Reconfigurable Intelligent Surface Assisted NOMA
Empowered Integrated Sensing and Communication
Jiakuo Zuo∗ and Yuanwei Liu†
∗ School of Internet of Things, Nanjing University of Posts and Telecommunications, Nanjing 210003, China.
† School of Electronic Engineering and Computer Science, Queen Mary University of London, London E1 4NS, U.K.
E-mail: zuojiakuo@njupt.edu.cn, yuanwei.liu@qmul.ac.uk.
Abstract—This paper exploits the potential of reconfigurable
intelligent surface (RIS) to improve radar sensing in a non-
arXiv:2208.04786v4 [cs.IT] 28 Sep 2022
orthogonal multiple access (NOMA) empowered integrated sens-
ing and communication (NOMA-ISAC) network. The objective
is to maximize the minimum radar beampattern gain by jointly
optimizing the active beamforming, power allocation coefficients
and passive beamforming. To tackle the formulated non-convex
problem, we propose an efficient joint optimization algorithm
by invoking alternating optimization, successive convex approx-
imation (SCA) and sequential rank-one constraint relaxation
(SRCR) algorithm. Numerical results show that the proposed RIS
assisted NOMA-ISAC system, with the aid of the proposed scheme,
outperforms the RIS assisted ISAC system without NOMA.
Index Terms—Beamforming optimization, integrated sensing
and communication, non-orthogonal multiple access, reconfig-
urable intelligent surface.
I. I NTRODUCTION
In conventional wireless communication, communication
quality of service (QoS) is the main indicator. However, in the
upcoming beyond fifth generation (B5G) and sixth generation
(6G) wireless networks, sensing service will play a more
important role than ever before [1]. The goal of future wireless
communication is to provide services based on both sensing and
communication functionalities. Toward this trend, integrated
sensing and communication (ISAC) has emerged and attracted
growing attention in both academia and industries.
To boost the communication and sensing capabilities, various
key technologies can be exploited in ISAC, such as: millimeter
wave [2], terahertz [3] and massive multiple-input multiple-
output (MIMO) [4]. However, the communication users will
suffer from severe interference when the system is overloaded,
which is not considered in these works. Non-orthogonal mul-
tiple access (NOMA) is a good solution as it can multi-
plex communication users and mitigate the inter-user interfer-
ence. The authors in [5] first proposed a NOMA empowered
ISAC (NOMA-ISAC) system, where the superimposed NOMA
communication signal was utilized to perform radar sensing.
Considering that the sensing signal can also be information-
bearing, the authors in [6] used multiple beams of the sensing
signal to deliver extra multicast information and detect radar
targets simultaneously. The authors in [7] proposed a novel
NOMA aided joint radar and multicast-unicast communication
system, where the base station (BS) transmits superimposed
This work was supported by Key Laboratory of Universal Wireless
Communications (BUPT), Ministry of Education, P.R.China under Grant
KFKT-2022105 and China Postdoctoral Science Foundation under Grant
2021M693699.
multicast and unicast messages to the radar users and com-
munication users, while detecting the radar user targets. Since
the bandwidth has been occupied by different applications,
the ISAC networks can not utilize all the spectrum for both
radar detection and wireless communication. To overcome this
difficulty, the authors in [8] proposed an uplink NOMA-assisted
semi-ISAC system, where the total bandwidth was divided into
ISAC bandwidth and communication-only bandwidth.
In conventional ISAC systems, radar sensing relies on line-
of-sight (LoS) links between the BS and radar targets. How-
ever, in practical scenarios, the radar targets are likely to be
distributed in the non-LoS (NLoS) region of the BS. As a
remedy, reconfigurable intelligent surface (RIS) can be inte-
grated into ISAC systems. RIS comprises a large number of
passive elements and can reconfigure the signal propagation by
adjusting the phase shifts. Also, the RIS can provide virtual
LoS links to the ISAC systems, potentially resulting in better
sensing performance [9].
Inspired by the aforementioned discussion, we investigate
the RIS assisted NOMA-ISAC (RIS-NOMA-ISAC) system.
To the best of our knowledge, the joint optimization design
for the considered system has not been studied yet. In this
paper, the communication users and radar targets are assumed
to be at the NLoS areas of the BS. The RIS is deployed to
create reflection and virtual LoS links for communication and
target sensing. A novel alternating algorithm is proposed to
optimize the active beamforming, power allocation coefficients
and passive beamforming.
Notations: C M×1 denotes a complex vector of size M, diag(x)
denotes a diagonal matrix whose diagonal elements are the
corresponding elements in vector x. The (m, n)-th element
of matrix X is denoted as [X]m,n . xH and XH denote the
conjugate transpose of vector x and matrix X, respectively.
The notations Tr(X) and rank(X) denote

the trace and rank
of matrix X, respectively. CN 0, σ 2 represents a random
complex variable following the distribution of zero mean and
σ 2 variance.
II. S YSTEM M ODEL AND P ROBLEM F ORMULATION
As shown in Fig. 1, the considered RIS-NOMA-ISAC system
consists of a dual-functional base station (BS) equipped with
NT antennas, 2K single-antenna users, an uniform linear
array (ULA)-RIS with M reflecting elements and L radar
targets. We assume that the direct links from the BS to the
communication users and radar targets are blocked, which is
 order in each cluster. Particularly, user U (k, n) first decodes
the signal of user U (k, f ) and then subtracts this signal from
its observation to decode its own information. Therefore, the
achievable rate for user U (k, n) to decode the signal of user
U (k, f ) is given by  
2
H
 a k,f g k,n ΘGw k 
Rk,f →n = log2 1 + iner iter + σ 2  , (3)
Ik,n + Ik,n
2
iner H iter
where Ik,n = ak,n gk,n ΘGwk and Ik,n =
Fig. 1: System model of the RIS-NOMA-ISAC system PK H
2
e
k6=k gk,n ΘGwek .
a typical application scenario when RISs are needed or used. If the above decoding is successful, then user U (k, n)
Therefore, in our system, we assume that all the communication removes the signal sk,f from yk,n to further decode its own
users and radar targets are at the NLoS areas of the BS. We signal sk,n . The corresponding individual achievable rate is
carefully deploy the RIS to ensure that all the targets are in given by  2

LoS areas of the RIS. H
 a k,n g k,n ΘGw k 
To improve the spectral efficiency and reduce the system
Rk,n = log2 1 + iter + σ 2 . (4)
load, we assume that the 2K users are grouped into K clusters Ik,n
by employing user clustering techniques. As a result, there Accordingly, user U (k, f ) directly decodes its own signal by
are two types of users in each cluster, namely the RIS- treating the signal of user U (k, n) as interference. Therefore,
near user (RNU) and the RIS-far user (RFU). Generally, the the achievable rate for user U (k, f ) to decode its own signal
RNUs are much closer to the RIS than the RFUs. In each can be expressed as  
2
cluster, the NOMA protocol is applied during transmission. a g H
ΘGw
 k,f k,f k 
The cluster and user sets are denoted by K = {1, · · · , K} Rk,f →f = log2 1 + iner , (5)
iter
Ik,f + Ik,f + σ 2
and U = {1, · · · , 2K}, respectively. Moreover, denote by Uk
the set of users in cluster k, where U = ∪k∈ Uk , Uk ∩ Uk = iner H
2
iter
⊘ (k, k ∈ K, k 6= k). For notation simplicity, we denote the where Ik,f = ak,n gk,f ΘGwk and Ik,f =
RNU and RFU in the k-th cluster as user U (k, n) and user PK H
2
e
k6=k gk,f ΘGwek .
U (k, f ), respectively.
As a result, the individual achievable rate of user U (k, f ) is
A. Communication Model given by
The superimposed communication signal transmitted by the
Rk,f = min {Rk,f →n , Rk,f →f } . (6)
BS is given by B. Radar Detection Model
XK
√ √
Since all the potential targets are in the NLoS ares of the
x= wk ak,n sk,n + ak,f sk,f , (1)
k=1
BS, NLoS links from the BS can be exploited to perform radar
where wk ∈ C NT ×1 is the active beamforming vector for target sensing. However, we can utilize the reflection-LoS links
the k-th cluster, sk,i denotes the communication
 signal to created by the RIS to complete the sensing tasks. We consider
H
be sent to user U (k, i) with E sk,i sk,i = 1, ak,i is the using the RIS’s beampattern gain as the sensing performance
metric. The reflected signal at the RIS can be expressed
! as
corresponding power allocation coefficient, i ∈ {n, f }, k ∈ K. K
X

√ √
Let G ∈ C M×NT and gk,i ∈ C M×1 be the channel coefficients x = ΘG wk ak,n sk,n + ak,f sk,f . (7)
of the communication links BS → RIS and RIS → U (k, i), k=1
Therefore, the corresponding covariance matrix is given by
respectively. With the help of RIS, the signal received at user K
!

X
U (k, i) can be mathematically expressed as Rx = E xxH = ΘG wk wkH GH ΘH . (8)
K
H

X √ √
k=1
yk,i = gk,i ΘG wk ak,n sk,n + ak,f sk,f + zk,i , In our considered system, the communication signal is used
k=1 to perform radar target sensing, thus the RIS’s beampattern gain
(2) with respect to the q-th interested angle, i.e,!θqTg , is given by
whereh Θ = diag (v) is the i RIS’s diagonal phase shift matrix, XK
RIS RIS RIS


v = ejθ1 ejθ2 · · · ejθM is the passive beamforming vector PθqTg (wk , v) = aH θqTg ΘG wk wkH GH ΘH a θqTg ,
RIS k=1
of the RIS with θm ∈ [0, 2π) denoting the phase shift
of the (9)
m-th reflecting element, m ∈ M, and zk,i ∼ CN 0, σ 2 is the h iT
j 2πd sin(θ) j 2πd (M−1) sin(θ)
additive white Gaussian noise (AWGN), M = {1, 2, · · · , M } where a (θ) = 1, e λ ,··· ,e λ is the
is the reflecting elements set. steering vector at the RIS with angle θ, d denotes the an-
In the proposed RIS-NOMA-ISAC system, successive inter- tenna spacing, λ denotes the n carrier wavelength, o q ∈ Q ,
ference cancellation (SIC) is applied for the users to decode {1, 2, · · · , Q}, and Qθ = θ1Tg , θ2Tg , · · · , θQTg
is the set of
their signals in the same cluster. We assume a fixed decoding interested sensing angles.
C. Maximize The Minimum Beampattern Gain smooth one. Finally, the original problem (10) can be recast
The objective of this paper is to maximize the minimum equivalently as follows:
beampattern gain towards Q interested angles by jointly op- max χ, (16a)
χ>0,ak ,Wk <0,V<0
timizing the active beamforming and power allocation co- " K
! #
X
efficients at the BS, and the passive beamforming at the s.t. Tr VΥq Wk ΥH q > χ, (16b)
RIS. Accordingly, the optimization problem is formulated as k=1
K
X
max minPθqTg (wk , v) , (10a)
ak ,wk ,v q∈Q Tr (Wk ) 6 Pmax , (16c)
min
s.t. Rk,i > Rk,i , (10b) k=1

K
[V]m,m = 1, (16d)
X
kwk k2 6 Pmax , (10c) rank (Wk ) = 1, (16e)
k=1 rank (V) = 1, (16f)
ak,n + ak,f = 1, ak,i ∈ (0, 1) , (10d)
(10d), (13), (14), (15), (16g)
RIS
θm ∈ [0, 2π) , (10e) where i ∈ {n, f } , k ∈ K, m ∈ M, q ∈ Q.
min
where Rk,i is the minimum QoS requirement, Pmax is the
T
maximum transmit power at the BS, ak = [ak,n ak,f ] , i ∈ A. Joint Active Beamforming And Power Allocation Coeffi-
{n, f } , k ∈ K, m ∈ M. cients Optimization
In this subsection, we aim to optimize the active beamform-
III. P ROPOSED S OLUTION ing and power allocation coefficients by solving problem (16)
In this section, we elaborate on how to solve problem (10). with given passive beamforming matrix V. In particular, we can
We first transform problem (10) into a more tractable form. To obtain the optimal {Wk } and {ak } by solving the following
facilitate the design, we define Wk = wk wkH , where Wk < 0 optimization problem:
and rank (Wk ) = 1, k ∈ K. Similarly, we define V = vvH , max χ, (17a)
χ>0,ak ,Wk <0
which satisfies V < 0 and rank (V) = 1. Then, the beampattern s.t. (10d), (13), (14), (15), (16b), (16c), (16e). (17b)
gain in (9) can be rewritten
" as: ! # The difficulty to solve problem (17) is the nonconvex
K
X
PθqTg (Wk , V) = Tr VΥq Wk Υq , H
(11) constraints (13), (14), (15), and (16e). We first handle con-
n  o K=1 straint (13) by introducing a new slack variable ηk , then we
where Υq = diag aH θq
Tg
G. have
(
H
2 ak,n Tr (Wk Hk,n ) > ηk2 , (18)
Moreover, the quadratic terms gk,i ΘGwk and 2 min iter 2


2 ηk > rk,n Ik,n + σ , (19)
H
gk,i ΘGwek in (3), (4) and (5) can be rewritten as: P

iter K
 where Hk,n = ΓH k,n VΓk,n and Ik,n = k6=k Tr We
e k Hk,n .
2   Then, by applying the Schur complement theory [10], (18)

 gH ΘGwk = Tr VΓk,i Wk ΓH ,
k,i k,i can be expressed in linear matrix inequality form
2   (12)

 gk,i ΘGwe = Tr VΓk,i We ΓH
H  
k k k,i , ak,n ηk
< 0, ∃ηk > 0 (20)
n o ηk Tr (Wk Hk,n )
H
where Γk,i = diag gk,i G, k 6= k̃.
Furthermore, by using the successive convex approximation
By exploiting the above definitions, constraints in (10b) can (SCA) approach based on First-order Taylor approximation,
be equivalently reformulated as follows: (19) can be approximated as followed:

min iter 2



ak,n Tr VΓk,n Wk ΓH k,n > rk,n Ik,n + σ , (13) ηek2 + 2ηek (ηk − ηek ) > rk,n
min iter
Ik,n + σ2 , (21)

min iner iter 2


ak,f Tr VΓk,n Wk ΓH k,n > rk,f Ik,n + Ik,n + σ , (14) (t ) (t )
where ηek is a fixed point and can be updated by ηek 1 = ηk 1 ,

min iner iter 2

t1 is the iteration index.
ak,f Tr VΓk,f Wk ΓH k,f > rk,f Ik,f + Ik,f + σ , (15) Next, we tackle the constraints (14) and (15). Substituting
  ak,f = 1 − ak,n into (14) and (15), we have:
iner
where Ik,n = ak,n Tr VΓk,n Wk ΓH k,n ,  
  min Tr(Wk Hk,n ) iter 2
iner
Ik,f = ak,n Tr VΓk,f Wk ΓH and Ik,i iter
= rk,f min
rk,f
− Ik,n − σ
k,f
PK   min min
> ak,n Tr (Wk Hk,n ) , (22)
H min Rk,i
− 1, i ∈ {n, f } , rk,f + 1
k6=k Tr VΓk,i We
e k Γk,i , rk,i = 2
k ∈ K.  
min Tr(Wk Hk,f ) iter
Furthermore, we introduce an auxiliary variable χ to trans- rk,f min
rk,f
− Ik,f − σ2
min + 1
> ak,n Tr (Wk Hk,f ) , (23)
form the nonsmooth objective function in problem (10) into a rk,f
 PK

where Hk,f = ΓH iter Algorithm 1 Proposed joint active beamforming and power
k,f VΓk,f and Ik,f = k6=k Tr We
e k Hk,f .
We note that the functions ak,n Tr (Wk Hk,n ) and allocation coefficients optimization algorithm
ak,n Tr (Wk Hk,f ) in (22) and (23) are nonconvex. To deal 1:
(0) (0) (0)
Initialize βk,1 , βk,2 , ηek , k ∈ K, and set t1 = 0.
with these non-convexity, we adopt the arithmetic-geometric 2: repeat
mean inequality [11]. Specifically, ak,n Tr (Wk Hk,n ) and 3: t1 = t1 + 1;
ak,n Tr (Wk Hk,f ) can be approximated as 4:
(t )
update Wk 1 , ak,n
(t1 ) (t )
and ηk 1 by solving problem (28)
2
βk,1 a2k,n (Tr (Wk Hk,n )) (t −1) (t −1) (t −1)
with given βk,11 , βk,21 , ηek 1 ;
ak,n Tr (Wk Hk,n ) 6 + , Tk,1 , (t )

2 2βk,1 (t ) (t ) (t ) Tr Wk 1 Hk,n (t )
(24) 5: update ηek 1 = ηk 1 , βk,11 = (t )
1
and βk,21 =
ak,n
βk,2 a2k,n (Tr (Wk Hk,f ))2 
(t )
Tr Wk 1 Hk,f

ak,n Tr (Wk Hk,f ) 6 + , Tk,2 , ;
2 2βk,2 (t )
1
ak,n
(25) 6: until the objective value of problem (28) converge.
where βk,1 and βk,2 are fixed points. The equality in (24) 7: Output: Wk and ak , k ∈ K.
Tr(Wk Hk,n )
and (25) will always hold if βk,1 = ak,n and βk,2 =
Tr(Wk Hk,f )
ak,n .
Based on the aforementioned transformations and approxi- eigenvalue of V(t2 ) , ε(t2 ) ∈ [0, 1] is a relaxation parameter
mations, constraints given in (22) and (23) can be approximated in the t2 -th iteration. We can increase ε(t2 ) from 0 to 1
as follows   sequentially via iterations to gradually approach a rank-one
min Tr(Wk Hk,n ) iter solution. After each iteration, the relaxation parameter can be
rk,f min
rk,f
− Ik,n − σ2
min + 1
> Tk,1 , (26) updated as
!
 rk,f  λmax V(t2 +1)
(t2 +1)
+ρ (t +1)
min Tr(Wk Hk,f )
rk,f iter
− Ik,f − σ2 ε ←− min 1, 2
, (31)
min
rk,f Tr V(t2 +1)
> Tk,2 , (27)

min + 1
rk,f where λmax V(t2 ) is the largest eigenvalue of V(t2 ) and ρ(t2 )
denotes the step size.
Finally, let us turn our attention to the rank-one con-
As a result, in the t2 -th iteration, the optimization problem
straint (16e). To address this issue, we exploit the semidefinite
that needs to be solved is given as follows
relaxation (SDR) technique by removing the rank-one con-
max χ, (32a)
straints from the problem formulation. As a result, the relaxed χ>0,V<0
problem is give as s.t. (13), (14), (15), (16b), (16d), (30). (32b)
max χ, (28a) Problem (32) is a semidefinite program (SDP) problem and
χ>0,0<ak,n <1,Wk <0
can be solved by the CVX tool [13]. The procedure for
s.t. (16b), (16c), (20), (21), (26), (27). (28b) optimizing passive beamforming is sketched in Algorithm 2.
In the following theorem, we will verify the tightness of the
SDR problem. Algorithm 2 Proposed passive beamforming optimization al-
Theorem 1: If the relaxed problem (28) is feasible, then the gorithm
solutions {Wk } obtained by solving the problem (28) always
satisfy rank (Wk ) 6 1, k ∈ K. 1: Initialize V(0) and ρ(0) . Set ε(t2 ) = 0 and t2 = 0.
Proof : Similar proof can be found in [11]. 2: repeat 
Theorem 1 represents the fact that we can obtain the rank-one 3: Solve problem (32) with ε(t2 ) , V(t2 ) to obtain V∗ ;
solutions of problem (17) by solving the convex problem (28). 4: if problem (32) is solvable
To facilitate the understanding of the proposed algorithm, we 5: Update V(t2 +1) = V∗ ;
summarize it in Algorithm 1. 6: Update ρ(t2 +1) = ρ(0) ;
B. Passive Beamforming Optimization 7: else
8: Update V(t2 +1) = V(t2 ) ;
For any given {Wk } and {ak }, the passive beamforming (t2 )
9: update ρ(t2 +1) = ρ 2 ;
optimization problem is given by
10: end
max χ, (29a)
χ>0,V<0 11: Update ε(t2 +1) via (31);
s.t. (13), (14), (15), (16b), (16d), (16f). (29b) 12: Update t2 = t2 + 1;
Tr(V(t2 ) )
Now, the remaining non-convexity in problem (29) lies in 13: until λ is below a predefined threshold and the
max (V 2 )
(t )
the rank-one constraint (16f). According to the sequential rank- objective value of problem (32) converges.
one constraint relaxation (SRCR) algorithm [12], the constraint 14: Output: V.
rank (V) = 1 can betransformed
 equivalently
 as:
H (t2 ) (t2 ) (t2 )
emax V Vemax V > ε Tr (V) , (30) C. Proposed Algorithm, Complexity and Convergence
(t2 )
where V
is the obtained solution in the t2 -th iteration, To facilitate the understanding of the proposed
emax V(t2 ) is the eigenvector corresponding to the maximum algorithm for solving problem (16), we summarize it in
 1

0.9

0.8
Desired

normalized beampattern gain

0.7 M=36

0.6

0.5 M=30

0.4

0.3

0.2

0.1

0
-90 -80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90
Angle (degree)

Fig. 2: The simulated RIS-NOMA-ISAC system geometry Fig. 3: Obtained beampattern gain with NT = 9
Algorithm 3. The convergence of Algorithm 3 is analyzed
as follows. Algorithm 1 and Algorithm 2 converge to a
KKT stationary solution of problem (17) and problem (29),
respectively, which has been proved in [11] and [12]. In
addition, the objective value of problem (16) is non-decreasing
after each iteration and the radar beampattern gain is upper
bounded. Therefore, the proposed algorithm is guaranteed to
converge. The complexities
  of Algorithm 1 and Algorithm2
max 4√ 1
are O T1 max {Ncont , NT } NT log ̟1
  √ 
and O T2 max
max {Ncont + 1, M }4 M log ̟12 ,
respectively, where ̟1 and ̟2 are the solution Fig. 4: Illumination power of the RIS with NT = 9 and
accuracies, T1max and T2max are the number of iterations, M = 30.
 = 3K
Ncont  +Q+ 1. The whole complexity of Algorithm
 3is and 45o , and the corresponding radius are set to 90 m, 90 m
max 4√ 1
T1 max {Ncont , NT } NT log ̟1 + and 80 m, respectively. With the angles of sensing targets, the
O T3max  
4√
 ,
desired beampattern can be defined as
T2max
max {Ncont , M } M log ̟12

where T3max is the number of iterations for Algorithm 3. 1, θT − ∆θ ∆θ
2 6 θ 6 θT + 2 , θT ∈ {−45 , 0 , 45 } ,
o o o
P (θ) =
0, otherwise,
Algorithm 3 Proposed joint optimization algorithm (33)
1: Initialize V(0) and set t3 = 0. where ∆θ = 6o is the desired beam  width, θ is
2: repeat the element of angle grid − π2 : 100 π
: π2 , and the in-
3: t3 = t3 + 1; terested sensing angle set can be defined as Qθ =

4:
(t ) (t )
update Wk 3 and ak 3 by Algorithm 1 with V(t3 −1) ; θ|θT − ∆θ ∆θ o o
2 6 θ 6 θT + 2 , θT ∈ {−45 , 0 , 45 } .
o

5:
(t ) (t )
update V(t3 ) by Algorithm 2 with Wk 3 and ak 3 ;
6: until the objective value of problem (16) converges.
7: Output: Wk , ak , and V, k ∈ K.
IV. S IMULATION R ESULTS
In this section, the performances of the proposed algorithm
for RIS-NOMA-ISAC system are evaluated through numerical
simulations. The simulated RIS-NOMA-ISAC system geometry
is shown in Fig. 2. We assume that there are three clusters
and three radar targets. The RIS is located in the origin of the
coordinate, while the BS is located at (−40, 10) meter (m). In
each cluster, the RNU and the RFU are randomly distributed on
the half circles centered at (0, 0) m with radius of rk,n and rk,f ,
respectively, where rk,n ∈ [20, 25] m and rk,f ∈ [80, 85] m.
Let θk,i denote the angle from the RIS to user U (k, i) and
we further assume that {θ1,i } in cluster 1, {θ2,i } in cluster 2,
and {θ3,i } in cluster 3, are randomly distributed in the angle
ranges (−30o , −20o ], (20o , 30o ], and (60o , 70o ], respectively,
i ∈ {n, f }. Without loss of generality, let θk,n = θk,f , k ∈ K.
The angle from the RIS to the three targets are set to −45o , 0o
All the channels follow the Rician fading, which can be
modeled as in [14]. The path loss at a reference distance of
one meter is set to −30 dB, the path loss exponents are set
to 2.2, the Rician factors are set to 3, the noise power is set
to −90 dBm, the normalized spacing between two adjacent
antennas(elements) is set as λd = 0.5. The minimum QoS
min
requirement for RNUs and RFUs are set to Rk,n = 0.5
min
bits/s/Hz and Rk,f = 0.1 bits/s/Hz, respectively. The maximum
transmit power is set to 35 dBm.
In Fig. 3, we plot the normalized beampattern gain obtained
over one random channel realization. The desired beampattern
gain is obtained via (33). As illustrated in Fig. 3, the proposed
scheme can achieve the dominant peaks of the beampattern
gain in the angles of interest, i.e., −45o, 0o and 45o . Moreover,
at the target directions, the achievable beampattern gains for
M = 36 are always higher than that for M = 30, which
implies that the performance of the proposed scheme can be
improved by increasing the number of the RIS elements. In
Fig. 4, we show that the normalized illumination power [15]
of the RIS on different angles and locations over one random
 10 -4 10-3
2
RIS-ISAC RIS-ISAC: NT=9
12
RIS-NOMA-ISAC 1.8 RIS-NOMA-ISAC: NT=9
11 RIS-ISAC RIS-ISAC: NT=3
RIS-NOMA-ISAC 1.6

Minimum Beampattern gain

RIS-NOMA-ISAC: NT=3
10
10-4
1.4
9 6.5

Beampattern gain
6
1.2 5.5
8
5
1 4.5
7 Underloaded: N T =9
44 45 46
0.8
6
Overloaded: N T =3 0.6
5

4 0.4

3 0.2

24 28 32 36 -90 -80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90

Number of RIS reflecting elements: M Angle (degree)

Fig. 5: Obtained minimum beampattern gain by different Fig. 6: Obtained beampattern gain by different systems with
schemes M = 36
channel realization. From Fig. 4, we can see that there are three results were presented to demonstrate the superiority of the
brightest region towards the target directions. This is intuitive proposed RIS-NOMA-ISAC system.
since the RIS attempts to steer both the active and passive R EFERENCES
beamforming towards the targets to maximize the minimum [1] J. A. Zhang, M. L. Rahman, K. Wu, X. Huang, Y. J. Guo, S. Chen,
beampattern. and J. Yuan, “Enabling joint communication and radar sensing in mobile
networks—a survey,” IEEE Commun. Surv. Tut., vol. 24, no. 1, pp. 306–
To evaluate the performance of the proposed RIS- 345, Oct. 2022.
NOMA-ISAC system, the RIS assisted conventional ISAC [2] Z. Gao, Z. Wan, D. Zheng, S. Tan, C. Masouros, D. W. K. Ng, and
(RIS-ISAC) system without NOMA is considered. The S. Chen, “Integrated sensing and communication with mmwave massive
mimo: a compressed sampling perspective,” 2022, [Online]. Available:
achievable
 rate of user k can
2
 be obtained by: Rk = https://arxiv.org/abs/2201.05766.
|gkH ΘGwk | [3] C. Chaccour, W. Saad, O. Semiari, M. Bennis, and P. Popovski, “Joint
log2 1 + P2K H 2 , where gk is the channel co-
i6=k |gi ΘGwk | +σ
2 sensing and communication for situational awareness in wireless THz
efficients from the RIS to user k, k ∈ {1, 2, · · · , 2K}. The joint systems,” 2021, [Online]. Available: https://arxiv.org/abs/2111.14044.
[4] L. You, X.-H. Qiang, C. G. Tsinos, F. Liu, W. Wang, X. Gao, and B. E.
active and passive beamforming optimization problem for RIS- Ottersten, “Beam squint-aware integrated sensing and communications for
ISAC system can also be solved by our proposed algorithms. hybrid massive MIMO LEO satellite systems,” 2022, [Online]. Available:
In Fig. 5, we compare the minimum beampattern gain obtained https://arxiv.org/abs/2203.00235.
[5] Z. J. Wang, Y. Liu, X. Mu, Z. Ding, and O. A. Dobre, “NOMA em-
by the RIS-NOMA-ISAC and RIS-ISAC systems versus M . powered integrated sensing and communication,” IEEE Commun. Lett.”,
The results are obtained over 100 random channel realizations. vol. 26, no. 3, pp. 677–681, March 2022.
First, it is observed that the achievable minimum beampat- [6] Z. J. Wang, Y. Liu, X. Mu, and Z. Ding, “NOMA inspired interference
cancellation for integrated sensing and communication,” 2021, [Online].
tern gain monotonically increases with M . This is expected Available: https://arxiv.org/abs/2206.00377.
since more reflecting elements can provide higher possible [7] X. Mu, Y. Liu, L. Guo, J. Lin, and L. H. Hanzo, “NOMA-aided joint
beamforming gain towards the radar targets, thereby increasing radar and multicast-unicast communication systems,” IEEE J. Sel. Areas
Commun., vol. 40, no. 6, pp. 1978–1992, June 2022.
beampattern gain. Second, the proposed RIS-NOMA-ISAC [8] C. Zhang, W. Yi, Y. Liu, and L. H. Hanzo, “Semi-integrated-sensing-and-
system outperforms the RIS-ISAC system under overloaded communication (semi-ISaC): From OMA to NOMA,” 2022, [Online].
and underloaded cases. This is because the RIS-NOMA-ISAC Available: https://arxiv.org/abs/2204.11245.
[9] X. Song, D. Zhao, H. Hua, T. X. Han, X. Yang, and J. Xu, “Joint trans-
system can mitigate the inter-user interference by SIC and mit and reflective beamforming for IRS-assisted integrated sensing and
provide more degrees of freedoms. for radar sensing. However, communication,” 2022 IEEE Wireless Communications and Networking
the RIS-ISAC system cannot well mitigated the inter-user Conference (WCNC), pp. 189–194, 2022.
[10] X. Xie, F. Fang, and Z. Ding, “Joint optimization of beamforming, phase-
interference. Fig. 6 compares the beampattern gain obtained by shifting and power allocation in a multi-cluster IRS-NOMA network,”
different systems. We observe that the beampattern gain of all IEEE Trans. Veh. Technol., vol. 70, no. 8, pp. 7705–7717, Aug. 2021.
schemes have peaks towards target directions and our proposed [11] X. Sun, N. Yang, S. Yan, Z. Ding, D. W. K. Ng, C. Shen, and Z. Zhong,
“Joint beamforming and power allocation in downlink NOMA multiuser
RIS-NOMA-ISAC system has a stronger peak compared with MIMO networks,” IEEE Trans. Wireless Commun., vol. 17, no. 8, pp.
the RIS-ISAC system. On the other hand, the proposed scheme 5367–5381, Aug. 2018.
can obtain higher beampattern gains at the worst angles. [12] P. Cao, J. S. Thompson, and H. V. Poor, “A sequential constraint relax-
ation algorithm for rank-one constrained problems,” 2017 25th European
Signal Processing Conference (EUSIPCO), pp. 1060–1064, 2017.
V. CONCLUSIONS [13] M. Grant and S. Boyd, “CVX: Matlab software for disciplined convex
We have proposed a novel RIS-NOMA-ISAC system, where programming, version 2.1,” http://cvxr.com/cvx, Mar. 2014.
the RIS is deployed to serve the communication users and [14] Y. Li, M. Jiang, Q. Zhang, and J. Qin, “Joint beamforming design
in multi-cluster MISO NOMA reconfigurable intelligent surface-aided
assist radar sensing. A joint optimization problem over ac- downlink communication networks,” IEEE Trans. Commun., vol. 69,
tive beamforming, power allocation coefficients, and passive no. 1, pp. 664–674, Jan. 2021.
beamforming was formulated with the aim of maximizing the [15] R. S. P. Sankar, S. P. Chepuri, and Y. C. Eldar, “Beamforming in inte-
grated sensing and communication systems with reconfigurable intelligent
minimum beampattern gain. Due to the nonconvex nature of surfaces,” 2022, [Online]. https://arxiv.org/abs/2206.07679.
the formulated problem, an alternating optimization based algo-
rithm was proposed to solve the original problem. Simulation