This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2017.2728608, IEEE
Transactions on Vehicular Technology
1

On the Performance of NOMA-Based Cooperative

Relaying Systems over Rician Fading Channels
Ruicheng Jiao, Linglong Dai, Jiayi Zhang, Richard MacKenzie, and Mo Hao

Abstract—Non-orthogonal multiple access (NOMA) is a strong users at the transmitter side. In this way, the transmitter
promising technique for the fifth generation (5G) wireless com- sends the superposition of signals with different power levels
munications. As users with good channel conditions can serve and the receiver applies successive interference cancellation
as relays to enhance the system performance by using successive
interference cancellation (SIC), the integration of NOMA and (SIC) to strong users to realize multi-user detection [5], [6].
cooperative relaying has recently attracted increasing interests. Such non-orthogonal resource allocation enables NOMA to
In this paper, a NOMA-based cooperative relaying system is accomodate more users and makes it promising to address
studied, and an analytical framework is developed to evaluate the 5G requirement of massive connectivity, with the cost of
its performance. Specifically, the performance of NOMA over controllable increase of complexity in receiver design due to
Rician fading channels is studied, and the exact expression of the
average achievable rate is derived. Moreover, we also propose an SIC [5].
approximation method to calculate the achievable rate by using In NOMA systems, the use of SIC implies that strong users
the Gauss-Chebyshev Integration. Numerical results confirm that have prior information about the messages of other users, so
our derived analytical results match well with the Monte Carlo essentially they are able to serve as cooperative relays. More-
simulations.
over, cooperative relaying is able to significantly enhance the
Index Terms—5G, non-orthogonal multiple access (NOMA), system performance of cellular networks [7]. Thus, combining
Rician fading channels, cooperative relaying, achievable rate. cooperative relaying and NOMA is promising to improve the
throughput of future 5G wireless networks, and has attracted
I. I NTRODUCTION
increasing interests recently [8]. Specifically, a cooperative
T is highly expected that future 5G networks should achieve
I a 10-fold increase in connection density, i.e., 106 con-
nections per square kilometers [1]. Non-orthogonal multiple
NOMA transmission scheme was proposed in [9], where
strong users decode the signals that are intended to others and
serve as relays to improve the performance of weak users.
access (NOMA) has been proposed as a promising candidate Another NOMA-based cooperative scheme was proposed in
to realize such an aggressive 5G goal [2]–[5]. NOMA is [10], where the performance of a NOMA-based decode-and-
foundamentally different from conventional orthogonal multi- forward relaying system under Rayleigh fading channel was
ple access (OMA) schemes such as FDMA, TDMA, OFDMA, studied. However, most of existing NOMA schemes only
etc., since it allows multiple users to simultaneously transmit consider the Rayleigh fading channel, which is suitable for
signals using the same time/frequency radio resources but rich scattering scenarios without line of sight (LOS), while
different power levels [3]–[5]. The key advantage of NOMA little attention has been drawn to the more general Rician
is to explore the extra power domain to further increase the fading channel, which takes both LOS and non LOS (NLOS)
number of supportable users. Specifically, users are identified into consideration. In some typical 5G application scenarios,
by their channel conditions, those with good channel con- such as massive machine-type communications (mMTC) and
ditions are called strong users and others are called weak Internet of things (IoT), “users” may be low-cost sensors
users. For the sake of fairness, less power are allocated to deployed in a small area, where both LOS and NLOS exist,
which can be better modeled by the Rician fading channel.
Copyright (c) 2015 IEEE. Personal use of this material is permitted.
However, permission to use this material for any other purposes must be
In this paper, we investigate the performance of the NOMA-
obtained from the IEEE by sending a request to pubs-permissions@ieee.org. based cooperative relaying transmission scheme in [10] under
This work was supported in part by the International Science & Technology Rician fading channels1 . Evaluating system performance under
Cooperation Program of China (Grant No. 2015DFG12760), the National
Natural Science Foundation of China (Grant No. 61571270), the Royal
Rician fading channel is rather challenging as the probability
Academy of Engineering under the UK-China Industry Academia Partnership density function of Rician distribution variables consists of
Programme Scheme (Grant No. UK-CIAPP \49), and the British Telecom and Bessel function, which makes it difficult to calculate the
Tsinghua SEM Advanced ICT LAB. (Corresponding author: Linglong Dai.)
R. Jiao and L. Dai are with the Tsinghua National Laboratory for In-
average achievable rate through integration. In order to derive
formation Science and Technology (TNList), Department of Electronic En- the exact expression of the achievable rate, we propose an
gineering, Beijing 100084, China (e-mails: jiaors16@mails.tsinghua.edu.cn, analytical method using Taylor expansion of Bessel function
daill@tsinghua.edu.cn).
J. Zhang is with the School of Electronics and Information Engi-
and incomplete Gamma function. However, the complexity
neering, Beijing Jiaotong University, Beijing 100044, China (e-mail: jiay- of the incomplete Gamma function makes it still difficult to
izhang@bjtu.edu.cn). get the exact values, so we further propose an approximation
R. MacKenzie is with the BT Innovate and Design, Adastral Park, Ipswich,
UK (e-mail: richard.mackenzie@bt.com).
M. Hao is with the Tsinghua SEM Advanced ICT LAB, Tsinghua Univer- 1 Simulation codes are provided to reproduce the results presented in this
sity, Beijing 100084, China (e-mail: haom@sem.tsinghua.edu.cn). paper: http://oa.ee.tsinghua.edu.cn/dailinglong/publications/publications.html.

0018-9545 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2017.2728608, IEEE
Transactions on Vehicular Technology
2

(a) (b)

Fig. 1. System models of two cooperative relaying systems: (a) Traditional cooperative relaying systems; (b) NOMA-based cooperative relaying systems.

method using Gauss-Chebvshev Integration to simplify the where nSR and nSD denote the additive white Gaussian noise
calculation. Finally, simulations confirm that our analytical (AWGN) with zero mean and variance σ 2 . The destination
results match well with the Monte Carlo results. only decodes symbol s1 by treating symbol s2 as noise, while
The rest of the paper is organized as follows. The system the relay acquires symbol s2 from (1) using SIC. Thus, the
model of the NOMA-based cooperative relaying system is received signal-to-interference plus noise ratios (SINRs) for
introduced in Section II. In Section III, we provide a detailed symbols s1 and s2 at the relay can be respectively obtained as
analysis for the achievable rate of the system and acquire both
|hSR | 2 a1 Pt
accurate and approximated results. Section IV provides numer- γSR
1
= , (4)
ical results to validate the theoretical analysis and section V |hSR | 2 a2 Pt + σ 2
concludes the paper. |hSR | 2 a2 Pt
γSR
2
= , (5)
II. S YSTEM M ODEL σ2
and the received SINR for symbol s1 at the destination is
As illustrated in Fig. 1 (a) and Fig. 1 (b), we consider
obtained as
a simple cooperative relaying system (CRS) consisting of a
source (S), a decode-and-forward relay (R) which works in |hSD | 2 a1 Pt
γSD = . (6)
half-duplex mode, and a destination (D). We assume that all |hSD | 2 a2 Pt + σ 2
links between them (i.e., S-to-D, S-to-R, and R-to-D) are In the second time slot, only the relay transmits the decoded
available. The independent Rician fading channel coefficients symbol s2 with full power Pt to the destination. Assuming
of S-to-D, S-to-R, and R-to-D links are denoted as hSD , hSR , that the relay can perfectly decode symbol s2 in the first time
and hRD , with the average powers of ΩSD , ΩSR , and ΩRD , slot [5], the received signal at the destination in the second
respectively. It is also assumed that ΩSD < ΩSR , since in time slot can be expressed as
general the path loss of the S-to-D link is usually worse than
r RD = hRD Pt s2 + n RD,
p
that of the S-to-R link [10]. (7)
In the traditional CRS presented in Fig. 1 (a), the source where nRD is the AWGN with zero mean and variance σ 2 ,
transmits s1 to the relay and destination in the first time slot. and the received SINR for symbol s2 in (7) can be obtained
Then in the second time slot, the relay transmits s1 to the as
destination. In this way, the destination only receives one |hRD | 2 Pt
signal in two time slots. γRD = . (8)
σ2
In the NOMA-based CRS showed in Fig. 1 (b), the destina-
tion is able to receive two different signals in two time slots, As the expressions for received signals and SINRs are
so it outperforms the traditional CRS in terms of throughput. already acquired, we will calculate both the exact and approx-
Specifically, in the first time slot, the source transmits the imated achievable rates in the NOMA-based CRS in the next
superposition of two different data symbols s1 and s2 to the section.
relay and the destination as follows: III. ACHIEVABLE R ATE A NALYSIS AND A PPROXIMATION
t = a1 Pt s1 + a2 Pt s2,
p p
(1) In this section, we first derive the exact expression of the
average achievable rate of the NOMA-based CRS over Rician
where si denotes the i-th data symbol with normalized fading channel. As the exact value of achievable rates are
power E[|si | 2 ] = 1, Pt is the total transmit power, and ai difficult to calculate, we further propose an approximation
is the power allocation coefficient. It is noted that a1 + a2 = 1, method using Gauss-Chebyshev Integration to simplify the
and a1 > a2 due to Ω2SD < Ω2SR [5]. Thus, the received signals numerical calculation.
r SR and r SD at the relay and the destination in the first time
slot are respectively expressed as A. Achievable Rate Analysis
r SR = hSR ( a1 Pt s1 + a2 Pt s2 ) + nSR,
p p
(2) In this subsection, we analyze the average achievable rate
of s1 and s2 . Let λ SD , |hSD | 2 , λ SR , |hSR | 2 , λ RD , |hRD | 2 ,
r SD = hSD ( a1 Pt s1 + a2 Pt s2 ) + nSD,
p p
(3) and ρ , Pt /σ 2 , where ρ represents the transmit SNR. As

0018-9545 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2017.2728608, IEEE
Transactions on Vehicular Technology
3

both the relay and the destination must successfully decode The final value will not change as k or n increases, so the
s1 and s2 , the rates of these two signals should be lower than infinite summation in (11) is convergent.
the rates of both links calculated by Shanon formula, so the Similarly, we can obtain the CDF of z2 as follows:
achievable rate is the minimum of the rates of two different X∞ X ∞

links. According to [10], we can obtain the achievable rates G(z2 ) =1 − Aw Ay Hy (k)Γ(n + 1, aw z2 )
Hw (n) B
B
Cs1 and Cs2 of signals s1 and s2 respectively as n=0 k=0
ay
1 ( ) × Γ(k + 1, z2 )
Cs1 = min log2 (1 + γSD ), log2 (1 + γSR 1
) a2
2 ∞ X ∞ ay
Hy (k)n!k!e−(aw + a2 )z2
X
1   =1 − Aw Ay Hw (n) B
B
= log2 1 + min{λ SD, λ SR } ρ −
2 k=0 n=0
1   k ai ( ay ) j
n X
log2 1 + min{λ SD, λ SR } ρa2 , (9) X w a2 i+j
2 × z2 , (14)
i=0 j=0
i! j!

1 ( ) where the parameters in (14) are similarly defined as those in

Cs2 =min log2 (1 + γSR2
), log2 (1 + γRD )
2 (11).
1 
= log2 1 + min{a2 λ SR, λ RD } ρ .

(10) After the CDF of z1 , min{λ SR, λ SD } has been obtained
2 as (11), we can substitute it into (9), and then the average
Let z1 , min{λ SR, λ SD }, z2 , min{a2 λ SR, λ RD }. According to achievable rate Cs1 of the signal s1 as shown in (9) can be
[11], we can get the cumulative distribution function (CDF) expressed as
of z1 as 
1 ∞
∞ X∞ Cs1 = [log2 (1 + z1 ρ) − log2 (1 + z1 ρa2 )]dF (z1 )
X 2 0
F (z1 ) = 1 − Ax Ay B Hy (k)Γ(n + 1, a x z1 )
Hx (n) B "  ∞  ∞ #
1 1 − F (z1 ) 1 − F (z1 )
k=0 n=0 = ρ dz1 − ρa2 dz1 .
× Γ(k + 1, ay z1 ) 2 ln(2) 0 1 + z1 ρ 0 1 + z1 ρa2
∞ X ∞  ∞ (15)
(a) 1 − F (z1 )
Hy (k)n!k!e−(a x +ay )z1
X
= 1 − Ax Ay Hx (n) B
B Let D( ρ) = ρ dz1 , and substitute (11) into
0 1 + z1 ρ
k=0 n=0 D( ρ), we have
Xn X k j
aix ay i+j  ∞
× z , (11) D( ρ) = ρ
1 − F (z1 )
dz1
i=0 j=0
i! j! 1 1 + z1 ρ
0
∞ X ∞ n X k j
where Bx (n) = (K xn (1 + K x ) n )/(Ωnx (n!) 2 ), By (k) = (Kyk (1 +
X X aix ay
= Ax Ay Hx (n) B
B Hy (k)n!k! ×
Ky ) k )/(Ωky (k!) 2 ), a x = (1 + K x )/Ωx , ay = (1 + Ky )/Ωy , k=0 n=0 i=0 j=0
i! j!
Ax = a x e−K x , Ay = ay e−Ky , B Hx (n) = Bx (n)/a xn+1 , B Hy (k) =  ∞ i+j −(a x +ay )z1
z1 e
By (k)/ayk+1 . The subscript x denotes the S-to-D link, y denotes d(z1 ρ)
the S-to-R link, w denotes the R-to-D link, and K is the Rician 0 1 + z1 ρ
∞ X ∞ n X k j
factor. Note that the expansion form of incomplete Gamma (b)
X X aix ay
function is used for the second equality (a) of (11). = Ax Ay Bx (n) By (k)n!k! ×
H H
k=0 n=0 i=0 j=0
i! j!ρi+j
Then, we prove the convergence of the infinite summation
 ∞ i+j −
a x +a y
t
in (11) as follows. t e ρ

Proof: Let Px = (K x (1 + K x ))/Ωx , Q y = (Ky (1 + dt, (16)

0 1+t
Ky ))/Ωy , we have
where (b) is obtained by setting t = z1 ρ.
Γ(n + 1, a x z1 ) Γ(n + 1) Now we have the following Lemma 1 to calculate the
< < 1, (12)  ∞ i+j − a x ρ+a y t
n! n! t e
Γ(k + 1, ay z1 ) Γ(n + 1) integral dt in (16).
< < 1, (13) 0 1+t
n! n!
then Lemma 1. For m ∈ Z∗ and β > 0, we have
∞
∞ X
 ∞ m −βt
t e
dt = eβ m!Γ(−m, β),
X
Ax Ay Bx (n) By (k)Γ(n + 1, a x z1 )Γ(k + 1, ay z1 ) (17)
1+t
H H
0
k=0 n=0  ∞ −t
∞ X ∞ k e
X Pxn Q y Γ(n + 1, a x z1 ) Γ(k + 1, ay z1 ) where Γ(−m, β) = dt denotes the incomplete
= Ax Ay β t
m+1
k=0 n=0
n! k! n! k! Gamma function.
∞ X ∞ k
X Pxn Q y Proof: Let x = β(1 + t), we have
< Ax Ay  ∞ m −βt  ∞
k=0 n=0
n! k! t e eβ (x − β) m e−x
Px +Q y dt = m dx. (18)
= Ax Ay e . 0 1+t β β x

0018-9545 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2017.2728608, IEEE
Transactions on Vehicular Technology
4

Then we define: B. Achievable Rate Approximation

(x − β) m In this subsection, we propose an approximation method us-
Jm (x) = , (19)
x
 ing Gauss-Chebyshev Integration [12] to simplify the numer-
eβ ∞
ical calculation of the incomplete Gamma function Γ(−m, β).
I (x) = m Jm (x)e−x dx. (20)
β β However, Gauss-Chebyshev Integration is used on the limited
On the one hand, by substituting (19) into (20), we have interval [−1, 1], while the integral intervals in incomplete
 ∞ Gamma functions of (23) and (25) are infinite intervals. Thus,
eβ (x − β) m
I (x) = − m d(e−x ) we set t = 2 β x1 −1 and convert the incomplete Gamma function
β β x as
eβ (x − β) m −x ∞ eβ 
=− m e  + m × 1 m 1 1 2β p
β x β β Γ(−m, β) = ( ) √ (t + 1) m−1 e− t +1 1 − t 2 dt
 ∞ 2β −1 1 − t 2
(x − β) m−1 (mx − x − β) −x n
e dx 1 πX 2l − 1
β x 2 = ( )m (cos( π) + 1) m−1
 ∞ 2 β n l=1 2n
eβ
= m Jm−1 (x)e−x dx. (21) − 2β
2l − 1
β β ×e cos( 2l−1 π )+1
2n π)|,
| sin( (26)
We can observe from (21) that as long as β is a root of 2n
Jm (x), the integral can be successively calculated by using where n is the approximation order. Substituting (26) into the
(21) for m times. On the other hand, we know that Jm (x) = exact expression of the achievable rate (23), we can finally
(x − β) m /x = x m−1 + am−2 x m−2 + ... + a0 + (−1) m β m /x, so obtain the approximation of (23) as
after m times of integration by part, we will have 1  
 ∞ Cs1 = D( ρ) − D( ρa2 ) , (27)
eβ (x − β) m 2 ln(2)
I (x) = − m d(e−x )
β β x where ∞ X∞ n X k j
 ∞ X X (i + j)! aix ay
eβ β m  (m) D( ρ) = Ax Ay Hx (n) B
B Hy (k)n!k! ×
=− m x m−1 + ... + a0 + (−1) m d(e−x ) i! j! ρi+j
β β x k=0 n=0 i=0 j=0
 ∞ ! i+j X
π
n
eβ
a x +a y 1 2l − 1
β m  (m) ×e ρ π) + 1) i+j−1
=− m (−1) m d(e−x ) a x +a y (cos(
2n
β β x 2 n l=1
ρ
= eβ m!Γ(−m, β),
a x +a y
(22) 2 ρ
− 2l − 1
cos( 2l−1 π )+1
where (·) (m) denotes m-order derivation. ×e 2n π)|,
| sin( (28)
2n
Substitute (17) in Lemma 1 into (16), we can get the final and D( ρa2 ) shares the same form as D( ρ). Similarly, (25) can
exact expression of Cs1 as be approximated as
1
Cs1 = (D( ρ) − D( ρa2 )), (23) 1 X∞ X ∞ Xn X k
(i + j)!
2 ln(2) Cs2 = Aw Ay BHw (n) B
Hy (k)n!k!
where 2 ln(2) k=0 n=0 i=0 j=0
i! j!
∞ X ∞ i+j n
i (a /a ) j a w +a y /a
πX
!
aw 1
X
D( ρ) = Ax Ay Hx (n) B
B Hy (k)n!k! ×
y 2
e ρ
2

ρi+j a +a /a n l=1
k=0 n=0 2 w ρy 2
n X k j
X (i + j)! aix ay a x ρ+a y a x + ay 2
a w +a y /a 2
ρ
× e Γ(−i − j, ), 2l − 1 − 2l − 1
i=0 j=0
i! j! ρ i+j ρ (cos(
2l−1
π) + 1) i+j−1 × e cos( 2n π )+1 | sin( π)|.
2n 2n
(24) (29)
and D( ρa2 ) shares the same form as D( ρ). Thus, the approximated achievable rates (27) and (29) can
Similarly, we can derive the exact expression of Cs2 as be conveniently calculated numerically, and their accuracy will
1 X∞ X ∞ Xn X k be validated by the simulation results in the next section.
Cs2 = Aw Ay Hy (k)n!k! ×
Hw (n) B
B
2 ln(2) IV. N UMERICAL R ESULTS AND S IMULATIONS
k=0 n=0 i=0 j=0
i (a /a ) j a w +a y /a In this section, we compare the analytical results obtained
(i + j)! aw y 2 2 aw + ay /a2
× e ρ Γ(−i − j, ). in the previous Section III with Monte Carlo simulations
i! j! ρi+j ρ to validate their accuracy. Specifically, 105 realizations of
(25)
Rician distribution random variables are generated, and the
Although we have derived the exact expressions of the approximation order for Gauss-Chebyshev Integration is set
achievable rates of s1 and s2 in (23) and (25) respectively, as 100.
such expressions are very complicated, since the incomplete Fig. 2 presents the achievable rate performance of s1 , s2 and
Gamma function is difficult to calculate. Thus, it is still the corresponding sum rate of the NOMA-based CRS against
difficult to get the exact values of the achievable rates, which the power allocation coefficient a2 . In the model of Rician fad-
motivates us to propose an approximation method to solve this ing channel, the parameter Ω denotes the average power gain
problem in the next subsection.

0018-9545 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2017.2728608, IEEE
Transactions on Vehicular Technology
5

6 up for the decrease of s1 ’s achievable rate, which causes the

decrease of the sum rate, as shown in Fig. 2. Thus, there exists
5 an optimal power allocation coefficient to maximize the sum
Average Achievable Rate(bit/s/Hz)

rate, which is an interesting research topic deserving further

4 signal 1 (simulated)
signal 1 (analytical) investigation in the future.
signal 2 (simulated) Fig. 3 compares the achievable rates of the traditional CRS
3 signal 2 (analytical)
sum rate (simulated) and the NOMA-based CRS against the transmit SNR, where
sum rate (analytical) we set a2 = 0.4, ΩSD = 9, ΩRD = 36, and ΩSR = 144. We find
2
that the simulation results and analytical results are consistent,
1
and the NOMA-based CRS achieves higher achievable sum
rate than the traditional CRS, since NOMA-based CRS can
0
transmit two signals in two slots, while traditional CRS can
0 0.1 0.2 0.3 0.4 0.5
Power Allocation Coefficient a2 only transmit one signal during the same time.
Fig. 2. Achievable rates for the NOMA-based CRS over Rician fading V. C ONCLUSIONS
channels. In this paper, we have investigated the performance of a
NOMA-based cooperative relaying system by deriving the
8
exact analytical expressions of the achievable rates. Moreover,
7 an efficient approximation method using Gauss-Chebyshev
Integration for the achievable rates was also proposed, which
6
Achievable Rate (bit/s/Hz)

enables the sum series of the achievable rate expressions

5 converge quickly. Simulation results have verified that our
4
derived analytical results match well with the Monte Carlo
signal 1 (simulated), CRS using NOMA
signal 2 (simulated), CRS using NOMA simulations, and the NOMA-based CRS is able to achieve
3 sum rate (simulated), CRS using NOMA higher achievable rate than the traditional CRS.
signal 1 (analytical), CRS using NOMA
2 signal 2 (analytical), CRS using NOMA R EFERENCES
sum rate (analytical), CRS using NOMA
1 sum rate (simulated), traditional CRS [1] F. Boccardi, R. W. Heath, A. Lozano, T. L. Marzetta, and P. Popovski,
“Five disruptive technology directions for 5G,” IEEE Commun. Mag.,
0 vol. 52, no. 2, pp. 74–80, Feb. 2014.
0 5 10 15 20 25 30
Transmit SNR (dB) [2] L. Dai, B. Wang, Y. Yuan, S. Han, C.-L. I, and Z. Wang, “Non-
orthogonal multiple access for 5G: Solutions, challenges, opportunities,
Fig. 3. Achievable rates comparison between the NOMA based CRS and and future research trends,” IEEE Commun. Mag., vol. 53, no. 9, pp.
traditional CRS. 74–81, Sept. 2015.
[3] F. Fang, H. Zhang, J. Cheng, and V. Leung, “Energy-efficient resource
allocation for downlink non-orthogonal multiple access network,” IEEE
of the channel [13], which is usually determined by distances Trans. Commun., vol. 64, no. 9, pp. 3722–3732, Sept. 2016.
[4] Y. Sun, D. Ng, Z. Ding, and R. Schober, “Optimal joint power and
between transceivers. In our system, Ωi (i ∈ {SD, SR, RD}) subcarrier allocation for full-duplex multicarrier non-orthogonal multiple
denotes the average power gain of link SD, link SR, and link sccess systems,” IEEE Trans. Commun., vol. 65, no. 3, pp. 1077–1091,
RD, mainly reflecting the impacts of distances from S to D, Mar. 2017.
[5] Y. Saito, A. Benjebbour, Y. Kishiyama, and T. Nakamura, “System-level
from S to R, and from R to D, respectively. Thus, we set performance evaluation of downlink non-orthogonal multiple access
ΩSD = 9 < ΩSR = ΩRD = 36 [10], because the distances from (NOMA),” in Proc. IEEE Annu. Symp. PIMRC, London, U.K., Sept.
S to R and from R to D are usually smaller than the distance 2013, pp. 611–615.
[6] L. Liu, C. Yuen, Y. L. Guan, and Y. Li, “Capacity-achieving iterative
from S to D, and thus link SR and link RD have higher average LMMSE detection for MIMO-NOMA systems,” in Proc. IEEE ICC
power gains than link SD. According to [11], other parameters 2016, Kuala Lumpur, Malaysia, May 2016, pp. 1–6.
are set as SNR=20 dB, KSR = KRD = 5, and KSD = 2. From [7] J. N. Laneman, D. N. Tse, and G. W. Wornell, “Cooperative diversity
in wireless networks: Efficient protocols and outage behavior,” IEEE
Fig. 2, we can observe that the derived analytical results using Trans. Inf. Theory, vol. 50, no. 12, pp. 3062–3080, Dec. 2004.
Gauss-Chebyshev Integration match well with the simulation [8] Y. Liu, Z. Ding, M. Elkashlan, and H. V. Poor, “Cooperative non-
results. In addition, as a2 increases, s2 will get more power and orthogonal multiple access with simultaneous wireless information and
power transfer,” IEEE J. Sel. Areas Commun., vol. 34, no. 4, pp. 938–
its achievable rate increases accordingly, while the achievable 953, Apr. 2016.
rate of s1 decreases. Moreover, the sum rate of two signals [9] Z. Ding, M. Peng, and H. V. Poor, “Cooperative non-orthogonal multiple
first increases and then slowly decreases with the increase of access in 5G systems,” IEEE Commun. Lett., vol. 19, no. 8, pp. 1462–
1465, Aug. 2015.
a2 . Actually, we can see from (10) that when a2 is small, [10] J.-B. Kim and I.-H. Lee, “Capacity analysis of cooperative relaying
the achievable rate of s2 is mainly determined by link SR, systems using non-orthogonal multiple access,” IEEE Commun. Lett.,
as a2 λ SR will always be smaller than λ RD . Due to SIC, s2 vol. 19, no. 11, pp. 1949–1952, Nov. 2015.
[11] M. R. Bhatnagar, “On the capacity of decode-and-forward relaying over
will have no interference in link SR, so increasing a2 will rician fading channels,” IEEE Commun. Lett., vol. 17, no. 6, pp. 1100–
largely increase the achievable rate of s2 , and thus increase the 1103, Jun. 2013.
sum rate. However, when a2 increases, a2 λ SR will be larger [12] E. Hildebrand, Introduction to Numerical Analysis. New York, NY, USA:
Dover, 1987.
than λ RD at last, and the rate of link RD will slowly become [13] A. Abdi, C. Tepedelenlioglu, M. Kaveh, and G. Giannakis, “On the
the determinant factor, which is not influenced by a2 . As a estimation of the K parameter for the Rice fading distribution,” IEEE
result, the increase of s2 ’s achievable rate finally cannot make Commun. Lett., vol. 5, no. 3, pp. 92–94, Mar. 2001.

0018-9545 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.