You are on page 1of 11

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

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCOMM.2016.2594759, IEEE
Transactions on Communications

Energy-Efficient Resource Allocation for Downlink


Non-Orthogonal Multiple Access (NOMA) Network
Fang Fang, Haijun Zhang, Member, IEEE, Julian Cheng, Senior Member, IEEE, Victor C.M. Leung, Fellow, IEEE

Abstract—Non-orthogonal multiple access (NOMA) is a for mobile traffic data volume is expected to be 500-1,000
promising technique for the fifth generation mobile communica- times larger in 2020 than that in 2010 [2]. To further meet over-
tion due to its high spectral efficiency. By applying superposition whelming requirement of data rates, various new techniques
coding and successive interference cancellation techniques at the
receiver, multiple users can be multiplexed on the same sub- have been proposed in recent years, and these techniques
channel in NOMA systems. Previous works focus on subchannel include massive multiple-input multiple output (MIMO) [3],
assignment and power allocation to achieve the maximization millimeter wave communications [4], LTE-U [5], C-RAN [6],
of sum rate; however, the energy-efficient resource allocation SON [7] and non-orthogonal multiple access (NOMA) [8]. A-
problem has not been well studied for NOMA systems. In this mong them, NOMA takes advantage of spectrum efficiency by
paper, we aim to optimize subchannel assignment and power
allocation to maximize the energy efficiency for the downlink allowing multiple users to occupy the same subchannel, which
NOMA network. Assuming perfect knowledge of the channel is different from the resource allocation in OFDM systems [9]–
state information at base station, we propose a low-complexity [11]. By applying successive interference cancellation (SIC) in
suboptimal algorithm which includes energy-efficient subchannel NOMA systems, superposition coded signal can be correctly
assignment and power proportional factors determination for decoded and demodulated at the receiver [12]–[15]. Therefore,
subchannel multiplexed users. We also propose a novel power al-
location across subchannels to further maximize energy efficiency. NOMA has been well considered as a promising candidate for
Since both optimization problems are non-convex, difference of the next generation mobile communication systems.
convex programming is used to transform and approximate the
original non-convex problems to convex optimization problems.
Solutions to the resulting optimization problems can be obtained A. Related Works and Motivation
by solving the convex sub-problems iteratively. Simulation results
show that the NOMA system equipped with the proposed Since the basic concept of NOMA was introduced and the
algorithms yields much better sum rate and energy efficiency cell-edge user throughput performance improvement was pre-
performance than the conventional orthogonal frequency division sented in [16], NOMA has attracted much research attention.
multiple access scheme. The NOMA system has also been envisioned as a key tech-
Index Terms—Difference of convex, energy efficiency, non- nology in the fifth generation mobile communication systems
orthogonal multiple access, orthogonal frequency division multi- [17]. In [18], the author discussed an application of combining
ple access, resource allocation, successive interference cancella- NOMA with MIMO technologies. Various aspects of resource
tion. allocation have been investigated in NOMA systems [19]–
[22]. By using fractional transmit power allocation (FTPA)
I. I NTRODUCTION among users and equal power allocation across subchannels,
the authors in [19] compared system-level performance of the
In the fourth generation mobile communication systems NOMA system with the OFDMA system and showed that the
such as long-term evolution (LTE) and LTE-Advanced [1], overall cell throughput, cell-edge user throughput, and the de-
orthogonal frequency division multiple access (OFDMA) has gree of proportional fairness of NOMA are all superior to those
been widely adopted to achieve higher data rate. The demand of OFDMA scheme. In [20], the same authors also showed
that NOMA still achieves higher gains than OFDMA scheme
Fang Fang and Julian Cheng are with School of Engineering, The U-
niversity of British Columbia, Kelowna, BC, VIX 1V7, Canada (e-mail: even with the error propagation in SIC. Though it is simple
fang.fang@alumni.ubc.ca, julian.cheng@ubc.ca). to implement, FTPA fails to optimally allocate power among
Haijun Zhang is with the Department of Electrical and Computer Engineer- multiplexed users on each subchannel. In [21], a new power
ing, the University of British Columbia, Vancouver, BC, V6T 1Z4, Canada.
Haijun Zhang is also with the National Mobile Communications Research allocation scheme based on water filling was proposed to
Laboratory, Southeast University. (e-mail: dr.haijun.zhang@ieee.org). achieve high spectral efficiency. The authors in [22] proposed
Victor C.M. Leung is with the Department of Electrical and Computer cooperative relay system based on NOMA and showed the
Engineering, The University of British Columbia, Vancouver, BC, V6T 1Z4,
Canada (e-mail: vleung@ece.ubc.ca). improvement of the spectral efficiency. A greedy subchannel
This paper has been presented in part at the IEEE International Conference and power allocation algorithm was proposed for the NOMA
on Communications (ICC 2016), Kuala Lumpur, Malaysia. system in [23], and a cooperative NOMA transmission scheme,
This work was supported by the Natural Sciences and Engineering Research
Council of Canada (Grants RGPIN-2014-06119, RGPAS-462031-2014), the where some users have prior information of the other users’
National Natural Science Foundation of China (Grant 61271182, 61471025), message, was proposed to improve spectrum efficiency in [24].
and the Open Research Fund of National Mobile Communications Research The multiple-input and multiple-output (MIMO) NOMA de-
Laboratory, Southeast University (No. 2016D07). (Corresponding author:
Haijun Zhang.) sign for small packet transmission and the multi-user detection
for uplink grant-free NOMA systems were investigated in [25]

0090-6778 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCOMM.2016.2594759, IEEE
Transactions on Communications

and [26], respectively. The energy-efficient power allocation as a difference of two convex functions (DC). Since
was investigated for NOMA systems in [27], [28]. In [27], the problem is non-convex, in order to use the DC
using statistical channel state information at the transmitter, programming approach to obtain its suboptimal solution,
the authors proposed a near optimal power allocation scheme we prove the convexity of the sub-functions in the ob-
to maximize the system energy efficiency. [28] is conference jective function. A convex approximation expression is
version of this paper. In [28], the authors only discussed the found by exploiting the structure of DC programming
user scheduling for NOMA system to maximize the energy problem. Due to the considerable computing complexity
efficiency. In this paper, we also investigated power allocation of the global optimal solution, we propose a suboptimal
scheme to further increase the system energy efficiency. approach to obtain an energy-efficient power allocation
Although several recent works have been considered for scheme by iteratively solving the convex sub-problems.
subchannel and power allocation in NOMA systems [16], [19], The proposed algorithms are evaluated by extensive simu-
[20], these papers mainly focused on sum rate maximization. lations. Numerical results show that the NOMA system with
However, with the exponential growth of wireless data traffic, proposed subchannel assignment and power allocation algo-
energy consumption of wireless networks has been rapidly rithms outperforms the OFDMA system in terms of sum rate
increasing. Therefore, saving transmit energy for a block of and energy efficiency. We further show that the performance
bits is an important and practical consideration. To the best of the power allocation among multiplexed users is better than
of the authors’ knowledge, the resource allocation problem both equal power allocation and the FTPA scheme.
that maximizes the system energy efficiency has not been well
studied for the NOMA systems.
C. Paper Organization
B. Contributions The rest of the paper is organized as follows. Section
In this paper, we consider a downlink NOMA wireless II presents the NOMA system model and formulates the
network where the base station (BS) assigns subchannels to optimization problem. In Section III, we propose a two-sided
multiple users, allocates different powers among users who matching algorithm for subchannel assignment and determine
share the same subchannel, and allocates the power across the power proportional factors for multiplexed users. Section
subchannels. Assuming the BS knows perfect channel state IV introduces the DC programming approach, and a sub-
information, we investigate the energy-efficient subchannel optimal power allocation scheme is proposed to maximize
assignment and power allocation algorithms for the NOMA the system energy efficiency. Performance of the proposed
systems with the constraints of maximum BS transmit power algorithms is evaluated in Section V by simulations. Finally,
and the maximum number of users allocated on the same sub- Section VI concludes the paper.
channel. The main contributions of this paper are summarized
as follows.
D. Notation
• We focus on the energy efficiency aspect of resource
allocation in a downlink NOMA network and use bits The following notation is adopted in the rest of the paper.
per Joule to measure the system energy efficiency per- Lowercase and uppercase bold fonts denote vectors and matri-
formance [29]. We formulate the subchannel assignment ces, respectively. Inequalities between vectors are component-
and power allocation problem for the downlink NOMA wise inequalities. The transpose and the conjugate (Hermitian)
T H
network as an energy efficiency optimization problem, transpose operations are denoted by (·) and (·) , respective-
which is an NP-hard. To obtain an energy-efficient re- ly. ∥·∥1 is the l1 (Euclidean) vector norms [30]. The gradient
source allocation scheme, we decouple subchannel as- of f (x) at point x0 is denoted by ∇f (x0 ).
signment and subchannel power allocation from each
other. A maximum BS transmit power is provided. The II. S YSTEM M ODEL AND P ROBLEM F ORMULATION
maximum number of users can be multiplexed on the
same subchannel is limited to two in order to reduce the A. System Model
computing complexity of the SIC receiver. As a result, the Figure 1 shows a downlink NOMA network. A BS transmits
proposed resource allocation algorithms are only valid for its signals to M user terminals (UTs) through N subchannels,
the two user case. and SIC is employed at the receiver of UTs. We denote m
• For subchannel assignment, we first assume equal power as index for the mth mobile user where m ∈ {1, 2, · · · , M }
allocation across subchannels, then we formulate the and denote n as index for the nth subchannel where n ∈
subchannel assignment as a two-sided matching problem {1, 2, · · · , N }. In the cell, M users are uniformly distributed
between the subchannels and the users. To maximize in a circular region with radius R. The total bandwidth of the
the energy efficiency of the NOMA system, we propose system, BW , is equally divided into N subchannels where
a low-complexity suboptimal matching scheme for sub- the bandwidth of each subchannel is Bsc = BW/N . Let
channel assignment (SOMSA). The complexity of the Mn ∈ {M1 , M2 , · · · , MN } be the number of users allocated
SOMSA is compared to the optimal solution obtained on the subchannel n (SCn ) and the power allocated to the
through exhaustive searching. lth user on SCn is denoted by pl,n . Then the subchannel
• Based on the proposed subchannel assignment scheme, ∑n
M
and BS power constraints are given by pi,n = pn and
the power allocation across subchannels is formulated i=1

0090-6778 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCOMM.2016.2594759, IEEE
Transactions on Communications

pl,n |hl,n |2
DL DL

DL
pl,n Hl,n
UT DL SIN Rl,n = =
DL UT ∑
M n ∑n
M
pi,n |hl,n |2

DL
σn2 +

DL
1+ pi,n Hl,n

D
L
UT UT
UT
BS i=1,i̸=l i=1,i̸=l
(3)
where σn2 = E[|zl,n |2 ] is the noise power on SCn and
UT
UT UT

Hl,n , |hl,n |2 /σn2 represents the channel response normalized


Fig. 1. System model of a downlink NOMA single cell network. by noise (CRNN) of the lth user on SCn .
Based on the Shannon’s capacity formula, the sum rate of

N SCn is given by
pn = Ps , where pn and Ps are, respectively, the allocated ∑
Mn ∑Mn ( )
n=1 pl,n Hl,n
power on SCn and the total transmitted power of the BS. In Rn = Bsc log2 (1 + SIN Rl,n ) = Bsc log2 1 +
1 + Il,n
NOMA systems, we assume that the BS has full knowledge l=1 l=1
(4)
of the channel state information. According to the NOMA
where Il,n is the interference that U Tl,n receives from the
protocol [16], multiple users can be allocated to the same
other users on the SCn , which can be expressed as
subchannel with SIC technique. A block fading channel is
considered in the system model, where the channel fading of ∑
Mn
Il,n = pi,n Hl,n . (5)
each subchannel remains the same, but it varies independently
i=1,i̸=l
across different subchannels. Based on the parameters and
constraints of the system, the BS needs to assign multiple In NOMA systems, the SIC process is implemented at UT
users (with different power levels) to different subchannels receiver to reduce the interference from the other users on
and allocate different powers across subchannels. Considering the same subchannel. The optimal decoding order for SIC is
Mn users are allocated on SCn , the symbol transmitted by the increasing order of CRNNs. Based on this order, any user
the BS on each subchannel SCn can be expressed as can successfully and correctly decode the signals of the other
users with smaller CRNN values. Thus, the interference from

Mn
√ the users having poorer channel condition can be cancelled
xn = pi,n si (1)
and removed by the user who has better channel condition.
i=1
In order to maximize the sum rate of SCn , NOMA protocol
where si is the modulated symbol of the ith user on SCn , allocates higher power to the users with lower CRNN [16], i.e.,
which is denoted by U Ti,n 1 . The received signal at the lth for two users U Ti,n and U Tj,n sharing the same SCn with
user on SCn is CRNNs |Hi,n | ≥ |Hj,n |, we always set pi,n ≤ pj,n to achieve
yl,n = hl,n xn + zl,n user fairness. This assumption is widely used in the NOMA

Mn scheme [19], [20]. Consider that Mn users are allocated on
√ √ (2)
= pl,n hl,n sl + pi,n hl,n si + zl,n SCn with CRNNs order
i=1,i̸=l
|H1,n | ≥ |H2,n | ≥ · · · ≥ |Hl,n | ≥ |Hl+1,n | ≥ · · · ≥ |HMn ,n | .
−1
where hl,n = gl,n · P L (d) is the coefficient of SCn from (6)
the BS to U Tl,n , and where gl,n is assumed to have Rayleigh According to the optimal SIC decoding order, User l can
fading channel gain, and P L−1 (d) is the path loss function successfully decode and remove the interference symbols from
between the BS and U Tl,n at distance d. The impact of users i > l. However, the interference symbol from User i
users’ channel conditions on the performance gain of NOMA (i < l) cannot be removed and will be treated as noise by
over OFDMA was studied in [31]. In this work, the authors User l. Therefore, the SINR of User l with SIC at receiver
presented that the performance gain of the NOMA over can be written as
OFDMA will increase when the difference of channel gain ^Rl,n = pl,n Hl,n
SIN . (7)
of users becomes larger. The authors in [32] showed that the ∑
l−1
distances between BS and UTs will affect the performance of 1+ pi,n Hl,n
i=1
the NOMA system. In this paper, we assume these ( distances
) of
Then the data rate of the lth user on SCn can be expressed
different users are known by BS. Let zl,n ∼ CN 0, σn2 be the
as
additive white Gaussian noise (AWGN) with zero mean and  
variance σn2 . In a downlink NOMA network, each subchannel
 
can be shared by multiple users. Each user on SCn receives  pl,n Hl,n 
Rl,n (pl,n ) = Bsc log2 1 + . (8)
its signals as well as interference signals from the other users  ∑
l−1 
1+ pi,n Hl,n
on the same subchannel. Therefore, without SIC at receiver, i=1
the received signal-to-interference-plus-noise ratio (SINR) of Therefore, the overall sum rate of NOMA systems can be
the lth user on the SCn is written by written as
1 Without causing notational confusion, U T
i,n denotes the ith user on ∑N ∑Mn ∑
N
SCn , while U Tm denotes the mth user in the whole cell, where m ∈ R= Rl,n (pl,n ) = Rn (pn ) . (9)
{1, 2, · · · , M }. n=1 l=1 n=1

0090-6778 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCOMM.2016.2594759, IEEE
Transactions on Communications

B. Problem Formulation and find proportional factor for multiplexed users on each
In this subsection, we formulate the energy-efficient sub- subchannel. Based on the efficient subchannel assignment,
channel assignment and power allocation as an optimization we then focus on the energy-efficient power allocation across
problem. For energy-efficient communication, it is desirable subchannels within the constraint of total transmit power of
to maximize the amount of transmitted data bits with a unit BS.
energy, which can be measured by energy efficiency. For each
subchannel in the NOMA system, given assigned power pn on III. E NERGY-E FFICIENT S UBCHANNEL M ATCHING FOR
SCn and additional circuit power consumption pc , the energy NOMA S YSTEMS
efficiency over SCn is defined as In this section, we investigate the energy-efficient matching
Rn algorithm for subchannel assignment in the NOMA network.
En = . (10) For the optimization problem (13), it can be shown that the
pc + pn
subchannel assignment and power allocation for subchannels
Then the overall energy efficiency of the system can be given are coupled with each other in terms of energy efficiency. Due
by to the considerable complexity of global optimum solution,
∑N
we decouple subchannel assignment and power allocation
E= En . (11)
to obtain a suboptimal solution. We first propose a greedy
n=1
subchannel-user matching algorithm by assuming equal power
For the downlink NOMA network, SIC technique is well is allocated on each subchannel, in which each power pro-
investigated in [12], [15]. The implementation complexity of portional factor βn is also determined to allocate different
SIC at the receiver increases with the maximum number of powers to the multiplexed users on the same subchannel. We
the users allocated on the same subchannel. In order to keep define the parameter βn as the proportional factor of assigned
the receiver complexity comparatively low, we consider the power to the user who performs SIC on SCn . By decomposing
simple case where only two users can be allocated on the the objective function into difference of convex functions, the
same subchannel. This assumption is important because it also suboptimal matching scheme for subchannel assignment is
restricts the error propagation. In this case, given that the two decided by a DC programming approach.
users sharing SCn with CRNNs |H1,n | ≥ |H2,n |, the sum rate
of SCn can be expressed as
A. Subchannel Matching Problem Formulation
Rn (pn ) =W1,n log2 (1 + βn pn H1,n ) To describe the dynamic matching between the users and
( ) the subchannels, we consider subchannel assignment as a two-
(1 − βn ) pn H2,n (12)
+ W2,n log2 1 + sided matching process between the set of M users and the
1 + βn pn H2,n set of N subchannels. Considering only two users can be
where βn is the power proportional factor for the two users on multiplexed on the same subchannel due to the complexity of
SCn . Generally, βn is used for the user who performs SIC on decoding, following [16] we assume M = 2N . We say U Tm
SCn and βn ∈ (0, 1). The optimal power proportional factor and SCn are matched with each other if U Tm is allocated
can be decided within our proposed subchannel assignment on SCn . Based on the perfect channel state information, the
scheme. In (12), Wi,n represents the weighted bandwidth of preference lists of the users and subchannels can be denoted
the ith user. To obtain an energy-efficient resource allocation by
scheme for this system, we formulate the energy efficiency T
P F U T = [P F U T (1), · · · , P F U T (m), · · · , P F U T (M )]
optimization problem as T
P F SC = [P F SC(1), · · · , P F SC(n), · · · , P F SC(N )]
∑N
Rn (pn ) (15)
max (13) where P F U T (m) and P F SC(n) are the preference lists
pn >0 p + pn
n=1 c of U Tm and SCn , respectively. We say U Tm prefers SCi to
SCj if U Tm has higher channel gain on SCi than that on

N
SCj , and it can be expressed as
subject to C1 : Rl,n (pn ) ≥ Rmin ; C2 : p n = Ps
n=1 SCi (m) ≻ SCj (m) . (16)
(14)
where C1 guarantees user minimum data rate constraint and For an example, we consider four users and two subchannels
Rmin is denoted as minimum data rate determined by quality with channel gain matrix
of service (QoS) requirement. The constraint C2 ensures [ ]
the maximum BS power constraint. Since this optimization H = 0.197, 0.778; 0.437, 0.143; 0.322, 0.545; 0.272, 0.478
problem is non-convex and NP-hard, it is challenging to find where row index denotes the users and column index denotes
the global optimal solution within polynomial time. To solve the subchannels. Therefore, we have preference list of the users
this problem efficiently, we will treat subchannel assignment as
and subchannel power allocation separately. Assuming equal [ ]T [ ]T
power is allocated to the subchannels, we first match sub- P F U T (1) = 2 1 , P F U T (2) = 1 2
[ ]T [ ]T
channels to multiple users to maximize the energy efficiency P F U T (3) = 2 1 , P F U T (4) = 2 1

0090-6778 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCOMM.2016.2594759, IEEE
Transactions on Communications

and preference list of the subchannels as a process of finding the preferred matching pair for each user
[ ]T and subchannel.
P F SC(1) = 2 3 4 1
[ ]T
P F SC(2) = 1 3 4 2 . Algorithm 1 Suboptimal Matching for Subchannel Assign-
ment
We say SCn prefers user set qm to user set qn (qn , qm is 1: Initialize the matched list SM atch (n) to record user-
denoted as subsets of {1, 2, · · · , M }) if the users in set qm s matched on SCn for all the subchannels ∀n ∈
can provide higher energy efficiency than users in set qn on {1, 2, · · · , N }.
SCn , and we represent this scenario as 2: Initialize preference lists P F U T (m) and P F SC (n)
En (qm ) > En (qn ) , qm , qn ⊂ {U T1 , U T2 , · · · , U TMn } . for all the users ∀m ∈ {1, 2, · · · , M } and all the sub-
(17) channels ∀n ∈ {1, 2, · · · , N } according to CRNNs.
Matching theory has been studied in [33], [34], where various 3: Initialize the set of unmatched users SU nM atch to record
properties and types of preferences have been discussed. users who has not been allocated to any subchannel.
Based on the preference lists of users and subchannels, the 4: while {SU nM atch } is not empty do
subchannel assignment problem is formulated as a two-sided 5: for m = 1 to M do
matching problem [33], [34]. 6: Each user sends matching request to its most pre-
Definition 1: (Two-sided Matching) Consider users and ferred subchannel n̂ according to P L U T (m).
subchannels as two disjoint sets, M = {1, 2, · · · , M } and 7: if |SM atch (n̂)| < 2 then
N = {1, 2, · · · , N }. A two-to-one, two-sided matching M is 8: Sub-channel n̂ adds user m to SM atch (n̂), and
a mapping from all the subsets of users M into the subchannel removes user m from {SU nM atch }
set N satisfying U Tm ∈ M and SCn ∈ N 9: end if
1) M(U Tm ) ∈ N . 10: if |SM atch (n̂)| = 2 then
2) M−1 (SCn ) ⊆ M . 11: a) Find power proportional factor βn for every
3) |M(U Tm )| = 1, |M−1 (SCn )| = 2. two users in Sqm , Sqm ⊂ {Smatch (n̂), m} by
4) SCn ∈ M(U Tm ) ⇔ U Tm ∈ M−1 (SCn ). using (18), or exhaustive search method or DC
programming algorithm in Section IV.
Condition 1) states that each user matches with one sub- 12: b) Subchannel n̂ selects a set of 2 users Sqm
channel, and Condition 2) represents each subchannel can be satisfying maximum energy efficiency En̂ (qm ) ≥
matched with a subset of users. Condition 3) states that the En̂ (qn ), qm , qn ⊂ {Smatch (n̂), m}.
number of users can be allocated on each subchannel is limited 13: c) Subchannel n̂ sets Smatch (n̂) = qm , and
to two. Condition 4) means U Tm and SCn are matched with rejects other users. Remove the allocated users
each other. from {SU nM atch }, add the unallocated user to
Definition 2: (Preferred Matched Pair) Given a matching {SU nM atch }.
M that U Tm ∈ / ( M−1 (SCn )) and SCn ∈ / M(U Tm ). If 14: d) The rejected user removes subchannel from their
En (Snew ) > En M−1 (SCn ) where Snew ⊆ {U Tm } ∪ S preference lists.
and S = M−1 (SCn ), where S is the user set has been 15: end if
assigned to SCn , Snew becomes the preferred users set for 16: end for
subchannel n and (U Tm , SCn ) is a preferred matched pair. 17: end while
Based on the above definition, we will describe in Section III.B
the matching action between the users and the subchannels.
If each subchannel has to select the best subset of users to 1) SOMSA Algorithm Description: Algorithm 1 describes
allocate, it will cause considerable complexity especially when the proposed SOMSA scheme to maximize the system energy
the number of users is large. Because the optimal solution efficiency. This algorithm includes initialization and match-
requires to search all the possible combinations of the users ing procedures. In the initialization step, preferences lists of
to maximize energy efficiency. To reduce the complexity, a subchannels and users are decided according to the channel
suboptimal matching algorithm is proposed for subchannel state information, and SM atch (n), ∀n ∈ {1, 2, · · · , N } and
assignment as follows. SU nM atch are initialized to record the allocated users on
SCn and unallocated users of the system, respectively. In
the matching procedure, at each round, each user sends the
B. Suboptimal Matching for Subchannel Assignment Algorith- matching request to its most preferred subchannel. Accord-
m in NOMA ing to the preferred list of each user (P F U T (m), ∀m ∈
In this section, we propose a suboptimal matching algorithm {1, 2, · · · , M }) which is a list of subchannels ordered by
for subchannel assignment. The main idea of this matching decreasing channel gains, the mth user will find the first non-
model is that each user sends the matching request to its zero entry in P F U T (m) and send matching request to the
most preferred subchannel according to its preference list. But corresponding subchannel. The subchannel accepts the user
this preferred subchannel has the right to accept or reject directly if the number of allocated users on this subchannel
the user according to energy efficiency that the all users is less than two. When the number of the allocated users
can provide on this subchannel. Based on the equal power equals to two, only the subset of users that can provide higher
allocation across subchannels, the user selection algorithm is energy efficiency will be accepted or it will be rejected. This

0090-6778 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCOMM.2016.2594759, IEEE
Transactions on Communications

matching process will terminate when there is no user left to be through searching all βn values in (0, 1) with a sufficiently
matched. After that, the allocated user and the corresponding small step size. Therefore, the optimal power proportional
subchannels in the preference list are set to zero. The proposed factors for the multiplexed users can be obtained. However, the
SOMSA converges to a stable matching after a limited number computational complexity of the exhaustion method is much
of iterations [34]. higher than FTPA. Therefore, in the following, we consider
2) Complexity Analysis: The optimal subchannel assign- a suboptimal but efficient DC programming to allocate power
ment scheme can only be obtained by searching over all pos- among multiple users to maximize the energy efficiency.
sible combinations of users and selecting one that maximizes
the system energy efficiency. If we have M users and N IV. E NERGY-E FFICIENT P OWER A LLOCATION ACROSS
subchannels (M = 2N ). The scheduler needs to search (2N 2N
)!
S UBCHANNELS FOR NOMA SYSTEMS
combinations. The time complexity of exhaustive searching
As mentioned in Section III, equal power allocation is
is O( (2N
2N
)!
). In order to compare the complexity of different
assumed across subchannels in SOMSA. In order to fur-
algorithms, we take natural logarithm of the complexity. The
ther improve the energy efficiency of the NOMA system,
logarithm complexity is O(ln((2N )!) − N ) = O(ln((2N )!)).
we consider obtaining the energy-efficient subchannel power
By using the Stirling’s formula, ln(n!) = n ln n−n+O(ln(n)),
allocation instead of equal power allocation. In this section,
the logarithm complexity of the exhaustive searching can be
we introduce DC programming approach and discuss its
written by O(N ln N ). In the SOMSA algorithm, the complex-
application in finding power proportional factors as well as
ity of the worst case is O(N 2 ). Taking natural logarithm of
power allocation across subchannels.
the complexity, the logarithm complexity is O(ln N ). Since
O(ln N ) < O(N ln N ) and actual complexity of SOMSA
is much less than the complexity of the worst case, the A. DC Programming
complexity of SOMSA algorithm is much less than the optimal DC programming approach has been studied recently to
subchannel assignment scheme. It can be shown that for a solve non-convex optimization problems [35]. It is shown that
small number of users (M = 4), the SOMSA will yield the DC programming can be applied if the objective function can
identical results from the exhaustive search. be written as a minimization of a difference of two convex
functions, which is represented as
C. Power Allocation Between Multiplexed Users on Each
min q (x) = f (x) − g (x) (19)
Subchannel x∈χ

In Algorithm 1, it is required to determine the power T


where x = [x1 , x2 , · · · xL ] and χ is the convex set; f (x)
proportional factor βn for every two subchannel users. In this
and g (x) are continuous, convex or quasi-convex [35]. In
section, we will first review the existing fractional transmit
general, the problem defined (19) is non-convex. However,
power allocation scheme and exhaustive searching method.
it can be solved sub-optimally by using Algorithm 2. The
Then we will introduce a new energy-efficient power allocation
key idea of Algorithm 2 is to convert a non-convex problem
algorithm based on DC programming in Section IV.B. It will
to convex sub-problems by using successive convex approxi-
be shown in Section V that the new algorithm can result in
mations. In this algorithm, ε is the difference tolerance and
improved energy efficiency.
1) Fractional transmit power allocation: According to the
the term( −g)(x) in the ( objective
)( function
) (19) is replaced
by −g x(k) − ∇g T x(k) x − x(k) in (20). The convex
SINR expression in (7), the transmit power allocation to one
optimization problem in (20) can be solved by using standard
user affects the achievable sum rate as well as the energy
algorithms from convex optimization theory [36]–[38], i.e.,
efficiency on each subchanel. Due to its low computational
interior point method and sequential quadratic programming.
complexity, FTPA is widely adopted in OFDMA systems and
In this paper, sequential quadratic programming is used in the
NOMA systems [15], [16]. In the FTPA scheme, the transmit
simulations.
power of U Tm on SCn is allocated according to the channel
gains of all the multiplexed users on SCn , which is given as Algorithm 2 Iterative, Suboptimal Solution for DC Problems
−α [36]
(Hl,n )
pl,n = pn M (18) (0)
∑n −α (x (k+1)
1: Initialize , set) iteration
( number
) k = 0.
(Hi,n ) 2: while q x − q x(k) > ε do
i=1
3: Define convex approximation of q (k) (x) as
where α (0 ≤ α ≤ 1) is a decay factor. In the case α = 0, ( ) ( )( )
it corresponds to equal power allocation among the allocated q̂ (k) (x) = f (x) − g x(k) − ∇g T x(k) x − x(k)
users. From (18), it is clear that when α increases, more power (20)
is allocated to the user with poorer CRNN. Note that the 4: Solve the convex problem
same decay factor should be applied to all subchannels and
transmission times. x(k+1) = arg min q̂ (k) (x) (21)
x∈χ
2) Exhaustive searching method: In finding power pro-
portional factor βn , the method of exhaustion also can be 5: k ←k+1
exploited for βn ∈ (0, 1). The optimal value can be found 6: end while

0090-6778 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCOMM.2016.2594759, IEEE
Transactions on Communications

The convergence of Algorithm 2 can be easily proved by


( ) ( ) ( ) ( )

N
q x(k) = q̂ (k) x(k) ≥ q̂ (k) x(k+1) ≥ q x(k+1) subject to C1 : Rl,n (pn ) ≥ Rmin ; C2 : pn = Ps
( (k) ) ( (k) ) (22) n=1
where q x = q̂ (k)
x is the kth iteration step, and (28)
( (k) ) ( (k+1) )
q̂ (k)
x ( ≥) q̂ (k)
x can be obtained by (21). There- where Rl,n (pl,n ) is defined in (8). Since Rl,n (pl,n ) is a
fore, q x(k) monotonically decreases when k increases. linear function respect to assigned power pn on SCn . The
Under an additional assumption that f (x) and g (x) are constraint C1 can be converted to pn > pn,min , where
continuous and differentiable on the constraint set. In this case, pn,min is the minimum assigned power on SCn determined by
Algorithm 2 always returns a stationary point of q (x), which Rmin . Condition C2 in (28) guarantees BS power constraint.
may not be the global optimal solution [35]. Note that the optimization problem in (27) is non-convex
with respect to pn . However, the representation of (27) is
similar to the DC problem representation. Thus (27) can be
B. DC Programming to Obtain Power Proportional Factor βn rewritten as (29) and (30) at the top of next page. Where
T
Considering two users U T1 and U T2 that are to be mul- P = [p1 , p2 , · · · , pn , · · · , pN ] represents the allocated pow-
tiplexed over SCn with CRNNs H1,n ≥ H2,n and weighted ers on the subchannels. Problem (27) can be written as
bandwidths W1,n , W2,n . According to the principle of SIC
min Q (P) = min F (P) − G (P)
decoding sequences, U T1 can cancel the interfering power P≻0 P≻0
term of U T2 , whereas U T2 treats the symbol power U T1
as noise. The problem of finding βn to maximize energy
efficiency of SCn can be formulated as subject to C1 : P ≻ Pmin ; C2 : ∥P∥1 = Ps (31)
T
W1,n log2 (1 + βn pn H1,n ) where Pmin = [p1,min , p2,min , · · · , pn,min , · · · , pN,min ]
max and P ≻ Pmin means all the elements in P are larger than the
βn ∈(0,1) pc + pn
( ) (23) corresponding elements in Pmin , pn > pn,min . Proposition
(1−β )p H2,n
W2,n log2 1 + 1+βnn pnnH2,n 1 on the next page proves convexity of F (P) and G (P),
+ Therefore, the DC programming approach can be applied to
pc + pn
realize energy-efficient power allocation using Algorithm 3 on
which can be rewritten as
( ) the next page. Once the power allocation over subchannels is
1+p H obtained, we replace the equal power allocation with our new
W1,n log2 (1 + βn pn H1,n ) + W2,n log2 1+βnnpn H2,n2,n
max . power allocation scheme to achieve higher energy efficiency
βn ∈(0,1) pc + pn
(24) of the system.
In order to use the DC programming approach, we can convert
Algorithm 3 DC Programming Algorithm for Power Alloca-
(24) to DC representation
tion across Subchannels
( ) 1: Initialize P(0) , set iteration number k = 0. The Objective
1+pn H2,n
W1,n log2 (1 + βn pn H1,n ) W2,n log2 1+βn pn H2,n function Q( (P), convex
) ) F (P) and G (P) .
functions
(
min − − (k)
βn ∈(0,1) pc + pn pc + pn 2: while Q P
(K+1)
−Q P > ε do
(25) 3: Define convex approximation of G(k) (P) at P(k) as
or ( ) ( )( )
min (f (βn ) − g (βn )) (26) Q(k) (P) = F (P)−G P(k) −∇GT P(k) P − P(k)
βn ∈(0,1)
(32)
W1,n log2 (1+βn pn H1,n )
where f (βn ) = − pc +pn and g (βn ) = 4: Solve the convex problem
( 1+pn H2,n
)
W2,n log2 1+βn pn H2,n
, and both terms are convex functions P(k) = arg min Q(k) (P) (33)
pc +pn pn ≥pn,min , ∥P∥1 =Ps
with respect to βn because ∇2 f (βn ) > 0 and ∇2 g (βn ) > 0.
Therefore, the DC programming approach can be used to find 5: k ←k+1
βn by replacing x with βn in Algorithm 2. 6: end while
( )
In Algorithm 3, ∇G P(k) is the gradient of G(P) at
C. Subchannel Power Allocation by DC Programming
the point P(k) and it is calculated by (34) at the top of
Given the subchannel-user matching scheme and power
this page. Since (32) and the power domain are convex,
proportional factors on different subchannels by Algorithm 1,
problem (33) can be solved by either the interior point
the optimization problem in (13) can rewritten as
 ( method
)  or the sequential quadratic programming. In order
∑  W1,n log (1 + βn pn H1,n ) W2,n log2 1+βn pn H2,n 
N 1+pn H2,n to use the DC programming approach, the quasi-convexity
2 of F (P) and G (P) needs to be established. It is easy to
max +
pn ≥0  pc + pn pc + pn  ∑
N
n=1
show that f (P) = − W1,n log2 (1 + βn pn H1,n ) −
(27) n=1

0090-6778 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCOMM.2016.2594759, IEEE
Transactions on Communications

 ( )
N 
W1,n log2 (1 + βn pn H1,n ) W2,n log2 1+βn pn H2,n 
1+pn H2,n

min − +
pn ≥0  pc + pn pc + pn 
n=1
{ N N ( )}
∑ W1,n log (1 + βn pn H1,n ) ∑ N
W2,n log2 (1 + pn H2,n ) ∑ W2,n log2 (1 + βn pn H2,n )
min − 2
− + .
pn ≥0 pc + pn pc + pn pc + pn
n=1 n=1 n=1
(29)
Let
∑ N ( )
W1,n log2 (1 + βn pn H1,n ) ∑ W2,n log2 (1 + pn H2,n ) ∑
N N
W2,n log2 (1 + βn pn H2,n )
F (P) = − − ; G (P) = − .
n=1
pc + pn n=1
pc + pn n=1
pc + pn
(30)

( ) ∑ N W βn H2,n
1,n log2 (1 + βn pn H2,n ) − (pc + pn ) (1+βn pn H2,n ) ln 2
∇G P(k) = 2 . (34)
n=1 (pc + pn )


N
W2,n log2 (1 + pn H2,n ) and g (P) = 62

n=1
∑N 60
− W2,n log2 (1 + βn pn H2,n ) are convex since ∇2 f (P)

Total sum rate of system (Mbps)


n=1 58
and ∇2 g (P) are positive semi-definite matrices.
Proposition 1: If −f1 (pn ) = W1,n log2 (1 + βn pn H1,n ) + 56

W2,n log2 (1 + pn H2,n ) and −g1 (pn ) = 54


W2,n log2 (1 + βn pn H2,n ) are strictly concave in pn ,
−F1 (pn ) = −f 1 (pn ) −g1 (pn )
pc +pn and −G1 (pn ) = pc +pn are quasi-
52

NOMA−DC
concave with constant pc . Inspired by [39], we can prove 50 NOMA−EQ
OFDMA
Proposition 1 as follows.
48
10 20 30 40 50 60
Number of users per BS
Proof : Denote the α-sublevel sets of function −F1 (pn ) as
Fig. 2. Sum rate of the system versus different number of users.
Sα = {pn > 0| − F1 (pn ) ≥ α} . (35)

Based on the Proposition 1, −F1 (pn ) is strictly quasi-concave


if and only if Sα is strictly convex for any α. In this case, with two users. In OFDMA schemes, each user can only be as-
when α < 0, there are no points satisfying −F1 (pn ) = α. signed to one subchannel. In the simulations, we compare the
Therefore, Sα is strictly convex when α ≤ 0. When α > 0, we performance of NOMA systems with OFDMA systems, both
can rewrite Sα as Sα = {pn > 0|α (pc + pn ) + f1 (pn ) ≤ 0}. with resource allocation algorithms. For the subchannel power
Since f (pn ) is strictly convex in pn , Sα is therefore also allocation, we compare our proposed suboptimal algorithm
strictly convex. Hence, −F1 (pn ) and −G1 (pn ) are strictly with equal power allocation scheme based on our proposed
quasi-concave. Therefore, F1 (pn ) and G1 (pn ) are quasi- subchannel assignment scheme. FTPA for multiplexed users
convex. As a result, F (P) and G (P) are quasi-convex.  on subchannel is also compared with our proposed algorithms.
In our simulations, we set BS peak power, Ps , to be 41
dBm and circuit power consumption pc = 1 W [40]. The
V. S IMULATION R ESULTS AND D ISCUSSION
maximum number of users is 60 and σn2 = BW N N0 , where
In this section, simulation results are presented to evaluate N0 = −174 dBm/Hz is the AWGN power spectral density.
the performance of the proposed resource allocation algo- For simplicity, we consider each user has the same weighted
rithms for NOMA systems through extensive Monte Carlo bandwidth (W1,n = W2,n = BW/N ). In the simulations, we
simulations. In the simulations, we consider one base station set the value of α as 0.4 [19].
located in the cell center and the user terminals are uniformly In Fig. 2, the performance of total sum rate is evaluated
distributed in a circular range with radius of 500 m. We set with the number of users M (M varies from 10 to 60). We
the minimum distance between users to be 40 m, and the set difference tolerance ε = 0.01, and the bandwidth is limited
minimum distance from users to BS is 50 m. The bandwidth to 5 MHz. It is shown that the total sum rate increases when
is 5 MHz. Let M users be randomly distributed in the cell. the number of the users grows. As the number of users grows
In NOMA systems, to reduce demodulating complexity of the larger, the sum rate continues to increase, but the rate of growth
SIC receiver, we consider each subchannel is only allocated becomes slower, as expected from the Shannon’s formula in

0090-6778 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCOMM.2016.2594759, IEEE
Transactions on Communications

calculating the sum rate. From Fig. 2, we observe that the 60

performance of NOMA system with the proposed resource


50
allocation algorithms, including subchannel assignment and

Total sum rate of system (Mbps)


power allocation, is much better than the OFDMA scheme. For 40
example, when the number of users is 30, the sum rate of the
proposed algorithm (NOMA-DC) 2 is 12.5% more than that of 30
55.5
OFDMA scheme, and the sum rate of equal power allocation
20
(NOMA-EQ) is 11.9% more than that of OFDMA scheme. 55
10.6 10.7
That is because in OFDMA scheme, one subchannel can only 10
NOMA−DC
NOMA−EQ
be used by one user. As a result, BS cannot fully use the OFDMA

spectrum resources. For different subchannel power allocation 0


0 2 4 6 8 10 12
schemes, the sum rate of NOMA-DC is higher than that of Power of BS with 10 users (Watt)

NOMA-EQ. Figure 3 shows the energy efficiency versus the


number of users with the same constraints of Fig. 2. It can Fig. 4. Sum rate of the system versus BS power.
be observed that the energy efficiency also increases when 7
x 10
the number of users grows. The trend of curve is similar to 4

the sum rate curves due to the energy efficiency expression. 3.5

Energy effiency of system (bits/Joule)


From this figure, the performance of our proposed subchannels 3
and power allocation is much more energy-efficient than the
2.5
OFDMA scheme. Our proposed subchannel power allocation
through the DC programming achieves better performance 2

than the equal power allocation. When the number of users 1.5

is 30, the energy efficiency of NOMA-DC is 33% more than 1


that of the OFDMA scheme and 19% more than NOMA-EQ. NOMA−DC
0.5 NOMA−EQ
In Fig. 4, the performance of total sum rate versus BS power OFDMA

with a fixed circuit power of pc = 1 W, the total number of 0


0 2 4 6 8 10 12
Power of BS with 10 users (Watt)
users M = 10, and the BS power is from 1 W to 12 W. In Fig.
4, the sum rate of the system increases as the BS power grows.
Fig. 5. Energy efficiency of the system versus BS power.
In NOMA systems, our proposed algorithm using DC for
subchannel power allocation performs better than equal power
allocation. Both algorithms outperforms the OFDMA system.
For larger BS power levels, NOMA-DC achieves much better
Figure 5 illustrates the total energy efficiency versus BS power
performance than NOMA-EQ and OFDMA.
with the fixed circuit power of pc = 1 W, and the number of
users is M = 10, and the BS power ranges from 1 W to 12 W. x 10
7

It shows that the total energy efficiency first increases from 0 3


NOMA−DC
when BS transmit power increases. After the power reaches a NOMA−EQ
OFDMA
Energy effiency of system (bits/Joule)

certain level, the total energy efficiency begins to decreases. 2.5

That is because there is a tradeoff between transmission


capacity and power consumption for the energy-efficient power 2

allocation. From Fig. 5, it is seen that NOMA-DC can achieve


better performance than NOMA-EQ and OFDMA schemes. 1.5

2 NOMA-DC uses DC programming to allocate power across subchannels.


1

7
x 10 0.5
5 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Pc/Ps with 10 users
4.5
Total Energy Efficiency of System (bits/Joule)

4 Fig. 6. Energy efficiency of the system versus Pc/BS power.


3.5

3
Figure 6 shows the total energy efficiency versus circuit
power to BS power ratio pc /Ps . The system energy efficiency
2.5
decreases when the ratio pc /Ps increases. With the fixed BS
2
power of 12 W, the system performs less energy-efficient
1.5 NOMA−DC−DC Pc=1w
NOMA−DC−EQ Pc=1w
when the circuit power increases. According to the definition
1
OFDMA
of energy efficiency, its value will become smaller when
10 15 20 25 30 35 40 45 50 55 60
Number of UE per BS pc increases. However, the NOMA system equipped with
the proposed resource allocation algorithms still outperforms
Fig. 3. Energy efficiency of the system versus different number of users. the OFDMA system. In Fig. 7, FTPA among multiplexed

0090-6778 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCOMM.2016.2594759, IEEE
Transactions on Communications

10

5
x 10
the proposed subchannel power allocation achieves better
Energy efficiency of system (bits/Joule) 4.5
performance than the equal power allocation scheme.
4
R EFERENCES
3.5
[1] 3GPP TR 36.913 (V8.0.0), “Requirements for further advancements for
3 E-UTRA (LTE-Advanced),” June 2008.
[2] Qualcomm, “3GPP RAN Rel-12 and beyond,” Qualcomm, Tech. Tep.,
2.5
2012.
2 NOMA−DC−DC Pc=1w
[3] Z. Gao, L. Dai, Z. Wang, and S. Chen, “Spatially common sparsity based
NOMA−DC−EQ Pc=1w adaptive channel estimation and feedback for FDD massive MIMO,”
1.5 OFDMA IEEE Trans. Signal Process., vol. 63, no. 23, pp. 6169–6183, Dec. 2015.
NOMA−FTPA−EQ Pc=1w
[4] X. Gao, L. Dai, S. Han, C.-L.I, and R. W. Heath, “Energy-efficient
1
10 15 20 25 30 35 40 45 50 55 60 hybrid analog and digital precoding for mmWave MIMO systems with
Number of users per BS large antenna arrays,” IEEE J. Sel. Areas Commun., vol. 34, no. 4, pp.
998–1009, Apr. 2016.
[5] H. Zhang, X. Chu, W. Guo, and S. Wang, “Coexistence of Wi-Fi and
Fig. 7. Energy efficiency of the system versus different number of users.
heterogeneous small cell networks sharing unlicensed spectrum,” IEEE
Commun. Mag., vol. 53, no. 3, pp. 158–164, Mar. 2015.
[6] H. Zhang, C. Jiang, J. Cheng, and V. C. M. Leung, “Cooperative
users with equal subchannel power scheme is compared with interference mitigation and handover management for heterogeneous
cloud small cell networks,” IEEE Wireless Commun., vol. 22, no. 3,
the NOMA-DC, NOMA-EQ and OFDMA schemes. Fig. 7 pp. 92–99, June 2015.
shows that the energy efficiency increases as the number [7] H. Zhang, C. Jiang, Q. Hu, and Y. Qian, “Self-organization in disaster
of users grows. NOMA-DC-DC3 performs the best among resilient heterogeneous small cell networks,” IEEE Network, vol. 30, no.
2, pp. 116–121, Mar. 2016.
those schemes. When user number is 20, the energy efficiency [8] J. Thompson, X. Ge, H. Wu, R. Irmer, H. Jiang, G. Fettweis, and S.
of NOMA-DC-DC is 35% more than that of the OFDMA Alamouti, “5G wireless communcation systems: Prospects and chal-
scheme and 30% more than NOMA-FTPA-EQ4 . NOMA-DC- lenges,” IEEE Commun. Mag., vol. 52, no. 2, pp. 62–64, Feb. 2014.
[9] H. Zhang, C. Jiang, X. Mao, and H. Chen, “Interference-limit resource
EQ5 achieves 12% more than OFDMA schemes in terms of optimization in cognitive femtocells with fairness and imperfect spec-
energy efficiency. trum sensing,” IEEE Trans. Veh. Technol., vol. 65, no. 3, pp. 1761–1771,
Mar. 2016.
[10] H. Zhang, C. Jiang, N. Beaulieu, X. Chu, X. Wang, and T. Quek,
“Resource allocation for cognitive small cell networks: A cooperative
VI. C ONCLUSION bargaining game theoretic approach,” IEEE Trans. Wireless Commun.,
vol. 14, no. 6, pp. 3481–3493, June 2015.
By assigning only two users to the same subchannel, we [11] H. Zhang, C. Jiang, N. C. Beaulieu, X. Chu, X. Wen, and M. Tao,
proposed energy-efficient resource allocation algorithms for a “Resource allocation in spectrum-sharing OFDMA femtocells with
downlink NOMA wireless network. These algorithms include heterogeneous services,” IEEE Trans. Commun., vol. 62, no. 7, pp.
2366–2377, July 2014.
subchannel assignment, power proportional factors determi- [12] K. Higuchi and Y. Kishiyama, “Non-orthogonal access with random
nation for multiplexed users and power allocation across beamforming and intra-beam SIC for cellular MIMO downlink,” in IEEE
subchannels. By formulating subchannel assignment problem VTC Fall, pp. 1–5, Sept. 2013.
[13] Y. Endo, Y. Kishiyama, and K. Higuchi, “A study on transmission power
as a two-sided matching problem, we proposed the SOMSA control considering inter-cell interference for non-orthogonal access with
algorithm to maximize the system energy efficiency. Power MMSE-SIC in cellular uplink,” in IEICE RCS2012, pp. 19–24, Sept.
proportional factors for the multiplexed users on each subchan- 2012.
[14] J. Umehara, Y. Kishiyama, and K. Higuchi, “Enhancing user fairness
nel are determined by SOMSA. In the power allocation across in non-orthogonal access with successive interference cancellation for
subchannels scheme, since the objective function is non- cellular downlink,” Proc. of ICCS, Nov. 2012.
convex, DC programming was utilized to approximate the non- [15] N. Otao, Y. Kishiyama, and K. Higuchi, “Performance of non-orthogonal
access with SIC in cellular downlink using proportional fair-based
convex optimization problem into the convex sub-problem. resouce allocation,” ISWCS, pp. 476–480, Aug. 2012.
Therefore, a suboptimal power allocation across subchannels [16] Y. Saito, Y. Kishiyama, A. Benjebbour, T. Nakamura, A. Li, and K.
was obtained by solving the convex sub-problems iteratively. Higuchi, “Non-orthogonal multiple access (NOMA) for cellular future
radio access,” Proc. IEEE 77th Veh, Technol. Conf., vol. 62, no. 7, pp.
Based on the resource scheduling from proposed SOMSA 1–5, Dresden, Germany, June 2013.
algorithm, further improvement in the system energy efficiency [17] L. Dai, B. Wang, Y. Yuan, S. Han, C.-L. I, and Z. Wang, “Nonorthogonal
was achieved by the proposed subchannel power allocation multiple access for 5G: Solutions, challenges, opportunities, and future
research trends,” IEEE Commun. Mag., vol. 53, no. 9, pp. 74–81, Sept.
scheme. Through extensive simulations, the performance of 2015.
the proposed algorithms for resource allocation was compared [18] Z. Ding, Y. Liu, J. Choi, Q. Sun, M. Elkashlan, and H. V. Poor, “Ap-
with the OFDMA system. It was shown that the total sum plication of non-orthogonal multiple access in LTE and 5G networks,”
submitted in IEEE Commun. Mag., 2016.
rate and energy efficiency of NOMA system are much higher [19] Y. Saito, A. Benjebbour, Y. Kishiyama, and T. Nakamura, “System-level
than the OFDMA scheme. The proposed power allocation for performance evaluation of downlink non-orthogonal multiple access
subchannel users outperforms the FTPA scheme. Moreover, (NOMA),” in IEEE PIMRC, pp. 611–615, Sept. 2013.
[20] A. Benjebbour, A. Li, Y. Saito, Y. Kishiyama, A. Harada, and T.
Nakamura, “System-level performance of downlink NOMA for future
3 NOMA-DC-DC uses DC programming approach to determine the power
LTE enhancements,” in IEEE Globecom Workshops, pp. 66–70, Dec.
proportional factors and allocate different powers acrose the subchannels. 2013.
4 NOMA-FTPA-EQ uses FTPA to determine the power proportional factors,
[21] M. R. Hojeij, J. Farah, C. A. Nour, and C. Douillard, “Resource
and equal power allocation acrose the subchannels. allocation in downlink non-orthogonal multiple access (NOMA) for
5 NOMA-DC-DC uses DC programming approach to determine the power future radio access,” IEEE 81st Vehicular Technology Conference (VTC
proportional factors and equal power allocation acrose the subchannels. Spring), pp. 1–6, Glasgow, UK, May 2015.

0090-6778 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCOMM.2016.2594759, IEEE
Transactions on Communications

11

[22] J.-B. Kim and I.-H. Lee, ”Capacity analysis of cooperative relaying Haijun Zhang (M’13) is a Postdoctoral Research
systems using non-orthogonal multiple access,” IEEE Commun. Lett., Fellow in in Department of Electrical and Computer
vol. 19, no. 11, pp. 1949–1952, Nov. 2015. Engineering, the University of British Columbia
[23] P. Parida and S. Das, “Power allocation in OFDM based NOMA system: (UBC), Vancouver, Canada. He received his Ph.D.
A DC programming approach,” in IEEE Globecom Workshops, pp. degree in Beijing University of Posts Telecommuni-
1026–1031, Dec. 2014. cations (BUPT). From September 2011 to September
[24] Z. Ding, M. Peng, and H. V. Poor, “Cooperative non-orthogonal multiple 2012, he visited Centre for Telecommunications
access in 5G systems,” IEEE Commun. Lett., vol. 19, no. 8, pp. 1462– Research, King’s College London, London, UK, as
1465, June 2015. a Visiting Research Associate, supported by China
[25] Z. Ding, L. Dai, and H. V. Poor, “MIMO-NOMA design for small packet Scholarship Council. Dr. Zhang has published more
transmission in the internet of things,” accepted, IEEE Access, 2016. than 70 papers and authored 2 books. He serves as
[26] B. Wang, L. Dai, T. Mir, and Z. Wang, “Joint user activity and data Editor of Journal of Network and Computer Applications, Wireless Networks,
detection based on structured compressive sensing for NOMA” IEEE Telecommunication Systems, and KSII Transactions on Internet and Infor-
Commun. Lett., vol. 20, no. 7, pp. 1473–1476, July 2016. mation Systems. He also serves as Leading Guest Editor of ACM/Springer
[27] Q. Sun, S. Han, C.-L. I, and Z. Pan, “Energy efficiency optimization Mobile Networks & Applications (MONET) Special Issue on “Game Theory
for fading MIMO non-orthogonal multiple access systems,” Proc. IEEE for 5G Wireless Networks”. He serves as General Chair of GameNets’16,
ICC 2015, pp. 2668–2673, London, UK, June 2015. and served as Symposium Chair of the GameNets’14 and Track Chair of
[28] F. Fang, H. Zhang, J. Cheng, and V. C. M. Leung, “Energy efficiency ScalCom’15. His current research interests include 5G, Resource Allocation,
of resource scheduling for non-orthogonal multiple access (NOMA) NOMA, LTE-U, Heterogeneous Small Cell Networks and Ultra-Dense Net-
wireless network,” Proc. IEEE ICC 2016, Kuala Lampur, Malaysia, May works.
2016.
[29] C. U. Saraydar, N. B. Mandayam, and D. J. Goodman, “Pricing and
power control in a multicell wireless data network,” IEEE J. Sel. Areas
Commun., vol. 19, no. 10, pp. 1883–1892, Oct. 2001.
[30] R. A. Horn and C. R. Johnson, Matrix Analysis. New York, NY,
Cambridge University Press, USA, 1985.
[31] Z. Ding, P. Fan, and H. V. Poor, “Impact of user pairing on 5G Julian Cheng (S’96, M’04, SM’13) received a
nonorthogonal multiple access downlink transmissions,” IEEE Trans. B.Eng. degree (with first-class honors) in electrical
Veh. Technol., Sept. 2015. engineering from the University of Victoria, Victo-
[32] Z. Ding, Z. Yang, P. Fan, and H. V. Poor, “On the performance of ria, BC, Canada, in 1995, a M.Sc.Eng. degree in
non-orthogonal multiple access in 5G systems with randomly deployed mathematics and engineering from Queens Univer-
users,” IEEE Sig. Process. Lett., vol. 21, no. 12, pp. 1501–1505, Dec. sity, Kingston, ON, Canada, in 1997, and a Ph.D.
2014. degree in electrical engineering from the University
[33] A. Roth and M. Sotomayor, Two-Sided Matching: A Study in Game of Alberta, Edmonton, AB, Canada, in 2003. He
Theoretic Modeling and Analysis. Cambridge University Press, 1992. is currently a Full Professor (with tenure) in the
[34] S. Bayat, R. Louie, Z. Han, B. Vucetic, and Y. Li, “Distributed user as- School of Engineering, Faculty of Applied Science
sociation and femtocell allocation in heterogeneous wireless networks,” at The University of British Columbia (Okanagan
IEEE Trans. Commun., vol. 62, no. 8, pp. 3027–3043, Aug. 2014. campus) in Kelowna, BC, Canada. Previously he worked for Bell Northern
[35] N. Vucic, S. Shi, and M. Schubert, “DC programming approach for Research and Northern Telecom (later known as NORTEL Networks). His
resource allocation in wireless networks,” Proc. Int. Symp. on Modeling current research interests include digital communications over fading chan-
and Optimization in Mobile, Ad Hoc and Wireless Networks (WiOpt), nels, statistical signal processing for wireless applications, optical wireless
pp. 380–386, June 2010. communications, and 5G wireless networks.
[36] D. P. Beretsekas, Nonlinear Programming. Athena Scientific, Belmont, Dr. Cheng co-chaired the 12th Canadian Workshop on Information Theory
MA, USA, 1999. (CWIT 2011), the 28th Biennial Symposium on Communications (BSC 2016),
[37] A. Ben-Tal and A. Nemirovski, Lectures on Modern Convex Optimiza- and the 6th EAI International Conference on Game Theory for Networks
tion: Analysis, Algorithms, and Engineering Applications. MPS-SIAM (GameNets 216). He currently serves as an Editor of IEEE COMMUNICA-
Series on Optimization, Philadelphia, PA, USA, 2001 TIONS LETTERS, IEEE TRANSACTIONS ON WIRELESS COMMUNI-
[38] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge Uni- CATIONS, and IEEE Access. He served as a Guest Editor for a special issue
versity Press, New Yourk, NY, USA, 2004. of the IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS
[39] G. Miao, N. Himayat, and G. Y. Li, “Energy-efficient link adaptation on optical wireless communications. He is also a Registered Professional
in frequency-selective channels,” IEEE Trans. Commun., vol. 58, no. 2, Engineer in the Province of British Columbia, Canada. Currently, he serves
pp. 545–554, Feb. 2010. as a Vice President of the Canadian Society of Information Theory.
[40] D. Ng, E. Lo, and R. Schober, “Energy-efficient resource allocation in
OFDMA systems with large numbers of base station antennas,” IEEE
Trans. Wireless Commun., vol. 11, no. 9, pp. 3292-3304, Sept. 2012.

Victor C. M. Leung (S’75, M’89, SM’97, F’03) is


a Professor of Electrical and Computer Engineering
and holder of the TELUS Mobility Research Chair
at the University of British Columbia (UBC). His
research is in the areas of wireless networks and
Fang Fang is currently pursuing the Ph.D. degree mobile systems. He has co-authored more than 900
at the School of Engineering, the University of technical papers in archival journals and refereed
British Columbia (UBC), Kelowna, BC, Canada. She conference proceedings, several of which had won
received the B.A.Sc. and M.A.Sc. degrees in Elec- best-paper awards. Dr. Leung is a Fellow of the
tronic Information Science and Engineering from Royal Society of Canada, a Fellow of the Cana-
Lanzhou University, China, in 2010 and 2013, re- dian Academy of Engineering and a Fellow of the
spectively. Her current research interests include 5G, Engineering Institute of Canada. He is serving on the editorial boards of
Resource Allocation for Non-orthogonal Multiple IEEE JSAC-SGCN, IEEE Wireless Communications Letters, IEEE Access and
Access Network (NOMA), Performance Analysis several other journals. He has provided leadership to the technical program
for NOMA and Heterogeneous Small Cell Network. committees and organizing committees of numerous international conferences.
Dr. Leung was the recipient of the 1977 APEBC Gold Medal, NSERC
Postgraduate Scholarships from 1977-1981, a 2012 UBC Killam Research
Prize, and an IEEE Vancouver Section Centennial Award.

0090-6778 (c) 2016 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.

You might also like