSI sharing is a promis-
estimation error model is adopted to model imperfect Hd , H1 ing direction. Besides, the information carrying capacity of
and H2 , which is written by Him = Hp +He . In the model, Him JRM&SRM should be probed in to provide design guideline
and Hp respectively denote the estimated and the real channel and insights. JRM&SRM-based multi-user communications
matrix; He represents the channel error matrix with each entry call for a through and deep investigation.
following a zero-mean Gaussian variable with variance (δσ)2
[36], [39]–[41]. This means that the channel estimation errors VII. C ONCLUSION
is propositional to the channel noise variance. Based on the This paper aimed to contribute to this growing area of
channel estimation error model, we simulate the BER of the research by exploring RIS for information transfer. A gen-
proposed designs under severe channel estimation errors and eral concept named RM with more flexible transmit signals,
compared it with the design employing the perfect CSI. Sim- reflecting patterns, and bit mapping methods was proposed.
ulation results as illustrated in Fig. 7 show that those designs RM was classified into JRM and SRM (JRM&SRM) according
are all affected by channel estimation errors. Even though the to the fact that the transmitter and RIS jointly transmit infoma-
proposed JRM&SRM with CJMSR are more sensitive to the tion or not. Both analysis and simulations results showed that
channel estimation errors, they still show great performance the optimized JRM&SRM surely outperform existing designs.
improvement compared to RIS-C in the presence of the same To enhance transmission reliability, this paper proposed a
level of channel estimation errors. discrete optimization-based joint signal mapping, shaping, and
reflecting design for JRM&SRM to minimize the system BER
VI. I MPLEMENTATION C HALLENGES with a given transmit signal candidate set and a given reflecting
AND F UTURE D IRECTIONS pattern candidate set. To further improve the performance,
The implementation challenges of RM lies in that the we proposed to alternatively optimize the signal shaping and
synchronization between the transmitter and RIS, the fast reflecting in continuous fields. In the reflecting design with
reflecting modification at the RIS, the acquisition and sharing given transmit signal sets, multiple reflecting patterns for
of precise CSI for DJMSR&CJMSR. First, to encode the reflecting and carrying information were jointly optimized.
information together, the transmitter and RIS should be per- In the signal shaping design with given reflecting patterns,
fectly synchronized. This is challenging since RIS do not have the transmit signal sets for all reflecting patterns were jointly
information processing capacity on the surface. It requires optimized. The computational complexity of all proposed
additional synchronization signal receive circuit at the con- designs were quantitatively in the number of multiplications.
troller. Second, JRM&SRM have to modify the reflecting The performance of the proposed designs were investigated in

Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on May 18,2026 at 05:32:05 UTC from IEEE Xplore. Restrictions apply.
 2560 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 38, NO. 11, NOVEMBER 2020

various systems with and without considering CSI estimation [19] S. Zhang and R. Zhang, “Capacity characterization for intelligent
errors. It was found that JRM&SRM achieve much better reflecting surface aided MIMO communication,” IEEE J. Sel. Areas
Commun., early access, Jun. 8, 2020, doi: 10.1109/JSAC.2020.3000814.
performance than existing schemes in all cases. Moreover, [20] R. Karasik, O. Simeone, M. Di Renzo, and S. Shamai (Shitz), “Beyond
implementation challenges and future directions of RM were Max-SNR: Joint encoding for reconfigurable intelligent surfaces,” in
pointed out. We believe that our proposed generalized trans- Proc. IEEE ISIT, Los Angeles, CA, USA, Jun. 2020.
[21] W. Tang et al., “Wireless communications with reconfigurable intelligent
mission model and design methods can provide a significant surface: Path loss modeling and experimental measurement,” 2019,
reference to related research. arXiv:1911.05326. [Online]. Available: [Link]
[22] W. Tang et al., “Wireless communications with programmable
R EFERENCES metasurface: Transceiver design and experimental results,” 2018,
arXiv:1811.08119. [Online]. Available: [Link]
[1] M. D. Renzo et al., “Smart radio environments empowered by [23] W. Tang et al., “Wireless communications with programmable meta-
reconfigurable AI meta-surfaces: An idea whose time has come,” surface: Transceiver design and experimental results,” China Commun.,
EURASIP J. Wireless Commun. Netw., vol. 2019, no. 1, pp. 1–20, vol. 16, no. 5, pp. 46–61, May 2019.
May 2019. [24] W. Tang et al., “Programmable metasurface-based RF chain-free 8PSK
[2] M. Di Renzo et al., “Reconfigurable intelligent surfaces vs. Relay- wireless transmitter,” Electron. Lett., vol. 55, no. 7, pp. 417–420,
ing: Differences, similarities, and performance comparison,” 2019, Apr. 2019.
arXiv:1908.08747. [Online]. Available: [Link] [25] M. Cui, G. Zhang, and R. Zhang, “Secure wireless communication via
[3] Q.-U.-A. Nadeem, H. Alwazani, A. Kammoun, A. Chaaban, M. Debbah, intelligent reflecting surface,” IEEE Wireless Commun. Lett., vol. 8,
and M.-S. Alouini, “Intelligent reflecting surface-assisted multi-user no. 5, pp. 1410–1414, Oct. 2019.
MISO communication: Channel estimation and beamforming design,” [26] H. Shen, W. Xu, S. Gong, Z. He, and C. Zhao, “Secrecy rate
IEEE Open J. Commun. Soc., vol. 1, pp. 661–680, May 2020. maximization for intelligent reflecting surface assisted multi-antenna
[4] A. Taha, M. Alrabeiah, and A. Alkhateeb, “Enabling large intelli- communications,” IEEE Commun. Lett., vol. 23, no. 9, pp. 1488–1492,
gent surfaces with compressive sensing and deep learning,” 2019, Sep. 2019.
arXiv:1904.10136. [Online]. Available: [Link] [27] X. Yu, D. Xu, and R. Schober, “Enabling secure wireless
[5] Z.-Q. He and X. Yuan, “Cascaded channel estimation for large intelligent communications via intelligent reflecting surfaces,” in Proc. IEEE
metasurface assisted massive MIMO,” IEEE Wireless Commun. Lett., Global Commun. Conf. (GLOBECOM), Waikoloa, HI, USA,
vol. 9, no. 2, pp. 210–214, Feb. 2020. Dec. 2019, pp. 1–6.
[6] D. Mishra and H. Johansson, “Channel estimation and low-complexity [28] M. Jung, W. Saad, and G. Kong, “Performance analysis of large intelli-
beamforming design for passive intelligent surface assisted MISO wire- gent surfaces (LISs): Uplink spectral efficiency and pilot training,” 2019,
less energy transfer,” in Proc. IEEE Int. Conf. Acoust., Speech Signal arXiv:1904.00453. [Online]. Available: [Link]
Process. (ICASSP), Brighton, U.K., May 2019, pp. 4659–4663. [29] R. Y. Mesleh, H. Haas, S. Sinanovic, C. W. Ahn, and S. Yun, “Spatial
[7] J. Ye, S. Guo, and M.-S. Alouini, “Joint reflecting and precoding modulation,” IEEE Trans. Veh. Technol., vol. 57, no. 4, pp. 2228–2241,
designs for SER minimization in reconfigurable intelligent surfaces Jul. 2008.
assisted MIMO systems,” IEEE Trans. Wireless Commun., early access, [30] M. Di Renzo, H. Haas, A. Ghrayeb, S. Sugiura, and L. Hanzo,
May 20, 2020, doi: 10.1109/TWC.2020.2994455. “Spatial modulation for generalized MIMO: Challenges, opportuni-
[8] C. Huang, A. Zappone, G. C. Alexandropoulos, M. Debbah, and ties, and implementation,” Proc. IEEE, vol. 102, no. 1, pp. 56–103,
C. Yuen, “Reconfigurable intelligent surfaces for energy efficiency in Jan. 2014.
wireless communication,” IEEE Trans. Wireless Commun., vol. 18, no. 8, [31] S. Guo, H. Zhang, and M.-S. Alouini, “MIMO capacity with
pp. 4157–4170, Aug. 2019. reduced RF chains,” 2019, arXiv:1901.03893. [Online]. Available:
[9] M. Fu, Y. Zhou, and Y. Shi, “Intelligent reflecting surface for downlink [Link]
non-orthogonal multiple access networks,” in Proc. IEEE Globecom [32] S. Guo, H. Zhang, P. Zhang, P. Zhao, L. Wang, and M.-S. Alouini,
Workshops (GC Wkshps), Waikoloa, HI, USA, Dec. 2019, pp. 1–6. “Generalized beamspace modulation using multiplexing: A breakthrough
[10] X. Yu, D. Xu, and R. Schober, “MISO wireless communication systems in mmWave MIMO,” IEEE J. Sel. Areas Commun., vol. 37, no. 9,
via intelligent reflecting surfaces : (Invited Paper),” in Proc. IEEE/CIC pp. 2014–2028, Sep. 2019.
Int. Conf. Commun. China (ICCC), Changchun, China, Aug. 2019, [33] W. Yan, X. Yuan, and X. Kuai, “Passive beamforming and information
pp. 735–740. transfer via large intelligent surface,” IEEE Wireless Commun. Lett.,
[11] E. Basar, “Transmission through large intelligent surfaces: A new vol. 9, no. 4, pp. 533–537, Apr. 2020.
frontier in wireless communications,” in Proc. Eur. Conf. Netw. Com- [34] N. Serafimovski, M. Di Renzo, S. Sinanovic, R. Y. Mesleh, and H. Haas,
mun. (EuCNC), Valencia, Spain, Jun. 2019, pp. 112–117. “Fractional bit encoded spatial modulation (FBE-SM),” IEEE Commun.
[12] E. Basar, “Reconfigurable intelligent surface-based index modulation: Lett., vol. 14, no. 5, pp. 429–431, May 2010.
A new beyond MIMO paradigm for 6G,” IEEE Trans. Commun., vol. 68, [35] S. Guo, H. Zhang, S. Jin, and P. Zhang, “Spatial modulation via
no. 5, pp. 3187–3196, May 2020. 3-D mapping,” IEEE Commun. Lett., vol. 20, no. 6, pp. 1096–1099,
[13] E. Basar, M. Di Renzo, J. De Rosny, M. Debbah, M.-S. Alouini, Jun. 2016.
and R. Zhang, “Wireless communications through reconfigurable [36] S. Guo, H. Zhang, P. Zhang, D. Wu, and D. Yuan, “Generalized
intelligent surfaces,” IEEE Access, vol. 7, pp. 116753–116773, 3-D constellation design for spatial modulation,” IEEE Trans. Commun.,
Aug. 2019. vol. 65, no. 8, pp. 3316–3327, Aug. 2017.
[14] V.-D. Nguyen, M. D. Renzo, A. Mansour, A. Coatanhay, and [37] D. Tse and P. Viswanath, Fundamentals of Wireless Communication.
N. Linh-Trung, “A closed-form expression of the BER of reconfigurable Cambridge, U.K.: Cambridge Univ. Press, 2005.
antenna aided space shift keying (SSK),” in Proc. 19th Int. Symp. [38] K. Zeger and A. Gersho, “Pseudo-gray coding,” IEEE Trans. Commun.,
Commun. Inf. Technol. (ISCIT), Sep. 2019, pp. 350–355. vol. 38, no. 12, pp. 2147–2158, Dec. 1990.
[15] S. Hu, F. Rusek, and O. Edfors, “Beyond massive MIMO: The potential [39] S. Guo, H. Zhang, P. Zhang, and D. Yuan, “Link-adaptive
of data transmission with large intelligent surfaces,” IEEE Trans. Signal mapper designs for space-shift-keying-modulated MIMO systems,”
Process., vol. 66, no. 10, pp. 2746–2758, May 2018. IEEE Trans. Veh. Technol., vol. 65, no. 10, pp. 8087–8100,
[16] S. Hu, F. Rusek, and O. Edfors, “The potential of using large antenna Oct. 2016.
arrays on intelligent surfaces,” in Proc. IEEE 85th Veh. Technol. Conf. [40] S. Guo, H. Zhang, P. Zhang, and D. Yuan, “Adaptive mapper
(VTC Spring), Sydney, NSW, Australia, Jun. 2017, pp. 1–6. design for spatial modulation with lightweight feedback over-
[17] S. Hu, F. Rusek, and O. Edfors, “Capacity degradation with model- head,” IEEE Trans. Veh. Technol., vol. 66, no. 10, pp. 8940–8950,
ing hardware impairment in large intelligent surface,” in Proc. IEEE Oct. 2017.
Global Commun. Conf. (GLOBECOM), Abu Dhabi, UAE, Dec. 2018, [41] S. Guo, H. Zhang, P. Zhang, S. Dang, C. Liang, and M.-S. Alouini,
pp. 1–6. “Signal shaping for generalized spatial modulation and generalized
[18] M. Jung, W. Saad, Y. Jang, G. Kong, and S. Choi, “Performance analysis quadrature spatial modulation,” IEEE Trans. Wireless Commun., vol. 18,
of large intelligent surfaces (LISs): Asymptotic data rate and chan- no. 8, pp. 4047–4059, Aug. 2019.
nel hardening effects,” 2018, arXiv:1810.05667. [Online]. Available: [42] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge, U.K.:
[Link] Cambridge Univ. Press, 2004.

Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on May 18,2026 at 05:32:05 UTC from IEEE Xplore. Restrictions apply.
 GUO et al.: REFLECTING MODULATION 2561

Shuaishuai Guo (Member, IEEE) received the B.E. Jia Ye (Student Member, IEEE) received the [Link].
and Ph.D. degrees in communication and informa- degree in communication engineering from South-
tion systems from the School of Information Science west University, Chongqing, China, in 2018. She is
and Engineering, Shandong University, Jinan, China, currently pursuing the master’s degree with the King
in 2011 and 2017, respectively. He visited The Uni- Abdullah University of Science and Technology
versity of Tennessee at Chattanooga (UTC), USA, (KAUST), Saudi Arabia. Her main research interests
from 2016 to 2017. He worked as a Post-Doctoral include the performance analysis and modeling of
Research Fellow at the King Abdullah University wireless communication systems.
of Science and Technology (KAUST), Saudi Arabia,
from 2017 to 2019. He is currently working as a Full
Professor at Shandong University. His research inter-
ests include wireless multiple-input–multiple-output communications, optical
wireless communications, and reconfigurable intelligent surface (RIS).
Shuheng Lv (Student Member, IEEE) received the
B.E. degree in automation from the School of Elec-
trical and Information Engineering, Tianjin Univer-
sity, China, in 2017. He is currently pursuing the
master’s degree with Shandong University, China.
His research interests include terahertz communi-
cation technologies and reconfigurable intelligent
surface (RIS).

Haixia Zhang (Senior Member, IEEE) received the

B.E. degree from the Department of Communica-
tion and Information Engineering, Guilin University
of Electronic Technology, China, in 2001, and the
[Link]. and Ph.D. degrees in communication and
information systems from the School of Informa-
tion Science and Engineering, Shandong University,
China, in 2004 and 2008, respectively. From 2006 to
2008, she was an Academic Assistant with the
Institute for Circuit and Signal Processing, Technical
University of Munich. From 2016 to 2017, she was Peng Zhang (Member, IEEE) received the Ph.D.
a Visiting Professor at the University of Florida, USA. She is currently degree in communication and information systems
a Full Professor with Shandong University. Her current research interests from Shandong University, Jinan, China, in Decem-
include cognitive radio systems, cooperative (relay) communications, resource ber 2012. He is currently an Associate Professor
management, space-time process techniques, mobile edge computing, and with the School of Computer Engineering, Weifang
smart communication technologies. University, Shandong, China. His current research
Dr. Zhang serves on editorial boards of the IEEE I NTERNET OF T HING interests include advanced coded modulation, mas-
J OURNAL, the IEEE W IRELESS C OMMUNICATIONS L ETTERS , and China sive multiple-input–multiple-output systems, spatial
Communications. She has been actively participating in many professional modulation, cross-layer design, cooperative commu-
services, serving as TPC member, session chair, and invited speaker for nications, and 5G wireless communications.
conferences.

Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on May 18,2026 at 05:32:05 UTC from IEEE Xplore. Restrictions apply.