1

Reconfigurable Intelligent Surface (RIS) Aided

Multi-User Networks: Interplay Between
NOMA and RIS
Yuanwei Liu, Xidong Mu, Xiao Liu, Marco Di Renzo, Zhiguo Ding, and Robert Schober
arXiv:2011.13336v1 [cs.IT] 26 Nov 2020

Abstract

This article focuses on the exploitation of reconfigurable intelligent surfaces (RISs) in multi-user
networks employing orthogonal multiple access (OMA) or non-orthogonal multiple access (NOMA),
with an emphasis on investigating the interplay between NOMA and RIS. Depending on whether the
RIS reflection coefficients can be adjusted only once or multiple times during one transmission, we
distinguish between static and dynamic RIS configurations. In particular, the capacity region of RIS
aided single-antenna NOMA networks is characterized and compared with the OMA rate region from
an information-theoretic perspective, revealing that the dynamic RIS configuration is capacity-achieving.
Then, the impact of the RIS deployment location on the performance of different multiple access schemes
is investigated, which reveals that asymmetric and symmetric deployment strategies are preferable for
NOMA and OMA, respectively. Furthermore, for RIS aided multiple-antenna NOMA networks, three
novel joint active and passive beamformer designs are proposed based on both beamformer based and
cluster based strategies. Finally, open research problems for RIS-NOMA networks are highlighted.

Y. Liu and X. Liu are with the School of Electronic Engineering and Computer Science, Queen Mary University of London,
London E1 4NS, UK, (email: {yuanwei.liu, x.liu}@qmul.ac.uk).
X. Mu is with School of Artificial Intelligence and Key Laboratory of Universal Wireless Communications, Ministry of
Education, Beijing University of Posts and Telecommunications, Beijing, China (email: [email protected]).
M. Di Renzo is with Université Paris-Saclay, CNRS, CentraleSupélec, Laboratoire des Signaux et Systèmes, 3 Rue Joliot-Curie,
91192 Gif-sur-Yvette, France. (email: [email protected]).
Z. Ding is with the School of Electrical and Electronic Engineering, The University of Manchester, Manchester, UK (e-mail:
[email protected]).
R. Schober is with the Institute for Digital Communications, Friedrich-Alexander-University Erlangen-Nürnberg (FAU),
Germany (e-mail: [email protected]).
 2

I. I NTRODUCTION

Reconfigurable intelligent surfaces (RISs) [1–4] have been proposed as promising candidate technolo-
gies for next-generation wireless networks due to their potential to control the propagation of wireless
signals through customized reflections and refractions [1]. More particularly, by adjusting the phases
and amplitudes of the reconfigurable RIS elements, desired signals can be enhanced and undesired
interferences can be mitigated to realize a so-called ‘Smart Radio Environment’ [1, 5]. As a result, the
coverage and spectral efficiency of existing wireless networks can be significantly improved. In contrast to
conventional active relaying technologies, RISs consume much less energy and are less expensive since
they do not employ radio frequency (RF) chains nor other high power consumption components [6].
RISs can be seamlessly integrated into existing wireless communication networks to realize different
objectives, including the reduction of the transmit power consumption [7], the mitigation of inter-user
interference [8], the avoidance of signal blockage in millimeter-wave (mmWave) communications [9],
and the facilitation of advanced modulation techniques [10]. Motivated by these benefits of RISs, we
focus our attention on investigating RIS aided multi-user networks. Compared to RIS aided single-user
networks, simultaneously serving multiple users in the time/frequency/power domains introduces new
challenges for smartly coordinating the interference to efficiently exploit the spatial degrees of freedom
(DoFs) facilitated by the joint active and passive beamforming at the base station (BS) and the RIS,
respectively. To begin with, we provide a brief introduction to the integration of RISs into multi-user
networks and the combination of multiple access (MA) schemes with two types of RIS configurations.

A. RIS Aided Multi-User Networks Exploiting Different MA Schemes

Fig. 1 illustrates an RIS aided multi-user network employing non-orthogonal multiple access (NOMA)
and orthogonal multiple access (OMA), respectively. Specifically, OMA is further divided into time
division multiple access (TDMA) and frequency division multiple access (FDMA). For RIS configuration,
two options are considered, namely static RIS configuration and dynamic RIS configuration. For static RIS
configuration, the reflection coefficients of the RIS can be adjusted only once during one transmission.
Then, users are distinguished in the power/time/frequency domain for a static resource allocation policy
based on the adopted MA strategy, as illustrated in the top right of Fig. 1. For dynamic RIS configuration,
the RIS reflection coefficients are reconfigured N times during one transmission. As a result, the RIS
can create an artificial fading channel. This unique feature enabled by the RIS allows dynamic resource
allocation for the adopted MA scheme, as illustrated in the bottom right of Fig. 1.
 3

Power

Power

Power
Direct Links RIS
Reflection Links

Time
me T
Time T
Time

Static RIS configuration Static RIS configuration Static RIS configuration

NOMA TDMA FDMA

Power

Power
BS

Power
…
… Ċ Ċ Ċ

1 2 Ċ N Time 1 2 Ċ N Tim
Time 1 2 Ċ N Time
T

Dynamic RIS configuration Dynamic RIS configuration Dynamic RIS configuration

RIS aided multi-user network NOMA TDMA FDMA

Fig. 1. Illustration of an RIS aided multi-user network employing NOMA, TDMA, and FDMA under static and dynamic RIS
configuration.

B. Motivation for Applying NOMA in RIS Aided Multi-user Networks

Inspired by the aforementioned discussion of MA in RIS aided wireless networks, we focus on NOMA
in the following. On the one hand, NOMA is capable of providing efficient MA in RIS aided multi-user
networks, thus improving the spectral efficiency and enhancing massive connectivity. On the other hand,
RISs introduce the following benefits in existing NOMA networks: 1) The reflection links of RISs can
enhance the performance of existing NOMA networks by providing additional signal diversity without
consuming extra time slots and energy. 2) RISs increase the design flexibility of NOMA networks,
resulting in the transition from ‘channel condition-based NOMA’ to ‘quality of service (QoS)-based
NOMA’. More specifically, in conventional NOMA, the successive interference cancelation (SIC) decoding
order of the users is generally determined by the given ‘dumb’ channel conditions [11]. As a ‘channel
changing’ technique, RISs can enhance or degrade the channel quality of individual users by adjusting
the reflection coefficients and the deployment locations. This enables a ‘smart’ NOMA design. 3) The
use of RISs can reduce the constraints in multiple-antenna NOMA design. For the conventional multiple-
input multiple-output (MIMO) NOMA design, constraints on the numbers of antennas at the transmitters
and receivers may have to be satisfied [12]. With the aid of RISs, such constraints can be relaxed due
to the additional passive array gains provided by RISs. Given the potential benefits facilitated by the
interplay between NOMA and RIS, researchers have begun to heavily investigate RIS-NOMA networks.
Nevertheless, the application of NOMA in RIS aided networks introduces numerous new challenges,
which provide the main motivation for this article.
 4

Our main contributions can be summarized as follows.

• The fundamental information-theoretic capacity/rate limits of RIS aided multi-user single-antenna

networks are characterized and compared for NOMA and OMA employing different RIS configu-
rations, where the dynamic RIS configuration is shown to be capacity-achieving.
• The optimal deployment of RISs is discussed and deployment strategies for different MA schemes
are presented.
• Novel joint beamformer designs are proposed for RIS aided multiple-antenna NOMA networks,
where beamformer based strategies and cluster based strategies are developed.

II. C APACITY C HARACTERIZATION : A N I NFORMATION -T HEORETIC P ERSPECTIVE

In this section, we discuss the information-theoretic capacity limits of the RIS aided multi-user broad-
cast channel (BC), revealing that NOMA with dynamic RIS configuration maximizes the capacity region
and is the capacity-achieving transmission strategy.
We focus on an RIS aided BC where one single-antenna base station (BS) serves K single-antenna
users assisted by an RIS, as illustrated in Fig. 1. It is known that the capacity region of the BC with
single-antenna transmitter and single-antenna receivers is achieved by NOMA transmission, i.e., invoking
superposition coding (SC) at the transmitter and SIC at the receivers. The capacity region contains all
achievable rate tuples of the users under a constraint on the total transmit power. In contrast to the
conventional BC without RIS, the reflection coefficients at the RIS and the wireless resource allocation
at the BS have to be jointly optimized to characterize the capacity region of the considered RIS aided
BC. Next, we discuss the capacity region characterization of the RIS aided BC for static and dynamic
RIS configuration, respectively.

A. Static RIS Configuration

In this case, the reflection coefficients of the RIS are fixed for the entire transmission, as illustrated
in Fig. 1. For given reflection coefficients, the considered RIS aided BC is identical to the conventional
BC. As a result, for the static RIS configuration, the capacity region of the RIS aided BC can be
characterized by exhaustively considering all possible reflection coefficients, and taking the union of all
achievable NOMA rate tuples. The size of the search space is determined by the number of reflection
elements and the resolution of the RIS phase shifters. Furthermore, the rate regions achieved by TDMA
and FDMA for static RIS configuration can be similarly characterized based on the achievable rate tuples
for each scheme.
 5

B. Dynamic RIS Configuration

For dynamic RIS configuration, the RIS reflection coefficients are assumed to be reconfigured more than
once during the transmission, as illustrated in Fig. 1. The capacity region for dynamic RIS configuration
is characterized by the union of all achievable NOMA rate tuples over all possible combinations of N
sets of reflection coefficients. However, determining this capacity region is prohibitively complex since
the size of the set of all possible reflection coefficient combinations grows exponentially with N .
To tackle this obstacle and to determine a performance upper bound for RISs, we consider the ideal case
where N tends to infinity, which means that the time duration needed to adjust the RIS is neglected, i.e.,
the RIS configuration can be updated in real time. In this case, time sharing can be carried out among all
possible reflection coefficients. Thus, the capacity region for infinitely large N can be obtained by taking
the convex hull of the capacity regions achieved by all possible static RIS configurations. Moreover, it can
be verified that, for dynamic RIS configuration, the capacity region for infinitely large N always contains
the capacity region for finite N , which in turn contains the capacity region for static RIS configuration.
Thus, the dynamic RIS configuration with NOMA is capacity-achieving for the considered RIS aided
multi-user BC. The TDMA/FDMA rate region for dynamic RIS configuration can be obtained in a similar
manner.

C. Discussions and Insights

6
NOMA, Dynamic RIS Configuration, infinitely large N
TDMA/FDMA, Dynamic RIS Configuration, infinitely large N
NOMA, Static RIS Configuration
5 TDMA/FDMA, Static RIS Configuration

NOMA
Rate at user 2 (bit/s/Hz)

TDMA/FDMA
3

NOMA capacity
2 TDMA/FDMA rate region improvement
region improvement by dynamic RIS
by dynamic RIS configuration

1 configuration

0
0 0.5 1 1.5 2 2.5 3
Rate at user 1 (bit/s/Hz)

Fig. 2. Illustration of the boundaries of the NOMA capacity region and the TDMA/FDMA rate region for an RIS aided two-user
BC for static and dynamic RIS configuration. The full parameter settings can be found in [13].

We continue by providing a numerical example for an RIS aided two-user BC to demonstrate the
fundamental capacity/rate limits. We assume that the direct links and the reflection links suffer from
 6

Rayleigh and Rician fading, respectively. For a random channel realization, we compare in Fig. 2 the
NOMA capacity region of the RIS aided two-user BC with the rate region achieved by TDMA/FDMA
under the two considered RIS configurations [13]. For dynamic RIS configuration, we evaluate the case
when N tends to infinity. As expected, the NOMA capacity region contains the TDMA/FDMA rate region
for both RIS configurations. Moreover, we observe that the capacity/rate gain introduced by dynamically
adjusting the RIS is much more pronounced for TDMA/FDMA than for NOMA. This highlights an
interesting capacity-complexity tradeoff: although NOMA introduces additional complexity for carrying
out SIC at the users, it requires less hardware complexity for the RISs as dynamic configuration is
generally not needed. To fully reap the RIS benefits, OMA schemes require a much more sophisticated
RIS hardware design than NOMA, i.e., real-time RIS control. On the other hand, SIC is avoided in OMA
due to the orthogonal transmission. The tradeoff between hardware complexity and network capacity is
an interesting topic for further investigation. Note that, in this section, the capacity was characterized
for the RIS aided single-antenna BC. The capacity limits and the capacity-achieving strategies for RIS
aided networks with multiple-antenna transmitters and receivers are still largely unknown and constitute
an interesting topic for future work.

III. RIS D EPLOYMENT D ESIGN FOR M ULTIPLE ACCESS

Having characterized the fundamental capacity limits of RIS aided multi-user networks, we focus our
attention on the RIS deployment design for different MA schemes. Besides adjusting the reflection coef-
ficients, the RIS deployment location is another important DoF affecting system performance. Therefore,
we need to jointly consider the deployment location of the RIS and the wireless resource allocation of
the users to fully reap the benefits of RISs. In this section, we provide some initial results for the joint
deployment and MA design for RIS aided multi-user networks.
We first discuss the interplay between RIS deployment and NOMA for enabling a ‘QoS-based NOMA’
operation. For ease of exposition, we consider a two-user NOMA network. For the conventional NOMA
network without RIS, the user with the higher direct channel gain is designated as the strong user, and
the other is the weak user. RIS deployment can have two different objectives:

• Consolidation: On the one hand, an RIS can be deployed in the vicinity of the strong user to further
enhance its channel quality. By doing so, the channel disparity between the two users can be further
enlarged, and NOMA yields higher performance gains over OMA. This consolidation-based RIS
deployment strategy is suitable for the case when the strong user has a stricter QoS requirement
than the weak user.
 7

• Reverse: On the other hand, an RIS can be deployed near the weak user to enhance its channel quality
and to reverse the original ‘channel-condition based’ decoding order without RIS. The reverse-based
RIS deployment strategy is appealing for the case where the QoS requirements and decoding orders
without RIS are not aligned.
For TDMA and FDMA, where orthogonal transmission is employed, the RIS can be deployed in the
vicinity of the users experiencing extremely poor channel conditions to enhance the efficiency of the
allocated resources. Additionally, another effective RIS deployment strategy for TDMA/FDMA is based
on the users’ distributions. For instance, deploying the RIS in the vicinity of those users who are close
together with highly correlated channels can simultaneously improve their channel conditions.

6
S-NOMA S-FDMA D-TDMA

w 2 =[0.25 0.25 0.25 0.25] w 1 =[0.1 0.2 0.3 0.4]

4
y(m)

User1 User2 User3 User4

0
30 35 40 45
x(m)

Fig. 3. Illustration of the weighted-sum-rate optimal horizontal RIS deployment location for different MA schemes for two
different user rate weight vectors w1 and w2 . The BS is located at (0, 0, 5) m. The heights of the RIS and the users are 1.5
m and 0 m, respectively. The RIS deployment location can be varied between 30 m and 45 m along the x-axis. The other
parameter settings can be found in [14].

Fig. 3 illustrates the optimal RIS deployment locations for different MA schemes for maximization
of the weighted sum rate in a four-user network [14], where the direct BS-user links are assumed to be
blocked and only the reflection BS-RIS-user links are present and follow the Rician fading channel model.
In particular, NOMA and FDMA with static RIS configuration, denoted by S-NOMA and S-FDMA, are
considered along with TDMA with dynamic RIS configuration, referred to as D-TDMA. For D-TDMA,
each user is served sequentially with different RIS reflection coefficients to ensure its channel quality is
maximized by exploiting the dynamic RIS configuration. Fig. 3 illustrates for different user rate weights
that an asymmetric RIS deployment strategy is preferable for NOMA, while a symmetric RIS deployment
strategy is superior for both FDMA and TDMA regardless of the type of RIS configuration. This
phenomenon can be explained as follows. An asymmetric RIS deployment strategy leads to pronounced
 8

channel differences between the users, such that NOMA is capable of achieving a significant performance
enhancement. In contrast, similar channel conditions are preferable for FDMA and TDMA, which results
in a symmetric RIS deployment.
The joint RIS deployment and MA design still holds many open research problems. For example,
given that the users generally move in the network, dynamically modifying the RIS deployment location
(e.g., by mounting the RIS on unmanned aerial vehicles (UAVs) and unmanned ground vehicles (UGVs))
and selecting the desired MA and RIS configurations can be a promising solution for satisfying the
users’ time-variant QoS requirements. However, conventional standard optimization methods may be
ineffective for handling the resulting complicated optimization problems. In this context, the application
of advanced mathematical tools, such as machine learning (ML) employing a long-term performance
metric, is expected to be beneficial.

IV. RIS A IDED M ULTIPLE -A NTENNA NOMA N ETWORKS : J OINT B EAMFORMER D ESIGN

In this section, we consider RIS aided multiple-antenna NOMA networks, where multiple antennas are
equipped at transmitters and receivers to further enhance the spectral efficiency by exploiting the spatial
DoFs. Depending on whether one beamformer is used to serve one user or multiple users, we broadly
classify the current multiple-antenna NOMA transmission into two categories, namely beamformer based
strategies and cluster based strategies [12]. For each strategy, we propose novel joint beamformer designs
for the considered RIS aided multiple-antenna NOMA networks.

A. Beamformer Based RIS-NOMA: a Novel Equivalent Reconfigurable Channel Inspired Design

RIS
R

RIS SIC

User m signal
Subtract user m’s
detection with
signal
Rnom
Cascaded BS-
Equivalent RIS-user channel
Transformation User n signal
BS Direct BS-user
Direct links detection with
channel Rnon
User n
BS
User m signal
User n detection with
User m Rmom
Equivalent
User m
reconfigurable channels

Fig. 4. Beamformer based strategy for RIS aided multiple-antenna NOMA networks: a novel equivalent reconfigurable channel
inspired design.
 9

The introduction of RISs poses new challenges for beamformer based multiple-antenna NOMA strate-
gies as the channel qualities of the users are determined by both the direct links and the reflection links,
as shown on the left hand side of Fig. 4. In order to address this issue, we propose a novel equivalent
reconfigurable channel inspired design, where the equivalent channel gains facilitated by the direct links
and the reflection links are exploited for joint beamformer design, as shown on the right hand side of
Fig. 4. More particularly, let us consider a two-user setup, where the users are indexed by m and n.
The equivalent reconfigurable channels between the BS and each user result from the combination of the
cascaded BS-RIS-user channels reconfigured by the RIS and the direct BS-user channel, as illustrated
in Fig. 4. Let User n and User m be the strong and weak NOMA users, respectively. User m directly
decodes its own signal received through the equivalent reconfigurable channel by treating the signal of
User n as interference at an achievable rate of Rm→m . In contrast, User n will first decode the signal of
User m at a decoding rate of Rn→m , before decoding its own signal at an achievable rate of Rn→n with
User m’s signal subtracted. Note that, for a given user decoding order, SIC can be successful carried
out at User n provided that its decoding rate is no less than the achievable rate at User m. Compared
to conventional networks, the condition of successful SIC in RIS-NOMA networks depends on both the
active beamformer at the BS and the passive beamformer at the RIS, which causes the joint beamformer
design to be highly coupled and non-trivial to solve.
The proposed equivalent reconfigurable channel inspired design illustrated in Fig. 4 has the following
advantages for facilitating joint beamformer design. On the one hand, for a given passive beamformer at
the RIS, we can map the RIS-NOMA network to a conventional network with the equivalent reconfigurable
channel. Known efficient solutions for conventional NOMA networks can then be exploited for active
beamformer design at the BS for beamformer based RIS-NOMA strategies. On the other hand, it has
been shown that under some special conditions (e.g., quasi-degraded channels [12]), NOMA is capable
of achieving the same performance as dirty paper coding (DPC). Based on this important insight, low
complexity or even closed-form solutions for the RIS passive beamformer design can be obtained by
exploiting the special conditions, where NOMA can achieve high performance, as design objectives for
the equivalent reconfigurable channel.
Nevertheless, we note that the joint optimal decoding order and beamformer design is still challenging,
especially for the case when the RIS can be dynamically adjusted, and requires further research. Addi-
tionally, in practice, RISs employ finite resolution phase shifters [15], and for passive beamformer design
with discrete phase shifts, efficient algorithms have to be developed and sophisticated mathematical tools
have to be invoked.
 10

B. Cluster Based RIS-NOMA: Centralized and Distributed Design

The key idea behind cluster based RIS-NOMA is to partition the users into several different clusters
and serve the users in each cluster with one common beamformer. Conventionally, the MIMO NOMA
channel can be decomposed into multiple single-input single-output (SISO) NOMA channels, by adopting
transmit beamforming at the BS and zero-forcing detection at each multiple-antenna user. However, such
a decomposition relies on specific relationships between the number of equipped antennas at the BS
and the number of users [12], which limits the practicality of this cluster based strategy. Fortunately,
the deployment of RISs can relax the aforementioned specific relationships since RISs are capable of
providing additional passive array gains. In other words, we can shift part of the signal processing tasks
from the BS and users to the RIS, by configuring it to carry out inter-cluster interference cancelation or
intra-cluster signal enhancement. Inspired by this insight, we propose two novel designs for cluster based
RIS-NOMA, namely 1) a centralized RIS enabled design; and 2) a distributed RIS enhanced design. In
the following, we will compare the advantages and disadvantages of both designs.

Centralized RIS

BS

User 1 U 1
User
ĂĂ
Ă Ă ĂĂ
Ă Ă
User 2 User L1 User 2 User L2

ĂĂ ĂĂ
Userr 1
U
ĂĂ
User 2 User LM

Fig. 5. Cluster based strategy for RIS aided multiple-antenna NOMA networks: a centralized RIS enabled design.

1) Centralized RIS Enabled Design: A possible implementation of the centralized RIS enabled design
is illustrated Fig. 5, where the users are partitioned into several clusters enabled by the passive beamformer
at one centrally deployed RIS. The proposed centralized design for the cluster based RIS-NOMA strategy
is appealing for the case when there are no direct links between the BS and the users or when the direct
links are highly correlated (e.g., in mmWave communications). In these scenarios, it is impossible to carry
out user clustering through active beamforming at the BS, and the additional passive beamformer provided
by the RIS comes to the rescue. Nevertheless, in the centralized design, an extremely large number of RIS
reflection elements may be required to realize a high passive array gain for user clustering. Furthermore,
 11

the passive beamformer design requires the accurate instantaneous channel state information (CSI) of the
entire network, which is non-trivial to obtain, especially when the numbers of users and reflection elements
are large. Therefore, robust beamformer design for imperfect CIS is a promising research direction.

Distributed Distributed
RIS m RIS 1

User Lm ĂĂ
ĂĂ
User L1
User 2 User 2
User 1 User 1

ĂĂ
Distributed ĂĂ
Distributed
RIS M
RIS 2
BS

User 1 User 1
ĂĂ User 2 User 2 ĂĂ

User LM User L2

Fig. 6. Cluster based strategy for RIS aided multiple-antenna NOMA networks: a distributed RIS enhanced design.

2) Distributed RIS Enhanced Design: In this design, multiple distributed RISs are deployed close to
different NOMA user clusters which have been already separated by the active beamformer at the BS, as
illustrated in Fig. 6. More particularly, each RIS aims to enhance the performance of one specific user
cluster within its own local coverage. The distributed RIS enhanced design for the cluster based RIS-
NOMA strategy is suitable when the users are widely distributed. In this case, the multiple distributed
RISs can be deployed as ‘add-on’ promoters for existing conventional NOMA networks. An advantage
of this setup is that the impact of a given RIS on other unintended clusters is small due to the relatively
large distances. Due to this property, the passive beamformer design at each RIS depends mainly on
the CSI of its dedicated cluster of users. However, the BS in the proposed distributed design has to
simultaneously exchange information with multiple RISs, which introduces new challenges for the RIS
control link and for protocol design especially when employing dynamic RIS configuration.

V. C ONCLUSIONS AND R ESEARCH O PPORTUNITIES

In this article, RIS aided multi-user networks have been considered with a particular focus on the appli-
cation of NOMA. Considering static and dynamic RIS configurations, the information-theoretic capacity
limits and optimal RIS deployment strategies were investigated for RIS aided single-antenna NOMA
networks. Dynamic RIS configuration is necessary to maximize the capacity gain and an asymmetric
RIS deployment is preferable for NOMA to facilitate significant channel differences between the users.
For RIS aided multiple-antenna NOMA networks, efficient joint active and passive beamformer designs
for different application scenarios were proposed for beamformer based and cluster based strategies. For
 12

beamformer based strategies, a novel equivalent reconfigurable channel inspired design was advocated,
which facilitates the joint active and passive beamformer design by exploiting insights from conventional
NOMA networks. For cluster based strategies, a centralized RIS enabled design and a distributed RIS
enhanced design were developed, and their respective advantages and disadvantages were elaborated.
Nevertheless, there are still many research opportunities for the design of future RIS-NOMA networks.
Three particularly interesting problems are as follows:

• Evaluation of Impact of Node Distribution Using Stochastic Geometry: As a powerful math-

ematical tool, stochastic geometry is capable of capturing the spatial randomness of large-scale
RIS-NOMA networks to provide practical RIS deployment guidelines. Since the reflection links
depend on the spatial locations of the BSs, RISs, and users, performance evaluation is a difficult
task. Additionally, multi-cell RIS-NOMA networks necessitate the redesign of the user association
due to the flexible decoding order facilitated by RISs.
• Optimal Decoding Order and Beamformer Design: The reconfigurable channels facilitated by
RISs enable a flexible decoding order design for RIS-NOMA networks. For RIS aided single-antenna
NOMA networks, the decoding order depends on the users’ channel qualities, which are affected
by the RIS passive beamformer. For RIS aided multiple-antenna NOMA networks, the problem
becomes rather complicated since additional constraints imposed by the SIC decoding rate need to
be satisfied, which causes the decoding order and the active and passive beamformers to be highly
coupled. Therefore, determining the optimal decoding order and beamformer design for RIS-NOMA
networks is a challenging task, which requires further research.
• Efficient Machine Learning Algorithm Design: In future ML-empowered RIS-NOMA networks
for 5G and beyond, the design goals will be more ambitious than simply optimizing the network’s
capacity or delay, i.e., a single metric using single-component optimization. ML algorithm design
is likely to probe further first by striking a trade-off between capacity, delay, and power, followed
by finding all Pareto-optimal solutions for the resulting challenging multi-objective optimization
problems.

R EFERENCES

[1] M. D. Renzo et al., “Smart radio environments empowered by reconfigurable intelligent surfaces: How it works, state of
research, and road ahead,” IEEE J. Sel. Areas Commun., vol. 38, no. 11, 2020, pp. 2450–2525.
[2] Y. Liang et al., “Large intelligent surface/antennas (LISA): Making reflective radios smart,” J. Commun. Inf. Networks, vol.
4, no. 2, 2019, pp. 40–50.
[3] Q. Wu and R. Zhang, “Towards smart and reconfigurable environment: Intelligent reflecting surface aided wireless network,”
IEEE Commun. Mag., vol. 58, no. 1, 2020, pp. 106–112.
 13

[4] C. Huang et al., “Holographic MIMO surfaces for 6G wireless networks: Opportunities, challenges, and trends,” IEEE
Wireless Commun., vol. 27, no. 5, 2020, pp. 118–125.
[5] M. Di Renzo et al., “Smart radio environments empowered by AI reconfigurable meta-surfaces: An idea whose time has
come,” EURASIP J. Wireless Commun., 2019.
[6] E. Basar et al., “Wireless communications through reconfigurable intelligent surfaces,” IEEE Access, vol. 7, 2019, pp.
116 753–116 773.
[7] C. Huang et al., “Reconfigurable intelligent surfaces for energy efficiency in wireless communication,” IEEE Trans. Wireless
Commun., vol. 18, no. 8, 2019, pp. 4157–4170.
[8] J. Yuan et al., “Intelligent reflecting surface-assisted cognitive radio system,” IEEE Trans. Commun., Early Access, 2020.
[9] P. Mursia et al., “RISMA: Reconfigurable intelligent surfaces enabling beamforming for IoT massive access,” IEEE J. Sel.
Areas Commun., Early Access, 2020.
[10] E. Basar et al., “Reconfigurable intelligent surface-based index modulation: a new beyond MIMO paradigm for 6G,” IEEE
Trans. Commun., vol. 68, no. 5, 2020, pp. 3187–3196.
[11] Y. Liu et al., “Nonorthogonal multiple access for 5G and beyond,” Proc. of the IEEE, vol. 105, no. 12, 2017, pp. 2347–2381.
[12] Y. Liu et al., “Multiple-antenna-assisted non-orthogonal multiple access,” IEEE Wireless Commun., vol. 25, no. 2, 2018,
pp. 17–23.
[13] X. Mu et al., “Capacity and optimal resource allocation for IRS-assisted multi-user communication systems,”
https://arxiv.org/abs/2001.03913, 2020.
[14] X. Mu et al., “Joint deployment and multiple access design for intelligent reflecting surface assisted networks,”
https://arxiv.org/abs/2005.11544, 2020.
[15] Z. Ding, R. Schober, and H. V. Poor, “On the impact of phase shifting designs on IRS-NOMA,” IEEE Wireless Commun.
Lett., vol. 9, no. 10, 2020, pp. 1596–1600.