See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/360501749

Performance Analysis of IRS-aided Short-Packet NOMA Systems over

Nakagami-m Fading Channels

Preprint · May 2022

DOI: 10.36227/techrxiv.19709767.v1

CITATIONS READS

0 97

4 authors, including:

Lei Yuan Nan Yang

Lanzhou University Australian National University
36 PUBLICATIONS   133 CITATIONS    225 PUBLICATIONS   6,008 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

URLLC: Short-packet Communication View project

performance analysis on IRS View project

All content following this page was uploaded by Lei Yuan on 16 May 2022.

The user has requested enhancement of the downloaded file.

Performance Analysis of IRS-aided Short-Packet NOMA
Systems over Nakagami-m Fading Channels
This paper was downloaded from TechRxiv (https://www.techrxiv.org).

LICENSE

CC BY 4.0

SUBMISSION DATE / POSTED DATE

05-05-2022 / 10-05-2022

CITATION

yuan, lei; du, qiuyu; yang, nan; fang, fang; yang, nan (2022): Performance Analysis of IRS-aided Short-Packet
NOMA Systems over Nakagami-m Fading Channels. TechRxiv. Preprint.
https://doi.org/10.36227/techrxiv.19709767.v1

DOI

10.36227/techrxiv.19709767.v1
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, JUNE 2022
Performance Analysis of IRS-aided Short-Packet NOMA
Systems over Nakagami-m Fading Channels
Lei Yuan, Member, IEEE, Qiuyu Du, Nan Yang, Senior Member, IEEE, Fang Fang, Member, IEEE, Nan Yang
Abstract—This paper analyzes the performance of an intelligent
reflecting surface (IRS)-aided short-packet communication system over
Nakagami-m fading channels. In this system, the base station adopts
non-orthogonal multiple access (NOMA) to transmit to a near user and
a far user, and its transmission to the far user is aided by the IRS.
Both continuous phase shifts (CPSs) and discrete phase shifts (DPSs) are
considered for the IRS. By using order statistic and moment matching,
we approximate the end-to-end channel gain to follow the Gamma
distribution and then derive the approximate closed-form expressions
for the average block error rate (BLER) in the high signal-to-noise ratio
(SNR) regime and then conduct a refined analysis of the interplay among
the minimum common blocklength, optimal power allocation, and the
related parameters of IRS (e.g., the number of elements of the IRS) for
achieving target BLERs at two users. Finally, numerical and simulation
results demonstrate the accuracy of theoretical analysis. They also show
that the IRS-aided short-packet NOMA system outperforms the system
without IRS, and that DPSs with 3-bit quantization achieve a similar
performance as the optimal CPSs.
Index Terms—Non-orthogonal multiple access, intelligent reflecting
surface, short-packet communications, Nakagami-m fading.
I. I NTRODUCTION
Intelligent reflecting surface (IRS), which consists of a large num-
ber of passive reflective elements, has recently been recognized as
one of the most promising techniques for the beyond fifth-generation
(B5G) wireless networks, due to its capability to dynamically alter
wireless channels and enhance the spectrum efficiency of wireless
systems [1]. A prototype of the IRS-aided communication system
consisting of 1100 controllable elements of the IRS was designed in
[2], where the field trials verified the great potential of the IRS for
the B5G wireless networks.
The performance of IRS-aided communications has been widely
analyzed over the past few years. For example, [3] derived the symbol
error probability and outage probability (OP) of the IRS-aided system
with the instantaneous channel state information (CSI) over Rician
fading. In [4], the authors analyzed the OP and ergodic capacity (EC)
of the IRS-aided system over Nakagami-m fading. Additionally, the
impact of phase errors at the IRS on the symbol error probability was
analyzed in [5]. Compared with the instantaneous CSI, the statistical
CSI is much easier to be obtained in practice. Motivated by this,
[6] analyzed the OP of an IRS-aided single-antenna system with the
statistical CSI over Rician fading.
Apart from IRS, non-orthogonal multiple access (NOMA) has also
been identified as a promising technology for the B5G wireless
networks. Thus, it is natural to integrate IRS and NOMA into the
design of emerging wireless systems. Against this background, [7]
studied the impact of optimal continuous phase shifts (CPSs) and
This work was supported in part by the National Key R&D Program of
China (2017YFE0118900). (Corresponding author: Lei Yuan.)
L. Yuan, Q. Du, and N. Yang are with the School of Information Science
and Engineering, Lanzhou University, Lanzhou, 730000, Gansu, China. (e-
mail: yuanl@lzu.edu.cn; duqy20@lzu.edu.cn; nyang20@lzu.edu.cn).
N. Yang is with the School of Engineering, Australian National University,
Canberra, ACT 2600, Australia (e-mail: nan.yang@anu.edu.au).
F. Fang is with the Department of Electrical and Computer Engineering
and the Department of Computer Science, Western University, London, ON
N6A 3K7, Canada (e-mail: fang.fang@uwo.ca).
1
random discrete phase shifts (DPSs) on the OP performance in an
IRS-aided downlink two-user NOMA system. Moreover, [8] designed
an energy-efficient scheme for an IRS-aided multiple-input single-
output NOMA system. In addition, by utilizing the moment matching
method to approximate the received signal-to-noise ratios (SNRs), the
OP of an IRS-aided uplink two-user NOMA system over Nakagami-
m fading was derived in [9].
Short-packet communications (SPC) is a major feature of ultra-
reliable and low-latency communications (URLLC), which is a vital
scenario of the B5G wireless networks. Thus, it is worthwhile to
explore the potential of using SPC, IRS and NOMA to accomplish
URLLC. For SPC, the maximal achievable rate (MAR) depends on
the blocklength and block error rate (BLER) [10]. It was further
pointed out in [11] that the performance metrics used in the infinite
blocklength regime, such as the OP and EC, cannot accurately predict
the performance of SPC, and the optimal design of SPC systems
relies on a refined analysis of the interplay among the BLER, MAR,
and blocklength. Following [11], [12] and [13] studied the interplay
among the average BLER, minimum blocklength, and optimal power
allocation of short-packet NOMA systems in the single-antenna
and multiple-antenna scenarios, respectively. Also, considering the
advantage of IRS, [14] and [15] analyzed the MAR and average
BLER of IRS-aided SPC systems over various fading. To the best
knowledge of the authors, the performance of IRS-aided short-packet
NOMA systems has not been investigated until very recently by [16],
where the authors considered random and optimal CPSs.
We note that there are three limitations in [16]. First, [16] only
considered Rayleigh fading, which is not a generalized fading model.
Second, [16] did not optimize the blocklength and power allocation
coefficients for the considered system. Third, [16] did not touch
DPSs at the IRS, which is a more feasible option than CPSs for
practical use. In order to overcome these limitations, we analyze the
BLERs of an IRS-aided short-packet NOMA system over generalized
Nakagami-m fading, which encompasses Rayleigh fading as a special
case. This analysis allows us to examine the impacts of both CPSs and
DPSs on the average BLERs, optimal power allocation, and minimum
common blocklength in the considered system. Using numerical and
simulation results, we validate our theoretical analysis, demonstrate
the superiority of the IRS-aided short-packet NOMA system over the
system without IRS, and show the impact of quantization of DPSs
on the achieved performance, especially compared to CPSs.
Notations: |x| and arg(x) denote the modulus and angle of a
complex number x, respectively. fX (x), FX (x) and X = E[X]
denote the probability density function (PDF), cumulative distribu-
tion function (CDF) and expectation of a random variable (RV)
∫∞ −t2
X, respectively. Q(x) = x √12π e 2 dt is the Gaussian Q-
function. Γ(m) is the Gamma function. U(a, b) denotes the uniform
distribution over the interval (a, b). Nakagami(m, Ω) denotes the
Nakagami-m distribution with shape parameter m and average power
Ω. Gamma(m, Ω) denotes the Gamma distribution, whose PDF is
m m−1 −Ωx
fX (x) = Ω x Γ(m)e . CN (0, 1) and N (0, 1) denote circularly-
symmetric complex Gaussian and real Gaussian distributions with
zero mean and unit variance, respectively.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, JUNE 2022 2

In order to determine the distribution of the received SNR at UE-f ,

we need to derive the distribution of |Gc |2 . To this end, we define
∆ ∑
IRS H′ = N n=1 e
jϕn
|h1n | |h2n | and present the distribution of H ′ in
the following lemma.
IRS Controller Lemma 1: When N is large and under the assumptions that
∆ ∆
E[|h1n |2 ] = E[|h2n |2 ] = 1, a1 = E [|h1n |] < 1, a2 = E [|h2n |] < 1,
h1n , h2n , and ϕn are mutually independent RVs, and ϕn follows the
UE-f same distribution ϕ with zero-mean and p-order circular moments
φp = E[ejpϕ ], we can approximate H ′ = U + jV as a non-
circularly symmetric complex Gaussian RV, where U and V are
independent RVs, U ∼ N (µ, σU 2
), V ∼ N (0, σV2 ), µ = N φ1 a1 a2 ,
BS σU = 2 (1 + φ2 − 2φ1 a1 a2 ), and σV2 = N2 (1 − φ2 ).
2 N 2 2 2

Proof : This lemma can be proved aided by [5, Appendix A]. 

UE-t
Fig. 1: The illustration of the considered IRS-aided downlink NOMA
system.
II. S YSTEM M ODEL
This work considers an IRS-aided downlink NOMA system, as
depicted in Fig. 1, where a single-antenna BS communicates with two
single-antenna users. UE-t and UE-f represent the near user and the
far user from the BS, respectively. Due to the long distance between
the BS and UE-f , a passive reflective IRS with N elements is used
According to [5], if the IRS adopts the optimal CPSs, both φ1
and φ2 equal 1. In this case, substituting φ1 = φ2 = 1 into Lemma
1, H ′ is actually approximated as a real Gaussian RV, i.e., U ∼
N (N a1 a2 , N (1 − a21 a22 )), and V is zero. Differently, if the IRS
adopts the optimal DPSs with q quantization bits, φ1 and φ2 equal
sin(2−q π ) sin(2−q+1 π )
2−q π
and 2−q+1 π , respectively. After obtaining H ′ , we can
use the moment matching method, which is shown in Lemma 2, to
derive the distribution of |Gc |2 .
Lemma 2: When X is a non-negative [ ] RV with the first moment
E [X] and the second moment E X 2 , we can approximate X as
2

to improve the transmission quality between the BS and UE-f . Gamma(m, η), where the shape parameter is m = E X(E[X])
[ 2 ]−(E[X])2
We assume that all links undergo independent and identically and scale parameter is η = E X 2 E[X] . Moreover, Gamma RVs
[ ]−(E[X])2
distributed (i.i.d.) Nakagami-m small-scale fading and have the satisfy the scaling property, i.e., the RV cY ∼ Gamma (m, cη).
same path-loss exponent β. The channel coefficient between the nth Proof : This lemma 
∆ −β/2  can be proved with the aid of [9].
element of the IRS and the BS is denoted by g1n = d1 h1n for all  −β/2 2
Given |Gc | = (d1 d2 )
2
(U + jV ) + |gf | , by using both
n ∈ {1, · · · , N }, where d1 denotes the distance between the BS and
the center of the IRS and |h1n | ∼ Nakagami (m1 , 1). Similarly, we Lemma 1 and Lemma 2, we can derive and present the distribution
denote the channel coefficient between the nth element of the IRS and of |Gc |2 in the following theorem.
UE-f by g2n = d2
∆ −β/2
h2n , where d2 denotes the distance between Theorem 1: For large N , the distribution of |Gc |2 is approximated
the center of the IRS and UE-f and |h2n | ∼ Nakagami (m2 , 1). The as the Gamma distribution, which is characterized by two parameters
(E[|Gc |2 ])2 E[|Gc |2 ]
channel coefficient between the BS and UE-k, k ∈ {t, f }, is denoted mc = 2 and ηc = 2 , i.e.,
E[|Gc |4 ]−(E[|Gc |2 ]) E[|Gc |4 ]−(E[|Gc |2 ])
∆ −β/2
by gk = dk hk , where dk denotes the distance between the BS [ ] [ ]
|Gc | ∼ Gamma(mc , ηc ). E |Gc | and E |Gc | are given by
2 2 4
and UE-k and |hk | ∼ Nakagami (mk , 1). (A.1) and (A.2), respectively.
∆
(√According √ to the principles
) of NOMA, the BS transmits s = Proof: Please refer to Appendix A.
αt Ps xt + αf Ps xf to two users, where Ps is the transmit 2) Distributions of the received SINRs at UE-t and UE-f: At
power at the BS, xk denotes the normalized signal sent to UE-k, αk each time slot, we define UE-1 as the user which possesses a
denotes the power allocation coefficient for UE-k, and αt + αf = 1. larger end-to-end channel gain, and define α1 as its power allocation
Thus, the received signals at UE-t and UE-f are expressed as coefficient. According to the rules of NOMA, the BS allocates a
yt = gt s + wt , (1) lower transmit power to UE-1, i.e., α1 < 0.5. For example, if UE-t
possesses a larger channel gain, then UE-t and αt are UE-1 and
and α1 , respectively. We further define G1 and x1 as the end-to-end
(∑N )
yf = g1n g2n ejΘn + gf s + wf = Gc s + wf , (2) channel coefficient and signal for UE-1, respectively. Then, we have
n=1 |G1 |2 = max(|Gc |2 , |gt |2 ). Similarly, we define UE-2 as the user
∆ ∑N with a smaller end-to-end channel gain, and define α2 , G2 , and x2
respectively, where Gc = n=1 g1n g2n e
jΘn
+ gf , Θn ∈ [0, 2π)
denotes the phase adjustment introduced by the nth element of the as the power allocation coefficient, end-to-end channel coefficient,
IRS, and wk denotes the complex additive white Gaussian noise and signal for UE-2, respectively.
which follows CN (0, σ 2 ). By using successive interference cancellation (SIC), UE-1 first
1) Statistics of the end-to-end channel Between BS and UE-f: We recovers x2 by treating x1 as interference. The instantaneous signal-
assume that the instantaneous CSI is available at the system. Thus, to-interference-plus-noise ratio (SINR) of decoding x2 at UE-1 is
coherent combination and the optimal CPSs at the IRS, given by written as
α2 |G1 |2
Θn = arg (hf ) − arg (h1n ) − arg (h2n ), can be used to maximize γ1,2 = , (3)
the received SNR at UE-f . which causes high implementation α1 |G1 |2 + ρ−1
complexity in practice. To reduce such complexity, [5] suggested where ρ = Ps /σ 2 is the transmit SNR. If x2 is successfully decoded
to use DPSs, although quantization errors are inevitable. With this and removed, then the instantaneous SNR of decoding x1 at UE-1 is
use, the optimal DPSs at the IRS are given by Θn = arg (hf ) − written as
arg (h1n ) − arg (h2n )+ϕn , where ϕn denotes the quantization error γ1,1 = α1 ρ|G1 |2 . (4)
which is modeled by uniform distribution. For q quantization bits,
ϕn ∼ U (−2−q π, 2−q π) [5]. At UE-2, x2 is decoded directly where x1 is treated as interference.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, JUNE 2022 3

( ) ( )
ηc x ηt x
Then the received SINR of decoding x2 at UE-2 is given by γ mc , ρ(α2 −α1 x)
γ mt , ρ(α2 −α1 x)
α2 |G2 |2 Fγ12 (x) = , (13)
γ2,2 = . (5) Γ (mc )Γ (mt )
α1 |G2 |2 + ρ−1 and
( ) ( )
III. P ERFORMANCE A NALYSIS γ mc , ηc x
γ mt , ηt x
α1 ρ α1 ρ
In this section, we analyze the average BLERs in the considered Fγ11 (x) = , (14)
Γ (mc )Γ (mt )
IRS-aided short-packet NOMA system and accordingly evaluate the
impacts of CPSs and DPSs on such BLERs, the optimal power alloca- where ηt = mt /dβt and the power allocation coefficients need to
tion, and the minimum common blocklength. Given the blocklength satisfy the condition α2 − α1 x > 0. By using such expressions, the
L, the BLER ε, and the SNR γ, the MAR R (L, γ, ε) in bits per approximate average BLER of decoding x2 at UE-2, denoted by ε2 ,
channel use (bpcu) is written as [10] is derived as
√ ( )( )
( ) γ (mc , ηc ϱ2 ) γ (mt , ηt ϱ2 )
V (γ) −1 log2 L ε2 = ε2,2 ≈ 1 − 1 − 1− , (15)
R (L, γ, ε) = C (γ) − Q (ε) + O , (6) Γ (mc ) Γ (mt )
L L o2
where ϱ2 = ρ(α2 −α 1 o2 )
. Moreover, the approximate average BLER
where
( C (γ) = log
) 2 (1 + γ) denotes the Shannon capacity, V (γ) =
of decoding x1 at UE-1, denoted by ε1 , is derived as
1 − (1 + γ)−2 (log2 e)2 is the channel dispersion,
( ) Q−1
(·) is the
inverse of the Gaussian Q-function, and O log2 L
is the remaining ε1 ≈ ε1,1 + ε1,2
L
γ (mc , ηc ϱ1 ) γ (mt , ηt ϱ1 ) + γ (mc , ηc ϱ2 ) γ (mt , ηt ϱ2 )
terms of order logL2 L , which can be omitted when L ≥ 100 [12]. ≈ ,
In this work, we define Ti and Li as the number of data bits Γ (mc ) Γ (mt )
(16)
and blocklength for UE-i, respectively, i ∈ {1, 2}. Then, the MAR
o1
at UE-i is expressed as Ri = Ti /Li , and the instantaneous BLER where ϱ1 = ρα 1
.
corresponding to the link with the instantaneous SNR γi,j , denoted In the high SNR regime, by using γ(s,x) xs
→ 1s for x → 0 [17] and
by εi,j , j ∈ {1, 2}, , is approximated as (11), the asymptotic expressions for the average BLER at both users
( )
C (γi,j ) − Rj are derived. Specifically, the asymptotic average BLER of decoding
εi,j ≈ Q √ . (7) x2 at UE-2, denoted by ε∞ 2 , is derived as
V (γi,j ) /Lj  (η ϱ )mc
Furthermore, for wireless fading channels, the average BLER εi,j  c 2
 Γ(mc +1) , mt > mc ,

can be written as ∞ (ηt ϱ2 )mt
∫ ∞ ( ) ε2 ≈ , mt < mc , (17)


Γ(mt +1)
C (x) − Rj  m
ϱ2 t
Q √
mt
εi,j ≈ fγi,j (x)dx, (8) Γ(mt +1)
mt
(ηt + ηc ) , mt = mc .
0 V (x) /Lj
Moreover, the approximate average BLER of decoding x1 at UE-1,
where fγi,j (x) denotes the PDF of γi,j .
denoted by ε∞
1 , is derived as
It is noted that in URLLC the perfect SIC cannot be guaranteed ( t +mc )
and εi,j is a very small value, e.g., 10−3 to 10−5 , the average BLER ηtmt ηcmc ϱm + ϱm t +mc
ε∞
1 ≈ 1 2
. (18)
of decoding x1 at UE-1 is approximated as Γ (mt + 1) Γ (mc + 1)
ε1 =ε1,2 + (1 − ε1,2 ) ε1,1 ≈ ε1,2 + ε1,1 . (9) Based on (17) and (18), the diversity order at UE-2 and UE-1 are
In (order to find a closed-form expression for εi,j , we approximate obtained as
√ ) {
Q (C (γi,j ) − Rj ) / V (γi,j ) /Lj as [12] log ε∞ mt , mt 6 mc ,
 − lim 2
= (19)
ρ→∞ log ρ m c , mt > m c ,

 1, γi,j 6 vj ,
Zi,j (γi,j ) = 1
− κ j (γ i,j − oj ), v j < γi,j < µj , (10) and
log ε∞


2
− lim 1
= mt + mc , (20)
0, γi,j > µj , ρ→∞ log ρ
√ √ ( )
where κj = Lj / 2π 2 2Rj
− 1 , oj = 2Rj − 1, vj = oj − 2κ1 j , respectively.
1
and µj = oj + 2κj . In this case, the average BLER εi,j can be B. Power Allocation and Blocklength Optimization at High SNRs
approximated as We denote εth
i as the average BLER threshold at UE-i, and apply
∫ ∞ ∫ µj them into (17) and (18) to obtain the minimum blocklength Li,min
(a) (b)
εi,j ≈ Zi,j (x)fγi,j (x)dx = κj Fγi,j (x)dx ≈ Fγij (oj ) , of UE-i at high SNRs as
0 vj
(11) T
L2,min ≈ ( 2 ) (21)
where Fγi,j (x) is the CDF of γi,j , step (a) is obtained by log2 1+α1+ξ
1ξ
using the partial integration method, and step (b) is obtained
by and
∫ b using the first-order a+b
Riemann integral approximation given by
f (x)dx ≈ (b − a)f ( ) [12]. L1,min ≈ (
T1
) , (22)
a 2
( ( )mt +mc ) mt +m
1
c
A. Average BLERs log2 α1 ρ ψε1 − ρ
th ξ
+1
By applying the order statistics, the expressions for Fγij are
 ( )
γ mc , ρ(α2η−α
cx where ψ = Γ(mηtt+1)Γ(m
mt η mc
c
c +1)
and
Fγ22 (x) = 1 − 1 −
1 x)
  ρ 1
Γ (mc ) 
 (ε̄th
2 Γ (mt + 1)) t ,
m mt < mc , (23a)


 ( ) (12)  ηt
ρ th 1
γ mt , ρ(α2η−αtx ξ=
 (ε̄2 Γ (mc + 1)) mc , mt > mc , (23b)
× 1 −
1 x)
, 
 ηc
Γ (mt ) 
 1
ρ(ε̄th
2 Γ (mt + 1) / (ηt
mt
+ ηcmc )) mt , mt = mc . (23c)
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, JUNE 2022 4

100
Algorithm 1: Power Allocation Algorithm
Input: input parameters N , T1 , T2 , mt , mf , m1 , m2 , dt , df ,
d1 , d2 , β, q, ρ, ε̄th th
1 , ε̄2 , and tolerance µ. 10-2
Output: Optimal power allocation coefficient α1∗ . UE-2, Without IRS
α− +α+

Average BLER
−
1 Initialize:α ← 0, α ← 0.5, and α̂ ←
+
2
; UE-1, Without IRS
UE-2, N=32, Eq.(15)
2 while |L2,min − L1,min | > µ do 10-4 UE-2, N=32, Eq.(17)

3 if L2,min − L1,min < 0 then UE-1, N=32, Eq.(16)

UE-1, N=32, Eq.(18)
4 Set α− ← α̂; UE-2, N=96, Eq.(15)
UE-2, N=96, Eq.(17)
5 else 10-6
UE-1, N=96, Eq.(16)
6 Set α+ ← α̂; UE-1, N=96, Eq.(18)
UE-1, Simulation
7 end UE-2, Simulation
− + 10-8
8 Set α̂ ← α +α 2
and compute L2,min − L1,min based on 45 50 55 60 65 70 75

(21) and (22); Transimit SNR (dB)

9 end
∗ Fig. 2: Average BLER versus ρ for N = 32, N = 96, and the system
10 Set α1 ← α̂ ;
∗
without the use of IRS.
11 return α1 .

In Fig. 3, we study the impact of DPSs on the average BLER. In

By calculating the derivative of L2,min with respect to α1 , we this figure, we use α1 = 0.2, mt = 4, and dt = 10. Specifically, Fig.
obtain 3(a) plots the average BLER versus ρ for the optimal CPSs, the DPSs
∂L2,min T2 ξ with q = 1 or q = 3, and N = 64. We observe from this subfigure
≈ ( ( ))2 > 0, (24) that, on the one hand, the analytical curves corresponding to the DPSs
∂α1 1+ξ
(1 + α1 ξ) log2 1+α 1ξ
ln 2 are consistent with simulation results, which verifies our theoretical
which shows that L2,min is an increasing function of α1 . Then by analysis. On the other hand, the BLER performance of DPSs with
calculating the derivative of L1,min with respect to α1 , we obtain 3-bit quantization approaches to that of the optimal CPSs, implying
that using the 3-bit quantizer is sufficient to achieve the near-optimal
∂L1,min T1 ϑ
≈− <0 (25) performance in practice. Furthermore, we examine the impact of the
∂α1 (α1 ϑ + 1) (log2 (α1 ϑ + 1))2 ln 2 number of quantization bits on the required number of elements of
with IRS for achieving a target BLER. Since the impact on two users
( ) 1
ϑ = ρ ψεth
mt +mc
, (26) is similar, we only plot the average BLER versus the number of
1
elements of the IRS for UE-2 in Fig. 3(b). As shown in this subfigure,
which shows that L1,min is a decreasing function of α1 . Therefore, to achieve the BLER of 10−5 at UE-2, the required numbers of
the minimum common blocklength L∗ is obtained by solving L∗ = elements of the IRS for DPSs with 1-bit, 2-bit, and 3-bit quantizers
L1,min = L2,min . are 169, 119, and 110, respectively. This observation demonstrates
Based on [13], the optimal power allocation coefficient α1∗ , corre- that, compared to using the 1-bit quantizer, using the 3-bit quantizer
sponding to L∗ , can be obtained by using the bisection search shown can save 59 elements of the IRS for achieving ε2 = 10−5 , while
in Algorithm 1. After doing so, we substitute α1∗ into (21), L∗ can the tradeoff between the required number of quantization bits and
be expressed as the required number of elements of the IRS depends on practical
T
L∗ ≈ ( 2 ) (27) implementation complexity.
1+ξ
log2 1+α ∗ξ Fig. 4 shows the impact of DPSs on Li,min for different α1 in the
1
high transmit SNR regime. The parameters are mt = 3.5, dt = 15,
−5
IV. N UMERICAL R ESULTS εth th
1 = ε2 = 10 , and q = 1, 2, 3. In Fig. 4(a), we assume N = 128
In this section, we provide numerical and simulation results to and ρ = 67 dB. Using Theorem 1, we obtain mc = 4.3, 6.5, and
validate our analysis in Section III, and offer some useful insights 7.2 for q = 1, 2, and 3, respectively. Since mt < mc , according
into system design. The simulation parameters are set as follows: to (21) and (23a), the curve corresponding to L2,min is unchanged
β = 4, m1 = m2 = 4, mf = 1, d1 = 25, d2 = 10, df = 30, for different values of q. Moreover, we observe from this subfigure
T1 = T2 = 200, L1 = L2 = 300, and µ = 10−3 . Other parameters that L∗ decreases from 176 to 158 when q increases from 1 to 3. In
are specified in each figure. Fig. 4(b), we assume N = 64 and ρ = 72 dB. Similarly, we obtain
To show the accuracy of our derived expressions for the average mc = 2.3, 3.1, and 3.3 for q = 1, 2, and 3, respectively. Since
BLER, Fig. 2 plots the average BLER versus ρ for the considered mt > mc , according to (21), (22), and (23b), we find that the value
IRS-aided short-packet NOMA system with the optimal CPSs, where of q affects both L2,min and L1,min . Furthermore, we observe from
α1 = 0.2, mt = 3.5, dt = 15 and N = 32 or 96. We first observe this subfigure that L∗ decreases from 740 to 203 when q increases
that, the analytical curves derived from (15) and (16) match perfectly from 1 to 3. These observations show that quantization errors impose
with simulation results over the entire transmit SNR range, and the a profound impact on achieving low-latency transmission for the case
asymptotic curves derived from (17) and (18) accurately predict the of mt > mc .
average BLERs in the high transmit SNR regime. These observations
demonstrate the correctness of our analysis. Second, we observe from V. C ONCLUSION
this figure that, compared to the system without IRS, the IRS-aided This paper analyzed the performance of IRS-aided short-packet
system achieves lower BLERs for both users at the same transmit NOMA systems over Nakagami-m fading channels. We first analyzed
SNR. Third, we observe that when the number of elements of the IRS the statistical property of the end-to-end channel in the considered
increases, the performance improvement becomes more profound, at IRS-aided system, based on which we derived the approximate
the cost of increasing implementation complexity. closed-form expressions for the average BLER at both users, as
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, JUNE 2022 5

100 10-1
UE-2, DPS, q=1
UE-1, DPS, q=1
DPS, q=3 Furthermore, if Z2 = b+Z1 , where b is a non-zero constant number
(i.e., Z2 ∼ N (b, σZ
DPS, q=2 2
10-1 UE-2, DPS, q=3
UE-1, DPS, q=3 10-2
DPS, q=1
1
)), then the pth moment of Z2 is
Simulation
UE-2, CPS
10-2 UE-1, CPS ∑
p [ ]
Average BLER

Average BLER
E [Z2p ] = E [(b + Z1 )p ] = Cpk bk E Z1p−k .
UE-2, Simulation
10-3
UE-1, Simulation (A.4)
10-3
k=0
10-5 10-4
According to [9], if Z3 ∼ Nakagami (mz , Ωz ), then the pth moment
2
10-4 1.5

( )
1

10-5 10 -5 of Z3 is
0.5
Γ mz + P2
64 64.5 65
E [Z3p ] = . (A.5)
10-6
56 58 60 62 64 66 68 70 72 10-6
40 60 80 100 120 140 160 180 200 Γ (mz ) (mz /Ωz )P /2
Transimit SNR (dB) Number of elements of IRS
sin(2−q+i−1 π )
(a) (b) By using φi = 1 for the optimal CPSs, φi = 2−q+i−1 π for the
optimal DPSs, and Lemma 1, we obtain the parameters of U1 and V1
Fig. 3: Impact of DPSs on the average BLER. (a) Average BLER versus
ρ for N = 64. (b) Average BLER at UE-2 versus N for q = 1, 2, 3.
in different cases. Then, by substituting (A.3), (A.4), and (A.5) into
(A.1) and (A.2), we derive E[|Gc |2 ] and E[|Gc |4 ], respectively.
600 900
L2,min
Minimum blocklength (channel uses)

Minimum blocklength (channel uses)

550
L1,min , DPS, q=1
800 R EFERENCES
500 L , DPS, q=3
2,min
L1,min , DPS, q=2 700
450
L , DPS, q=3
L
1,min
, DPS, q=3 [1] Q. Wu, S. Zhang, B. Zheng, C. You and R. Zhang, “Intelligent reflecting
1,min L2,min , DPS, q=2
400 L*
600 surface-aided wireless communications: a tutorial,” IEEE Trans. Com-
L1,min , DPS, q=2
350 500 mun., vol. 69, no. 5, pp. 3313-3351, May 2021.
L2,min , DPS, q=1
300
[2] X. Pei et al., “RIS-aided wireless communications: prototyping, adaptive
400 L1,min , DPS, q=1
250 L*
beamforming, and indoor/outdoor field trials,” IEEE Trans. Commun.,
200
300 vol. 69, no. 12, pp. 8627-8640, Dec. 2021.
150
200 [3] A. M. Salhab and M. H. Samuh, “Accurate performance analysis of
reconfigurable intelligent surfaces over Rician fading channels,” IEEE
100 100
0.05 0.1 0.15 0.2 0.25 0.3 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0.11 Wireless Commun. Lett., vol. 10, no. 5, pp. 1051-1055, May 2021.
1 1 [4] D. Selimis, K. P. Peppas, G. C. Alexandropoulos and F. I. Lazarakis,
“On the performance analysis of RIS-empowered communications over
(a) (b)
Nakagami-m fading,” IEEE Commun. Lett., vol. 25, no. 7, pp. 2191-
Fig. 4: Minimum blocklength versus α1 for (a) mt < mc and (b) mt > 2195, July 2021.
mc . [5] M.A. Badiu and J. P. Coon, “Communication through a large reflecting
surface with phase errors,” IEEE Wireless Commun. Lett., vol. 9, no. 2,
well as the asymptotic expressions for the average BLER at high pp. 184-188, Feb. 2020.
[6] P. Xu, W. Niu, G. Chen, Y. Li and Y. Li, “Performance analysis of RIS-
transmit SNRs. Using the asymptotic expressions, we quantified assisted systems with statistical channel state information,” IEEE Trans.
the relationship among the minimum common blocklength, optimal Veh. Technol., vol. 71, no. 1, pp. 1089-1094, Jan. 2022.
power allocation, and the parameters of the IRS for achieving the [7] Z. Ding, R. Schober and H. V. Poor, “On the impact of phase shifting
target BLERs at two users. Finally, numerical and simulation results designs on IRS-NOMA,” IEEE Wireless Commun. Lett., vol. 9, no. 10,
pp. 1596-1600, Oct. 2020.
validated our theoretical analysis, and demonstrated that (i) IRS-aided [8] F. Fang, Y. Xu, Q. -V. Pham and Z. Ding, “Energy-Efficient Design of
short-packet NOMA systems outperform the systems without IRS in IRS-NOMA Networks,“IEEE Trans. Veh. Technol., vol. 69, no. 11, pp.
terms of reliability and (ii) DPSs with 3-bit quantization are sufficient 14088-14092, Nov. 2020.
to achieve the near-optimal performance. Considering that the statis- [9] B. Tahir, S. Schwarz and M. Rupp, “Analysis of uplink IRS-assisted NO-
MA under Nakagami-m fading via moments matching,” IEEE Wireless
tical CSI is much easily obtained at the BS and reduces feedback
Commun. Lett., vol. 10, no. 3, pp. 624-628, Mar. 2021.
latency, the design of IRS-aided short-packet NOMA systems with [10] Y. Polyanskiy, H. V. Poor and S. Verdu, “Channel coding rate in the
statistical CSI will be investigated in the future work. finite blocklength regime,” IEEE Trans. Info. Theory, vol. 56, no. 5, pp.
2307-2359, May 2010.
A PPENDIX A [11] G. Durisi, T. Koch, J. Ostman, Y. Polyanskiy and W. Yang, “Short-packet
P ROOF OF T HEOREM 1 communications over multiple-antenna Rayleigh-fading channels,” IEEE
Trans. Commun., vol. 64, no. 2, pp. 618-629, Feb. 2016.
Let us define U1 = (d1 d2 )−β/2 U and V1 = (d1 d2 )−β/2 V , [12] Y. Yu, H. Chen, Y. Li, Z. Ding, and B. Vucetic, “On the performance of
and then rewrite |Gc |2 as |Gc |2 = (U1 + |gf |)2(+ V12 . Using
) the non-orthogonal multiple access in short-packet communications,” IEEE
scaling property, we find that |gf | ∼ Nakagami mf , d−β , U1 ∼ Commun. Lett., vol. 22, no. 3, pp. 590-593, Mar. 2018.
( ) ( f )
N (d1 d2 )−β/2 µ, (d1 d2 )−β σU
2
, and V1 ∼ N 0, (d1 d2 )−β σV2 . [13] D. D. Tran, S. K. Sharma, S. Chatzinotas, “Short-packet communications
for MIMO NOMA systems over Nakagami-m fading: BLER and
Thus, we obtain the expectation of |Gc | and |Gc | as
2 4
minimum blocklength analysis,” IEEE Trans. Veh. Technol., vol. 70, no.
[ ] [ ] [ ] [ ] 4, pp. 3583-3598, Apr. 2021.
E |Gc |2 = E |gf |2 + E U12 + 2E [|gf |] E [U1 ] + E V12 (A.1)
[14] R. Hashemi, S. Ali, N. H. Mahmood and M. Latva-aho, “Average rate
and and error probability analysis in short packet communications over RIS-
[ ] [ ] [ ] [ ] [ ] aided URLLC systems,” IEEE Trans. Veh. Technol., vol. 70, no. 10, pp.
E |Gc |4 =E |gf |4 + E U14 + 6E |gf |2 E U12 10320-10334, Oct. 2021.
[ ] [ ] [15] H. Ren, K. Wang and C. Pan, “Intelligent reflecting surface-aided
+ 4E |gf |3 E [U1 ] + 4E [|gf |] E U13 URLLC in a factory automation scenario,” IEEE Trans. Commun., vol.
[ ] [ ] [ ] [ ]
+ 2E V12 E |gf |2 + 2E V12 E U12 70, no. 1, pp. 707-723, Jan. 2022.
[ 2] [ ] [16] T. -H. Vu, T. -V. Nguyen, D. B. Da Costa and S. Kim, “Intelligent
+ 4E V1 E [|gf |] E [U1 ] + E V14 , (A.2) reflecting surface-aided short-packet non-orthogonal multiple access
systems,” IEEE Trans. Veh. Technol., doi: 10.1109/TVT.2022.3146856.
respectively. According to [18], if a real Gaussian RV Z1 follows [17] I. Gradshteyn and I. Ryzhik, Table of Integrals, Series, and Products,
N (0, σZ2
1
), then the pth moment of Z1 is 7th ed. Academic Press, Mar. 2007.
{ ( σ )p [18] H. Stark and J. Woods, Probability and Random Processes with Appli-
p! Z
√1 , when p is even, cations to Signal Processing and Communications, Jul. 2001.
E [Z1 ] =
p (p/2)! 2 (A.3)
0, when p is odd.

View publication stats