Wireless Networks

https://doi.org/10.1007/s11276-018-1746-8 (0123456789().,-volV)(0123456789().,-volV)

Performance analysis of decode-and-forward partial relay selection

in NOMA systems with RF energy harvesting
Tran Manh Hoang1 • Nguyen Trung Tan1 • Nguyen Huy Hoang1 • Pham Thanh Hiep1

 Springer Science+Business Media, LLC, part of Springer Nature 2018

Abstract
In this paper, we investigate a dual hop communication decode-and-forward scheme relay system where a source node
wants to transmit simultaneously two symbols to two desired destinations with the help of one selected energy constraint
relay node. The power for operation of relay is come from the ambient radio frequency energy harvesting and the non-
orthogonal multiple access technology is applied. We mathematically evaluate the impact of partial relay selection on the
system performance by considering the undecodable probability of symbol for which symbols can not be decoded at the
relay node or two destinations. Furthermore, the undecodable probability and the ergodic capacity are analyzed under the
effect of imperfect and perfect successive interference cancellation (SIC). The results of theoretical analysis is similar to
the simulation results, especially in the high region of transmit power. It verifies the correctness of mathmatical closed-
form in our analysis. The results also show that the performance of the system are significantly influenced by the efficiency
of SIC technology and the number of relay nodes.

Keywords NOMA  Partial relay section  Energy harvesting  Perfect and imperfect successive interference cancellation

1 Introduction envisioned as a key component in fifth generation (5G)

mobile systems, it has also been applied to cooperative
Non-orthogonal multiple access (NOMA) has been proven relay systems [3, 4]. In [3], the cooperative NOMA system
as a promising technique for the fifth generation (5G) with buffer-aided relaying was studied. Assuming that the
mobile networks due to its superior spectral efficiency [1]. relay node possesses a buffer, the authors proposed an
Unlike orthogonal-multiple-access (OMA), NOMA allows adaptive transmission scheme in which the system adap-
multiple users to pair and share the same radio resources, tively chooses its working mode in each time slot. The
either in time, frequency or in code. The key idea of authors in [4] proposed and investigated a dual-hop coop-
NOMA is to explore power domain, where users are served erative relaying scheme using NOMA, where two sources
at different power levels [2]. More importantly, NOMA has communicate with their corresponding destinations simul-
recently been recognized as a promising multiple access taneously over the same frequency band and via a common
(MA) technique to significantly improve the spectral effi- relay. After receiving symbols transmitted from both
ciency of mobile communication networks. It is also been sources with different allocated powers, the relay forwards
a superposition coded composite signal to the destinations
& Pham Thanh Hiep by using NOMA technique.
phamthanhhiep@gmail.com On the other hand, despite of advantages of simultane-
Tran Manh Hoang ous wireless information and power transfer (SWIPT), the
tranmanhhoang@tcu.edu.vn research on it is lack, especially studying of join of SWIPT
Nguyen Trung Tan and NOMA relay systems. There are several researches on
trungtan68@gmail.com NOMA systems combining radio frequency (RF) energy
Nguyen Huy Hoang harvesting (EH) as follows. The general combination of
hoangnh@mta.edu.vn NOMA and EH was investigated in [5–7]. SWIPT of
1
NOMA networks where users are spatial randomly located,
Le Quy Don Technical University, Hanoi, Vietnam

123
 Wireless Networks

was also investigated [8]. The system in which the users denotes the complex channel coefficient between Rn and
closer to the source act as EH relays to help the users Di . Since the path loss and shadowing effect of g1 are more
farther was analyzed. The outage performance of NOMA- severe than g2 , we have XRn D1 \XRn D2 which is important
EH relaying networks with antenna selection at the source for performing the successive interference cancellation
and splitting power at the relay was discussed in [7]. To the (SIC) [10]. The relay and the user respectively receive the
best of our knowledge, the aforementioned work, the EH signal from the source and the best relay, hence the
on the partial relay selection (PRS) combining with NOMA channel state information (CSI) between the source and the
technique, has not been discussed. Therefore, we decide the relay is available at the relay, the CSI between the best
main contributions of this paper are as follows: relay and the user is available at the user.
In this paper, we consider the PRS scheme where relay
• We propose a NOMA system where the relay node
node is selected based only on the instantaneous knowl-
harvests energy from the radio frequency by using time
edge of the channel pertaining to the first hop. The source
switching scheme to support the operation of forward-
terminal continuously monitors the quality of its connec-
ing information to both of users.
tivity with the relays via the transmission of local feed-
• In order to evaluate the performance of partial selection
backs. From this information, the source selects the best
relay scheme, we derive closed-form expressions of
link S ! Rn for data transmission. As multiple relay nodes
undecodable probability and ergodic capacity. Unde-
form a group, one best relay Rb is selected before trans-
codable probability is defined as probability the desired
mitting. This PRS strategy is expressed as1
information at the best relay node and/or the destination
nodes can not be decoded successfully because of low b ¼ arg max jh1;n j2 : ð1Þ
n¼1; 2...;N
signal to noise ratio (SNR).
• The performance of NOMA-EH system under the We assume that the harvested power is consumed by the
imperfect SIC and perfect SIC is analyzed and com- relays for forwarding signals to D1 and D2 . The processing
pared with each other and with the orthogonal power required for the transmitting - receiving circuitry of
frequency division multiple access (OFDMA). relay is generally negligible compared to the power used
• According to the channel gains, the detailed power for signal transmission. The time switching (TS) architec-
allocation coefficient for is settled for fairness in ture for EH [11]2 is applied. Specifically, the energy is
performance of users harvested from the received information signal for a
The rest of this paper is organized as follows Sect. 2 duration of aT in each block, where a is the fraction of the
describes the system model of partial relay selection in block time in which the relay harvests energy from the
NOMA system with RF EH. The mathematical analysis of received information signal, and T is the block time in
the system is represented in Sect. 3. Numerical results are which a certain information is transmitted from the source
shown in Sect. 4 to examine the performance of our pro- to the users. Hence, the harvested energy is given by [12].
 2
posed system. Finally, the paper is concluded in Sect. 5. Eh ¼ aTgPS h1;b  ; ð2Þ

where gis the energy conversion efficiency, depends on the

2 System model quality of energy harvesting electric circuitry, 0\g  1,
and PS is transmission power of S.
We assume that the duration time for relay to receive the
We consider a NOMA downlink system consisted of one signal from the source and to transmit the received signal to
source (S), two users ( D1 and D2 ) as shown in Fig. 1. The both of users is the same and equals ð1  aÞT=2. From (2),
source transmits the signal to the users via several relay the transmission power of relay node is calculated as
nodes Rn with n 2 f1; . . .; N g which are equipped one  2
Eh 2agPS h1;b 
antenna and operate in a half-duplex mode. In this paper, PR ¼ ¼ : ð3Þ
ð1  aÞT=2 1a
we only consider the case of partial relay selection.
Moreover, all relays have no fixed power supply and are The operation on the system is briefly described as
powered by wireless energy transfer from the source. The follows. There are two time slots involving in each trans-
channels from the S to Rn and from Rn to Di with mission block. All blocks are normalized to one unit.
i 2 f1; 2g, exhibit frequency, flat Rayleigh block fading.
As shown in Fig. 1, h1;n  CN ð0; X1;n Þ is the complex 1
The sequence of X are referred to as order statistics of the sequence
channel coefficient between the S and Rn . of X1;n .
wR ; wDi  CN ð0; 1Þ are the additive white Gaussian noise 2
The proposed analysis approach can be applied for the power
(AWGN) at Rn and Di , respectively. gi  CN ð0; XRn Di Þ spitting EH model.

123
Wireless Networks

pffiffiffiffiffiffi pffiffiffiffiffi pffiffiffiffiffi

yD i ¼ PR gi ð a1 x1 þ a2 x2 Þ þ wDi ; ð7Þ

where gi denotes the channel gain between the selected

relay and Di .
According to the SIC principle, the D1 decodes symbol
x1 while treating x2 as noise. From (7), the signal-to-in-
terference-plus-noise ratio (SINR) at D1 is given by
a 1 PR j g 1 j 2
c2u1 ¼ : ð8Þ
Fig. 1 Partial relaying selection in NOMA downlink relay network a 2 PR j g 1 j 2 þ 1
with RF EH
It should be noticed that x1 and x2 coexist at D2 .
During the first time slot, the source node transmits the Therefore, the SIC is needed to decode own symbol x2 . To
pffiffiffiffiffi pffiffiffiffiffi perform SIC, D2 decodes high-power symbol x1 by treating
superimposed mixture, a1 x1 þ a2 x2 , where xi and ai
denote the signal and the power allocation coefficient of the low-power symbol x2 as noise, and cancels it using SIC
user i, respectively. It should be noted that a1 þ a2 ¼ 1, to obtain symbol x2 . Thus, the received SINR for x1 at D2 is
and without loss of generality, we assume that a1  a2 [13]. given by
The baseband received signal at Rn is thus given by a1 PR jg2 j2
pffiffiffiffiffiffi pffiffiffiffiffi pffiffiffiffiffi c2u1 !u2 ¼ : ð9Þ
ynR ¼ PS h1;n ð a1 x1 þ a2 x2 Þ þ wR : ð4Þ a2 PR jg2 j2 þ 1
Since the key idea of NOMA technique is to use the After cencelling x1 , D2 decodes its own message, x2 ,
power domain for multiple accesses, i.e. users are served at with the SNR as
different power levels, and to adopt the superposition code
c2u2 ¼ a2 PR jg2 j2 : ð10Þ
at the transmitter. We use the SIC principle to decode
signals at the receivers, such as the stronger signal is
decoded and removed from the superposed signals, and
then the weaker signal is decoded to obtain the desired 3 Performance analysis
information. Thus, in this paper, we only consider SIC
process at the best relay node and the D2 because the D1 , In this section, we discuss on undecodable probability and
which is allocated higher transmit power, decodes the ergodic capacity of partial relay selection scheme NOMA
superimposed mixture signal and obtains its own signal system with RF EH. The undecodable probability can be
while considering the other signal as noise. refered as the probability that symbols x1 and x2 are not
When the SIC is employed at the Rn , the best relay, Rb , able to be decoded at D1 and D2 . In additional, a symbol is
firstly decodes symbol x1 by treating the symbol x2 as not able to be decoded if the received SNR or SINR is
noise. Then, it performs SIC to obtain symbol x2 . Thus, the lower than the threshold value, c.
received signal-to-interference-plus-noise ratio (SINR) for
symbol x1 and signal-to-noise ratio (SNR) for symbol x2 at 3.1 Undecodable probability of x1
Rb are written as
 2 We denote Perror as the event that the best relay or D1 or D2
a1 PS h1;b  x1
c1u1 ¼  2 ; ð5Þ cannot decode x1 successfully because the SINR/SNR is
a2 PS h1;b  þ 1 lower than the threshold value, cth1 . From (5), (8), and (9),
 2 we can write the expression of Perror as in (11) [14],
c1u2 ¼ a2 PS h1;b  : ð6Þ x1

In the first time slot, Rn processes messages x1 , x2 and in

the second time slot, the selected relay sends the message
pffiffiffiffiffiffipffiffiffiffiffi pffiffiffiffiffi 
PR a1 x1 þ a2 x2 to two users D1 and D2 . Hence, the
received signal at Di is described as follows.

123
 Wireless Networks

8  2 9
Z1 n
>
>
> a1 PS h1;b  >
> t2 t2 o
>
>
>  2 \cth1 ; > >
>
> Perror
x1 ¼1 Pr Y [ ; Z [ fX ð xÞdx
>
> a P h1;b  þ 1 >
> x x
>
>
2 S >
> t1
< 2 = ð13Þ
error
Px1 ¼ Pr or a 1 P R j g 1 j Z1 Z1 h t i
> 2
\c th1 >; 2
> > ¼1 1  FY fZ ðzÞfX ð xÞdxdz;
> a 2 PR j g 1 j þ 1
>
>
>
>
> x
>
> 2 >
>
>
> a P
1 R 2 j g j >
> t1 t2
>
: or 2
\c th1 > ;
x

a2 PR j g 2 j þ 1
8  2 9 ð11Þ where
>
>   >
>
> a1 P S h1;b
>  [ c ; >
> cth1 cth1
>
>   2 th1 >
> t1 ¼ and t2 ¼ :
>
> a P h þ 1 >
> PS ða1  a2 cth1 Þ PS /ða1  a2 cth1 Þ
>
>
2 S 1;b >
>
< 2 =
¼ 1  Pr a P
1 R 1 j g j From (13) and after some manipulation, we have
> 2
[ cth1 ; >;
>
> a 2 P R jg 1 j þ 1 >
>   Z1  
>
> >
> XN
>
> 2 >
> n1 N n b nx
>
> a 1 P R jg 2 j >
> Perror ¼1 ð1Þ exp   dx ;
>
: 2
[ cth1 > ;
x1
n X1 x X1
a2 PR jg2 j þ 1 n¼1
t1
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl
ffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
2r1 w
where cth1 ¼ 21a  1 and r1 is the target data rate of D1 .
 2 ð14Þ
Let h1;b  ¼ X, jg1 j2 ¼ Y, and jg2 j2 ¼ Z be the channel
gains from the source to the best relay and from the best where b ¼ Xt22 þ Xt23 . Unfortunately, a closed-form expres-
relay to D1 and D2 , respectively. In this paper, we assume sion for w in (14) is difficult to be derived, thus we employ
that all channel coefficients are modeled as independent the approximation method. By using expanded Taylor’s
Rayleigh-distributed random variables (RVs). Therefore,   P t ð1Þk ak
series, it follows that exp  ax ¼ Nk¼0 k! x , where
RVs jh1;b j2 , jg1 j2 , and jg2 j2 have exponential distributions, Nt 2 f1; . . .; 1g. Then, we obtain
with
  XNt Z1  k  
XN ð1Þk b nx
n1 N n nx w¼ exp  dx
fX ðxÞ ¼ ð1Þ expð Þ; k! x X1
n¼1 n X1 X1 k¼0
t1 ð15Þ
1 y y X
Nt k k k1  
fY ðyÞ ¼ expð Þ; FY ðyÞ ¼ 1  expð Þ; ð1Þ b 1 nt1
X2 X2 X2 ¼ Ek ;
k¼0
k! t1 X1
1 z
fZ ðzÞ ¼ expð Þ:
X3 X3 where Ek ðÞ is the exponential integral function [9]. Sub-
where f and F are respectively the cumulative density stituting (15) into (14), we get the expression of Perror
x1 as
function and probability density function, X1 ¼ XN X      
Nt
ð1Þn ð1Þk bk N n 1 k1 nt1
Perror
x1 ¼1 Ek :
Efjh1;b j2 g; X2 ¼ Efjg1 j2 g and X3 ¼ Efjg2 j2 g are means of k! n X1 t1 X1
n¼0 k¼0
random variables. We can rewrite (11) to become (12)
8 9 ð16Þ
> cth1 >
>
> X[ >
>
> PS ða1  a2 cth1 Þ >>
>
As we can see from (14), if the transmit power lies in a
>
< >
= high region, it leads to t1 ¼ PS 1 ðac1th1a2 c Þ ! 0. Thus, we
error
c th1
Px1 ¼ 1  Pr XY [ ; ð12Þ th1

>
> PS /ða1  a2 cth1 Þ >
> apply [9, (3.324.1)] to obtain the final expression of Perror
x1
>
> cth1 >
>
>
> >
> as
: XZ [ ;
PS /ða1  a2 cth1 Þ
XN  sffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffi!
N 4nb 4nb
2ag
where / ¼ 1a . As can be seen from (12), the outage Perror
x1 1 ð1Þn1 K1 ; ð17Þ
n¼1 n X1 X1
always occurs if cth1 [ aa12 . Hence, we need to allocate more
power for symbol x1 meaning a1 [ a2 cth1 is required. By where K1 ðÞ is the first-order modified Bessel function of
using the conditional probability in [15], we can rewrite the second kind.
(12) as

123
Wireless Networks

3.2 Undecodable probability of x2 3.2.2 In case of imperfect SIC

In this paper, we consider two scenarios, the first scenario In this scenario, the symbol x1 is incompletely removed
is perfect SIC, and the second one is imperfect SIC at both and becomes the interference. The SINRs of symbol x2 at
Rn and D2 . Rb and D2 are respectively given as
 2
a2 PS h1;b 
3.2.1 In case of perfect SIC c1u2 ¼  2 ; ð23Þ
a1 q1 PS h1;b  þ 1
In this NOMA system, the SIC technique is carried out at
a2 PR jg2 j2
D2 to remove signal x1 before detecting its own message. c2u2 ¼ ; ð24Þ
In case SIC is perfect at D2 and Perror is denoted as the a1 q2 PR jg2 j2 þ 1
x2
event that the best relay or D2 cannot correctly decode where 0  qi  1 with i 2 f1; 2g denotes the level of
symbol x2 because the SNR is lower than the threshold residual interference due to the imperfect SIC at Rb and D2 .
2r2
value, cth2 ¼ 21a  1 where r2 is the target data rate of D2 . Especially, qi ¼ 1 and qi ¼ 0 refer to the cases of without
Then, the expression of Perror
x2 is given by SIC and perfect SIC, respectively. From (23) and (24), we
have the undecodable probability of x2 for the case of
Perror
x2 ¼ Prðminðc1u2 ; c2u2 Þ  cth2 Þ imperfect SIC shown in (25).
  2
¼ 1  Pr a2 PS h1;b  [ cth2 ; a2 PR jg2 j2 [ cth2 : "  2 ! #
a2 PS h1;b  a2 PR jg2 j2
PISIC
x2 ¼ Pr min  2 ; 2
 cth2 :
ð18Þ a1 PS q1 h1;b  þ 1 a1 PR q2 jg2 j þ 1

Substituting PR from (3) into (18), we can rewrite as ð25Þ

 
cth2 cth2 ð1  aÞ After solving (25), we get the result as
Perror
x2 ¼ 1  Pr X [ ; Z [ : ð19Þ
a2 P S X2a2 agPS Z1  
d
Based on the conditional probability [15], we can express PISIC
x2 ¼1 1  FZ fX ð xÞdx; ð26Þ
x
(19) as c

Z1   where
b
Perror
x2 ¼1 1  FZ fX ð xÞdx; ð20Þ cth2
x c¼ ;
a PS ða2  a1 q1 cth2 Þ
cth2
th2 ð1aÞ
where a ¼ ac2th2PS and b ¼ c2a . By substituting the d¼ :
2 agPS /PS ða2  a1 q2 cth2 Þ
cumulative distribution functions (CDF) and probability
density function (PDF) of X and Y into (20) and then using Then, with some further manipulations, we can rewrite (26)
expanded Taylor series, we obtain the expression of Perror into (27).
x2
as in (21). X
N   Z1    
N n d nx
PISIC
x2 ¼1 ð1Þn1 exp  exp  dx
  Z1     n X1 X3 x X1
X
N
N n b nx n¼1
c
Perror
x2 ¼ 1  ð1Þn1 exp  exp  dx      k1  
n X1 X3 x X1 X
N
N n X Nt
ð1Þk d k 1 nc
n¼1
a ¼1 ð1Þn1 Ek :
  X      n X1 k¼0 k! X3 c X1
X
N
N n Nt ð1Þk b k 1 k1 na n¼1
¼1 ð1Þn1 Ek : ð27Þ
n¼1 n X 1 k¼0 k! X 3 a X 1

ð21Þ In high-power region, we have the approximate expression

of PISIC
x2 as
Similar to undecodable probability of x1 , when the transmit
power lies in a high region, it will lead to a ¼ a2 Pcth21 ! 0. XN  rffiffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffi
S ISIC n1 N 4nd 4nd
error
Hence, the approximate expression of Px2 is given by Px2  1  ð1Þ K1 :
n¼1 n X1 X3 X1 X3
XN  rffiffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffi
error n1 N 4bn 4bn ð28Þ
Px2  1  ð1Þ K1 :
n¼1 n X1 X3 X1 X3
ð22Þ

123
 Wireless Networks

3.3 Ergodic capacity X

N  sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi!
n1 N 4nn 4nn
FP ðnÞ ¼ 1  ð1Þ K1 :
n¼1 n a2 /PS X1 X2 a2 /PS X1 X2
3.3.1 In case of perfect SIC
ð34Þ
In this section, we investigate the ergodic capacity of the With the help of the [9, eq. (7.811.5), (9.34.3)] and after
system, the achievable sum rate of the NOMA downlink is some manipulation we have.
given as   0 !
 x1 þ C
sum ¼ C  x2 : 1  aX N
n1 N 31 n 

C ð29Þ I1 ¼ ð1Þ G13 :
D1 D2
2 ln 2 n¼1 n /PS X1 X2 0;3; 1
4 4
The achievable capacity at the D1 is given as ð35Þ
1a  0 1
CxD11 ¼ log2 ð1 þ min c1u1 ; c2u1 Þ   0
2 ( )! 1  aX N
n1 N 31 @ n 
 A:
 2 I2 ¼ ð1Þ G13
¼
1a
log2 1 þ min
a1 PS h1;b 
;
a1 PR jg1 j2
:
2 ln 2 n¼1 n a2 /PS X1 X2 0; 3; 1
 2 2 4 4
2 a2 PS h1;b  þ 1 a2 PR jg1 j þ 1
ð36Þ
ð30Þ
The achievable capacity at the D2 is given as
In practice, the amount of harvested energy at the relay is
always small, hence the transmit power of relay is much 1a  
CDx22 ¼ log2 1 þ minfc1u1 ; c2u2 g : ð37Þ
lower than that of the source, and then the assumption that 2
SINR at the destinations is lower than the SINR at the relay Similarly, we get the ergodic capacity of the D2 is given as
is reasonable.   0 !
 2 x2 1 aX N
n1 N 31 nb 
a1 PS h1;b  a1 PR j g 1 j 2 CD2 ¼ ð1Þ G13 : ð38Þ
 2 [ : 2 ln 2 n¼1 n X1 X3 0;3; 1
4 4
a2 PS h1;b  þ 1 a 2 PR j g 1 j 2 þ 1
Combining Eqs. (35), (36) and (41) we have the sum
Hence, from (30) we have ergodic capacity of the EH-NOMA downlink relay system.
!
1a PR j g 1 j 2 a 1
CxD11 ¼ log2 1 þ 3.3.2 In case of imperfect SIC
2 PR j g 1 j 2 a 2 þ 1
 2 ! ð31Þ
1a /PS h1;b  jg1 j2 þ 1 The ergodic capacity of imperfect SIC is similar to that of
¼ log2  2 : perfect SIC.
2 /PS h1;b  jg1 j2 a2 þ 1
ISIC ¼ C
C x2 =ISIC :
 x1 þ C ð39Þ
sum D1 D2
With the instantaneous capacity is derived in (31), we
x
 1 is derived by Eqs. (30)–(36) in the above
obtain the ergodic capacity of the D1 as where the C D1
   section.
1a  2 2
 x1
CD1 ¼ E 
log2 1 þ /PS h1;b jg1 j  x2 =ISIC can be given as
2 The C D2
    
1a  2 x =ISIC 1a
E log2 1 þ a2 /PS h1;b  jg1 j2 CD22 ¼ log2 1 þ minfc1;u2 ; c2;u2 g ; ð40Þ
2 2
Z1 Z1 From (23) and (24) we have
1a 1FQ ðnÞ 1a 1FP ðnÞ
¼ dn  dn; "  2 !#
2 ln 2 1þn 2 ln 2 1þn x =ISIC 1a a2 PS h1;n  a2 PR jg2 j2
0 0 CD22 ¼ log2 1 þ min  2 ;
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} 2 2
a1 PS q1 h1;n  þ 1 a1 PR q2 jg2 j þ 1
I1 I2 " !#
1a a2 PR jg2 j2
ð32Þ ¼ log2 1 þ :
2 a1 PR q2 jg2 j2 þ 1
 2 2  2 2
where Q ¼ /PS h1;b  jg1 j and P ¼ a2 /PS h1;b  jg1 j . ð41Þ
Based on the condition probability we have the CDFs of Hence, the ergodic capacity of the x2 in case of imperfect
Q and P, respectively. SIC can be rewritten as
X
N  sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi!
N 4nn 4nn
FQ ðnÞ ¼ 1  ð1Þ n1
K1 : ð33Þ
n¼1 n /PS X1 X2 /PS X1 X2

123
Wireless Networks

x =ISIC 1a h  i 100

CD22 ¼ E log2 1 þ PR jg2 j2 ða1 q2 þ a2 Þ
2

Undecodable probability of symbol x1

1a h  i
 E log2 1 þ a1 q2 PR jg2 j2
2 N = 1, 2, 3
ð42Þ 10-1
Z1 Z1
x =ISIC 1  a 1  FU ðuÞ 1  a 1  FV ðvÞ
CD22 ¼ du  dv
2 ln 2 1þu 2 ln 2 1þv
0 0
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} 10-2
J1 J2

ð43Þ
Simulation
2 2 2 2 Analysis, Approximation
where U ¼ /PS jh1;n j jg2 j j, and V ¼ a1 q2 /PS jh1;n j jg2 j Analysis, Exact
2a
with j ¼ a1 q2 þ a2 , / ¼ 1a . 10-3
0 5 10 15 20 25 30 35 40
Based on the condition probability, the CDFs of U and Eb/N0 [dB]
V is respectively represented as.
XN   Fig. 2 Undecodable probability of x1 for the perfect SIC and different
N
FU ðuÞ ¼ 1  ð1Þn1 number of the relay nodes, N
n¼1 n
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi!
4nu 4nu 10 0
 K1 :
ða1 q2 þ a2 Þ/PS X1 X3 ða1 q2 þ a2 Þ/PS X1 X3
Undecodable probability of symbol x 2
ð44Þ
X
N  sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi!
N = 1, 2, 3
N 4nv 4nu
FV ðvÞ ¼ 1  ð1Þn1 K1 : 10 -1
n¼1 n a1 q2 /PS X1 X3 a1 q2 /PS X1 X3

ð45Þ
Replace (44) and (45) into (43) we have
  10 -2
1  aX N
n1 N
J1 ¼ ð1Þ
2 ln 2 n¼1 n Simulation
0 ! ð46Þ Analysis, Exact
n  Analysis, Approximation
31
G13  : 10 -3
ða1 q2 þ a2 Þ/PS X1 X3 0;3; 1 0 5 10 15 20 25 30 35 40
4 4
! E b /N0 [dB]
  0
1  aX N
N n 

J2 ¼ ð1Þn1 G31 :
2 ln 2 n¼1 n 13
a1 q2 /PS X1 X3 0;3; 1 Fig. 3 Undecodable probability of x2 for the perfect SIC and different
4 4 number of the relay nodes, N
ð47Þ
Let b and ð1  bÞ Hz denote the bandwidth assigned for
Finally, we have the ergodic capacity of the system in
D1 and the remaining bandwidth assigned for D2 , respec-
imperfect SIC as
tively, where ð0  b  1ÞHz. By using Equ. (7.4) of [13],
ISIC ¼ I1  I2 þ J1  J2:
C ð48Þ the sum capacity of the OMA system is given by
sum
" !#
where I1 ; I2 is given in Eqs. (35) and (36), respectively. 1a PS jh1 j2 PR jg1 j2
COMA ¼ b log2 1 þ min ;
2 b b
" !#
3.3.3 Ergodic capacity of OFDMA ð1  aÞð1  bÞ PS jh1 j2 PR jg2 j2
þ log2 1 þ min ; :
2 1b 1b
In order to compare with orthogonal frequency division
ð49Þ
multiple access (OFDMA), which is a well-known high-
capacity orthogonal multiple access (OMA) technique, the
ergodic capacity of OFDMA is described in this section
and the calculation result is compared in the following
section.

123
 Wireless Networks

10 0 The result calculated by Eqs. (16) and (21) is named as

Analysis-Exact, whereas the result calculated by using
Undecodablel probability of symbol x 2

closed-form, Eqs. (17)and (22),is named as Analysis-Ap-

N = 1, 2, 3
proximation. In this scenario, we consider the PRS
10 -1 scheme in which the number of relays is varied. It is rec-
1
= 0.01, 2
= 0.04 ognized from Figs. 2 and 3 that the performance gain is
increased by increasing the number of relays from 1 to 3.
The reason is the selected best relay provides the best
channel from the source to the relay in order to achieve
10 -2
Simulation imperfect SIC better decoding performance as well as higher RF EH from
Analysis imperfect SIC
Simulation perfect SIC
the source in the first phase. We can see that the simula-
Analysis perfect SIC tions results are in excellent agreement with the analytical
results in all cases.
10 -3
0 5 10 15 20 25 30 35 40 Figure 4 shows the comparison of the undecodable
Eb /N0 [dB] probability of x2 versus the average SNR for the cases of
perfect and imperfect SIC with the levels of residual
Fig. 4 Comparison of undecodable probability of x2 for the perfect interference, q, at the best relay and D2 are q1 ¼ 0:01 and
and imperfect SIC q2 ¼ 0:04, respectively. We should be reminded that D1
should decode only symbol x1 , while D2 has to correctly
4 Simulation results decode x1 first, and then use the SIC to obtain x2 . The
undecodable probability of symbol x2 is plotted according
In this section, several representative numerical results are to the average SNR in dB. The round and square markers
provided to evaluate the impact of the number of relays, the indicate the simulation results of the perfect and the
channel gain (distance or path loss), and the imperfect SIC imperfect SIC, respectively. From Fig. 4, we can know that
on the performance of NOMA systems with RF EH. The the remaining coefficient SIC significantly impacts the
system parameters are as follows. The target data rated performance of D2 . It is noticed that the simulation results
r1 ¼ 0:5 ðbpcuÞ and r2 ¼ 1 ðbpcuÞ, the energy harvesting are again in excellent agreement with the analytical results
fraction a ¼ 0:3. The channel gains XSRb ¼ XRb D1 ¼ 1 and in all cases, which validates our mathematical models.
XRb D2 ¼ 2, and the energy conversion efficiency g ¼ 1. We also investigate the impact of fraction of block time,
X Rb D 1
The power allocation coefficient satisfies X Rb D 2 ¼ aa21 which a, on the undecodable probability of symbol x2 with dif-
is used to keep the trade-off between system throughput ferent number of relay nodes, and depict the results in
and user fairness. Fig. 5. In this scenario, we assume that the SIC is perfect
Figures 2 and 3 plot the undecodable probability of and the transmit power from the source PS ¼ 20 dB. As
symbols x1 and x2 , respectively for the case of perfect SIC. shown in Fig. 5, there is the optimal a in the sense of

4.5
10 0 Simulation User 1
4 Analysis User 1
Simulation User 2
Undecodable probability of symbol x 2

Ergodic Capacity [bit/s/Hz]

3.5 Analysis User 2

P S = 20 [dB] Simulation Sum
Analysis Sum
3
N = 1, 2, 3

2.5

10 -1 2

1.5 N=1

0.5

0
10 -2 0 5 10 15 20 25 30 35 40
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Eb /N0 [dB]

Fig. 5 Impact of fraction of block time, a, on the undecodable Fig. 6 Ergodic capacity versus SNR with number of relay nodes N ¼
probability of symbol x2 1 and perfect SIC

123
Wireless Networks

5 relays. However, the sum of ergodic capacity is just

Simulation perfect SIC slightly improved when the number of relay nodes is high.
4.5 Analysis perfect SIC
Analysic imperfect SIC Compared with perfect SIC, the ergodic sum capacity of
4 Simulation imperfect SIC imperfect SIC is lower, and the gap of them increases when
Ergodic sum capacity [bit/s/Hz]

Sim OMA the Eb =N0 increases. The reason is, while Eb =N0 increas-
3.5
ing, the interference term due to imperfect SIC increases,
3 therefore the SINR as well as the ergodic capacity slowly
2.5
increase.
N=1, 2, 3 The ergodic capacity of OMA-EH is also depicted in
2 Fig. 7. For OMA, the equal power and equal bandwidth
1.5
are applied for each user. As shown in Fig. 7, in case of
OMA Sim N = 1, perfect SIC, NOMA outperforms OMA because the SNR of
1 2, OMA is higher than that of NOMA, however the bandwidth
3
0.5
is just a half of NOMA and the capacity is linear with
bandwidth. In case of imperfect SIC, the OMA outperforms
0
0 5 10 15 20 25 30 35 40 NOMA when NEb0 is high because of extremely decreasing of
E /N [dB] SINR of x2
b 0

Fig. 7 Sum of ergodic capacity of two users with PRS scheme when
the number of relays changes
5 Conclusions
minimal undecodable probability of symbol x2 , and these
In this paper, we have analyzed the influence of EH tech-
minimal undecodable probabilities as well as the optimal a
nique in NOMA relay networks. The PRS scheme was
depend on the number of relay nodes, N. When N increases,
applied to improve the system performance. The closed-
the minimal undecodable probability becomes smaller
form expression of the probability of unsuccessful decod-
while the optimal a becomes larger. The reason is, when
ing symbol x1 , x2 and ergodic capacity were derived. The
N increases, the SINR of symbol x2 , c1u2 , also increases
theoretical analysis was verified by the Monte Carlo sim-
because of the application of PRS. Thus, the data rate
ulation method. From the theoretical and simulation
between the source and the best relay increases, and then
results, some important conclusions can be summarized as
the transmission time from the source to the best relay
follows. (1) The represented simulation results verified the
becomes shorter meaning the duration time for EH, a, is
correctness of our mathematical analysis. (2) The unde-
extended. With the larger a, the transmission power of the
codable probability and the ergodic capacity were
best relay is higher, and then the SINR of symbol x2 at D2
improved by increasing the number of relay nodes, how-
also becomes better. Consequently, according to (18), the
ever the large number of relay nodes is unnecessary for
undecodable probability of symbol x2 decreases when the
significantly improving the system performance. (3) There
SINR of both the best relay and D2 increases.
is the optimal duration time of EH for the minimal unde-
In Fig. 6, we plot the ergodic capacity of D1 and D2
codable probability and it is directly proportional to the
versus Eb =N0 . We can see that the analysis result of the
number of relay nodes. (4) The proposed system was
ergodic capacity for D1 matches excellently with the sim-
analyzed in both perfect and imperfect SIC, and compared
ulation result at the high SNR. The reason is that, we
with OFDMA system, the impact of residual interference
changed the instantaneous capacity of the information
was discussed.
symbol x1 from (30) into (31) under condition of the
As future works, we will apply NOMA into multiple-
medium and high region power of the source. In addition,
relay multiple-antenna EH relay networks and analyze the
the upper bound curve of the ergodic capacity for D2 and
network performance over Nakagami-m fading channel or
sum of the ergodic capacity for all users are close to the
double Rayleigh fading channel. In addition, we will
simulation results due to approximation of the CDF of the
investigate the impact of imperfect CSI on the performance
information symbol x2 from (22).
of relay selection method.
In Fig. 7, the impact of the number of relay nodes on
SWIPT-NOMA is depicted through the sum of ergodic Acknowledgements The authors would like to thanks Dr. Tran Trung
capacity of D1 and D2 , for several number of relays, Duy for his comment to improve quality of this page.
N ¼ 1; 2; 3. We can undestand that, the sum of ergodic
capacity can be improved by setting up more number of

123
 Wireless Networks

References Tran Manh Hoang was born in

07/11/1977. He is received the
1. Wang, Y., Ren, B., Sun, S., Kang, S., & Yue, X. (2016). Analysis B.S. degree in Communication
of non-orthogonal multiple access for 5G. China Communica- Command at Telecommunica-
tions, 13(Supplement 2), 52–66. tions University, Ministry of
2. Dai, L., Wang, B., Yuan, Y., Han, S., Chih-Lin, I., & Wang, Z. Defense, Vietnam, in 2002, and
(2015). Non-orthogonal multiple access for 5G: Solutions, chal- the B.E. degree in Electrical
lenges, opportunities, and future research trends. IEEE Commu- Engineering from Le Quy Don
nications Magazine, 53(9), 74–81. Technical University, Ha Noi,
3. Luo, S., & Teh, K. C. (2017). Adaptive transmission for coop- Vietnam, in 2006. He obtained
erative NOMA system with buffer-aided relaying. IEEE Com- the M.E. degree in Electronics
munications Letters, 21(4), 937–940. Engineering from Posts and
4. Kader, M. F., Shahab, M. B., & Shin, S.-Y. (2017). Exploiting Telecommunications, Institute
non-orthogonal multiple access in cooperative relay sharing’’. of Technology, (VNPT), Viet-
IEEE Communications Letters, 21(5), 1159–1162. nam, in 2013. He is currently
5. Du, C., Chen, X., & Lei, L. (2017). Energy-efficient optimisation pursuing the Ph.D. degree at Le Quy Don Technical University,
for secrecy wireless information and power transfer in massive Hanoi, Vietnam. His research interests include energy harvesting,
MIMO relaying systems. IET Communications, 11(1), 10–16. Non-orthogonal Multiple Access, and signal processing for wireless
6. Sun, R., Wang, Y., Wang, X., & Zhang, Y. (2016). Transceiver cooperative communications.
design for cooperative non-orthogonal multiple access systems
with wireless energy transfer. IET Communications, 10(15), Nguyen Trung Tan was born in
1947–1955. Ha Nam, Viet Nam. He received
7. Han, W., Ge, J., & Men, J. (2016). Performance analysis for the B.S. in 2004 and M.S. in
NOMA energy harvesting relaying networks with transmit 2007 in Electrical Engineering
antenna selection and maximal-ratio combining over Nakagami- from Le Quy Don Technical
m fading. IET Communications, 10(18), 2687–2693. University, Ha Noi, Vietnam.
8. Liu, Y., Ding, Z., Elkashlan, M., & Poor, H. V. (2016). Coop- He subsequently received the
erative non-orthogonal multiple access with simultaneous wire- Ph.D. degree in 2013 in Elec-
less information and power transfer. IEEE Journal on Selected trical Engineering from Harbin
Areas in Communications, 34(4), 938–953. Institute of Technology, China.
9. Zwillinger, D. (2014). Table of integrals, series, and products. After his Ph.D., he worked at
Amsterdam: Elsevier. Center for Telecommunication
10. Pedersen, K. I., Kolding, T. E., Seskar, I., & Holtzman, J. M. Engineering, Le Quy Don
(1996). Practical implementation of successive interference can- Technical University, Vietnam.
cellation in DS/CDMA systems. In IEEE international confer- His research interests lie within
ence on universal personal communications, 1996. Record (Vol. the areas of wireless communication, cooperative communication,
1, pp. 321–325). IEEE cognitive radio.
11. Gu, Y., & Aı̈ssa, S. (2015). RF-based energy harvesting in
decode-and-forward relaying systems: Ergodic and outage Nguyen Huy Hoang received
capacities. IEEE Transactions on Wireless Communications, B.E., M.E. and Ph.D. degree in
14(11), 6425–6434. Electronic Engineering from Le
12. Michalopoulos, D. S., Suraweera, H. A., & Schober, R. (2015). Quy Don Technical University,
Relay selection for simultaneous information transmission and Vietnam, in 1986, 1997 and
wireless energy transfer: A tradeoff perspective. IEEE Journal on 2007, respectively. He is cur-
Selected Areas in Communications, 33(8), 1578–1594. rently lecturer at Le Quy Don
13. Benjebbour, A., Saito, K., Li, A., Kishiyama, Y., & Nakamura, Technical University, Vietnam.
T.(2016). Non-orthogonal multiple access (NOMA): Concept and His research concentrates on the
design. In Signal processing for 5G: Algorithms and implemen- area of wireless information,
tations, John Wiley and Sons (Chap. 7, pp. 143–168) communication technologies,
14. Yang, Z., Ding, Z., Fan, P., & Al-Dhahir, N. (2017). The impact microwave electronic circuit.
of power allocation on cooperative non-orthogonal multiple
access networks with swipt. IEEE Transactions on Wireless
Communications, 16(7), 4332–4343.
15. Papoulis, A., & Pillai, S. U. (2002). Probability, random vari-
ables, and stochastic processes. New York City: Tata McGraw-
Hill Education.

123
Wireless Networks

Pham Thanh Hiep received the Technical University, Ha Noi, Viet Nam. His reseach interests lie in
B.E. degree in Communications the area of wireless information and communications technologies.
Engineering from National
Defence Academy, Japan, in
2005; received the M.E. and
Ph.D. degree in Physics, Elec-
trical and Computer Engineer-
ing from Yokohama National
University, Japan, in 2009 and
2012, respectively. He was
working as associate researcher
at Yokohama National Univer-
sity, Yokohama, Japan from
2012 to 2015. Now, he is a
lecturer at Le Quy Don

123