4332 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 16, NO.
7, JULY 2017
The Impact of Power Allocation on Cooperative
Non-orthogonal Multiple Access Networks
With SWIPT
Zheng Yang, Student Member, IEEE, Zhiguo Ding, Senior Member, IEEE,
Pingzhi Fan, Fellow, IEEE, and Naofal Al-Dhahir, Fellow, IEEE
Abstract— In this paper, a cooperative non-orthogonal multiple
access (NOMA) network is considered, where a source com-
municates with two users through an energy harvesting relay.
The impact of two types of NOMA power allocation policies,
namely NOMA with fixed power allocation (F-NOMA) and
cognitive radio inspired NOMA (CR-NOMA), on the consid-
ered cooperative simultaneous wireless information and power
transfer (SWIPT) system is investigated. In particular, closed-
form expressions for the outage probability and their high SNR
approximations are derived to characterize the performance
of SWIPT-F-NOMA and SWIPT-CR-NOMA. These developed
analytical results demonstrate that the two power allocation
policies realize different tradeoffs between the reception relia-
bility, user fairness and system complexity. Compared with the
conventional SWIPT relaying networks with orthogonal multiple
access (OMA), the proposed NOMA schemes can effectively
reduce the outage probability, although all of them realize the
same diversity gain.
Index Terms— Non-orthogonal multiple access, simultaneous
wireless information and power transfer, cognitive radio, outage
probability, diversity gain.
Manuscript received June 4, 2016; revised November 2, 2016 and
February 21, 2017; accepted April 11, 2017. Date of publication May 2,
2017; date of current version July 10, 2017. The work of Z. Yang was
supported in part by the National Science and Technology Major Project
under Grant 2016ZX03001018-002, in part by the Huawei HIRP Project
under Grant YB201504, in part by the National Natural Science Foundation
of China under Grant 61571128, and in part by the 111 Project under Grant
111-2-14. The work of Z. Ding was supported in part by the U.K. EPSRC
under Grant EP/L025272/1 and in part by H2020-MSCA-RISE-2015 under
Grant 690750. The work of P. Fan was supported in part by the National
Science and Technology Major Project under Grant 2016ZX03001018-002,
in part by the Huawei HIRP Project under Grant YB201504, and in part
by the 111 Project under Grant 111-2-14. The work of N. Al-Dhahir
was supported by NPRP from the Qatar National Research Fund under
Grant NPRP 8-627-2-260. The associate editor coordinating the review of this
paper and approving it for publication was M. Vu. (Corresponding author:
Zheng Yang.)
Z. Yang is with the Institute of Mobile Communications, Southwest Jiaotong
University, Chengdu 610031, China, and also with the College of Photonic
and Electronic Engineering, Fujian Normal University, Fuzhou 350007, China
(e-mail: zyfjnu@163.com).
Z. Ding is with the School of Computing and Communications, Lancaster
University, Lancaster LA1 4YW, U.K. (e-mail: z.ding@lancaster.ac.uk).
P. Fan is with the Institute of Mobile Communications, Southwest Jiaotong
University, Chengdu 610031, China (e-mail: p.fan@ieee.org).
N. Al-Dhahir is with the Department of Electrical Engineering, University of
Texas at Dallas, Richardson, TX 75080 USA (e-mail: aldhahir@utdallas.edu).
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TWC.2017.2697380
1536-1276 © 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.
Authorized licensed use limited to: Khon Kaen University provided by UniNet. Downloaded on November 09,2020 at 02:43:25 UTC from IEEE Xplore. Restrictions apply.
I. I NTRODUCTION
N ON-ORTHOGONAL multiple access (NOMA) has
recently attracted lots of attention, since it can provide
high spectral efficiency and realize massive connectivity in
the fifth generation (5G) mobile communication network [1]
and [2]. In addition, a two-user special case of downlink
NOMA has been adopted by Third Generation Partnership
Project (3GPP) Long Term Evolution (LTE), termed multi-user
superposition transmission (MUST) [3]. Unlike conventional
orthogonal multiple access (OMA), multiple users in a NOMA
network are allowed to share the same orthogonal resources in
the time, frequency, and code domains. Meanwhile, successive
interference cancellation (SIC) is applied at the receiver side,
in order to remove the co-channel interference due to the
implementation of NOMA [4] and [5].
The impact of partial channel state information on the per-
formance of NOMA has been studied in [6], by assuming that
users are uniformly distributed in a disk. The performance of
NOMA in large-scale underlay cognitive radio (CR) networks
with randomly deployed users was studied in [7]. The impact
of user pairing on NOMA was investigated in [8], where two
types of power allocation schemes were proposed. One is
termed NOMA based on fixed power allocation (F-NOMA),
and the other is termed CR inspired NOMA (CR-NOMA).
The power allocation problem for realizing user fairness in
NOMA systems has been investigated in [9], and the work
in [10] considered the power allocation problem for NOMA
with practical modulation based on finite constellation sizes.
In [11], an amplify-and-forward (AF) relay equipped with a
single antenna was used to help the base station for serving
users equipped with multiple antennas in cooperative NOMA
networks.
Simultaneous wireless information and power trans-
fer (SWIPT) is a new paradigm, which can prolong the lifetime
of energy constrained networks. Therefore, SWIPT is an
important enabling technology to 5G communications, in order
to support its Internet of Things (IoT) functionality. The key
idea of SWIPT is to encourage nodes to harvest energy from
their received radio frequency information-bearing signals,
which was firstly proposed in [12]. However, it is not practical
for a receiver to decode information and harvest energy at the
same time, due to the practical limitation. Motivated by this
issue, Zhou et al. [13] have proposed two practical receiver
 YANG et al.: THE IMPACT OF POWER ALLOCATION ON COOPERATIVE NOMA NETWORKS WITH SWIPT 4333

architectures, i.e., power splitting (PS) and time switch-

ing (TS). The PS and TS protocol have been applied to
cooperative networks with AF relaying, and their outage
probability and throughputs have been analysed in [14]. In
[15] and [16], game theoretic approaches for PS have been
proposed in a multi-user cooperative decode-and-forward (DF)
and AF relaying network, respectively. Krikidis [17] has stud-
ied the outage performance and the average harvested energy
for SWIPT in large-scale cooperative networks. The work
in [18]–[20] investigates the outage probability for the multiple
relays with energy storage devices, where the Markov chain
model is used to describe the states of the batteries.
Since NOMA improves the spectral efficiency of coop-
erative networks, and SWIPT provides more incentives for
user cooperation, it is natural to consider the combination
of NOMA and SWIPT in cooperative networks. An example
for such an combination can be found in [21], where users
with strong channel conditions were used as energy harvesting
relays to improve the reception reliability for those users with
poor channel conditions. In this paper, a different cooperative
DF relaying scenario is considered, where one source com-
municates with two users via a dedicated relay which harvests
energy from the source, and the NOMA principle is used for
cooperative transmission. The main contribution of this paper
is three-fold, as listed in the following:
• We first consider the application of SWIPT to cooperative
NOMA with fixed power allocation, where the outage
probability and diversity gain are considered as criteria to
analyse the system performance. The developed closed-
form analytical results show that SWIPT-F-NOMA can
achieve better outage performance than SWIPT-OMA
under the condition that the power allocation coefficients
and the users’ targeted rates are chosen correctly.
• With the application of the cognitive radio concept, we
consider two types of CR-NOMA with different con-
straints as explained in the following. One is to meet
a fixed quality of service (QoS) requirement, such as
a targeted data rate, at the user with worse channel
condition (CR-NOMA(F)), and the other is to meet a
dynamic QoS requirement at the user with strong channel
conditions (CR-NOMA(D)), i.e., this user will experience
a larger rate in NOMA than that in OMA. Furthermore,
the SWIPT concept is applied into the two CR-NOMA
cooperative networks. The exact expressions for the out-
age probability and their high signal-to-noise ratio (SNR)
asymptotic results are derived. The analytical results
show that SWIPT-CR-NOMA(F) achieves better outage
performance for the user with poor channel conditions,
but results in a loss of the outage probability for the
user with strong channel conditions, compared to those
of SWIPT-OMA. With SWIPT-CR-NOMA(D), the user
with the strong channel conditions experiences the same
outage probability as the case with SWIPT-OMA, but the
user with the poor channel conditions experiences better
outage performance compared to the case with SWIPT-
OMA.
• Furthermore, the analytical results developed in this
paper demonstrate that the use of SWIPT-F-NOMA,
Fig. 1. System model of NOMA based cooperative network with SWIPT.
SWIPT-CR-NOMA(F), or SWIPT-CR-NOMA(D) will
not cause any diversity gain loss, compared to the case
that the relay uses its own battery for relay transmission
power, but these SWIPT-NOMA schemes will lead to
a loss of the outage probability. In addition, SWIPT-
CR-NOMA(F) achieves the best outage performance for
the user with the poor channel conditions, but the worst
outage probability for the user with the strong channel
conditions, among the three SWIPT-NOMA schemes.
II. S YSTEM M ODEL
Consider a cooperative network with one source (S), one
energy harvesting relay (R), and two users (U1 , U2 ). Assume
that all the nodes are equipped with a single antenna, and
there is no direct link between the source and the users,
i.e., the source communicates with the users via the energy
harvesting relay. All the wireless channels are assumed to
undergo quasi-static independent and non-identically Rayleigh
fading, as shown in Fig. 1.
The cooperative transmission consists of two time slots.
In the first time slot, the source S uses superposition coding
√ combine √two independent signals, x 1 and x 2 , as x =
to
γ1 P x 1 + γ2 P x 2 , where x k , k = 1, 2, is message for
Uk , P is the transmission power at source S, and γk is the
power allocation factor for x k , with the constraint γ1 +γ2 = 1.
Therefore, during the first time slot, the receive signal at relay
R is given by
  
y R = (1 − )( γ1 P x 1 + γ2 P x 2 )h + w, (1)
where  is the power splitting coefficient, w ∼ CN (0, σ 2 )
α
is the additive white Gaussian noise (AWGN), h = gd − 2 is
the channel coefficient between S and R, g denotes Rayleigh
fading from S to R with g ∼ CN (0, 1), d is the distance
between S and R, and α is the path loss exponent.
Without loss of generality, assume that the relay detects the
signal, x 1 , first. Therefore, the achievable rates for the relay
to detect signal x 1 is given by
1  (1 − )γ1 ρ|h|2 
R1 = log2 1 + , (2)
2 (1 − )γ2 ρ|h|2 + 1
where ρ = σP2 denotes the transmit SNR.
If the relay can successfully decode the message x 1 , i.e.,
R1 ≥ R1∗ , then it can remove the signal x 1 before detecting x 2 ,

Authorized licensed use limited to: Khon Kaen University provided by UniNet. Downloaded on November 09,2020 at 02:43:25 UTC from IEEE Xplore. Restrictions apply.
 4334 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 16, NO. 7, JULY 2017

where R1∗ denotes the targeted rate for U1 . This means that paper is to show that the use of NOMA can yield some
under the condition R1 ≥ R1∗ , SIC is successfully carried out significant outage performance gains, compared to OMA,
at the relay, and the message x 2 will be detected at the relay when both of them use the same SWIPT protocol. On the
with the following instantaneous rate other hand, the simplified SWIPT model can provide some
1   insights to understand the advantage of combining SWIPT
R2 = log2 1 + (1 − )γ2 ρ|h|2 . (3) with NOMA, which can offer some technical guidances to
2
implement SWIPT in practice. It is worth pointing out that
It is noted that the power splitting protocol is applied at the it is a promising future direction to study the impact of the
relay to harvest energy from the source. In addition, the relay energy for maintaining the user circuits on the performance of
uses the DF protocol to decode the signals from the source. SWIPT, which is beyond the scope of this paper.
Because of these reasons, the relay has to firstly ensure the Since both x 1 and x 2 are decoded correctly at the relay,
detection of the messages from the source, then it can carry the NOMA scheme can be used to serve the two users in the
out energy harvesting. In other words, only when the relay −α
second time slot. Firstly, denote h k = gk dk 2 , k = 1, 2, is
can successfully decode the signals from the source, it can the channel coefficient between the relay and Uk , where gk ∼
use the harvested energy to forward the signals to the users.
CN (0, 1) and dk are Rayleigh fading and the distance from the
Therefore, it is much more important for the relay to decode relay to Uk , respectively. Without loss of generality, assume
the information than performing energy harvesting. that the channel fading gains |h k |2 are ordered as |h 1 |2 ≤
In order to ensure that the relay can harvest the sufficient
|h 2 |2 . According to NOMA scheme, the relay retransmits
√ a
energy to power the relay transmission, we impose the fol- superimposed mixture of x and x as x = α P x +
lowing constrains, R1 = R1∗ and R2 = R2∗ , which imply √ 1 2 R 1 R 1
α2 PR x 2 , where αk , k = 1, 2, is power allocation factor for
that relay R can correctly decode both the signals x 1 and x 2 , Uk , with α1 > α2 and α1 + α2 = 1. The received signal at the
simultaneously, where R2∗ denote the targeted data rates for k-th user, k = 1, 2, is given by
U2 . With these imposed conditions, R1 = R1∗ and R2 = R2∗ ,  
i.e., ⎧  yk = ( α1 PR x 1 + α2 PR x 2 )h k + wk , (10)
2 
⎪ 1 log 1 + (1 − )γ1 ρ|h|
⎨ = R1∗
2 2
(1 − )γ2 ρ|h|2 + 1 where wk is AWGN with mean zero and variance σ 2 .
(4)
⎩ 1 log 1 + (1 − )γ ρ|h|2  = R ∗
⎪ Based on the NOMA principle, both users will decode x 1
2
2 2 2 first, since |h 1 |2 ≤ |h 2 |2 . Therefore, the achievable rate for U1
After some algebraic manipulations, (4) can be rewritten as to detect x 1 is given by
follows: 1  α1 ρ R |h 1 |2 
 ∗ ∗ R1N = log2 1 + , (11)
(1 − )ρ|h|2 (1 − 22R1 γ2 ) = 22R1 − 1 2 α2 ρ R |h 1 |2 + 1
∗ (5)
(1 − )ρ|h|2 γ2 = 22R2 − 1
where ρ R = σPR2 .
According to (5), it is easy to obtain the power allocation The achievable rate for U2 to detect x 1 is given by
factor γ2 and the power splitting factor  which are given by 
1 α1 ρ R |h 2 |2 
∗
22R2 −1
N
R1→2 = log2 1 + . (12)
γ2 = , (6) 2 α2 ρ R |h 2 |2 + 1
2 2R1∗ +2R2∗ −1
Note that if R1→2 N
≥ R1∗ , i.e., U2 can correctly decode the
and N
signal, x 1 , which means that under the condition R1→2 ≥ R1∗ ,
 ∗ ∗
22R1 +2R2 − 1 the SIC can be successfully carried out at U2 to remove the
 = max 0, 1 − , (7)
ρ|h|2 signal, x 1 . Therefore, the achievable rate for U2 to decode x 2
is given by
respectively, note that 0 < γ2 < 1, and  < 1. Hence  = 0
means the received signals from the source are fully used for 1  
R2N = log2 1 + α2 ρ R |h 2 |2 . (13)
information decoding, and the relay cannot harvest any energy. 2
Based on (1), the energy harvested at relay R is given We will investigate the outage performance of SWIPT-
by [14], and [15] NOMA with two types of power allocation schemes in the
PR = η P|h|2 , (8) following two sections. The motivation for considering the
outage as the performance evaluation criterion is explained in
where η is the energy harvesting coefficient. Note that the the following.
energy harvested from AWGN has been omitted. Recall that the bit error ratio (BER) is an important per-
Based on (6) and (7), the maximal transmission power at formance criterion which can be used to evaluate the errors
relay R can be obtained as follows: that occur in a transmission system. However, it is difficult to
  obtain the BER in closed-form expressions, since it depends on
PR = max 0, η(P|h|2 − εσ 2 ) , (9)
which coding and modulation schemes are used. Fortunately,
∗ ∗
where ε = 22R1 +2R2 − 1. the outage probability can be evaluated in a simpler way and
The reason for the simplified energy harvesting model in (9) provides a tight bound on the BER, when optimal coding with
is given as follows. On one hand, the contribution of this an infinite length is used [22].

Authorized licensed use limited to: Khon Kaen University provided by UniNet. Downloaded on November 09,2020 at 02:43:25 UTC from IEEE Xplore. Restrictions apply.
 YANG et al.: THE IMPACT OF POWER ALLOCATION ON COOPERATIVE NOMA NETWORKS WITH SWIPT 4335

Furthermore, the outage probability is defined as the event Lemma 1: The closed-form expression for the outage prob-
that when the instantaneous rate at the receiver falls below ability at U2 in SWIPT-F-NOMA is given by
a targeted date rate, which can be used to measure a prede-
e λρ  −i m−1 mε m−1−i
− mε m−1
fined QoS requirement for delay-sensitive networks. This is 
P2F = 1 − 2
because the throughput is a function of the channel fading (m) i λρ
i=0
quality, and evaluating the satisfied QoS requirement is quite  i+1 
challenging. Therefore, it is important to investigate the outage × 1 Ki+1 ( 1 )− i+1 i+1
2 Ki+1 ( 2 )+ 3 Ki+1 ( 3 ) ,
performance in wireless communication system, especially in (17)
delay-sensitive networks.  α
4m
max d1
where m is an integer, 1 = λρ , 2 =
 
4m
max (d1α +d2α ) 4m
max d2α
λρ , 3 = λρ .
III. NOMA BASED ON F IXED P OWER A LLOCATION Proof: By using steps similar to those used to prove
Theorem 1, and also using the fact that the PDF of |h|2 is
With F-NOMA, the power allocation factors for the users
given by
are constants and the strong user is allocated less power than
the weak user, as long as the order of the users’ channels is mm mx
f |h|2 (x) = x m−1 e− λ , (18)
still the same. In this case, the power allocation coefficients (m)λm
are not functions of the users’ channel fading gains, i.e., the
transmitter only needs to know the order of the users’ channel the Lemma is proved.
It can be observed from (17) that the closed-form expression
fading gains to perform NOMA.
Recall that the relay harvests energy from the source and can be found under the condition that m is an integer, and
the SIC is carried out at U2 to remove the signal x 1 , before it is difficult if m is not an integer. It is important to point
out that Rayleigh fading is a special case of Nakagami-m
detecting its own message. Therefore the coverage probability
for U2 can be defined as follows: fading with m = 1, as such, we adopt Rayleigh fading to
model the link with SWIPT in the remaining of this paper, in
 N  order to evaluate the outage performance with mathematical
≥ R1∗ , R2N ≥ R2∗ , PR > 0 .
F
P2 = Pr R1→2 (14)
tractability.
Similar to (14), the coverage probability at U1 in F-NOMA
The following theorem provides a closed-form expression for SWIPT relaying transmission is defined as follows:
the outage probability experienced by U2 and its approxima-
 
P1 = Pr R1N ≥ R1∗ , PR > 0 .
tion in the high SNR region, when F-NOMA relaying is used. F
(19)
Theorem 1: The use of F-NOMA SWIPT relaying ensures
that the following outage probability is achieved at U2 Recall that the cumulative density function (CDF) of |h 1 |2
is given by
− dρ ε  
α
P2F = 1 − e φ1 K1 (φ1 ) − φ2 K1 (φ2 ) + φ3 K1 (φ3 ) , (15) α α
F|h 1 |2 (x) = 1 − e−(d1 +d2 )x . (20)
  ∗
ε1 ε2 By using this CDF and also following the steps similar to
where
max = max η(α1 −α ε ) , ηα , α1 > α2 ε1 , ε1 = 22R1 −
 2 1 α 2α  those in the proof of Theorem 1, a closed-form expression for
∗ 4
max d d1 4
max d α (d1α +d2α )
1, ε2 = 22R2 − 1, φ1 = ρ , φ2 = ρ , the outage probability experienced by U1 in F-NOMA SWIPT
 α α
4
max d d2 relaying transmission and its high SNR approximation can be
φ3 = ρ , and Kv () denotes the modified Bessel
obtained as follows:
function of the second kind with order v [23]. The high SNR
αε
approximation for P2F is given by −d
P1F = 1 − e ρ φ4 K1 (φ4 )

1 d α (d1α + d2α ) ρ
dα  dα dα  ≈ ln , (21)
P2F ≈ ε −
max d1α ln α 1 α −
max d2α ln α 2 α . ρ
1 d (d1α + d2α )
α
ρ d1 + d2 d1 + d2

(16) ε1 4
1 d α (d1α +d2α )
where
1 = η(α1 −α 2 ε1 )
, α 1 > α2 ε 1 , and φ 4 = ρ .
Proof: See Appendix A.
The Outage Probability of OMA: OMA is a good
Due to the fact that line-of-sight path of the S-R link may
benchmarking scheme for the NOMA transmission protocol.
exist, we use Nakagami-m fadding for the S-R link to which
In cooperative SWIPT networks, if source communicates with
SWIPT is applied, and investigate the outage performance for
Ui , i = 1, 2, via an energy harvesting relay in an OMA
U2 in SWIPT-F-NOMA systems. Particularly, assume that the
mode, such as time division multiple access (TDMA), four
S-R link, denoted by h, follows a Nakagami-m distribution
time slots are needed in total to serve U1 and U2 . Similar
with a sharp parameter m and a variance λ, whereas the links
to (9), the maximal transmission power for the i -th user is
from the relay to the users follow Rayleigh fading.
given by
Most results in this paper can be extended to the case
 
with Nakagami-m fading. For example, the outage proba- PR,i = max 0, η(P|h|2 − εi σ 2 ) , (22)
bility for U2 in SWIPT-F-NOMA is given in the following
∗
Lemma. where εi = 24Ri − 1, i = 1, 2.

Authorized licensed use limited to: Khon Kaen University provided by UniNet. Downloaded on November 09,2020 at 02:43:25 UTC from IEEE Xplore. Restrictions apply.
 4336 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 16, NO. 7, JULY 2017

The achievable rate for the energy harvesting relay commu- considered as a primary user sharing the same spectrum with
nicating with Ui in OMA is given by a secondary user, which has strong channel conditions. The
1 advantage of this CR-NOMA is that the QoS requirement of
RiT = log2 (1 + ρ R,i |h i |2 ), (23) the weak user can be strictly guaranteed.
4
P
Recall that the channel fading gains have been ordered
where ρ R,i = σR,i
2 . as |h 1 |2 ≤ |h 2 |2 , which means U1 can be considered as
The use of the OMA strategy means that the outage prob- a primary user and should be served with high priority
ability at Ui can be defined as follows: based on the concept of CR networks [24]. Therefore, the
1
 rate at U1 , R1N , in (11) needs to satisfy the following
PiO = 1 − Pr log2 (1 + ρ R,i |h i |2 ) > Ri∗ , P|h|2 > εi σ 2 constraint:
4  

εi εi 1 α1 ρ R |h 1 |2
= 1 − Pr |h i | >
2
, |h| >
2
. (24) log2 1 + ≥ R1∗ , (27)
η(ρ|h|2 − εi ) ρ 2 α2 ρ R |h 1 |2 + 1
Based on (24) and following the steps similar to those in the which means that the maximal transmit power factor α2 can
proof for Theorem 1, the outage probability of U1 in OMA be used for U2 is given by
and its high SNR approximation are given by 
ρ R |h 1 |2 − ε1
α2 = max 0, . (28)
d α ε1 ρ R |h 1 |2 (1 + ε1 )
P1O = 1 − e− ρ φ5 K1 (φ5 )
Based on the relay transmission power PR in (9) and the
ε1 d α (d1α + d2α ) ηρ power allocation factor α2 in (28), a closed-form expression
≈ ln , (25)
ηρ ε1 d α (d1α + d2α ) for the outage probability at U1 in CR-NOMA can be obtained
 as follows:

4ε d α (d α +d α ) 
where φ5 = 1 1 2 ε ε1
ηρ P1C,I = 1 − Pr |h|2 > , |h 1 |2 >
Similarly, the outage probability of U2 in OMA and its high ρ η(ρ|h|2 − ε)
 ∞ (d α +d α )ε
SNR approximation are given by − 1 2 1 α

= 1− e η(ρx−ε) d α e−d x d x
d α ε2 ε
−  ρ
P2O = 1−e φ6 K1 (φ6 ) − φ7 K1 (φ7 ) + φ8 K1 (φ8 )
ρ αε
  −d
= 1 − e ρ φ9 K1 (φ9 ),
dα dα dα (29)
≈ ε − ε2 d1α ln α 1 α − ε2 d2α ln α 2 α , (26) 
4ε1 d α (d1α +d2α )
ρ d1 + d2 d1 + d2 where φ9 = .
  ηρ

4ε2 d α d1α

4ε2 d α (d1α +d2α ) At high SNR, P1C,I can be approximated as follows:
where φ6 = ηρ , φ7 = ηρ , and φ8 =
 ε1 d α (d1α + d2α ) ηρ

4ε2 d α d2α P1C,I ≈ ln . (30)
ηρ. ηρ ε1 d (d1α + d2α )
α

One can observe from (16) and (26), (21) and (25) that Based on (9), (13), and (28), the coverage probability of U2
F-NOMA based SWIPT can achieve the same diversity gain as in CR-NOMA(F) relaying SWIPT transmission is defined as
OMA based SWIPT. However, in Section V, both simulations follows:
and analytical results demonstrate that F-NOMA based SWIPT  ε1 
P2 = R2N ≥ R2∗ , ρ R > 0, |h 1 |2 >
C,I
can realize a better outage performance compared to the OMA . (31)
ρR
scheme. This is because the use of NOMA allows all the users
The following theorem provides an exact expression for the
to share the bandwidth resources, which yields higher spectral
outage probability experienced by U2 and its approximation
efficiency, compared to OMA.
at high SNR.
Theorem 2: The use of CR-NOMA with a fixed QoS con-
IV. C OGNITIVE R ADIO I NSPIRED NOMA
straint at U1 yields an outage probability at U2 as follows:
In this Section, we focus on two types of CR inspired
 
 dα ε
NOMA. The first one is to treat the user with poor connections α
− dρ ε τ τ  2d α diα e− ρ
to the relay as a primary user, and the other one is to treat the P2C,I = 1−e K1− 1
ρ ρ ηρ
user with strong connections to the relay as a primary user. i1 ,i2 ∈{1,2}
 ε2 (1+ε1 )   
gi1 ,i2 (s)
A. CR-NOMA With a Fixed QoS Requirement at the Weak × K0 2 ds, (32)
0 ρ
User
d α diα ε2 (1+ε1 )(s+ε1 )+d α diα s(s+ε1 )
Theorem 1 shows that SWIPT-F-NOMA can achieve better where gi1 ,i2 (s) = 2
ηs
1
, τ =
performance than SWIPT-OMA under the condition α1 > 4d α (d1α +d2α )(ε1 +ε2 (1+ε1 ))
η When ρ → ∞, the high SNR approx-
.
α2 ε1 , i.e., the power allocation factors and the users’ targeted imation of the outage probability is given by
rates need to be carefully chosen. In order to overcome this
disadvantage, we can consider NOMA as a special case of ln ρ
P2C,I ≈ . (33)
CR systems [8], where a user with poor channel conditions is ρ

Authorized licensed use limited to: Khon Kaen University provided by UniNet. Downloaded on November 09,2020 at 02:43:25 UTC from IEEE Xplore. Restrictions apply.
 YANG et al.: THE IMPACT OF POWER ALLOCATION ON COOPERATIVE NOMA NETWORKS WITH SWIPT 4337

Proof: See Appendix B. power allocation factor α1 in (35) is based on the condition
Comparing (33) to (26), we can see that the diversity gain at in (34), i.e., U2 should get a rate no less than that in OMA.
U2 in SWIPT-CR-NOMA(F) is identical to that in SWIPT- Theorem 3: The use of CR-NOMA with a dynamic QoS
OMA, but the use of CR-NOMA results in a higher outage requirement at the strong user yields an outage probability
probability, compares to the OMA case. The main reason at U1 as expressed in (37), shown at the bottom of this page,
ρ −2g(x)
is that U2 is regarded as a secondary user in SWIPT-CR- where ρ R,2 = η(ρz −ε2 ), ρ R = η(ρz −ε), H (x) = R,2g 2 (x) ,
NOMA(F), and it is served only after the QoS requirement √
ρ R x−ε1 ρ (ε +1)2 −2ρ R + (ρ R,2 ,ρ R )
of U1 is satisfied. g(x) = (ε 1 +1)x
, K (ρ R,2 , ρ R ) = R,2 1 2ρ 2R
,
  2
(ρ R,2 , ρ R ) = ρ R,2 (ε1 + 1)2 − 2ρ R + 4ρ 2R (ε12 + 2ε1 ).
B. CR-NOMA With a Dynamic QoS Requirement at the At high SNR, P1C,I I can be approximated as follows:
Strong User ln ρ
P1C,I I ≈ . (38)
Recall that by using this CR-NOMA inspired power allo- ρ
cation policy, the strong user can be served by the transmitter Proof: See Appendix C.
only when the weak user’s QoS is met. Especially, when the Note that the SWIPT-CR-NOMA(D) achieves similar out-
weak user’s targeted rate is very high, the transmitter has to age performance compared to SWIPT-OMA at high SNR, as
allocate all the power to support this user, and the strong user the approximated expression in (38) is quite similar to that
cannot be served. In order to ensure that the strong user in in (25) at high SNR. However, the simulation results in
NOMA systems can be also served, we impose a constraint Section V show that the outage probability of U1 in SWIPT-
that this user achieves at least the same throughput as that CR-NOMA(D) is always better than that in SWIPT-OMA. The
in OMA, then allocate all the remaining power to the weak main reason is that the power allocation factor, α1 , in (35) is
user. In other words, the user with strong channel conditions a function of the strong user’s channel gain, and the rate for
is regarded as a primary user, and consider a dynamic QoS U1 , R1N , in (11) is an increasing function of α1 , which can
requirement of this user which is to achieve a rate larger than further improve the weak user’s performance.
that of OMA, i.e.,
V. N UMERICAL R ESULTS
1   1
log2 1 + α2 ρ R |h 2 |2 ≥ log2 (1 + ρ R,2 |h 2 |2 ), (34) In this section, computer simulations are used to verify the
2 4 accuracy of the developed analytical results about the outage
which means the maximal power allocation factor α1 is given probability of NOMA based cooperative SWIPT networks.
by In the considerable network, we choose the energy harvesting
 efficiency η = 0.4, the path loss factor α = 4. The carrier
1 + ρ R,2 |h 2 |2 − 1
α1 = 1 − . (35) frequency, transmitter antenna gain, and receiver antenna gain
ρ R |h 2 |2 are set to be 915 MHz, 1 dBm, and 1 dBm, respectively.
Based on (35), the coverage probability of U2 in Fig. 2-Fig. 4 are obtained with the following distances, where
CR-NOMA with a dynamic QoS requirement at the strong user the distance between the source and the relay is given by
(SWIPT-CR-NOMA(D)) can be expressed as follows: d = 10 m, the distances from relay to U1 and U2 are given by
1  d1 = 25 m and d2 = 10 m, respectively, and the noise power
α1 ρ R |h 2 |2 
≥ R1∗ ,
C,I I
P2 = Pr log2 1 + is set to be −114 dBm. Fig. 5 is obtained with the following
2 α2 ρ R |h 2 |2 + 1 distances, where the distance between the source and the relay
1   
log2 1 + α2 ρ R |h 2 |2 ≥ R2∗ is given by d = 5 m, the distances from the relay to the U1 and
2
1 the U2 are given by d1 = 16 m and d2 = 5 m, respectively,
 
= Pr log2 1 + ρ R,2 |h 2 |2 ≥ R1∗ , and the noise power is set to be −84 dBm.
4 Fig. 2 shows the outage performance of F-NOMA with
1   
log2 1 + ρ R,2 |h 2 |2 ≥ R2∗ SWIPT in cooperative networks as a function of the transmitter
4 power P. It can be seen from Fig. 2 that SWIPT-F-NOMA
1   
= Pr log2 1 + ρ R,2 |h 2 |2 ≥ R2∗ . (36) achieves the same diversity gain as SWIPT-OMA, but the
4 NOMA scheme offers a constant performance gain in the
From (36), it is interesting to observe that U2 in SWIPT-CR- outage probability, compared to the OMA scheme. This is
NOMA(D) can experience the same coverage probability as because NOMA allows multiple users to share the same
that of SWIPT-OMA. The reason for this is that the maximal resources at the time, frequency and code domains via power

   
∞ (d1α +d2α )ε1  α 4d α (d1α + d2α )ε1 4d α (d1α + d2α )ε1
α −(d1α +d2α )K (ρ R,2 ,ρ R ) − −d α z − dρ ε
P1C,I I = 1−d e −e η(ρz−ε) e dz − e K1
ε2
ρ
ηρ ηρ
  ∞  K (ρ R,2 ,ρ R )
−diα H (x)−diα x α
− diα1
ε2 ε1
e 2 1 d x d α e−d z dz. (37)
i1 ,i2 ∈{1,2} ρ ρR

Authorized licensed use limited to: Khon Kaen University provided by UniNet. Downloaded on November 09,2020 at 02:43:25 UTC from IEEE Xplore. Restrictions apply.
 4338 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 16, NO. 7, JULY 2017

Fig. 2. SWIPT-F-NOMA with α1 = 0.75 and α2 = 0.25. Fig. 4. The outage probability of weak user in SWIPT-CR-NOMA(D) with
a dynamic QoS requirement at the strong user.

Fig. 3. SWIPT-CR-NOMA with a fixed QoS requirement at the weak user.

domain multiplexing. Furthermore, if the users’ targeted rates

and power allocation factors cannot satisfy α1 > α2 ε1 , the
outage will always occur, as stated in Theorem 1. In addition,
it is important to point out that the curves for the closed-form
expressions in (15), (21), (25) and (26) match Monte Carlo
simulation results perfectly.
In Fig. 3, the outage probability of SWIPT-CR-NOMA
with a fixed QoS constraint at the weak user is investi-
gated. Fig. 3 demonstrates that all the curves are parallel,
which means the achievable diversity order is the same for
SWIPT-CR-NOMA(F) and SWIPT-OMA. Compared to
SWIPT-OMA, it is interesting to find that the weak user
in SWIPT-CR-NOMA(F) can significantly improve its outage
performance, while the strong user’s outage probability is
deteriorated severely. This is because the power allocation Fig. 5. Outage probability comparision with/without SWIPT for CR-NOMA
and F-NOMA, where γ1 = α1 = 0.7, γ2 = α2 = 0.3, R1∗ = 0.5 bits/s/H z,
factor, α2 in (28) is a function of the weak user’s channel R2∗ = 1.5 bits/s/H z.
gain, and the strong user cannot get sufficient power, especially
when the weak user’s targeted rate is high. Again, Fig. 3 performance of U2 , since this strong user experiences the same
confirms that the analytical results developed in (29) and (32) outage performance as in SWIPT-OMA. As can be observed
perfectly match computer simulations. from Fig. 4, SWIPT-CR-NOMA(D) can achieve better outage
In Fig. 4, we focus on the outage performance of the user performance than SWIPT-OMA, although the same diversity
with weak channel conditions when the CR-NOMA power gain is realized by both schemes. Furthermore, Fig. 4 also
allocation policy is designed to meet the dynamic QoS con- demonstrates that the users’ targeted rates significantly affect
straint at the strong user. Note that there is no need to show the the performance gap between NOMA and OMA. Particularly,

Authorized licensed use limited to: Khon Kaen University provided by UniNet. Downloaded on November 09,2020 at 02:43:25 UTC from IEEE Xplore. Restrictions apply.
 YANG et al.: THE IMPACT OF POWER ALLOCATION ON COOPERATIVE NOMA NETWORKS WITH SWIPT 4339

as the difference between R1∗ and R2∗ increases, the outage Since the unordered channel fading gains |h̃ k |2 , k = 1, 2 are
performance gap between SWIPT-CR-NOMA(D) and SWIPT- exponentially distributed with different expectations, the CDF
OMA enlarges accordingly. In addition, computer simulations of |h 2 |2 is given by
also confirm the accuracy of the analytical results developed α α
F|h 2 |2 (y) = (1 − e−d1 y )(1 − e−d2 y ). (40)
in (33).
F
In Fig. 5, the outage probability difference between the Based on (40), the coverage probability P2 in (39) can be
NOMA schemes with SWIPT and without SWIPT is studied. calculated as follows:
Fig. 5 demonstrates that the diversity gain achieved by NOMA  ∞  d α
max d2α
max (d1α +d2α )
max 
e− ρx−ε + e− ρx−ε − e− ρx−ε
F 1
with SWIPT is the same as that achieved by NOMA without P2 =
ε
energy harvesting, but the use of SWIPT results in a outage ρ
α
probability loss. This phenomenon is expected since the use × d α e−d x d x. (41)
of SWIPT means that the relay transmission power which is Let t = ρx − ε, the above
F
P2 can be further evaluated as
harvested from the signals sent by the source is much smaller follows:
than that in the case without SWIPT. In addition, it can be dα ε 
e− ρ ∞ d1α
max d2α
max
seen from Fig. 5 that SWIPT-CR-NOMA(F) achieves the best
e− t + e− t
F
P2 =
outage performance for the weak user, but the worst outage ρ 0
probability for the strong user. (d α +d α )
max  α
− 1 2t −d t
−e d α e ρ dt
 αdα
 
VI. C ONCLUSIONS dα ε 4
d 4
max d α d1α
= e− ρ
max 1
K1
We have investigated the application of SWIPT to coop- ρ ρ
erative F-NOMA and CR-NOMA networks, and their outage    
4
max d α (d1α + d2α ) 4
max d α (d1α + d2α )
probabilities have been analyzed in this paper. Compared to − K1
SWIPT-OMA, SWIPT-F-NOMA can achieve better outage ρ ρ
   
performance if the users’ targeted rates and power allocation 4
max d α d2α 4
max d α d2α
coefficients are correctly chosen. For SWIPT-CR-NOMA(F), + K1 . (42)
ρ ρ
the outage performance of the user with strong channel
conditions is degraded compared to the OMA case, but the F
Substituting (42) into the outage probability P2F = 1 − P2 , the
outage performance of the weak user can be significantly first part of the theorem is proved.
improved. Furthermore, for CR-NOMA-SWIPT(D), the weak When z → 0, the Bessel function zK1 (z) can be approx-
user’ outage performance improves evidently and there is no imated by using series representation as follows [23, eq.
loss in the outage performance at the user with strong channel (8.446)]:
conditions, compared to SWIPT-OMA. ∞
 ( 2z )2k+1 z  z 
In this paper, the effect of power allocation on coop- zK1 (z) = 1 + ln + C
erative NOMA with SWIPT has been investigated, where k!(k + 2) 2
k=0
all the nodes are equipped with a single antenna, and an
important future direction is to consider the application of 1 ∞
( 2z )2l+1 z  
l
1  1
l+1
− +
multiple-input multiple-output technologies to the addressed 2 l!(n + l)! k k
l=0 k=1 k=1
cooperative SWIPT-NOMA networks. In addition, another z z2
promising future direction is that the addressed system model ln .
≈ 1+ (43)
2 2
in this paper can be extended to the case with multiple relays
to improve the reception reliability of NOMA transmission. When z → 0, the exponential function e−z can be approx-
The key step is to consider how many relays can correctly imated as follows:
decode the source signals and harvest energy from the source e−z ≈ 1 − z. (44)
simultaneously, then choose one relay from the qualified relays
By applying (43) and (44) into (15), P2F
can be approxi-
to serve two users with different power allocation schemes.
mated at high SNR as follows:

 εd α  2
d αdα
max d α d1α

A PPENDIX A F
P2 ≈ 1 − 1 − 1+
max 1
ln
P ROOF OF T HEOREM 1 ρ ρ ρ
  
Based on (14), the coverage probability of U2 can be 2
max d α (d1α + d2α )
max d α (d1α + d2α )
rewritten as follows: − 1+ ln
ρ ρ
  
F 1 α1 ρ R |h 2 |2   
P2 = Pr log2 1 + ≥ R1∗ , α
2
max d d2 α α

max d α d2
2 α2 ρ R |h 2 |2 + 1 + 1+ ln
ρ ρ
1    
log2 1 + α2 ρ R |h 2 |2 ≥ R2∗ , PR > 0 d α d α dα
2
 ≈ ε −
max d1α ln α 1 α −
max d2α ln α 2 α .
ε ρ d1 + d2 d1 + d2
= Pr (ρ|h|2 − ε)|h 2 |2 >
max , |h|2 > . (39)
ρ (45)

Authorized licensed use limited to: Khon Kaen University provided by UniNet. Downloaded on November 09,2020 at 02:43:25 UTC from IEEE Xplore. Restrictions apply.
 4340 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 16, NO. 7, JULY 2017

Therefore, the second part of the theorem is also proved. The Let t = ρz − ε and s = η(ρz − ε)x − ε1 , Q 2 in (50) can be
proof for the theorem is complete. rewritten as follows:
  ∞  ε1 +ε2 (1+ε1 ) diα ε2 (1+ε1 )x
A PPENDIX B α
η(ρz−ε) − η(ρz−ε)x−ε
2 −diα x
Q2 = di1 ε
e 1 1 dx
P ROOF OF T HEOREM 2 ε 1
i1 ,i2 ∈{1,2} ρ η(ρz−ε)
Based on (9) and (13), the coverage probability of U2 in α −d α z
×d e dz
(31) can be further expressed as follows: α
− dρ ε 
1  d α diα1 e ε2 (1+ε1 )
C,I   ε1  =
P2 = Pr log2 1 + α2 ρ R |h 2 |2 ≥ R2∗ , ρ R > 0, |h 1 |2 > ηρ
2 ρR i1 ,i2 ∈{1,2} 0
  
η(ρ|h|2 − ε)|h 1 |2 − ε1 ε  diα ε2 (1+ε1 )(s+ε1 )+diα s(s+ε1 ) dαt
= Pr |h 2 |2 ≥ ε2 , |h|2 > , ∞ exp − 2 1
ηst − ρ dt
|h 1 |2 (1 + ε1 ) ρ × ds.
t
ε 1  0
 
|h 1 |2 > . (46)
η(ρ|h|2 − ε) Q3

Recall that the users’ channel gains are ordered as |h 1 |2 ≤ (51)

C,I ∞ exp(−t −z 2 /(4t ))
|h 2 |2 , the above coverage probability P2 can be further By applying Kv (z) = 12 ( 2z )v 0 t v+1
dt in [23, eq.
rewritten as follows: (8.447)], the above integral Q 3 can be evaluated as follows:

ε ε1 + ε2 (1 + ε1 )   α α 
d di2 ε2 (1 + ε1 )(s + ε1 ) + d α diα1 s(s + ε1 )
C,I 2 2 2 2
P2 = Pr |h 2 | ≥ |h 1 | , |h| > , |h 1 | >
ρ η(ρ|h|2 − ε) Q 3 = 2K0 2 .
 ηρs
ε2 (1 + ε1 )|h 1 | 2 ε
+ Pr |h 2 |2 ≥ , |h|2 > , (52)
η(ρ|h|2 − ε)|h 1 |2 − ε1 ρ
Substituting (49), (51) and (52) into (47), and using the fact
ε1 2 ε1 + ε2 (1 + ε1 )
< |h 1 | < . (47) C,I
that P2C,I = 1 − P2 , the first part of the theorem is proved.
η(ρ|h|2 −ε) η(ρ|h|2 − ε)
The Gaussian-Chebyshev quadrature  [26] is applied to
Note that the joint probability density function (PDF) of |h 1 |2 ε (1+ε1 ) gi1 ,i2 (s)
and |h 2 |2 is given by [25] approximate the integral 0 2 K0 2 ρ ds in (32)
 −d α x −d α y
as follows:
f |h 1 |2 ,|h 2 |2 (x, y) = diα1 e i1 diα2 e i2 , (48)  ε2 (1+ε1 )   
gi1 ,i2 (s)
i1 ,i2 ∈{1,2} K0 2 ds
ρ
where 0 < x < y. 0
  
πε2 (1 + ε1 )  !!
CR
2i − 1 !!
n
Denote the first factor in P2 in (47) as Q 1 , which can be gi1 ,i2 (si )
evaluated by applying (48) as follows: ≈ sin π K0 2 ,
2n 2n ρ
  ∞ ∞  ∞ i=1
−d α y
Q1 = d α diα1 diα2 e i2 d y (53)
ε2 (1+ε1 )  
ε +ε (1+ε )
ε 1 2 1 x
i1 ,i2 ∈{1,2} ρ η(ρz−ε) where si = + cos 2i−1
2 2n π , and n is a complexity-
1
−diα x α
× e 1 d x e−d z dz accuracy tradeoff parameter.

 d α diα1 ∞
−(diα +diα )
ε1 +ε2 (1+ε1 ) α
Recall that the series representation of Bessel function
= e 1 2 η(ρz−ε) e−d z dz K0 (z) can be expressed as follows [23, eq. (8.447)]:
dα + diα2 ε
i1 ,i2 ∈{1,2} i1 ρ ∞ ∞
α  z  ( 2z )2k  z 2k
(k + 1)
− dρ ε K0 (z) = − ln( ) +
 diα1 e 4d α (diα1 + diα2 )(ε1 + ε2 (1 + ε1 )) 2 (k!)2 22k (k!)2
= k=0 k=0
d α + diα2 ηρ z
i1 ,i2 ∈{1,2} i1 ≈ − ln( ). (54)
  2
4d α (diα1 + diα2 )(ε1 + ε2 (1 + ε1 ))
× K1 Based on (44), (53) and (54), Q 2 in (51) can be approxi-
ηρ mated as follows:
 dα ε 
α
− dρ ε4d α (d1α + d2α )(ε1 + ε2 (1 + ε1 ))  2d α diα e− ρ  ε2 (1+ε1 )  gi ,i (s) 
=e Q2 = 1
K0 2 1 2
ds
ηρ ηρ ρ
 α α i1 ,i2 ∈{1,2} 0

4d (d1 + d2α )(ε1 + ε2 (1 + ε1 ))  d α diα1 (1 − dαε
× K1 . (49) ρ ) πε2 (1 + ε1 )
ηρ ≈−
ηρ 2n
i1 ,i2 ∈{1,2}
C,I
Denote the second factor in P2
in (47) as Q 2 , which can  n
! !
be evaluated by applying (48) as follows: × ! sin 2i − 1 π ! ln gi1 ,i2 (si )
2n ρ
  ∞  1 η(ρz−ε)
ε +ε2 (1+ε1 )  i=1
∞
−d α y
diα2 e i2 d y  πd α diα1 ε2 (1 + ε1 ) 
n
! !
Q2 = ε ε ε (1+ε )x
1 2 1 ≈ ! sin 2i − 1 π ! ln ρ .
i1 ,i2 ∈{1,2} ρ η(ρz−ε) η(ρz−ε)x−ε1 2ηn 2n ρ
α
i1 ,i2 ∈{1,2} i=1
α −di1 x α
× di1 e d x d α e−d z dz. (50) (55)

Authorized licensed use limited to: Khon Kaen University provided by UniNet. Downloaded on November 09,2020 at 02:43:25 UTC from IEEE Xplore. Restrictions apply.
 YANG et al.: THE IMPACT OF POWER ALLOCATION ON COOPERATIVE NOMA NETWORKS WITH SWIPT 4341

√
ρ R,2 (ε1 +1)2 −2ρ R + (ρ R,2 ,ρ R )
When ρ → ∞, Q 1 in (49) can be approximated as follows: where K (ρ R,2 , ρ R ) = ,
2ρ 2R
τ 4ρ   2
Q1 ≈ 1 − ln . (56) (ρ R,2 , ρ R ) = ρ R,2 (ε1 + 1)2 − 2ρ R + 4ρ 2R (ε12 + 2ε1 ).
4ρ τ Based on (48), the first factor in (60), Q 41 , can be calculated
According to (55) and (56), the outage probability of U2 at as follows:
high SNR can be approximated as follows:   ∞ ∞  ∞
−d α y
Q 41 = ε
diα2 e i2 d y
P2C,I = 1 − Q 1 − Q 2 i1 ,i2 ∈{1,2} ρ
2 K (ρ R,2 ,ρ R ) x

τ  πd α diα ε2 (1 + ε1 ) α
−d x α
≈ ln
4ρ
− 1 × diα1 e i1 d x d α e−d z dz
τ  ∞
4ρ
i1 ,i2 ∈{1,2}
2ηn  d α diα1 −(diα +diα )K (ρ R,2 ,ρ R )−d α z
! ! = α α e 1 2 dz

n
! ! d + di2 ρ ε2

× ! sin 2i − 1 π ! ln ρ i1 ,i2 ∈{1,2} i1

 ∞
! 2n ! ρ
α −(d1α +d2α )K (ρ R,2 ,ρ R )−d α z
i=1 =d
ε2
e dz. (61)
ln ρ
≈ . (57) ρ
ρ The second factor in (60), Q 42 , can be calculated by
Therefore, the second part of the theorem is proved, and the applying (48) as follows:
proof is complete.   ∞  K (ρ R,2 ,ρ R )  ∞ −d α y
Q 42 = ε

ε
diα2 e i2 d y
A PPENDIX C i1 ,i2 ∈{1,2} ρ
2 1
ρR H (x)
−d xα α
P ROOF OF T HEOREM 3 × diα1 e i1 d x d α e−d z dz
Based on (9), (22) and (35), the coverage probability of U1   ∞  K (ρ R,2 ,ρ R )
−d α H (x)−diα x
can be expressed as follows: = diα1 ε ε
e i2 1 dx
2 1
C,I I i1 ,i2 ∈{1,2} ρ ρR
P1 α −d α z
1  ×d e
α1 ρ R |h 1 |2   dz. (62)
∗
= Pr log2 1 + ≥ R 1 , ρ R > ρ R,2 > 0 The second factor in (58), Q 5 , can be evaluated as follows:
2 α2 ρ R |h 1 |2 + 1
   1 
Q4 Q5 = log2 (1 + ρ R |h 1 |2 ) ≥ R1∗ , ρ R > 0, ρ R,2 = 0
1  2
+ Pr log2 (1 + ρ R |h 1 |2 ) ≥ R1∗ , ρ R > 0, ρ R,2 = 0 .  ε1 ε ε2 

 2   = |h 1 |2 > , < |h| 2

<
Q5 η(ρ|h|2 − ε) ρ ρ

(58)  2ε
(d α +d α )ε
ρ − 1 2 1 α
= e η(ρz−ε) d α e−d z dz
The first factor in (58), Q 4 , can be expressed as follows: ε
 
ρ
∞ (d1α +d2α )ε1
Q 4 = Pr 1 + ρ R |h 1 |2 > (ε1 + 1)(1 + α2 ρ R |h 1 |2 ), − α
 = e η(ρz−ε) d α e−d z dz
ε
ρ
ρ R > ρ R,2 > 0  ∞ (d1α +d2α )ε1
 − e
− η(ρz−ε)
α
d α e−d z dz. (63)
1 + ρ R,2 |h 2 |2 − 1 ρ R |h 1 |2 − ε1
ε2
= Pr < , ρ
|h 2 | 2 (ε1 + 1)|h 1 |2
Similar to (49), the first integral in (63) can be obtained as
ε1
ρ R > ρ R,2 > 0, |h 1 |2 > follows:
ρR  ∞ (d α +d α )ε
 − 1 2 1 α
ε1 e η(ρz−ε) d α e−d z dz
= Pr |h 2 | > H (|h 1| ), ρ R > ρ R,2 > 0, |h 1 | >
2 2 2
, ε
ρR ρ
  
(59) α
− dρ ε 4d α (d1α + d2α )ε1 4d α (d1α + d2α )ε1
=e K1 .
ρ −2g(|h |2 ) ρ R |h 1 | −ε1
where H (|h 1|2 ) = R,2g 2 (|h |2 )1 , and g(|h 1 |2 ) = (ε
2
ηρ ηρ
2.
1 1 +1)|h 1 |
Recall that the users’ channel gains have been ordered as (64)
|h 1 |2 ≤ |h 2 |2 . Therefore, Q 4 in (59) can be further expressed Substituting (61)-(63), and (64) into (58), and using the fact
as follows: C,I I
that P1C,I I = 1 − P1 , the first part of the theorem is proved.
 
Q 4 = Pr |h 2 |2 > |h 1 |2 > K (ρ R,2 , ρ R ) Note that (ρ R,2 , ρ R ) in K (ρ R,2 , ρ R ) can be upper bounded
   as follows:
Q  2
 41
ε1  (ρ R,2 , ρ R ) = ρ R,2 (ε1 + 1)2 − 2ρ R + 4ρ 2R (ε12 + 2ε1 )
+ Pr |h 2 | > H (|h 1|2 ),
2
< |h 1 |2 < K (ρ R,2 , ρ R ) ,  
ρR 4ρ R,2 ρ R 4ρ 2R
   = (ε1 + 1) ρ R,2 −
4 2
+
Q 42 (ε1 + 1)2 (ε1 + 1)2
(60) < (ε1 + 1) (ρ R,2 + ρ R ) .
4 2
(65)

Authorized licensed use limited to: Khon Kaen University provided by UniNet. Downloaded on November 09,2020 at 02:43:25 UTC from IEEE Xplore. Restrictions apply.
 4342 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 16, NO. 7, JULY 2017

(d1α +d2α )ε1

Therefore, K (ρ R,2 , ρ R ) can be upper bounded by applying Similar to Q 41 ,
∞ −
e η(ρz−ε)
α
d α e−d z dz in Q 5 can be
ε2
(65) as follows: ρ
 bounded as follows:
ρ R,2 (ε1 + 1)2 − 2ρ R + (ρ R,2 , ρ R )  ∞ (d α +d α )ε
K (ρ R,2 , ρ R ) = ln ρ − 1 2 1 α 1
2ρ 2R 1− < ε e η(ρz−ε) d α e−d z dz < 1 − . (72)
ρ 2 ρ
ρ R,2 (ε1 + 1)2 − 2ρ R + (ε1 + 1)2 (ρ R,2 + ρ R ) ρ
<
2ρ 2R Combining (67), (68), (70), (71) with (72), and noting that
3(ε1 + 1)2 c1 P1C,I I = 1 − Q 41 − Q 42 − Q 5 , we have
< < , (66)
2ρ R ρ R,2 ln ρ
P1C,I I ≈ . (73)
ρ
where c1 = 3(ε12+1) .
2

Based on (66), Q 41 in (61) can be lower bounded as follows: Therefore, the second part of the theorem is proved, and the
proof is complete.
Q 41
 ∞ α α α
= dα
ε2
e−(d1 +d2 )K (ρ R,2 ,ρ R )−d z dz R EFERENCES
ρ
[1] “5G radio access: Requirements, concept and technologies,”
 ∞ −
(d1α +d2α )c1 α
−d z NTT DOCOMO, Inc., Tokyo, Japan, 5G White Paper, Jul. 2014.
α η(ρz−ε2 )
≥d
ε2
e dz [2] Proposed Solutions for New Radio Access, Mobile and Wireless Com-
ρ
munications Enablers for the 2020 Information Society (METIS), Deliv-
  α α  erable D.2.4, Helsinki Aalto Univ., Malte Schellmann, Huawei ERC,
α 4d α (d1α + d2α )c1 ε1
− dρ ε 4d (d1 + d2α )c1 ε1 Espoo, Finland, Feb. 2015,
=e K1 [3] “3rd generation partnership project (3GPP), study on downlink multi-
ηρ ηρ user superposition transmission for LTE, TSG RAN Meeting 67,”
 d α ε  d α (d1α + d2α )c1 ε1 d α (d1α + d2α )c1 ε1  Tech. Rep. RP-150496, Mar. 2015.
≈ 1− 1+ ln [4] Z. Ding, Z. Yang, P. Fan, and H. V. Poor, “On the performance of
ρ ηρ ηρ non-orthogonal multiple access in 5G systems with randomly deployed
ln ρ users,” IEEE Signal Process. Lett., vol. 21, no. 12, pp. 1501–1505,
≈ 1− . (67) Dec. 2014.
ρ [5] Y. Saito, A. Benjebbour, Y. Kishiyama, and T. Nakamura, “System-level
−(diα +diα )κ1 performance evaluation of downlink non-orthogonal multiple access
Since e 1 2 ≤ 1, Q 41 in (61) can be upper bounded (NOMA),” in Proc. IEEE Annu. Symp. Pers., Indoor Mobile Radio
as follows: Commun. (PIMRC), London, U.K., Sep. 2013, pp. 611–615.
 ∞ [6] Z. Yang, Z. Ding, P. Fan, and G. K. Karagiannidis, “On the performance
α α α α
Q 41 = d
ε2
e−(d1 +d2 )K (ρ R,2 ,ρ R )−d z dz of non-orthogonal multiple access systems with partial channel informa-
tion,” IEEE Trans. Commun., vol. 64, no. 2, pp. 654–667, Feb. 2016.
ρ
 ∞
[7] Y. Liu, Z. Ding, M. Elkashlan, and J. Yuan, “Nonorthogonal multiple
α access in large-scale underlay cognitive radio networks,” IEEE Trans.
≤ dα
ε2
e−d z dz Veh. Technol., vol. 65, no. 12, pp. 10152–10157, Dec. 2016.
ρ [8] Z. Ding, P. Fan, and H. V. Poor, “Impact of user pairing on 5G

dα ε nonorthogonal multiple-access downlink transmissions,” IEEE Trans.
− ρ2 1
=e ≈1− . (68) Veh. Technol., vol. 65, no. 8, pp. 6010–6023, Aug. 2016.
ρ [9] S. Timotheou and I. Krikidis, “Fairness for non-orthogonal multiple
access in 5G systems,” IEEE Signal Process. Lett., vol. 22, no. 10,
−diα H (x)
Since e 2 ≤ 1, Q 42 in (62) can be upper bounded as pp. 1647–1651, Oct. 2015.
[10] J. Choi, “On the power allocation for a practical multiuser superposition
follows:
  ∞  κ1
scheme in NOMA systems,” IEEE Commun. Lett., vol. 20, no. 3,
−diα x α pp. 438–441, Mar. 2016.
Q 42 ≤ diα1
ε2 ε1
e 1 d x d α e−d z dz [11] J. Men and J. Ge, “Non-orthogonal multiple access for multiple-antenna
i1 ,i2 ∈{1,2} ρ ρR relaying networks,” IEEE Commun. Lett., vol. 19, no. 10, pp. 1686–1689,
  ∞ ε1  Oct. 2015.
−diα −diα K (ρ R,2 ,ρ R ) α
=
ε2
e 1 ρR −e 1 d α e−d z dz. [12] L. R. Varshney, “Transporting information and energy simultaneously,”
in Proc. IEEE Int. Symp. Inf. Theory (ISIT), Toronto, ON, Canada,
i1 ,i2 ∈{1,2} ρ
Jul. 2008, pp. 1612–1616.
(69) [13] X. Zhou, R. Zhang, and C. K. Ho, “Wireless information and power
transfer: Architecture design and rate-energy tradeoff,” IEEE Trans.
Based on (67) and (68), Q 42 can be further bounded as Commun., vol. 61, no. 11, pp. 4754–4767, Nov. 2013.
follows: [14] A. A. Nasir, X. Zhou, S. Durrani, and R. A. Kennedy, “Relaying
protocols for wireless energy harvesting and information processing,”
ln ρ 1
0 < Q 42 < − . (70) IEEE Trans. Wireless Commun., vol. 12, no. 7, pp. 3622–3636, Jul. 2013.
ρ ρ [15] Z. Ding, S. M. Perlaza, I. Esnaola, and H. V. Poor, “Power allocation
 α α α   strategies in energy harvesting wireless cooperative networks,” IEEE
dαε 4d (d1 +d2 )ε1 4d α (d1α +d2α )ε1
Similar to (67), e− ρ
Trans. Wireless Commun., vol. 13, no. 2, pp. 846–860, Feb. 2014.
ηρ K 1 ηρ in [16] H. Chen, Y. Li, Y. Jiang, Y. Ma, and B. Vucetic, “Distributed power
Q 5 can be approximated as follows: splitting for SWIPT in relay interference channels using game theory,”
  α α 
IEEE Trans. Wireless Commun., vol. 14, no. 1, pp. 410–420, Jan. 2015.
α α α 4d (d1 + d2α )ε1 [17] I. Krikidis, “Simultaneous information and energy transfer in large-scale
− d ρ ε 4d (d1 + d2 )ε1
α
e K1 networks with/without relaying,” IEEE Trans. Commun., vol. 62, no. 3,
ηρ ηρ pp. 900–912, Mar. 2014.
[18] I. Krikidis, “Relay selection in wireless powered cooperative networks
ln ρ with energy storage,” IEEE J. Sel. Areas Commun., vol. 33, no. 12,
≈ 1− . (71)
ρ pp. 2596–2610, Dec. 2015.

Authorized licensed use limited to: Khon Kaen University provided by UniNet. Downloaded on November 09,2020 at 02:43:25 UTC from IEEE Xplore. Restrictions apply.
 YANG et al.: THE IMPACT OF POWER ALLOCATION ON COOPERATIVE NOMA NETWORKS WITH SWIPT 4343

[19] K.-H. Liu, “Performance analysis of relay selection for cooperative
relays based on wireless power transfer with finite energy storage,” IEEE
Trans. Veh. Technol., vol. 65, no. 7, pp. 5110–5121, Jul. 2016.
[20] Y. Gu, H. Chen, Y. Li, and B. Vucetic. (2016). “Distributed multi-relay
selection in accumulate-then-forward energy harvesting relay networks.”
[Online]. Available: http://arxiv.org/abs/1602.00339
[21] Y. Liu, Z. Ding, M. Elkashlan, and H. V. Poor, “Cooperative non-
orthogonal multiple access with simultaneous wireless information
and power transfer,” IEEE J. Sel. Areas Commun., vol. 34, no. 4,
pp. 938–953, Apr. 2016.
[22] D. N. C. Tse, P. Viswanath, and L. Zheng, “Diversity-multiplexing
tradeoff in multiple-access channels,” IEEE Trans. Inf. Theory, vol. 50,
no. 9, pp. 1859–1874, Sep. 2004.
[23] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and
Products, 6th ed. New York, NY, USA: Academic, 2000.
[24] J. A. Hussein, S. S. Ikki, S. Boussakta, and C. C. Tsimenidis, “Per-
formance analysis of opportunistic scheduling in dual-hop multiuser
underlay cognitive network in the presence of cochannel interference,”
IEEE Trans. Veh. Technol., vol. 65, no. 10, pp. 8163–8176, Oct. 2016.
[25] H. A. David and H. N. Nagaraja, Order Statistics, 3rd ed. New York,
NY, USA: Wiley, 2003.
[26] F. B. Hildebrand, Introduction to Numerical Analysis. New York, NY,
USA: Dover, 1987.
Pingzhi Fan (M’93–SM’9–F’15) received the Ph.D.
degree in electronic engineering from Hull Univer-
sity, U.K.
He is currently a Professor and the Director of
the Institute of Mobile Communications, Southwest
Jiaotong University, China. He has over 200 research
papers published in various academic English jour-
nals, including the IEEE, IEE, and IEICE, and eight
books (incl. edited). He is the inventor of 22 granted
patents. He was a recipient of the U.K. ORS Award,
the Outstanding Young Scientist Award from NSFC,
and was a chief scientist of a National 973 Research Project. He has served as
the general chair or the TPC chair of a number of international conferences.
He is the guest editor-in-chief, a guest editor, or an editorial member of several
international journals. He is the Founding Chair of the IEEE VTS BJ Chapter
and the IEEE ComSoc CD Chapter, and the Founding Chair of the IEEE
Chengdu Section. He also served as a Board Member of the IEEE Region 10,
the IET (IEE) Council, and the IET Asia-Pacific Region.
His research interests include high mobility wireless communications, 5G
technologies, wireless networks for big data, and signal design and coding.
He is an IEEE VTS Distinguished Lecturer from 2015 to 2017 and a fellow
of the IET, CIE, and CIC.

Zheng Yang (S’12) received the B.S. degree

in mathematics from Minnan Normal University,
Zhangzhou, in 2008, the M.S. degree in mathemat-
ics from Fujian Normal University, Fuzhou, China,
in 2011, and the Ph.D. degree in information and
communications engineering from Southwest Jiao-
tong University, Chengdu, China, in 2016.
Dr. Yang was a visiting Ph.D. student with the
School of Electrical and Electronic Engineering,
Newcastle University, Newcastle upon Tyne, U.K.,
in 2014. He is currently a Lecturer with the College
of Photonic and Electronic Engineering, Fujian Normal University. His
research interests include 5G networks, cooperative and energy harvesting
networks, and signal design and coding.

Zhiguo Ding (S’03–M’05–SM’15) received the

B.Eng. degree in electrical engineering from the
Beijing University of Posts and Telecommunica-
tions in 2000, and the Ph.D. degree in electri-
cal engineering from Imperial College London in
2005. From 2005 to 2014, he was with Queen’s
University Belfast, Imperial College London, and
Newcastle University. From 2012 to 2016, he was
an academic visitor with Princeton University. Since
2014, he has been a Chair Professor with Lancaster
University.
His research interests are 5G networks, game theory, cooperative and energy
harvesting networks, and statistical signal processing. He received the Best
Paper Award at the IET Communication Conference on Wireless, Mobile and Naofal Al-Dhahir (S’89–M’90–SM’98–F’08)
Computing in 2009, was an IEEE C OMMUNICATIONS L ETTERS Exemplary received the Ph.D. degree in electrical engineering
Reviewer in 2012, and held the EU Marie Curie Fellowship from 2012 to from Stanford University. From 1994 to 2003, he
2014. He is serving as an Editor for the IEEE T RANSACTIONS ON C OMMU - was a Principal Member of the Technical Staff with
NICATIONS , the IEEE T RANSACTIONS ON V EHICULAR T ECHNOLOGIES and the GE Research and AT&T Shannon Laboratory.
the Journal of Wireless Communications and Mobile Computing. He was an He is currently the Erik Jonsson Distinguished
Editor for the IEEE W IRELESS C OMMUNICATION L ETTERS and the IEEE Professor with the University of Texas at Dallas.
C OMMUNICATION L ETTERS from 2013–2016. He is a Leading Guest Editor He has co-authored over 325 papers with over
for the IEEE J OURNAL ON S ELECTED A REAS IN C OMMUNICATIONS Special 7800 citations. He is a co-inventor of 41 issued
Issue on Non-Orthogonal Multiple Access for 5G and a Guest Editor for the U.S. patents. He was a co-recipient of four IEEE
IEEE Wireless Communications Special Issue on Non-Orthogonal Multiple best paper awards. He is the Editor-in-Chief of the
Access. IEEE T RANSACTIONS ON C OMMUNICATIONS.

Authorized licensed use limited to: Khon Kaen University provided by UniNet. Downloaded on November 09,2020 at 02:43:25 UTC from IEEE Xplore. Restrictions apply.