Received June 7, 2019, accepted June 28, 2019, date of publication July 4, 2019, date of current version July

30, 2019.
Digital Object Identifier 10.1109/ACCESS.2019.2926757

Low-Complexity Power Allocation in NOMA

Systems With Imperfect SIC for Maximizing
Weighted Sum-Rate
XIAOMING WANG 1,2,3 , (Member, IEEE), RUILU CHEN3 ,
YOUYUN XU 1 , (Senior Member, IEEE),
AND QINGMIN MENG1
1 College of Telecommunication and Information Engineering, Nanjing University of Posts and Telecommunications (NJUPT), Nanjing 210003, China
2 Key Laboratory of Broadband Wireless Communication and Sensor Network Technology Ministry of Education, Nanjing University of Posts and
Telecommunications (NJUPT), Nanjing 210003, China
3 National Mobile Communications Research Laboratory, Southeast University, Nanjing 210096, China

Corresponding author: Xiaoming Wang (xmwang@njupt.edu.cn)

This work was supported in part by the National Key Research and Development Program of China under Grant 2016YFE0200200, in part
by the National Natural Science Foundation of China under Grant 61801240, Grant 61601243, and Grant 61772287, in part by the Natural
Science Foundation of the Jiangsu Providence under Grant BK20180753, in part by the Natural Science Foundation of the Jiangsu Higher
Education Institutions of China under Grant 17KJB510046, in part by the Open Research Fund of the Key Laboratory of Broadband
Wireless Communication and Sensor Network Technology, Ministry of Education, under Grant JZNY201702, and in part by the Open
Research Fund of the National Mobile Communications Research Laboratory, Southeast University, under Grant 2019D16.

ABSTRACT In this paper, we investigate the power allocation for maximizing weighted sum rate (WSR) in
downlink multiple carriers non-orthogonal multiple access (MC-NOMA) systems with imperfect successive
interference cancellation (SIC). We formulate the power allocation problem as a non-convex optimization
problem with the total power constraint of all sub-channels while considering often-neglected issues of SIC
error and power order constraints at users. First, we discuss that the optimization problem assuming receivers
can perform perfect SIC, and we provide a concavity condition of the WSR maximization problem for the
MC-NOMA system. When the concavity condition is not satisfied, a fractional quadratic transformation
is used to overcome the difficulty of problem non-convexity. Based on the transformation, we propose an
iterative power allocation algorithm. Then, we consider the SIC error and the power order constraints in
the optimization problem and present a power allocation method with imperfect SIC. Moreover, for both
the perfect and imperfect SIC, we derive some propositions of the optimal power allocation solution to
the WSR maximization problem and propose a low-complexity power allocation algorithm based on these
propositions. Finally, we provide a joint user scheduling and power allocation algorithm for maximizing
the WSR. The simulation results illustrate that the proposed resource allocation methods have a better
performance than the existing schemes.

INDEX TERMS Non-orthogonal multiple access, power allocation, weighted sum-rate, successive interfer-
ence cancellation, low-complexity.

I. INTRODUCTION frequency resources, which is able to provide higher spectral

Non-multiple access (NOMA) has been recognized as a
promising multiple access technique for the future wireless
networks to meet the demands for high data-rates, low
latency, and massive connectivity [1], [2]. In particular,
by using the successive interference cancelation (SIC),
NOMA can remove the co-channel interference among
the users, to exploit the channel diversity more effi-
ciently. NOMA allows multiple users utilizing the same
The associate editor coordinating the review of this manuscript and
approving it for publication was Xiaofei Wang.
efficiency and support more connections comparing with
conventional orthogonal multiple access (OMA) techniques,
such as orthogonal frequency division multiple access
(OFDMA). In addition, with the advantages of fairness,
compatibility and flexibility, NOMA has therefore attracted
increasing attention [2].
Recently, different issues in NOMA systems have been
addressed in the literatures. From an information theoretic
perspective, the capacity region for NOMA was studied
in [3]. Meanwhile, the performance analyses of NOMA sys-
tems were given in [4]–[6]. They show that NOMA can

94238 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see http://creativecommons.org/licenses/by/4.0/ VOLUME 7, 2019
X. Wang et al.: Low-Complexity Power Allocation in NOMA Systems With Imperfect SIC
offer considerable performance gains over OMA in terms
of sum-rate and outage probability. To realize the gains
promised by the theoretic works, a simple practical cod-
ing and modulation design was proposed in [7]. Mean-
while, apart from the investigations of NOMA performance,
some implementation issues of NOMA in wireless networks
were addressed in [8], [9], such as signalling overhead,
SIC error propagation, and combining NOMA with beam-
forming or cognitive radio. However, most of the former
works assumed that fixed resource allocation were adopted
in NOMA systems.
A. MOTIVATION AND RELATED WORK
Resource allocation, especially power allocation, plays an
important role in exploiting the potential performance gains
of both NOMA and transitional OMA schemes [10]. In order
to enhance the performance of the NOMA systems, different
resource allocation schemes have been investigated. Several
researchers studied resource allocation schemes in NOMA
systems starting from the scenario of single carrier NOMA
(SC-NOMA) (e.g., [11], [12]), or the scenario of two users
on each subcarrier (e.g., [14]–[18]). They tried their best
to improve system performance, such as sum-rate, energy-
efficiency, fairness and outage performance. The authors
in [11] proposed a dynamic power allocation scheme where
two users are paired to perform SC-NOMA for achieving
better outage performance. In [12], the authors investigated
an optimal power allocation scheme for SC-NOMA with
adaptive rates and α-fairness. In [13], the authors studied the
optimal power allocation for weighted sum-rate (WSR) maxi-
mization in SC-NOMA systems with power order and quality
of service (QoS) constraints. In [14], [15], user fairness was
considered in two-user NOMA systems, where a balance
between throughput and fairness was achieved. By applying
NOMA in multicarrier (MC-NOMA) systems, spectral effi-
ciency can be further improved due to the degrees of freedom
offered by multiuser diversity. For two-user MC-NOMA sys-
tems, several performance criteria were considered for power
allocation in [16], [17]. In [18], the authors provided a joint
power and subcarrier allocation with the target of maximiz-
ing WSR, and they still focused on two users multiplexed on
each subcarrier.
Afterwards, more and more researchers turned their focus
into the MC-NOMA where each subchannel can support
more than two users. In [19], the authors extended two-
user scenario to the user-pairing multi-user case to max-
imize the sum rate of NOMA systems while considering
the proportional fairness of users. In [20], two candidate
non-orthogonal techniques were considered in heterogeneous
cellular systems, where the power allocation and subcar-
rier allocation were investigated for the two non-orthogonal
schemes. Aiming to minimize the total transmit power of
NOMA systems, the authors in [21] proposed a relax-then-
adjust power allocation algorithm and the authors in [22]
proposed a joint subcarrier assignments and power alloca-
tion scheme. In [23], the authors tried their best to transform
VOLUME 7, 2019
original probabilistic and non-convex problem into a non-
probabilistic and convex one with the target of maximizing
energy efficiency with imperfect channel state informa-
tion. Considering the constraints of user QoS and queue
stability, literatures [24] and [25] investigated power allo-
cation in MC-NOMA networks using the stochastic opti-
mization method. In [26] and [27], for uplink and downlink
MC-NOMA systems respectively, the authors studied power
and subcarrier allocations to maximize WSR.
Meanwhile, combining with other types of wireless tech-
niques, resource allocations for NOMA have been gradu-
ally studied in different communication scenarios. In [28],
the authors proposed an architecture of NOMA-based fog
radio access networks to meet the heterogeneous demands
of users, and studied the power and subchannel allocation
problems in this systems. In [29], a joint access selection and
heterogeneous resource allocation in ultra-dense networks
with mobile edge computing was investigated. In [30], several
radio resource allocation problems were studied in visual-
ized software defined-based network architecture combing
coordinated multi-point and NOMA, aiming of maximizing
the total throughput. In [31], a Nash bargaining game of
power allocation and user clustering for uplink MC-NOMA
in device-to-device underlaid cellular networks was investi-
gated. In [32], a low-complexity power allocation was pro-
posed to maximize the sum-rate in the downlink NOMA
systems applying multiple-input multiple-output (MIMO)
transceivers. In [33], the authors studied a secure beam-
forming for MIMO-NOMA based cognitive radio network,
which can maximize the total secrecy rate. In [34] and [35],
the application of NOMA in millimeter wave communica-
tions was investigated, in which user scheduling and power
allocation were studied to improve cell capacity or sum-rate.
Obviously, for most of the above NOMA scenarios,
(weighted) sum-rate maximized power allocation is one of the
vital issues in resource allocation. However, the SIC issue was
usually ignored in design of the power allocation algorithms.
Most of the above works assumed that SIC at the receiver is
perfect, although it may not be easy to accomplish due to the
weak detection capabilities of simple mobile devices in prac-
tical systems [36]–[38]. Some outage-constrained power allo-
cation schemes were proposed in [39]–[41] to minimize the
total transmit power for downlink and uplink NOMA systems
with imperfect SIC respectively. Furthermore, to guarantee
the accuracy of SIC in the NOMA systems, the later decoded
users should be allocated lower power than the formers in SIC
process, which is called power order constraints [13], [16].
So far both the SIC error and the power order constraints were
not well taken into account in most of the existing works.
Particularly, the relationship between the optimal solutions to
WSR-maximized optimization problems of power allocation
in OMA and NOMA systems were not well studied.
B. CONTRIBUTIONS AND ORGANIZATION
In this paper, we investigate the power allocation in downlink
MC-NOMA systems aiming to maximize WSR. We focus on
94239
 X. Wang et al.: Low-Complexity Power Allocation in NOMA Systems With Imperfect SIC
the power allocation with given channel assignment, while
considering the SIC error and the power order constraints
of users. We formulate the power allocation problem as
a non-concave optimization problem with the sum power
constraints of all sub-channels and power order constraints
on each sub-channel. Note that, combining with other wire-
less technologies, our proposed schemes will be able to be
extended to a variety of NOMA application scenarios.
The main contributions of this paper can be summarized as
follows:
• We formulate the power allocation problem for
MC-NOMA systems as a non-convex optimization
problem, considering both SIC error at the receivers and
power order constraints of users on each subchannel.
• We solve the formulated non-convex weighted max-
imization problems by using a factional quadratic
transform and an iterative algorithm. To the best of
our knowledge, no such method for NOMA systems is
studied in the existing literatures.
• We obtain some propositions of the optimal power
allocation solution to the WSR maximization problem
with or without perfect SIC. And based on these propo-
sitions, we propose a low-complexity power allocation
algorithm for MC-NOMA.
• We provide a concavity condition that should be satis-
fied by the weighting factors of users in MC-NOMA
systems. In this case, the power optimization problem
can be solved directly by standard convex optimization
methods.
• We propose a joint user scheduling and power allocation
scheme for maximizing WSR in MC-NOMA systems
with imperfect SIC.
The rest of the paper is organized as follows. Section II
introduces the system model and formulates the problem as
a non-convex optimization problem. Section III discusses the
power allocation in MC-NOMA with perfect SIC, and solves
the problem with ideal assumptions. Section IV investigates
the optimization problem with imperfect SIC, in which we
study the power allocation in MC-NOMA with SIC error
and power order constraints. Section V studies the joint user
scheduling and power allocation scheme. Numerical results
are provided in Section VI, and the paper is concluded in
Section VII.
II. SYSTEM MODEL AND PROBLEM FORMULATION
A. SYSTEM MODEL
We consider a downlink MC-NOMA system where there
is one base station (BS) located in the center, and users
randomly and uniformly distributed in the cell. Suppose that
the total bandwidth is Btotal , and we divide it equally into N
orthogonal subchannels. We denote the set of subchannels as
N = {1, 2, · · · , N }. Thus, the bandwidth of each subchannel
is Bf = Btotal /N . Assume that each subchannel can support
at most K users. The set of users on subchannel n is denoted
by Kn = {Usern,1 , · · · , Usern,K }. Note that, we assume that
user scheduling has be given and focus on power allocation in
94240
Section II to Section IV. The joint user scheduling and power
allocation scheme is provided in Section V.
Since superposition coding is used at transmitter, the trans-
PK √signal on subchannel n can be expressed as xn =
mitted
k=1 pn,k Sn,k , where Sn,k and pn,k are the signal message
and the power allocated into Usern,k respectively, and Sn,k
satisfies E{|Sn,k |2 } = 1. We denote hn,k as the channel fading
between BS and Usern,k . Consequently, the corresponding
received signal of Usern,k can be expressed as
K
X √
yn,k = hn,k xn + zn,k = hn,k pn,i Sn,i + zn,k , (1)
i=1
where zn,k is the additive zero-mean Gaussian noise which
is dependent and identical distributed with variance σz2 ,
i.e., zn,k ∈ CN (0, σz2 ). Without loss of generality, we assume
that channel gains of on subchannel n are sorted as |hn,1 | ≥
|hn,2 | ≥ . . . ≥ |hn,K |, ∀n. In the existing literatures, most of
the studies assumed that the receiver can perform perfect SIC.
That is to say, for any user k, the signals of later K − k users
can be decoded and be removed on the same subchannel.
Based on it, a lot of work on resource allocation in NOMA
systems has been studied.
However, the issue of imperfect SIC is important but often
overlooked for resource allocation in NOMA system, which
is studied insufficiently. On the one hand, to perform SIC,
the NOMA users have to be allocated reasonable power
levels [13], [16], i.e.,
pn,k+1 > pn,k , ∀k ∈ {1, · · · , K − 1}, ∀n. (2)
The constraint (2) is a necessary condition to guarantee SIC
operation, since it is used for distinguishing between the
signal to be decoded and the remaining non-decoded message
signals [42]. On the other hand, due to the limited detection
ability of mobile users and the impact of modulation and
decoding schemes, the receiver cannot always perform per-
fect SIC operation, even if (2) is satisfied.
When the SIC can be not implemented perfectly, the later
K − k users have some residual parts, which will be seen as
noise for user k. In this case, the interference at user k on
subchannel n resulting from imperfect cancellation as In,kSIC ,
can be expressed as
K
X
SIC
In,k = |hn,k |2 pn,i |xi − x̃i |2
i=1,|hn,i |2 <|hn,k |2
K
X
= |hn,k |2 pn,i |xi − x̃i |2 , (3)
i=k+1
where x − x̃ means the difference between actual signal xi and
the estimated signal x̃i , and ε = E{|xi − x̃i |2 } [36]. Note that,
due to the limited detection ability of mobile users and the
impact of modulation and decoding schemes, the receivers
can not always perform perfect SIC operation [37]. In this
case, the signal to interference plus noise ratio (SINR) of
VOLUME 7, 2019
X. Wang et al.: Low-Complexity Power Allocation in NOMA Systems With Imperfect SIC

user k on subchannel n can be defined as

Imperf
SINRn,k
|hn,k |2 pn,k
= . (4)
|hn,k |2 k−1
PK
i=1 pn,i + ε|hn,k | i=k+1 pn,i + σz
2 2
P

And the corresponding rate for user k on subchannel n is

 
Imperf Imperf
Rn,k (P) = Bf log2 1 + SINRn,k (P) , (5)

where P = [pn,k ]N ×K is power allocation matrix.

B. PROBLEM FORMULATION FIGURE 1. Concavity of sum-rate maximization problem in NOMA

We aim to maximize WSR of the MC-NOMA system by systems (i.e., ωn,k = 1, ∀n, k).
designing the power allocation. Based on (2), we consider
a necessary condition to guarantee SIC operation, which is
given by C3 and C4 below. We denote ωn,k as weighting III. POWER ALLOCATION IN MC-NOMA WITH IDEAL SIC
factor of user k on subchannel n which is usually used to In this section, we consider an ideal scenario, assuming that
SIC =
the receivers can always perform perfect SIC, i.e., In,k
adjust the user fairness. The WSR maximization problem can
be formulated as 0, ∀n, k without regard to C3 and C4. The optimization prob-
lem can be reformulated as
K
N X
Imperf
X
max ωn,k Rn,k (P) X K
N X
P
n=1 k=1
max ωn,k RIdeal
n,k (P)
P
N X K n=1 k=1
s.t. C1, C2, (7)
X
s.t. C1: pn,k ≤ Ptotal ,
n=1 k=1
where RIdeal SIC
n,k (P) is given by (5) with In,k = 0.
C2: pn,k ≥ 0, ωn,k ≥ 0, ∀n, k,
C3: pn,k+1 ≥ αn,k pn,k , ∀k ∈ {1, · · · , K − 1}, ∀n, A. CONVEXITY ANALYSIS OF OPTIMIZATION PROBLEM
C4: αn,k ≥ 1, ∀n, k, (6) Almost all past works regarded the sum-rate maximization
problem in NOMA systems with perfect SIC as a non-convex
where Ptotal represents the total power of the BS. C1 indicates problem. However, the literature [43] succeeded in demon-
that the sum power of all users on all subchannels can not strating that sum-rate maximization problem (without con-
exceed the power limit of BS. C3 and C4 indicate that the sidering weighting factors) in SC-NOMA scenario is convex.
power assigned on each subchannel sequentially increases For WSR maximization problem in MC-NOMA systems,
with k. These two constraints can ensure that the power of we demonstrate concavity conditions and get Theorem 1 as
user k + 1 is greater than the power of user k by a certain follows.
amount on each subchannel. αn,k represents the minimum Theorem 1: Weighted sum-rate maximization problem (7)
requirement of the ratio of power levels between user k for MC-NOMA systems with perfect SIC is concave with
and user k + 1 on subchannel n. The value of α depends respect to variable P if and only if
on the detection capabilities of mobile devices, and simpler ! !
receivers may need larger α for interference cancellation. σz2 σz2
ωn,k+1 + qn,k ≤ ωn,k + qn,k , (8)
The problem (6) is not a standard convex optimization |hn,k |2 |hn,k+1 |2
problem and the global optimum can not be easily obtained
where qn,k = ki=1 pn,i .
P
in practice. In the following sections, we first consider an
ideal scenario assuming perfect SIC can be always realized at Proof: See Appendix A.
receivers. Although this ideal assumption is impractical, it is a To guarantee the concavity of (7), the concavity conditions
basic assumption in most of the existing studies, in which we in (8) should be satisfied. In this case, the concave problem
provide some insights for the design of MC-NOMA systems. (7) can be solved by the convex optimization tools. Note that,
Then, we consider more practically the optimization problem ωn,k = 1, ∀n, k, is a special case of Theorem 1. Substituting
with SIC error and power order constraints for MC-NOMA it into (8), we can see that the inequality always holds due to
system, and provide some power allocation algorithms. Par- |hn,k |2 ≥ |hn,k+1 |2 , which indicates that sum-rate maximiza-
ticularly, we propose a low-complexity algorithm for NOMA tion problem in MC-NOMA is concave. Fig. 1 illustrates the
systems both with and without perfect SIC based on derived concavity of the sum-rate maximization problem for a two-
propositions. user case with power p1 and p2 .

VOLUME 7, 2019 94241

 X. Wang et al.: Low-Complexity Power Allocation in NOMA Systems With Imperfect SIC

B. SOLUTION TO OPTIMIZATION PROBLEM The problem (14) is equivalent to

WITH IDEAL SIC
J
However, the conditions in (8) do not always hold in the max
X
(2θj Aj (x) − θj2 Bj (x))
p
real systems for varied weighting factors due to the het- x,θ
j=1
erogenous QoS requirements of users. Therefore, in the most
s.t. x ∈ X , θj ∈ R, (15)
cases, the optimization problem (7) is a non-convex problem.
We now solve the problem by converting it into a convex where θ = {θ1 , · ·p
· , θJ } denotes the auxiliary variable, and
form. the optimal θj∗ = Aj (x)/Bj (x) [45], [46].
We introduce extra auxiliary variables {γn,k }, and trans- According to Lemma 1, we rewrite the optimization
form the problem (7) into an equivalent epigraph form [44] as problem (13) by the following quadratic transform with intro-
N X
X K duced variables {θn,k }, as
max ωn,k Bf log2 (1 + γn,k ) N X
K 
P,γ X q
n=1 k=1 max 2θn,k ωn,k Bf |hn,k |2 pn,k (1 + γn,k )
s.t. C1, C2, P,γ ,θ
n=1 k=1
|hn,k |2 pn,k k
!
,

C5:γn,k ≤ ∀n, k. (9) X
|hn,k |2 k−1i=1 pn,i + σz
2 − θn,k
2 2
pn,i + σz2 ln 2 + g(γ , θ)
P
|hn,k |
i=1
It is easy to see that for a fixed P, the problem (9) is a concave s.t. C1, C2, (16)
problem with respect to γ = [γn,k ]N ×K . The Lagrangian
γn,k

function with dual variable µn,k for C5 can be given by where g(γ , θ) = ωn,k Bf log2 (1 + γn,k ) − ln 2 .
N X
K Theorem 2: For fixed γ and θ, Problem (16) is a concave
optimization problem with respect to variable P.
X
LP (γ , µ) = ωn,k Bf log2 (1 + γn,k )
n=1 k=1
Proof: See Appendix B.
N X K
! According to Theorem 2, we can compute the power allo-
X |hn,k |2 pn,k cation by Lagrangian dual method with introducing variable
− µn,k γn,k − .
|hn,k |2 k−1i=1 pn,i + σz
2 λ ≥ 0 for C1. The optimal power can be expressed by
P
n=1 k=1
(10) θn,k
2 ω B |h |2 (1 + γ )
n,k f n,k n,k
According to the first-order condition ∂L(γ ,µ)
= 0, we have
p∗n,k =  2 , (17)
∂γn,k
PK 2
θ
i=k n,i |h n,i |2 ln 2 + λ
the optimal dual variable as
where λ will be obtained by bisection or subgradient method.
 Pk−1 
ωn,k Bf / ln 2 ω n,k B f |h n,k |2 p n,i + σ 2
µ∗n,k = =
i=1 z
, The optimal θ ∗ is given by
1 + γn,k
∗ Pk
i=1 pn,i + σz ) ln 2
(|hn,k | 2 2 q
(11) ωn,k Bf |gn,k |2 pn,k (1 + γn,k )
θn,k =
∗
. (18)
(|hn,k |2 ki=1 pn,i + σz2 ) ln 2
P
where
|hn,k |2 pn,k We then propose an iterative power allocation for
γn,k
∗
= . (12) MC-NOMA systems with perfect SIC, as shown in
|hn,k |2 k−1i=1 pn,i + σz
2
P
Algorithm 1. We initialize the power pn,k = Ptotal /(NK ),
From the primal problem maxγ LP (γ , µ∗ ), we can transform ∀n, k. Compute iteratively γn,k and θn,k by (12) and (18), and
the optimization problem (9) to update the power allocation by (17).
N X K 
X  γn,k  C. LOW-COMPLEXITY POWER ALLOCATION
max ωn,k Bf log2 (1 + γn,k ) −
P,γ ln 2 WITH IDEAL SIC
n=1 k=1
ωn,k Bf |hn,k |2 pn,k (1 + γn,k )
 Before proposing the low-complexity power allocation algo-
+  rithm, we derive some propositions of optimal power alloca-
|hn,k |2 ki=1 pn,i + σz2 ln 2
P
tion with ideal SIC.
s.t. C1, C2. (13)
1) WSR IN SC-NOMA
The objective function in (13) is a sum-of-ratios form for the We first assume that the total power on each subchannel has
variable {pn,k }, which is also not a standard convex function. been obtained, so the optimization problem in MC-NOMA
Lemma 1: Define a sum-of-ratio optimization problem, systems is reduced to that in SC-NOMA systems. Omitting
J the subscripts of subchannels, we have the following propo-
X Aj (x)
max , s.t. x ∈ X . (14) sition for WSR maximization problem in SC-NOMA systems
x Bj (x) with perfect SIC.
j=1

94242 VOLUME 7, 2019

X. Wang et al.: Low-Complexity Power Allocation in NOMA Systems With Imperfect SIC

Algorithm 1 Power Allocation for Maximizing WSR in From (20), the sum-rate of NOMA for any power allo-
MC-NOMA Systems cation is always lower than or equal to the case that all
Initialization power is allocated to the user with maximum channel-gain.
1. Initialize iteration index s = 0, maximum iterations Normally, the inequation (a)Ptakes the equal sign when
sm , tm ; |h1 |2 = |hk |2 , ∀k, or |hk |2 k−1 i=1 pi  σz , ∀k, or P =
2
(0,0)
2. Initialize power allocation pn,k = Ptotal /(NK ), ∀n, k; [Ptotal , 0, · · · , 0]. Therefore, for any |h1 | ≥ |hk |2 , ∀k,
2
Iterative Algorithm we have that the sum-rate of NOMA with perfect SIC is
3. Repeat upper bounded by the rate with optimal power allocation
4. Compute variables γ (s) by (12); P∗ = [Ptotal , 0, · · · , 0].
5. t = 0; We then consider that only User K who has minimum
6. Repeat channel-gain occupies the subchannel. The rate is given by
7. Compute variables θ (s,t) by (18);
|hK |2 Ptotal
 
8. Update power allocation P(s,t) by (17); Bf log2 1 +
9. Set t = t + 1; σz2
K
!
10. Until convergence or t = tm ; σz2 + |hK |2 p1 Y σz2 + |hK |2 ki=1 pi
P
11. Set s = s + 1; = Bf log2
σz2
Pk−1
k=2 σz + |hK |
2 2
12. Until convergence or s = sm . i=1 pi
K
!
X |hK |2 pk
= RK (p1 ) + Bf log2 1 +
σz2 + |hK |2 k−1
P
k=2 i=1 pi
Proposition 1: When one of the following conditions K
!
(b) X |hk |2 pk
Cond-1 : ω1 = ω2 = · · · = ωK , ≤ R1 (p1 ) + Bf log2 1 + ,
σz2 + |hk |2 k−1
P
k=2 i=1 pi
Cond-2 : ω1 = max{ω1 , · · · , ωK }, (19) (21)
is satisfied, the WSR of SC-NOMA with perfect SIC is upper 2
 
where RK (p1 ) = Bf log2 1 + |hKσ|2 p1 . The inequation (b)
bounded by allocating all power only to the user with max- z

imum channel-gain. Meanwhile, when Cond-1 is satisfied, holds since that |hK |2 ≤ |hk |2 , ∀k < K ,
the WSR of SC-NOMA with perfect SIC is lower bounded by k−1
X k−1
X
allocating all power to the user with minimum channel-gain. ⇒ |hK |2 σz2 +|hK |2 |hk |2 pi ≤ |hk |2 σz2 +|hK |2 |hk |2 pi ,
Proof: Consider the SC-NOMA scenario with perfect SIC i=1 i=1
for ω1 = ω2 = · · · P = ωK = 1. For single carrier scenario, |hK |2 pk |hk |2 pk
we have Ptotal = K ⇒ ≤ .
k=1 pk . When only User 1 who has σz2 + |hK |2 k−1 σz2 + |hk |2 k−1
P P
maximum channel gain occupies the subchannel, the rate can i=1 pi i=1 pi
be expressed as From (21), the rate of NOMA is lower bounded by allocating
all power to the user with minimum channel-gain.
|h1 |2 Ptotal
 
Bf log2 1 + On the other hand, assuming that User 1 has the largest
σz2 weighting factor ω1P= max{ω1 , · · · , ωK }, the WSR can be
K express as RIdeal = K k=1 ωk Rk
Ideal . We have
!
σz2 + |h1 |2 p1 Y σz2 + |h1 |2 ki=1 pi ω
P
= Bf log2
σz2
Pk−1
k=2 σz + |h1 |
2 2 K
i=1 pi X
K
! RIdeal
ω ≤ ω1 RIdeal
k = ω1 RIdeal , (22)
X |h1 |2 pk k=1
= R1 (p1 ) + Bf log2 1 +
σz2 + |h1 |2 k−1
P
i=1 pi
PK Ideal . From the proof of Cond-1,
k=2 where RIdeal = k=1 Rk
K
!
(a) X |hk |2 pk we can know that the optimal solution to sum-rate maximiza-
≥ R1 (p1 ) + Bf log2 1 + , tion problem with objective function as RIdeal (P) is P∗ =
σz2 + |hk |2 k−1
P
k=2 i=1 pi [Ptotal , 0, · · · , 0].
(20) Normally, we have from (22) that the optimal Rω is
 2
 achieved to the optimal ω1 RIdeal at ωk = ω1 , ∀k. However,
where R1 (p1 ) = Bf log2 1 + |h1σ| 2p1 . The inequation (a) due to the special form of the optimal power solution, the con-
z
holds since that |h1 |2 ≥ |hk |2 , ∀k > 1, ditions ωk = ω1 , ∀k, are not necessary. Obviously, the opti-
k−1 k−1
mal RIdeal
ω is also obtained when P∗ = [Ptotal , 0, · · · , 0],
⇒ |h1 |2 σz2 +|h1 |2 |hk |2
X
pi ≥ |hk |2 σz2 +|h1 |2 |hk |2
X
pi , where Rω = ω1 RIdeal .
Ideal

i=1 i=1
2) WSR IN MC-NOMA
|h1 |2 pk |hk |2 pk
⇒ ≥ . We then consider the cases of MC-NOMA with perfect SIC
σz2 + |h1 |2 k−1 σz2 + |hk |2 k−1
P P
i=1 pi i=1 pi and we have the following proposition.

VOLUME 7, 2019 94243

 X. Wang et al.: Low-Complexity Power Allocation in NOMA Systems With Imperfect SIC

Proposition 2: When one of the following conditions Algorithm 2 Low-Complexity Power Allocation for
Maximizing WSR in MC-NOMA Systems
Cond-3 : ωn,1 = ωn,2 = · · · = ωn,K , ∀n, 1. If (Cond-3 is satisfied) or (Cond-4 is satisfied)
Cond-4 : ωn,1 = max{ωn,1 , · · · , ωn,K }, ∀n, (23) 2. Compute pn,1 by weighted water-filling method based
on |hn,1 |2 , ∀n, and set pn,k 0 = 0, ∀k 0 6= 1, ∀n.
is satisfied, the WSR of MC-NOMA with perfect SIC is upper 3. Else
bounded by OFDMA with selecting the user who has maxi- 4. Compute power allocation by Algorithm 1;
mum channel-gain on each subchannel and weighted water- 5. End
filling power allocation among the subchannels.
Proof: Consider the MC-NOMA scenario with perfect SIC
for ωn,1 = · · · P = ωn,K = 1, ∀n. The sum-rate on subchannel A. OPTIMIZATION PROBLEM WITHOUT POWER
K Ideal . For any Ptotal = PK p
n is RIdeal
n = R
k=1 n,k n k=1 n,k on ORDER CONSTRAINTS
subchannel n, we have RIdeal n (Pn ) ≤ R Ideal (P0 ), where P =
n n n To simplify the discussion, we first do not consider the con-
[pn,1 , pn,2 , · · · , pn,K ] and P0n = [Ptotal
n , 0, · · · , 0]. Denote the straints C3 and C4 in optimization problem. We can reformu-
optimal solution to the problem (7) as P∗ = {p∗n,k }, and define late the problem (6) as
RIdeal = N
P PK Ideal
n=1 k=1 Rn,k . We have N X
K
Imperf
X
N N max ωn,k Rn,k (P)
(d) P
∗ (c)
X X
Ideal
R (P ) = RIdeal
n (P∗n ) ≤ RIdeal
n (P∗∗ ), (24) n=1 k=1
n=1 n=1
s.t. C1, C2, (26)
Imperf
where P∗n = [p∗n,1 , p∗n,2 , · · · , p∗n,K ] and P∗∗
n = where Rn,k (P) is given in (5). The problem (26) is also not a
PK
[ k=1 pn,k , 0, · · · , 0]. The equation (c) holds due to the
∗ standard concave problem, and it is difficult to obtain reliable
independence between any two subchannels. The inequation concavity conditions with imperfect SIC.
(d) holds from Proposition 2, which takes the equal sign Similar to Subsection III-B, by introducing auxiliary vari-
ables {γn,k } and {θn,k }, we transform the problem (26) into a
n . It indicates that pn,k = 0, ∀k > 1, and
when P∗n = P∗∗ ∗

the optimal solution is Pn = [pn,1 , 0, · · · , 0]. In this case,

∗ ∗ concave problem as
the optimal power allocation in the MC-NOMA is obtained K
N X
by weighted water-filling method only based on h1 = Imperf-Q
X
max fn,k (P, γ , θ)
[h1,1 , h2,1 , · · · , hN ,1 ]T and ω1 = [ω1,1 , ω2,1 , · · · , ωN ,1 ]T . P,γ ,θ
n=1 k=1
On the other hand, assuming that user who has maximum s.t. C1, C2, (27)
channel gains on each subchannel (i.e., |hn,1 |2 ) has the largest
weighting factor (i.e., ωn,1 = max{ωn,1 , · · · , ωn,K }), we where fn,k
Imperf-Q
is defined as
have the WSR as q
Imperf-Q
N X
X K fn,k = 2θn,k ωn,k Bf |hn,k |2 pn,k (1 + γn,k )
RIdeal
ω = ωn,k RIdeal Imperf
n,k 2
−θn,k (In,k + |hn,k |2 pn,k ) ln 2 + g(γ , θ), (28)
n=1 k=1
N K N Imperf
and In,k denotes
X X X
≤ ωn,1 RIdeal
n,k = ωn,1 RIdeal
n . (25)
n=1 k=1 n=1
k−1 K
Imperf
X X
Similar to the proof of Cond-3, we have from (25) that the In,k = |hn,k |2 pn,i + |hn,k |2 ε pn,j + σz2 . (29)
optimal power allocation for maximizing RIdeal ω is also P∗n = i=1 j=k+1
[pn,1 , 0, · · · , 0], ∀n, where pn,1 is obtained by the weighted
∗ ∗
For a fixed γ and θ, the transformed problem (27) is a stan-
water-filling method among the different subchannels.
dard concave problem with respect to P. We can compute the
Based on Proposition 2, we then propose a low-
power allocation by Lagrangian dual method with introduced
complexity power allocation algorithm for maximizing WSR
variable β ≥ 0 for C1. Optimal power can be expressed as
in MC-NOMA systems with perfect SIC as shown in
Algorithm 2. θn,k
2 ω B |h |2 (1 + γ )
n,k f n,k n,k
p∗n,k =
2 , (30)
IV. POWER ALLOCATION IN MC-NOMA ξn,k + β
WITH IMPERFECT SIC where ξn,k is given by
In this section, we discuss the optimization problem
with imperfect SIC operation being performed at the K
X k−1
X
receivers, and propose a WSR maximized power allocation ξn,k = θn,i
2
|hn,i |2 ln 2 + ε θn,j
2
|hn,j |2 ln 2, (31)
scheme. i=k j=1

94244 VOLUME 7, 2019

X. Wang et al.: Low-Complexity Power Allocation in NOMA Systems With Imperfect SIC

and β will be obtained by bisection or subgradient method. Proof: Based on Proposition 3, we can prove Proposition 4
And the optimal γn,k is given by by a similar method to the proof of Proposition 2. The detail
is omitted.
|hn,k |2 pn,k Based on Proposition 4, for MC-NOMA systems with
γn,k
∗
= Imperf
. (32)
In,k imperfect SIC, we can obtain power allocation scheme by
Algorithm 1, in which we need to replace (12), (17) and (18)
When pn,k is held fixed, we can obtained the optimal γn,k and with (32), (30) and (33) to update γ , P and θ. In addition,
θn,k by (32) and based on Proposition 4, we can also adopt the low-complexity
q power allocation algorithm (i.e., Algorithm 2) in MC-NOMA
ωn,k Bf |hn,k |2 pn,k (1 + γn,k ) systems with imperfect SIC.
θn,k
∗
= Imperf
. (33)
(In,k + |hn,k |2 pn,k ) ln 2
B. POWER ALLOCATION IN MC-NOMA WITH SIC ERROR
Based on Proposition 1 and Proposition 2, we can obtain AND POWER ORDER CONSTRAINTS
the following propositions for SC-NOMA and MC-NOMA In this subsection, we add the necessary constraints C3 and
with imperfect SIC. C4 that have to be satisfied to realize SIC at receivers in
Proposition 3: When one of Cond-1 and Cond-2 is satis- NOMA systems into the optimization problem. In this case,
fied, the optimal solution of SC-NOMA with imperfect SIC is there is also a certain degree of probability of existing SIC
equal to the solution in SC-NOMA with perfect SIC. error. And we transform the problem (6) into the following
Proof: For SC-NOMA with imperfect SIC, we can also equivalent problem as
omit the subscripts of subchannels. With Cond-1 ω1 = ω2 =
· · · = ωK = 1, it is easy to obtain that the rate of User k with N X
X K
Imperf-Q
perfect SIC is greater than or equal to that with imperfect SIC, max fn,k (P, γ , θ)
P,γ ,θ
i.e., n=1 k=1
! s.t. C1, C2,
|h | 2p
RIdeal = Bf log2 1 +
k k C3: pn,k+1 ≥ αn,k pn,k , ∀k ∈ {1, · · · , K − 1}, ∀n,
k
σz2 + |hk |2 k−1
P
! i=1 pi C4: αn,k > 1, ∀n, k. (37)
2
|hk | pk Imperf
≥ Bf log2 1 + Imperf = Rk . (34) where fn,k
Imperf-Q
(P, γ , θ) is given by (28).
In,k
The set consisting of constraints C3 and C4 is a convex
According to Proposition 1, the sum-rate of SC-NOMA with set, since the constraints are affine functions of P which can
imperfect SIC is upper bounded by allocating all power keep convexity. Therefore, the problem (37) is a concave
to the user with maximum channel-gain. Denote RIdeal = optimization problem with respect to variable P for fixed
PK Ideal and RImperf = PK RImperf . We have γ and θ. We solve the problem (37) by similar methods to
k=1 Rk k=1 k
the above sections or the other standard convex optimization
|h1 |2 PT methods [44], which is not repeated here. Then we find the
 
RImperf ≤ RIdeal ≤ Bf log2 1 + . (35) optimal power allocation by Algorithm 1. We just need to
σz2
replace (17) with solving the concave problem (37). The
Normally, (35) takes the equal sign when  ε = 20. However,
 equal power allocation may not be a valid solution of the
it is obvious that RImperf (P̃) = Bf log2 1 + |h1σ| 2PT , where optimization problem, due to the power order constrains.
z
P̃ = [PT , 0, · · · , 0]. That is to say, the optimal value of the Therefore, to ensure that the value of initial power locals
sum-rate maximization problem in SC-NOMA systems with in the feasible region of the optimization problem, we set
(0,0) k −α k−1 )P
imperfect SIC is achieved at allocating all power to the user pn,k = (α (α K −1)N
total
, ∀n, k.
with maximum channel-gain.
For SC-NOMA with imperfect SIC, when Cond-2 is satis- C. CONVERGENCY AND COMPLEXITY ANALYSIS OF
fied, it is easy to have POWER ALLOCATION ALGORITHMS
1) CONVERGENCY
K
Imperf
X Imperf We denote the objective functions of the problems (7), (13)
Rω ≤ ω1 Rk = ω1 RImperf , (36)
and (16) as F Ideal-O (P), F Ideal-L (P, γ ) and F Ideal-Q (P, γ , θ).
k=1
In inner iteration of Algorithm 1, the auxiliary variable γ is
Imperf PK Imperf
k=1 ωk Rk
where Rω = . Similar to the proof of fixed. We have
Proposition 1, based on (35), we can verify Proposition 3.
F Ideal-Q (P(s,t+1) , γ (s) , θ (s,t+1) )
Proposition 4: When one of Cond-3 and Cond-4 is satis- (a)
≥ F Ideal-Q (P(s,t) , γ (s) , θ (s,t+1) )
fied, the optimal solution of MC-NOMA with imperfect SIC (b)
is equal to the solution in MC-NOMA with perfect SIC. ≥ F Ideal-Q (P(s,t) , γ (s) , θ (s,t) ). (38)

VOLUME 7, 2019 94245

 X. Wang et al.: Low-Complexity Power Allocation in NOMA Systems With Imperfect SIC

The inequation (a) holds, since that the optimal power allo- Algorithm 3 User Scheduling Algorithm for Maximizing
cation can be obtained by solving the transformed prob- WSR in MC-NOMA Systems
lem (16) for a fixed θ due to its concavity. And the Initialization
inequation (b) holds, since that θ (s,t+1) is computed by 1. Initialize the sets Un as the unallocated users on
(18) which is the optimal θ of the problem (16) when subchannel n ∈ N , and initialize (n, k) = (1, 1);
(0,0)
P = P(s,t) and γ = γ (s) . Thus, the objective func- 2. Initialize power allocation pn,k =
tion F Ideal-Q monotonically non-decreases after each inner (α k −α k−1 )Ptotal
, ∀n, k;
(α K −1)N
iteration. The value of F Ideal-Q is bounded, so the inner Iterative Algorithm
iteration will converge. From Lemma 1, at the point of 4. Repeat
∗
convergence with index t ∗ , we have F Ideal-L (P(s,t ) , γ (s) ) = 5. Repeat
∗ ∗
F Ideal-Q (P(s,t ) , γ (s) , θ (s,t ) ). In outer iteration, γ is computed 6. Compute Un = |Un | possible SINRs for finding
by (12), which is optimal for the problem (13). Similarly, the k-th user on subchannel n by
∗ ∗
we have F Ideal-O (P(s+1,t ) ) = F Ideal-L (P(s+1,t ) , γ (s+1) ) ≥ SINRun,k = P
pn,k
, ∀u;
∗ ∗ σ2
F Ideal-L (P(s,t ) , γ (s) ) = F Ideal-O (P(s,t ) ) at the (s + 1)-th

Pk−1 K z
i=1 pn,i +ε i=k+1 pn,i +
|hn,u |2
iteration, and the outer iteration will converge. This algorithm 7. Find k-th user on subchannel
 n as 
can obtain at least a local optimum, and simulation results uk,n = arg max{ωu Bf log2 1 + SINRun,k };
show that it is almost the same with global optimum. u∈Un
8. Set Un = Un \un,k and k = k + 1;
9. Until the number of multiplexed users k = K ;
2) COMPLEXITY
10. Set n = n + 1;
We evaluate the computational complexity of the proposed 11. Until n = N .
algorithms with respect to each iteration.
In each iteration of Algorithm 1, the computational com-
plexity is O(NK 2 L0 ), where L0 is iteration number of
bisection method or subgradient method for updating dual weighted rate. Then, for a fixed power allocation, we can
variable λ. When the bisection method is adopted, it gener- select the user who has the largest weighted rate from the
ally achieves a logarithmic convergence rate. Namely, L0 = unallocated users set Un and remove it from Un . Algorithm 3
log2 ((λup − λlow )/ς), where ς is convergence tolerance of has the complexity of O(NK (2U − K )).
iterations, and λup and λlow are initial upper and lower values.
When the subgradient method is adopted, the iteration for
Algorithm 4 Joint User Scheduling and Power Allocation
updating λ converges to the optimal solution in polynomial
Algorithm for Maximizing WSR in MC-NOMA Systems
time.
In Algorithm 2, there exist two cases during the implemen- Initialization
tations. If Cond-3 or Cond-4 is satisfied, the computational 1. Initialize iteration index i = 0, maximum iterations
complexity is just O(N ). Otherwise, the complexity in each imax ;
(0,0)
iteration is O(NK 2 L0 ). 2. Initialize power allocation pn,k =
(α k −α k−1 )Ptotal
(α K −1)N
, ∀n, k;
V. JOINT USER SCHEDULING AND POWER ALLOCATION Iterative Algorithm
FOR MAXIMIZING WSR IN MC-NOMA 4. Repeat
In above sections, we discussed power allocation in (i)
5. For given P(i) , compute user scheduling Kn ,
MC-NOMA assuming that user scheduling (i.e., channel ∀n ∈ N by Algorithm 3;
assignment) has been given. We now consider a user schedul- 6. Update power allocation P(i+1) by Algorithm 1 or
ing scheme for our proposed MC-NOMA system. Obviously, Algorithm 2;
the joint user scheduling and power allocation problem is a 7. Set i = i + 1;
mixed integer non-linear programming problem, in which the 8. Until convergency or i = imax .
global optimal solution can be obtained only by the exhaus-
tive search method with exponential complexity [23]. There-
fore, we initialize a power allocation, based on which we In Algorithm 4, we give a joint user scheduling and power
provide a user scheduling scheme as shown in Algorithm 3. allocation algorithm. we first compute the user scheduling
And then we propose the joint user scheduling and power Kn , ∀n ∈ N by Algorithm 3 with the given power allocation.
allocation scheme as shown in Algorithm 4. Then we update power allocation P according to the proposed
In Algorithm 3, we initialize the power allocation by power allocation algorithm (i.e., Algorithm 1 or Algorithm 2).
(0,0)
pn,k , ∀n, k. From (4) and (5), for fixed bandwidth Bf , noise The WSR keeps improving at each iteration in Algorithm 4
level σz2 and SIC error ε, the achievable rate of user k on and the convergence of the iterative procedure can be guaran-
subchannel n only depends on its channel condition hn,k , teed. This algorithm finds at least a locally optimal solution
which is independent of hn,k 0 , ∀k 0 6 = k. If we select any to the joint resource allocation problem. It has the complex-
user u as Usern,k , we can get SINRun,k and the corresponding ity of O (NKL1 (K (L0 − 1) + 2N )), where L1 is the iteration

94246 VOLUME 7, 2019

X. Wang et al.: Low-Complexity Power Allocation in NOMA Systems With Imperfect SIC
FIGURE 2. Weighted sum-rate for the proposed algorithm with imperfect
SIC versus the number of iterations (ωn = [1, 1, 1], ε = 0.1).
number of Algorithm 4. From the simulation results, we can
see that it converges within 10 iterations.
VI. SIMULATION RESULTS AND DISCUSSIONS
In this section, simulation results are presented to evaluate the
performance of the proposed algorithms. In the simulation,
BS is located in the cell center and U = 60 users are
distributed in a circular range with a radius of 1km. The
channels are modeled by the product of path-loss and inde-
pendent Rayleigh fading, in which the path loss is modeled as
38.46 + 35 log10 (d) where d is the distance between BS and
user [47]. The bandwidth is Btotal = 5MHz and the number
of subchannels is N = 5, thus Bf = Btotal /N = 1MHz. The
noise power is −104dBm. At most K = 3 users are allowed
to occupy one subchannel.
First, we test the performance of the proposed power
allocation with a fixed user scheduling (in Subsection VI-A
and Subsection VI-B). Without loss of generality, we set
the distances between BS and 3 users occupying subchan-
nel n as dn,1 < dn,2 < dn,3 , ∀n. We test the performance
for three types of selected weighting factors, i.e., ωn =
[ωn,1 , ωn,2 , ωn,3 ] = [1, 1, 1], ωn = [1, 0.1, 0.01] and
ωn = [0.01, 0.1, 1]. Note that ωn = [1, 1, 1] and ωn =
[1, 0.1, 0.01] are special weighting factors, which satisfy the
conditions in Propositions 4. ωn = [0.01, 0.1, 1] is selected to
represent other general weighting factors. For lack of space,
we just provide results with imperfect SIC. In following sub-
sections, we evaluate performances of the proposed schemes
without and with the power order constraints respectively.
Then, we evaluate the effect of user scheduling schemes on
our proposed power allocations (in Subsections VI-C).
A. POWER ALLOCATION PERFORMANCE WITHOUT
POWER ORDER CONSTRAINTS
In this subsection, we study the proposed schemes ignoring
the power order constraints.
In Fig. 2, we study the convergence of WSR for the
proposed power allocation algorithm (i.e., Algorithm 1) for
VOLUME 7, 2019
FIGURE 3. Weighted sum-rate versus total power Ptotal (ωn = [1, 1, 1]
and [1, 0.1, 0.01], ε = 0.1).
different total power limits of BS. We consider the total power
of BS as 20dBm, 30dBm and 40dBm respectively. We set
weighting factor ωn = [1, 1, 1] and SIC error ε = 0.1. It can
be observed that the proposed iterative power allocation algo-
rithm has good convergence performance and it can converge
within about 5 iterations.
Fig. 3 illustrates the WSR for different power allocation
strategies versus maximum transmitted power with ωn =
[1, 1, 1] and ω = [1, 0.1, 0.01]. Note that, we will com-
bine the results at ωn = [1, 1, 1] and ωn = [1, 0.1, 0.01],
since we find the results are always the same in such
two cases of ω in our simulation. We denote the proposed
power allocation scheme in Algorithm 1 and low-complexity
scheme in Algorithm 2 as NOMA-Prop and NOMA-Prop-L
respectively. To demonstrate the effectiveness of the pro-
posed algorithms, some other schemes are considered as
comparisons. For example, the equal power allocation for
MC-NOMA is denoted by NOMA-EQ. We also consider a
naive method denoted by NOMA-Naive, in which power allo-
cation is based on ideal method ignoring SIC error, in spite
of ε = 0.1. And then, we compare our proposed algorithms
to the power allocation scheme in [27], denoted by NOMA-
Exp method. Finally, we compare these NOMA algorithms
to OMA schemes, in which each subchannel is divided K
orthogonal frequency bands to avoid inter-user interference.
In the OMA schemes, we focus on optimal power alloca-
tion (denoted by OMA-OPT) and equal power allocation
(denoted by OMA-EQ). We can see that WSR increases
monotonically with Ptotal . The results also show that the
NOMA-Prop-L method can achieve the same performance
to the NOMA-Prop method while reducing the calculation
complexity. The NOMA-Naive and NOMA-Exp schemes
have the same performance to the proposed methods. It is
because that only the user who has highest channel gain on
each subchannel is allocated power, in which no interference
needs to be cancellated. It verifies Proposition 4, and indicates
that the proposed algorithms can achieve global optimum.
Besides, we observe that the proposed algorithms are superior
to the other compared algorithms. However, the NOMA-EQ
94247
 X. Wang et al.: Low-Complexity Power Allocation in NOMA Systems With Imperfect SIC
FIGURE 4. Weighted sum-rate versus total power Ptotal
(ωn = [0.01, 0.1, 1], ε = 0.1).
FIGURE 5. Weighted sum-rate versus SIC error ε (ωn = [1, 1, 1] and
[1, 0.1, 0.01], Ptotal = 30dBm).
scheme has a worse performance even than the OMA-OPT
scheme. It demonstrates that dynamic power allocation is
necessary and meaningful in NOMA systems.
Fig. 4 shows WSR versus maximum transmitted power
with ωn = [0.01, 0.1, 1]. It can be observed from this
figure that the proposed algorithms (i.e., NOMA-Prop and
NOMA-Prop-L) are also superior to the other algorithms. The
WSRs of NOMA-Naive and NOMA-Exp schemes are less
than the proposed schemes, since the SIC error is underuti-
lized to design power allocation.
Fig. 5 depicts WSR versus SIC error ε with ωn = [1, 1, 1]
and [1, 0.1, 0.01]. The results show that the performance of
proposed algorithms keeps invariable with SIC error increas-
ing. This is owing to the optimal solution is allocating power
to the user who has highest channel gain on each subchannel,
in which no SIC needs to be operated at any user. In addition,
the NOMA-EQ scheme has a better performance than the
OMA-OPT scheme and the OMA-EQ scheme when ε is
small, but it is worse than the two OMA schemes when ε is
large.
Fig. 6 provides the results of WSR with ωn =
[0.01, 0.1, 1]. It can be seen that the performance of all
the NOMA schemes decreases with increasing ε. However,
94248
FIGURE 6. Weighted sum-rate versus SIC error ε (ωn = [1, 1, 1] and
[1, 0.1, 0.01], Ptotal = 30dBm).
FIGURE 7. Power allocation for users on each subchannel (ε = 0.1).
Top: ωn = [1, 1, 1] and ωn = [1, 0.1, 0.01]; Bottom: ωn = [0.01, 0.1, 1].
the proposed algorithms have better WSR performance
than OMA schemes at any ε. Besides, the performance
of NOMA-Naive scheme drops more than the proposed
NOMA-Prop and NOMA-Prop-L schemes. With increas-
ing ε, the performance of the NOMA-EQ scheme will fall
into lower than the OMA schemes.
Fig. 7 displays the power allocation on each subchannel for
different ω. In this figure, the horizontal axis (n, k) represents
the index of user k on subchannel n after being sorted. When
weighting factors are ωn = [1, 1, 1] and [1, 0.1, 0.01], it is
clearly seen that the proposed schemes (NOMA-Prop and
NOMA-Prop-L) only allocate power to the first user on each
subchannel, i.e., (n, 1), ∀n, and the allocated power of the
two proposed schemes is almost identical. We can easily
obtained this result from Propositions 4. When weighting
factors are ωn = [0.01, 0.1, 1], allocated power depends on
channel gains, weighting factors and inter-user interference.
Besides, the power allocation results of NOMA-Prop-L and
NOMA-Prop are also almost identical.
B. POWER ALLOCATION PERFORMANCE
WITH POWER ORDER CONSTRAINTS
We provide the performance results of the proposed algorithm
with both SIC error and power order constraints. Unless
stated otherwise, the minimum requirement of the ratio of
VOLUME 7, 2019
X. Wang et al.: Low-Complexity Power Allocation in NOMA Systems With Imperfect SIC
FIGURE 8. Weighted sum-rate with power order constraints versus SIC
error ε (Ptotal = 40dBm, α = 2). Left: ωn = [1, 1, 1]; Right: ωn =
[0.01, 0.1, 1].
FIGURE 9. Power allocation for users on each subchannel
(Ptotal = 40dBm, ε = 0.1, α = 2). Top: ωn = [1, 1, 1];
Bottom: ωn = [0.01, 0.1, 1].
power levels between adjacent users is set as α = 2. Most
of the existing schemes did not consider the power order
constraints, and they might not always be able to ensure
carrying out SIC in physical layer. In this case, the (weighted)
sum-rate they declared may not be always achieved. Thus,
we just compare our proposed algorithm to OMA schemes.
Fig. 8 shows the WSR with power order constraints versus
SIC error ε, with ωn = [1, 1, 1] and ωn = [0.01, 0.1, 1].
Obviously, when power order constraints are considered,
the NOMA-Prop performance may not be always better than
OMA schemes, especially for imperfect SIC cases. Increas-
ing SIC error will lead to worse NOMA performance than
OMA schemes.
Fig. 9 illustrates the power allocation for users on each
subchannel with weighting factors ωn = [1, 1, 1] and
[0.01, 0.1, 1]. We can see that, power allocation satisfies
order constraints. Particularly, when ωn = [1, 1, 1], the opti-
mal solution is obtained at the boundary of feasible region of
optimization problem, just to satisfy pn,3 = αpn,2 = α 2 pn,1 .
Fig. 10 shows the WSR with power order constraints versus
ratio α of powers between adjacent users. It is clear that
VOLUME 7, 2019
FIGURE 10. Weighted sum-rate with power order constraints versus
α (ωn = [1, 1, 1], ε = 0.1).
the WSR performance of proposed algorithm decreases with
increasing α. The result is reasonable, since when α increase,
the power difference between the users becomes large,
i.e., feasible region of optimization problem is reduced. This
illustrates that if simple receiver devices can carry out SIC
only with large enough power difference, it is not beneficial
to the (weighted) sum-rate performance.
We have known that weighting factor ω can provide
fairness, however, fixed ω is considered in above sub-
sections. We now design ω by the fairness criterions
dynamically, such as proportional fairness. Based on the
proposed algorithms above, we further present a power
allocation method for MC-NOMA with fairness (denoted
τn,k
by NOMA-Prop-F). We set ωn,k (t) = Imperf , where t
R̃n,k (t)
Imperf
is the time index and R̃n,k (t)
is the average achiev-
able rate of user (n, k). Without loss of generality, we set
Imperf
τ1,1 = τ1,2 = · · · =  τN ,K . R̃n,k (t) is updated by
Imperf Imperf Imperf
R̃n,k (t) = T1 Rn,k (t) + 1 − T1 R̃n,k (t − 1), where T
denotes the average window size [48], [49]. We update ωn,k
after each power allocation, which is a low-complexity imple-
mentation of proportional fairness.
Fig. 11 shows overall sum-rate and cell-edge user rate
for NOMA-Prop and NOMA-Prop-F schemes. It can be
seen that the NOMA-Prop scheme has a better sum-rate
performance than the NOMA-Prop-F method. It is because
more power is allocated to the weaker users to ensure fair-
ness in the NOMA-Prop-F method. However, the NOMA-
Prop-F scheme is superior to the NOMA-Prop scheme in
the respect of cell-edge user rate, although it has a poor
average sum-rate for all users. To evaluate the fairness,
Imperf 2
P
(
P N K
k=1 R̃n,k )
we define fairness index as ξ = n=1
PN PK Imperf 2 ,
NK n=1 k=1 (R̃n,k )
where the fairness index ξ ∈ [1/(NK ), 1] and is 1 only
when each user has the same average rate [50]. We have
the fairness index of the two schemes as ξNOMA-Prop =
0.5964 and ξNOMA-Prop-F = 0.7511. The results illus-
trate that NOMA-Prop-F scheme has a higher fairness than
NOMA-Prop schemes. This illustrates that the rational design
94249
 X. Wang et al.: Low-Complexity Power Allocation in NOMA Systems With Imperfect SIC

FIGURE 11. Sum-rate versus total power Ptotal (ωn = [1, 1, 1], FIGURE 13. Weighted sum-rate for the joint resource allocation algorithm
Ptotal = 40dBm, ε = 0.1, α = 2). Left: Overall sum-rate; Right: Cell-edge versus the number of users (Ptotal = 40dBm, α = 2).
user rate.

Fig. 13 shows the WSR performance of proposed joint

resource allocation scheme versus the number of users U
with Ptotal = 40dBm and α = 2. It is observed that the
WSR increases with increasing number of users because
of increasing multi-user diversity gain. On the other hand,
the WSR performance of the proposed scheme decreases with
increasing SIC error ε.

VII. CONCLUSION
In this paper, we have studied the power allocation issue
in downlink MC-NOMA systems for maximizing WSR by
considering imperfect SIC. We have explicitly considered
the SIC error and the power order constraints of users in
power allocation problem. By transforming the optimiza-
FIGURE 12. Weighted sum-rate for the joint resource allocation algorithm tion problem into equivalent concave form by the frac-
versus the number of iterations (U = 60, ε = 0.1 and α = 2). tional quadratic method, we have proposed an iterative
power allocation algorithm. We have also analyzed the
optimal solution for both perfect and imperfect SIC, and
of weighting factors in NOMA can improve the fairness of have presented a low-complexity algorithm. We have pre-
users. sented a joint user scheduling and power allocation scheme
for the MC-NOMA systems. Simulation results show that
C. JOINT USER SCHEDULING AND POWER the proposed algorithm converges within a small num-
ALLOCATION PERFORMANCE ber of iterations. Also, the proposed methods can achieve
In this subsection, we evaluate the performance of proposed a superior WSR performance than the existing schemes.
joint user scheduling and power allocation algorithm. In the Then, by designing weighting factors, user fairness is
simulation, we divide the users (U = 60) in the cell into improved.
three types with different weighting factors as 0.01, 0.1 and 1
respectively. APPENDIX
In Fig. 12, we study the convergence of WSR for the A. PROOF OF THEOREM 1
proposed joint user scheduling and power allocation algo- The constraints of the problem (6) are linear, we only need to
rithm (i.e., Algorithm 4) for different total powers of base discuss the objective function. We first convert the objective
station. We consider the total power of BS as 20dBm, 30dBm function as
and 40dBm respectively. And we set the number of users N X
X K  
U = 60, SIC error ε = 0.1 and α = 2. It can be min n,k (P) .
−ωn,k RIdeal (39)
P
observed that the proposed joint user scheduling and power n=1 k=1
allocation algorithm has also good convergence performance
σ2 PN PK
k=1 ωn,k Rn,k .
Let Hn,k = |h z |2 and Rω (P) , Ideal
and it can converge within about 10 iterations for different n,k n=1
powers. Due to |hn,1 | ≥ |hn,2 | ≥ . . . ≥ |hn,K |, ∀n, we have

94250 VOLUME 7, 2019

X. Wang et al.: Low-Complexity Power Allocation in NOMA Systems With Imperfect SIC

Hn,1 ≤ Hn,2 ≤ . . . ≤ Hn,K , and then we rewrite Rω as χn,1 = ψn,1 > 0). From (41), we can see ψn,K ≥ 0. Mean-
  while, three cases should be discussed about dk as follows:
N K −1 j
X X
ωn,j Bf log2 (Hn,j +
X Case I: For all k ∈ {1, 2, · · · , K − 1}, dk ≥ 0. In this
Rω = pn,i )
case, similar to ωn,k = ωn,k+1 , we can easily get ψn,k > 0
n=1 j=1 i=1
  and ψn,k − ψn,k+1 ≥ 0. The convexity of (39) can be
N K
X X −1 j
X demonstrated.
− ωn,j+1 Bf log2 (Hn,j+1 + pn,i ) Case II: For all k ∈ {1, 2, · · · , K − 1}, dk < 0. We have
n=1 j=1 i=1 0 < ψn,1 < ψn,2 < · · · < ψn,K . Consequently, the sign
N K of χn,m depends on the number of ds . If m is odd, χn,m will
!
X X
+ Bf ωn,K log2 (Hn,K + pn,i )−ωn,1 log2 Hn,1 . be large than zero, otherwise χn,m < 0. Thus in this case,
n=1 i=1 we cannot guarantee the convexity of (39).
(40) Case III: dk ≥ 0 and dk 0 < 0 for k ∈ CA and k 0 ∈ CB ,
where CA ∪ CB = {1, 2, · · · , K − 1} and CA ∩ CB = ∅. When
Denote ψn,k as the second-order derivative of −Rω (P),
the number of the elements in CB is odd, χn,K will be less
which can be expressed as
than zero and (39) is not a convex problem. When the number
∂ 2 (−Rω (P)) ωn,K Bf of the elements in CB is even, we have χn,K > 0, however,
ψn,k = = χn,kend < 0 where kend0 is the maximum value in CB . In this
∂p2n,k (Hn,K + K 2 0
P
i=1 pn,i ) ln 2
case, the convexity of (39) cannot be guaranteed.
K −1
ωn,j ωn,j+1
 
X Bf In conclusion,
+ − , (41)
j=k
ln 2 (Hn,j +qn,j )2 (Hn,j+1 +qn,j )2 ωn,j ωn,j+1
Pj − Pj ≥ 0, (46)
(Hn,j + i=1 pn,i ) 2 (Hn,j+1 + i=1 pn,i )2
Pj 0 < k ≤ K,
where qn,j = i=1 pn,i . For 1 ≤ k
we have ∂ (−Rω )/∂pn,k = ∂ (−Rω )/(∂pn,k ∂pn,k 0 ), and for
2 2 2 is a sufficient and necessary condition for the convexity
n0 6 = n, 1 ≤ k ≤ K , 1 ≤ k 0 ≤ K , we have of (39). Thus, Theorem 1 can be demonstrated.
∂ 2 (−Rω )/(∂pn,k ∂pn0 ,k 0 ) = 0. We can derive the Hessian
matrix 9 as B. PROOF OF THEOREM 2
Denote a function fγIdeal
,θ (P) as
91 ...
 
0 0
 0 92 . . . 0  q
,θ (P) = 2θn,k ωn,k Bf |hn,k | pn,k (1 + γn,k )
fγIdeal 2
9(P) =  . . . ..  , (42)
 
 .. .. .. .  k
!
0 0 0 9K
X
− θn,k ln 2 |hn,k |
2 2
pn,i + σz ln 2 + g(γ , θ).
2

where 9n is a symmetric matrix as i=1

(47)
ψn,1 ψn,2 . . . ψn,K
 
 ψn,2 ψn,2 . . . ψn,K  It is known that the square-root function y(x) = x 1/2 is a non-
9n =  . .. .. .. , ∀n. (43)
 
. decreasing concave function of x. According the convexity
 . . . . 
property of composition function, the first term in (47) is
ψn,K ψn,K ... ψn,K a concave with respect to P. The other terms are the affine
The m-th order principal minor of matrix 9n is function of P. Due to θn,k ≥ 0, we have fγIdeal
,θ (P) is a concave
function. Therefore the objective function in (16) is concave
m−1
Y since it is a non-negative weighted summation of fγIdeal ,θ (P).
χn,m = ψn,m (ψn,s − ψn,s+1 ). (44)
In addition, the constraint set of C1 and C2 is convex. Thus,
s=1
for fixed γ and θ, the problem in (16) is concave.
Denoting dk = ψn,k − ψn,k+1 , from (41), we have
ωn,s Bf / ln 2 ωn,s+1 Bf / ln 2 REFERENCES
ds = Ps − . (45) [1] L. Dai, B. Wang, Y. Yuan, S. Han, C.-L. I, and Z. Wang, ‘‘Non-
2 (Hn,s+1 + si=1 pn,i )2
P
(Hn,s + i=1 pn,i ) orthogonal multiple access for 5G: Solutions, challenges, opportuni-
ties, and future research trends,’’ IEEE Commun. Mag., vol. 53, no. 9,
We find that χn,m is related to not only Hn,k , but also ωn,k pp. 74–81, Sep. 2015.
and qn,k . Particularly, when ωn,k = ωn,k+1 , for all k ∈ [2] Y. Liu, Z. Qin, M. Elkashlan, Z. Ding, A. Nallanathan, and L. Hanzo,
{1, 2, · · · , K − 1}, we have ψn,m > 0 and ψn,s − ψn,s+1 > 0. ‘‘Nonorthogonal multiple access for 5G and beyond,’’ Proc. IEEE,
vol. 105, no. 12, pp. 2347–2381, Dec. 2017.
And according to the principle of block matrix, the convexity
[3] P. Xu, Z. Ding, and X. Dai, ‘‘NOMA: An information theoretic perspec-
of (39) can be demonstrated. tive,’’ 2015, arXiv:1504.07751. [Online]. Available: https://arxiv.org/abs/
However, the users’ weighting factors {ωn,k } cannot be 1504.07751
always all the same in practical communication systems. [4] Y. Saito, A. Benjebbour, T. Nakamura, and Y. Kishiyama, ‘‘System-
level performance evaluation of downlink non-orthogonal multiple access
To guarantee the convexity of (39), all the order principal (NOMA),’’ in Proc. IEEE Int. Symp. Pers., Indoor Mobile Radio Commun.,
minors have to be greater than or equal to zero (where Sep. 2013, pp. 611–615.

VOLUME 7, 2019 94251

 X. Wang et al.: Low-Complexity Power Allocation in NOMA Systems With Imperfect SIC
[5] A. Benjebbovu, A. Li, Y. Kishiyama, A. Harada, T. Nakamura, and Y. Saito,
‘‘System-level performance of downlink NOMA for future LTE enhance-
ments,’’ in Proc. IEEE Proc. IEEE Global Telecommun. Conf. Workshop,
Dec. 2013, pp. 66–70.
[6] Z. Ding, Z. Yang, H. V. Poor, and P. Fan, ‘‘On the performance of
non-orthogonal multiple access in 5G systems with randomly deployed
users,’’ IEEE Signal Process. Lett., vol. 20, no. 12, pp. 1501–1505,
Dec. 2014.
[7] S.-L. Shieh and Y.-C. Huang, ‘‘A simple scheme for realizing the promised
gains of downlink nonorthogonal multiple access,’’ IEEE Trans. Commun.,
vol. 64, no. 4, pp. 1624–1635, Apr. 2016.
[8] A. Benjebbour, Y. Saito, A. Li, A. Harada, T. Nakamura, and Y. Kishiyama,
‘‘Concept and practical considerations of non-orthogonal multiple access
(NOMA) for future radio access,’’ in Proc. IEEE Intell. Signal Process.
Commun. Syst., Nov. 2013, pp. 770–774.
[9] Z. Ding, Y. Liu, J. Choi, Q. Sun, M. Elkashlan, C.-L. I, and H. V. Poor,
‘‘Application of non-orthogonal multiple access in LTE and 5G networks,’’
IEEE Commun. Mag., vol. 55, no. 2, pp. 185–191, Feb. 2015.
[10] S. M. R. Islam, M. Zeng, O. A. Dobre, and K.-S. Kwak, ‘‘Resource
allocation for downlink NOMA systems: Key techniques and open issues,’’
IEEE Wireless Commun., vol. 25, no. 2, pp. 40–47, Apr. 2018.
[11] Z. Yang, Z. Ding, P. Fan, and Z. Ma, ‘‘Outage performance for dynamic
power allocation in hybrid non-orthogonal multiple access systems,’’ IEEE
Commun. Lett., vol. 20, no. 8, pp. 1695–1698, Aug. 2016.
[12] P. Xu, K. Cumanan, and Z. Yang, ‘‘Optimal power allocation scheme for
NOMA with adaptive rates and alpha-fairness,’’ in Proc. IEEE Global
Telecommun. Conf., Dec. 2017, pp. 1–6.
[13] J. Wang, Q. Peng, H.-M. Wang, X. You, and Y. Huang, ‘‘Convexity
of weighted sum rate maximization in NOMA systems,’’ IEEE Signal
Process. Lett., vol. 24, no. 9, pp. 1323–1327, Sep. 2017.
[14] J. Choi, ‘‘Power allocation for max-sum rate and max-min rate proportional
fairness in NOMA,’’ IEEE Commun. Lett., vol. 20, no. 10, pp. 2055–2058,
Oct. 2016.
[15] F. Liu, P. Mähönen, and M. Petrova, ‘‘Proportional fairness-based power
allocation and user set selection for downlink NOMA systems,’’ in Proc.
IEEE Int. Conf. Commun., May 2016, pp. 1–6.
[16] J. Zhu, J. Wang, Y. Huang, S. He, X. You, and L. Yang, ‘‘On opti-
mal power allocation for downlink non-orthogonal multiple access sys-
tems,’’ IEEE J. Sel. Areas Commun., vol. 35, no. 12, pp. 2744–2757,
Dec. 2017.
[17] F. Liu and M. Petrova, ‘‘Dynamic power allocation for downlink
multi-carrier NOMA systems,’’ IEEE Commun. Lett., vol. 22, no. 9,
pp. 1930–1933, Sep. 2018.
[18] Y. Sun, D. W. K. Ng, R. Schober, and Z. Ding, ‘‘Optimal joint power
and subcarrier allocation for MC-NOMA systems,’’ in Proc. IEEE Global
Telecommun. Conf., Dec. 2016, pp. 1–6.
[19] Z. Q. Al-Abbasi and D. K. C. So, ‘‘Resource allocation in non-orthogonal
and hybrid multiple access system with proportional rate constraint,’’
IEEE Trans. Wireless Commun., vol. 16, no. 10, pp. 6309–6320,
Oct. 2017.
[20] M. Moltafet, N. M. Yamchi, M. R. Javan, and P. Azmi, ‘‘Comparison study
between PD-NOMA and SCMA,’’ IEEE Trans. Veh. Technol., vol. 67,
no. 2, pp. 1830–1834, Feb. 2018.
[21] L. Lei, D. Yuan, and P. Värbrand, ‘‘On power minimization for non-
orthogonal multiple access (NOMA),’’ IEEE Commun. Lett., vol. 20,
no. 12, pp. 2458–2461, Dec. 2016.
[22] X. Li, C. Li, and Y. Jin, ‘‘Dynamic resource allocation for transmit power
minimization in OFDM-based NOMA systems,’’ IEEE Commun. Lett.,
vol. 20, no. 12, pp. 2558–2561, Dec. 2016.
[23] F. Fang, H. Zhang, J. Cheng, S. Roy, and V. C. M. Leung, ‘‘Joint user
scheduling and power allocation optimization for energy-efficient NOMA
systems with imperfect CSI,’’ IEEE J. Sel. Areas Commun., vol. 35, no. 12,
pp. 2874–2885, Dec. 2017.
[24] H. Zhang, B. Wang, K. Long, A. Nallanathan, V. C. M. Leung, H. V. Poor,
and C. Jiang, ‘‘Energy efficient dynamic resource optimization in NOMA
system,’’ IEEE Trans. Wireless Commun., vol. 17, no. 9, pp. 5671–5683,
Sep. 2018.
[25] J. Wang, H. Xu, L. Fan, B. Zhu, and A. Zhou, ‘‘Energy-efficient joint
power and bandwidth allocation for NOMA systems,’’ IEEE Commun.
Lett., vol. 22, no. 4, pp. 780–783, Apr. 2018.
[26] G. Lv, X. Li, P. Xue, Y. Jin, and R. Shang, ‘‘Dynamic resource allocation for
uplink non-orthogonal multiple access systems,’’ IET Commun., vol. 12,
no. 6, pp. 649–655, Apr. 2018.
94252
[27] B. Di, L. Song, and Y. Li, ‘‘Sub-channel assignment, power allo-
cation, and user scheduling for non-orthogonal multiple access net-
works,’’ IEEE Trans. Wireless Commun., vol. 15, no. 11, pp. 7686–7698,
Nov. 2016.
[28] H. Zhang, Y. Qiu, G. K. Karagiannidis, X. Wang, A. Nallanathan,
and K. Long, ‘‘Resource allocation in noma based fog radio access
networks,’’ IEEE Wireless Commun., vol. 25, no. 3, pp. 110–115,
Jul. 2018.
[29] S. Seng, X. Li, H. Zhang, and H. Ji, ‘‘Joint access selection and heteroge-
neous resources allocation in UDNs with MEC based on non-orthogonal
multiple access,’’ in Proc. IEEE ICC Workshops, May 2018, pp. 1–6.
[30] M. Moltafet, S. Rezvani, M. R. Javan, E. A. Jorswieck, and N. Mokari,
‘‘Smart soft-RAN for 5G: Dynamic resource management in CoMP-
NOMA based systems,’’ 2018, arXiv:1804.03778. [Online]. Available:
https://arxiv.org/abs/1804.03778
[31] H. Zheng, S. Hou, Z. Song, Y. Hao, and H. Li, ‘‘Power allocation and user
clustering for uplink MC-NOMA in D2D underlaid cellular networks,’’
IEEE Wireless Commun. Lett., vol. 7, no. 6, pp. 1030–1033, Dec. 2018.
[32] M. Zeng and V. Fodor, ‘‘Sum-rate maximization under QoS constraint in
MIMO-NOMA systems,’’ in Proc. IEEE Wireless Commun. Netw. Conf.,
Apr. 2018, pp. 1–6.
[33] N. Nandan, S. Majhi, and H.-C. Wu, ‘‘Secure beamforming for MIMO-
NOMA-based cognitive radio network,’’ IEEE Commun. Lett., vol. 22,
no. 8, pp. 1708–1711, Aug. 2018.
[34] A. J. Morgado, K. M. S. Huq, J. Rodriguez, C. Politis, and H. Gacanin,
‘‘Hybrid resource allocation for millimeter-wave NOMA,’’ IEEE Wireless
Commun., vol. 24, no. 5, pp. 23–29, Oct. 2017.
[35] J. Cui, Y. Liu, Z. Ding, P. Fan, and A. Nallanathan, ‘‘Optimal user schedul-
ing and power allocation for millimeter wave NOMA systems,’’ IEEE
Trans. Wireless Commun., vol. 17, no. 3, pp. 1502–1517, Mar. 2018.
[36] H. Wang, Z. Zhang, and X. Chen, ‘‘Energy-efficient power allocation
for non-orthogonal multiple access with imperfect successive interference
cancellation,’’ in Proc. Wireless Commun. Signal Process. Conf., 2017,
pp. 1–6.
[37] A. Agrawal, J. G. Andrews, J. M. Cioffi, and T. Meng, ‘‘Iterative power
control for imperfect successive interference cancellation,’’ IEEE Trans.
Wireless Commun., vol. 4, no. 3, pp. 878–884, May 2005.
[38] A. Ahmad and M. Assaad, ‘‘Optimal power and subcarriers allocation in
downlink OFDMA system with imperfect channel knowledge,’’ Optim.
Eng., vol. 14, no. 3, pp. 477–499, Sep. 2013.
[39] D. Tweed, M. Derakhshani, T. Le-Ngoc, and S. Parsaeefard, ‘‘Outage-
constrained resource allocation in uplink NOMA for critical applications,’’
IEEE Access, vol. 5, pp. 27636–27648, 2017.
[40] S. Li, M. Derakhshani, and S. Lambotharan, ‘‘Outage-constrained robust
power allocation for downlink MC-NOMA with imperfect SIC,’’ in Proc.
IEEE Int. Conf. Commun., May 2018, pp. 1–7.
[41] F. Cui, Z. Qin, M. Zhao, G. Y. Li, and Y. Cai, ‘‘Rethinking outage con-
straints for resource management in NOMA networks,’’ IEEE J. Sel. Areas
Signal Process., vol. 13, no. 3, pp. 423–435, Jun. 2019.
[42] M. S. Ali, H. Tabassum, and E. Hossain, ‘‘Dynamic user clustering
and power allocation for uplink and downlink non-orthogonal multiple
access (NOMA) systems,’’ IEEE Access, vol. 4, pp. 6325–6343, 2016.
[43] Z. Yang, C. Pan, W. Xu, Y. Pan, M. Chen, and M. Elkashlan, ‘‘Power
control for multi-cell networks with non-orthogonal multiple access,’’
IEEE Trans. Wireless Commun., vol. 17, no. 2, pp. 927–942, Feb. 2018.
[44] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge, U.K.:
Cambridge Univ. Press, 2004.
[45] X. Zhou and J. Yang, ‘‘Global optimization for the sum of concave-
convex ratios problem,’’ J. Appl. Math., vol. 2014, no. 10, May 2014,
Art. no. 879739.
[46] K. Shen and W. Yu, ‘‘Fractional programming for communication
systems—Part I: Power control and beamforming,’’ IEEE Trans. Signal
Process., vol. 66, no. 10, pp. 2616–2630, May 2018.
[47] Parameters for FDD HeNB RF Requirements, document 3GPP TSG-RAN
WG4 R4-C092042, Simulation Assumptions, 2009.
[48] H. Kim and Y. Han, ‘‘A proportional fair scheduling for multicarrier
transmission systems,’’ IEEE Commun. Lett., vol. 9, no. 3, pp. 210–212,
Mar. 2005.
[49] X. Wang, F.-C. Zheng, X. You, and P. Zhu, ‘‘Energy-efficient resource
allocation in coordinated downlink multicell OFDMA systems,’’ IEEE
Trans. Veh. Technol., vol. 65, no. 3, pp. 1395–1408, Mar. 2016.
[50] H. Shi, R. V. Prasad, E. Onur, and I. G. M. M. Niemegeers, ‘‘Fairness
in wireless networks: Issues, measures and challenges,’’ IEEE Commun.
Surveys Tuts., vol. 16, no. 1, pp. 5–24, 1st Quart., 2014.
VOLUME 7, 2019
X. Wang et al.: Low-Complexity Power Allocation in NOMA Systems With Imperfect SIC

XIAOMING WANG (M’16) received the Ph.D. YOUYUN XU (M’02–SM’11) received the Ph.D.
degree in information and communication engi- degree in information and communication engi-
neering from the National Mobile Communica- neering from Shanghai Jiao Tong University
tions Research Laboratory, Southeast University, (SJTU), China, in 1999. He is currently a Professor
Nanjing, China, in 2016. He is currently a Lecturer with the Nanjing University of Posts and Telecom-
with the Nanjing University of Posts and Telecom- munications. He is also a part-time Professor with
munications (NJUPT), Nanjing. His research the Institute of Wireless Communication Tech-
interests include radio resource management, nologies, SJTU. He has over 20-year professional
green communications, and machine learning in experience of teaching and researching in commu-
communications. nication theory and engineering with research and
development achievement, such as the WCDMA Trial System under the
C3G Framework in China, in 1999, the B3G-TDD Trial System under the
FuTURE Framework in China, in 2006, and the Chinese Digital TV Broad-
casting System. His current research interests include new generation wire-
less mobile communication systems (LTE, IM-T Advanced, and related),
advanced channel coding and modulation techniques, multiuser information
theory and radio resource management, wireless sensor networks, cognitive
radio networks. He is a Senior Member of the Chinese Institute of Electronics
and a member of IEICE.

QINGMIN MENG received the Ph.D. degree

RUILU CHEN received the B.E. degree from the
Nanjing University of Posts and Telecommunica-
tions (NJUPT), Nanjing, China, in 2018. She is
currently pursuing the master’s degree with the
National Mobile Communications Research Labo-
ratory, Southeast University, Nanjing. Her research
interests include non-orthogonal multiple access,
radio resource management, and machine learning
in communications.
in communication and information system from
the National Mobile Communications Research
Laboratory, Southeast University, Nanjing, China,
in 2007. He is currently an Associate Professor
with the Nanjing University of Posts and Telecom-
munications (NJUPT), Nanjing. His research
interests include massive multiple-input multiple-
output (MIMO), radio resource management, and
cognitive radio networks.

VOLUME 7, 2019 94253