IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 19, NO.

2, SECOND QUARTER 2017 721

Power-Domain Non-Orthogonal Multiple

Access (NOMA) in 5G Systems:
Potentials and Challenges
S. M. Riazul Islam, Member, IEEE, Nurilla Avazov, Octavia A. Dobre, Senior Member, IEEE,
and Kyung-Sup Kwak, Member, IEEE

Abstract—Non-orthogonal multiple access (NOMA) is one of mobile communications technology is presently facing a new
the promising radio access techniques for performance enhance- challenge, giving birth to a hyper-connected society through
ment in next-generation cellular communications. Compared the emergence of fifth-generation (5G) services. With enor-
to orthogonal frequency division multiple access, which is
a well-known high-capacity orthogonal multiple access technique, mous potential for both consumers and industry, 5G is
NOMA offers a set of desirable benefits, including greater spec- expected to roll out by 2020. From the next-generation radio
trum efficiency. There are different types of NOMA techniques, access technology viewpoint, a step change in data speed and
including power-domain and code-domain. This paper primar- a significant reduction in end-to-end latency is a major con-
ily focuses on power-domain NOMA that utilizes superposition cern for 5G, since the rapid development of the mobile Internet
coding at the transmitter and successive interference cancella-
tion at the receiver. Various researchers have demonstrated that and the Internet of Things (IoT) exponentially accelerates the
NOMA can be used effectively to meet both network-level and demand for high data–rate applications. In particular, many
user-experienced data rate requirements of fifth-generation (5G) of the industry initiatives that have progressed with work on
technologies. From that perspective, this paper comprehensively 5G declare that the network-level data rate in 5G should be
surveys the recent progress of NOMA in 5G systems, reviewing 10-20 Gbps (that is, 10-20 times the peak data rate in 4G), and
the state-of-the-art capacity analysis, power allocation strategies,
user fairness, and user-pairing schemes in NOMA. In addition, the user-experienced data rate should be 1 Gbps (100 times
this paper discusses how NOMA performs when it is inte- the user-experienced data rate in 4G). They also set the latency
grated with various proven wireless communications techniques, (end-to-end round-trip delay) at 1 millisecond (one-fifth of the
such as cooperative communications, multiple-input multiple- latency in 4G).
output, beamforming, space–time coding, and network coding The underlying physical connection in a cellular network
among others. Furthermore, this paper discusses several impor-
tant issues on NOMA implementation and provides some avenues is called radio access technology, which is implemented by
for future research. a radio access network (RAN). A RAN basically utilizes
a channel access technique to provide the mobile terminals
Index Terms—Non-orthogonal multiple access (NOMA),
orthogonal multiple access (OMA), 5G, NOMA solutions, with a connection to the core network. The design of a suit-
NOMA performance, research challenges, implementation able multiple access technique is one of the most important
issues. aspects in improving the system capacity. Multiple access
techniques can broadly be categorized into two different
approaches [1], namely, orthogonal multiple access (OMA)
I. I NTRODUCTION and non-orthogonal multiple access (NOMA). An orthogonal
ROM analog phone calls through to all Internet Protocol
F services, including voice and messaging, each transition
has been encouraged by the need to meet the requirements
scheme allows a perfect receiver to entirely separate unwanted
signals from the desired signal using different basis functions.
In other words, signals from different users are orthogonal
of the new generation of mobile technology. Subsequently, to each other in orthogonal schemes. Time division multiple
access (TDMA) and orthogonal frequency-division multiple
Manuscript received March 28, 2016; revised July 16, 2016 and access (OFDMA) are a couple of examples of OMA schemes.
September 8, 2016; accepted October 15, 2016. Date of publication
October 25, 2016; date of current version May 31, 2017. This work was In TDMA, several users share the same frequency channel
supported in part by the National Research Foundation of Korea by the on a time-sharing basis. The users communicate in rapid suc-
Korean Government (Ministry of Science, ICT and Future Planning) under cession, one after the other, each using their assigned time
NRF-2014K1A3A1A20034987, and in part by the Natural Sciences and
Engineering Research Council of Canada (NSERC) through its Discovery slots. OFDMA allows multi-user communications through
Program. (Corresponding author: Kyung-Sup Kwak.) an orthogonal frequency-division multiplexing (OFDM) tech-
S. M. R. Islam, N. Avazov, and K.-S. Kwak are with the UWB nique in which subcarrier frequencies are chosen so that the
Wireless Communications Research Center, Inha University, Incheon 402-751,
South Korea (e-mail: [email protected]; [email protected]; subcarriers are orthogonal to each other. In contrast to OMA,
[email protected]). NOMA allows allocating one frequency channel to multiple
O. A. Dobre is with the Faculty of Engineering and Applied users at the same time within the same cell and offers a num-
Science, Memorial University, St. John’s, NL A1B 3X5, Canada (e-mail:
[email protected]). ber of advantages, including improved spectral efficiency (SE),
Digital Object Identifier 10.1109/COMST.2016.2621116 higher cell-edge throughput, relaxed channel feedback (only
1553-877X  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.
722 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 19, NO. 2, SECOND QUARTER 2017

the received signal strength, not exact channel state infor- The final section concludes this survey. Fig. 1 presents an
mation (CSI), is required), and low transmission latency (no overview of the organization and structure of this paper.
scheduling request from users to base station is required).
The available NOMA techniques can broadly be divided
into two categories, namely, power-domain and code-domain II. BASIC C ONCEPTS OF NOMA
NOMA. This paper focuses on the power-domain NOMA that There exist different NOMA solutions, which can primar-
superposes multiple users in power domain and exploits the ily be classified into two major approaches. Fig. 2 presents
channel gain difference between multiplexed users. At the a simple classification of the existing NOMA techniques.
transmitter side, signals from various users are superposed and Unlike power-domain NOMA, which attains multiplexing in
the resulting signal is then transmitted over the same chan- power domain, code-domain NOMA achieves multiplexing
nels (i.e., the same time-frequency resources). At the receiver in code domain. Like the basic code division multiple
sides, multiuser detection (MUD) algorithms, such as suc- access (CDMA) systems, code-domain NOMA shares the
cessive interference cancellation (SIC) are utilized to detect entire available resources (time/frequency). In contrast, code-
the desired signals. Although this paper primarily surveys domain NOMA utilizes user-specific spreading sequences
the power-domain superposition coding (SC)-based NOMA, that are either sparse sequences or non-orthogonal cross-
a brief discussion of other classes of NOMA is given in the correlation sequences of low correlation coefficient. This can
next section. be further divided into a few different classes, such as low-
OMA is a realistic choice for achieving good performance density spreading CDMA (LDS-CDMA) [4], [5], low-density
in terms of system-level throughput. However, due to the spreading-based OFDM (LDS-OFDM) [6], [7], and sparse
aforementioned upcoming wave, 5G networks require further code multiple access (SCMA) [8], [9]. The use of low-
enhancement in system efficacy. In this regard, researchers density spreading sequences helps LDS-CDMA to limit the
all over the globe have started investigating NOMA as impact of interference on each chip of basic CDMA sys-
a promising multiple access scheme for future radio access. tems. LDS-OFDM can be thought of as an amalgamation
NOMA achieves superior spectral efficiencies by SC at the of LDS-CDMA and OFDM, where the information sym-
transmitter with SIC at the receiver [2], [3]. On top of that, bols are first spread across low-density spreading sequences
the evolution of wireless networks to 5G poses new challenges and the resultant chips are then transmitted on a set of
for energy efficiency (EE), since the entire network will be subcarriers. SCMA is a recent code-domain NOMA tech-
ultra-dense. With an extreme increase in the number of infras- nique based on LDS-CDMA. In contrast to LDS-CDMA, the
tructure nodes, total energy consumption may simply surpass information bits can be directly mapped to different sparse
an acceptable level. Although substantial energy is basically codewords, because both bit mapping and bit spreading are
consumed by the hardware, NOMA has an inherent ability combined. When compared to LDS-CDMA, SCMA provides
to adapt the transmission strategy according to the traffic and a low complexity reception technique and offers improved
users’ CSIs. Thus, it can achieve a good operating point, where performances.
both spectrum efficiency and EE become optimum. There exist some other multiple access techniques, which
Over the past few years, NOMA has attracted a great deal are also closely-related to NOMA, including pattern division
of attention from researchers trying to meet 5G requirements. multiple access (PDMA) [10] and spatial division multiple
As a consequence, many research efforts in this field already access (SDMA) [11]–[14]. PDMA can be realized in vari-
exist. Research trends in NOMA include diverse topics, for ous domains. At the transmitter side, PDMA first maximizes
example, various performance analysis methods, fairness anal- the diversity and minimizes the overlaps among multiple users
ysis, EE, and user pairing. Many researchers are attempting in order to design non-orthogonal patterns. The multiplexing
to further enhance the performance of other existing wireless is then performed either in the code domain, spatial domain,
technologies, such as cooperative communications, multiple- or a combination of them. For SDMA, the working principle
input multiple-output (MIMO), light communications, and is inspired by basic CDMA systems. Instead of using user-
relay networks by using NOMA. However, NOMA in 5G is specific spreading sequences, SDMA distinguishes different
still in its infancy. At this stage, comprehensive knowledge users by using user-specific channel impulse responses (CIRs).
on the up-to-date research status of NOMA in 5G systems This technique is particularly useful for the cases where the
is extremely useful to researchers who want to do more in number of uplink users is considerably higher than the number
this area. In this regard, this paper examines the trends in of corresponding receiving antennas in BS. However, accu-
NOMA-based research and the various issues that need to rate CIR estimation becomes challenging for a large number
be addressed to transform radio access techniques through of users. The concept of software defined radio for multiple
NOMA innovation. access (SDR-MA) allows various forms of NOMA schemes
The rest of this paper is organized as follows. In to coexist [15]. This technique provides a flexible configu-
Section II, the basic concepts of NOMA are presented, ration of participating multiple access schemes in order to
whereas Section III discusses state-of-the-art NOMA research support heterogeneous services and applications in 5G. It
into solutions. Section IV evaluates the performance is worth noting that while the aforementioned list pro-
of various NOMA solutions. In Section V, several vides some insights into different forms of NOMA, it is
challenges to NOMA implementations are presented, not exhaustive, and the primary focus of this paper is on
and Section VI discusses other implementation issues. power-domain NOMA.
ISLAM et al.: POWER-DOMAIN NOMA IN 5G SYSTEMS 723

Fig. 1. The organization and structure of this paper.

and giving a speech to a group of people with different

backgrounds and aptitudes, such as a lecture in a class-
room. Suppose a professor is giving students information
through a lecture in the classroom. Since there are differ-
ences among the qualities and backgrounds of the students,
some of them receive most of the information, and others
receive only part of it. The lecture may be organized that it
proceeds at the pace of the student with the weakest back-
ground. However, in the ideal situation, the lecture can be
designed in such a way that the students with the most suit-
Fig. 2. A simple classification of NOMA techniques. able background obtain more information and the students
with a weaker background get at least the minimum amount
of information. This situation is an example of a broad-
In the following, a brief note about SC and SIC is presented, cast channel where a superposed lecture is being delivered
since these two basic techniques play important roles in by the speaker. To make SC practical, the transmitter must
understanding the class of NOMA on which this paper focuses. encode information relevant to each user. For example, for
Henceforth, this paper refers to power-domain NOMA simply a two-user scenario, the transmitter will have to contain two
by NOMA. point-to-point encoders that map their respective inputs to
complex-valued sequences of the two-user signal. In order
to show how SC is performed, a schematic diagram is given
A. Superposition Coding (SC) in Fig. 3, where the quadrature phase-shift keying (QPSK)
The SC that was first proposed in [16] is a technique constellation of user 1 with higher transmitting power is super-
of simultaneously communicating information to several posed on that of user 2 with lower transmitting power. It
receivers by a single source. In other words, it allows the can be mentioned that SC is a recognized non-orthogonal
transmitter to transmit multiple users’ information at the same scheme that attains the capacity on a scalar Gaussian
time. Examples of communications in a superposed fashion broadcast channel. Vanka et al. [17] introduced good strate-
include broadcasting a television signal to multiple receivers gies for SC, and proposed a design technique for SC by
724 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 19, NO. 2, SECOND QUARTER 2017

Fig. 3. An example of SC encoding (a) signal constellation of user 1 (b) signal constellation of user 2 (c) constellation of superposed signal.

using off-the-shelf single-user coding and decoding blocks.

In the superposition encoding phase, two point-to-point
encoders, f1 : {0, 1}2 1  → CT and f2 : {0, 1}2 2  → CT
TR TR

first map the respective input bits to two output bit sequences
S1 (n) and S2 (n), respectively, each of block length T. Here
R1 and R2 denote the transmission rates of user 1 and user 2,
respectively, and · represents the floor operator. C is the
code library. Then, a summation device provides an output
sequence as
 
X(n) = Pβ1 S1 (n) + Pβ2 S2 (n), (1)
where βi represents a fraction of the total power P assigned
Fig. 4. An example of SC decoding (a) decoding the signal of user 1
to user i, subject to the constraint on β1 + β2 = 1. (b) decoding the signal of user 2.

B. Successive Interference Cancellation (SIC)

To decode the superposed information at each receiver, a) Decode user 1’s message S1 (n) by using the single-user
2TR1
1 : C → {0, 1}
Cover first proposed the SIC technique [16]. SIC is conceiv- decoder g√ T .
able by exploiting specifications on the differences in signal b) Subtract Pβ1 h2 S1 (n) from the received signal Y2 (n)
strength among the signals of interest. The basic idea of SIC 
is that user signals are successively decoded. After one user’s Y2 (n) = Y2 (n) − Pβ1 h2 S1 (n), (2)
signal is decoded, it is subtracted from the combined signal where h2 is the complex channel gain at user 2.
before the next user’s signal is decoded. When SIC is applied, c) Decode user 2’s messages S2 (n) by applying another
one of the user signals is decoded, treating the other user TR2
single-user decoder g2 : CT → {0, 1}2 on Y2 (n).
signal as an interferer, but the latter is then decoded with
the benefit of the signal of the former having already been
removed. However, prior to SIC, users are ordered accord- C. A Typical NOMA Scheme
ing to their signal strengths, so that the receiver can decode This scheme considers a single-cell downlink scenario
the stronger signal first, subtract it from the combined sig- where there is a single base station (BS), and N users Ui ,
nal, and isolate the weaker one from the residue. Note that with i ∈ N = {1, 2, . . . , N}, and all terminals are equipped
each user is decoded treating the other interfering users as with a single antenna. Note that a similar uplink scenario
noise in signal reception. Fig. 4 presents the technique for can also be described, and a NOMA scheme can equally be
decoding the superposed signal (Fig. 3) at the receiving side. utilized there. The BS always sends data to all users simul-
Here, the constellation point of user 1 is decoded first from taneously, subject to the constraint of total power P. It is
the received signal. Then, the decoding of the constellation assumed that the wireless links experience independent and
point of user 2 is performed with respect to decoded constel- identically distributed (i.i.d.) block Rayleigh fading and addi-
lation point of user 1. To gain a deeper understanding of how tive white Gaussian noise (AWGN). The channels are sorted
SIC performs in wireless communications in general, and in as 0 < |h1 |2 ≤ |h2 |2 ≤ . . .≤ |hi |2 · · · ≤ |hN |2 , which indi-
OFDM (and MIMO systems) in particular, interested readers cates that user Ui always holds the ith weakest instantaneous
are referred to [18]. In brief, the particular process involved channel. The NOMA scheme allows simultaneous serving of
in decoding the superposed messages can be mathematically all users by using the entire system bandwidth (BW) to trans-
expressed as follows [17]: mit data with the help of SC at the BS and SIC decoding
TR
1) At user 1, a single-user decoder g1 : CT → {0, 1}2 1 techniques at the users. Here, user multiplexing is executed
decodes the message S1 (n) by treating S2 (n) as noise. in the power domain. The BS transmits a linear superpo-
2) User 2 performs the following steps to successively sition of N users’ data by allocating a fraction βi of the
recover its message from its received signal Y2 (n): total power to each Ui , i.e., the power allocated for the ith
ISLAM et al.: POWER-DOMAIN NOMA IN 5G SYSTEMS 725

Fig. 5. Multiple access schemes for a two-user scenario (a) NOMA (b) OMA.

user is Pi = βi P. On the receiving end, each user decodes the However, weak users are considerably less affected by the
signals of the weaker users, i.e., Ui can decode the signals for interferences from strong users, since relatively less trans-
each Um with m < i. The signals for weaker users are then mit powers are allocated to strong users and strong users are
subtracted from the received signal to decode the signal of user usually closer to the BS.
Ui , itself treating the signals of the stronger users, Um , with Fig. 5 presents the aforementioned NOMA scheme with
m > i, as interference, termed the “intra-cell interference” or two users. This figure also presents the OMA scheme to dis-
“intra-cluster interference”. The received signal at user Ui can close the particular advantages of a NOMA scheme over an
be represented as OMA one. With NOMA, the entire 1 Hz BW is simultaneously
used by two users. However, with OMA, user 1 uses α Hz
yi = hi x + wi . (3)
N  and the remaining (1−α) Hz is assigned to user 2. In NOMA,
Here, x = i=1 Pβ i Si is the superposed signal transmitted user 2 first performs SIC to decode the signal for user 1, since
by the BS, with Si being the signal for user Ui . Also, wi is the channel gain of user 2 is higher than that of user 1. The
the AWGN of user Ui with zero mean and variance σn2 . If decoded signal is then subtracted from the received signal of
signal superposition at the BS, and SIC at Ui , are carried out user 2. This resultant signal is eventually used to decode the
perfectly, the data rate achievable for user Ui for 1 Hz system signal for user 2. For user 1, SIC is not executed, and the
BW is given by1 signal is directly decoded. Thus, the achievable data rate for
  user 1 and user 2 are given by (5) and (6), respectively.
βi P|hi |2  
Ri = log2 1 +  . (4)
P|hi |2 N P1 |h1 |2
k=i+1 βk + σn
2
R1 = log2 1 + , (5)
Note that the data rate of user UN is RN = P2 |h1 |2 + σn2
 
log2 (1 + βN P|hN |2 /σ 2n ) , as this user successively decodes R2 = log2 1 + P2 |h2 |2 /σn2 . (6)
and cancels all other users’ signals prior to decoding its own
signal. Under OMA, the achievable data rate for user 1 and user 2 are
Also note that a strong user experiences a better chan- given by (7) and (8), respectively:
 
nel condition, but that does not mean the signal strength
R1 = αlog2 1 + P1 |h1 |2 /σn2 , (7)
is stronger. In fact, a lower transmit power is assigned to  
a strong user, and a weak user is assigned more power. P2 |h2 |2
Thus, the weak user’s signal is the strongest one at the R2 = (1−α)log2 1 + . (8)
σn2
user with the strongest channel. Therefore, NOMA does
not contradict the basic concept of SIC, that decoding It is clear from (5) and (6) that the NOMA scheme controls
of the strongest signal should be performed first. It is the throughput of each user by adjusting the power allocation
worth mentioning that strong users receive strong interfer- ratio, P1 /P2 . Thus, the overall throughput and user fairness are
ences (intra-cell interference) from weak users, since rel- closely related to the power allocation scheme. If an asymmet-
atively high transmit powers are allocated to weak users. ric channel, where the signal-to-noise ratios (SNRs) of the two
users are different, is considered, it can be numerically shown
1 The imperfect SIC causes error propagation in subsequent decoding of the
that the values of R1 and R2 , calculated from (5) and (6),
signals of NOMA users. Therefore, the effect of the said error propagation on
NOMA performances remains a concern. This issue will be briefly discussed respectively, are considerably much higher than those of R1
in Section VI. and R2 calculated from (7) and (8). This numerical comparison
726 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 19, NO. 2, SECOND QUARTER 2017

III. P OTENTIAL NOMA S OLUTIONS

This section presents an overview of the present and emerg-
ing NOMA research, in a categorized fashion, considered as
potential solutions to problems or issues associated with the
integration of NOMA in 5G. Detailed explanations and math-
ematical derivations of the techniques will be avoided, since
our major focus is to cover the core ideas of the state-of-the-art
NOMA research in 5G systems. Interested readers are referred
to the original articles for greater depth.

A. Impact of Path Loss

A substantial number of researchers have investigated the
performance of NOMA schemes to study the feasibility of
adopting this technique as a multiple access scheme for
5G systems. A survey by Higuchi and Benjebbour and the
references therein demonstrate that NOMA can be a promis-
Fig. 6. Capacity regions of NOMA and OMA for downlink. ing power-domain user multiplexing scheme for future radio
access [19]. In a cellular network with randomly deployed
users, the path loss performance of NOMA can be evalu-
is basically a special case of the multi-user channel capacity
ated under two situations. In the first scenario, each user
analysis [3]. Fig. 6 gives us the idea of a generalized capacity
has a targeted data rate determined by the assigned qual-
comparison of NOMA and OMA for two users. It shows that
ity of service (QoS). Here, the outage probability is an
the boundary of achievable rate pairs with NOMA is outside
ideal performance metric, since it measures the capability of
the OMA capacity region, in general. Therefore, NOMA is
NOMA to meet the users’ QoS requirements. In the other sce-
highly effective in terms of system-level throughput when the
nario, users’ rates are opportunistically allocated according to
channels are different for two users. On that, NOMA is con-
the channel conditions. In this situation, the achievable ergodic
sidered a promising multiple-access technique for future radio
sum rate can be investigated to evaluate NOMA performance.
access.
According to Ding et al. [20], if users’ data rates and assigned
NOMA can equally be applied in an uplink scenario. In the
power are chosen properly, NOMA can offer better outage
uplink, SIC is performed at the BS (see Fig. 7). For a two-user
performance than other OMA techniques. This study also
NOMA, the received signal at the BS is represented as
  showed that NOMA can achieve a superior ergodic sum rate.
y = h1 Pβ 1 S1 + h2 Pβ 2 S2 + w. (9) If the SNR is high, the outage probability of the ith user in
 a typical disk-shaped cell with radius RD can be given by:
Here, user Ui transmits the signal Si , with Pi = Pβ i being
τi i  ∗ i
i =
Pout η ψi ,
the transmission power. Also, w is the AWGN with zero mean (10)
and variance σw2 . With the help of SIC, the BS decodes the i

signals of U1 and U2 in two stages. In the first stage, the where τi = (i−1)!(N−i)!N!
, and η = R1D Ll=1 βl with

signal of U2 is decoded, treating the signal of U1 as noise. αp

βl = πl 1 − θl2 ( R2D θl + R2D )(1 + ( R2D θl + R2D ) ), and θl =
In the next stage, the receiver subtracts the decoded signal S2 ∗
from the received signal and then decodes the signal of U1 . 2L π ). Symbols L, αp , and ψi represent the complexity-
cos( 2n−1
2
accuracy trade-off parameter, the path-loss exponent, and the
Note that the signal of U2 is likely the strongest at the BS,
since it experiences the strongest channel. On that, this signal maximum SNR corresponding to the data rate of the ith user,
is decoded first and experiences interference from the user U1 . respectively. Also, with a sufficient number of users, N, and
It is worth noting that there are a couple of major differences adequate transmit SNR, ρ, NOMA can achieve the following
between uplink and downlink NOMA. First, in downlink, ergodic sum rate:

strong users successively decode and cancel the signals of Rerg = log2 ρlog2 {log2 N } . (11)
weak users, prior to decoding their own signals, in order to
achieve throughput gains. In uplink, however, the BS suc- According to the feedback framework of LTE, each user
cessively decodes and cancels the signals of strong users, measures the downlink channel by employing reference signals
prior to decoding the signals of weak users. Second, because and feeds back CSI in the form of predesigned transmission
of limited processing capabilities of mobile users and lack formats. This CSI feedback contains a rank indicator (RI),
of a centralized processing unit, a downlink is a case in a precoding matrix indicator (PMI), and a channel quality indi-
which the implementation of suitable multi-user detection and cator (CQI) [21]. Moreover, the rank is reported by each user
interference cancellation schemes is more challenging than it 2 Complexity-accuracy trade-off parameter: as the name suggests, the accu-
is in uplink. Third, as mentioned earlier in this section, strong racy increases at the expense of increased computational complexity (in
users are more affected by the intra-cluster interference in outage calculation). Therefore, a trade-off between complexity and accuracy
is required in order to calculate the outage probability. The calculations of
downlink. In uplink, however, weak users are more likely to outage probabilities for NOMA users might be frequently needed for precise
be affected by the interferences from strong users. resource allocation.
ISLAM et al.: POWER-DOMAIN NOMA IN 5G SYSTEMS 727

Fig. 7. NOMA in an uplink scenario.

at a certain time interval (i.e., semi-static). The BS can then user broadcasts the combination of the (N − 1) messages.
use this feedback information for various purposes, including One can note that the power allocation coefficients at each
power control. During the channel measurement, full trans- time slot are different based on local channel conditions. This
mission power of BS and no interference from intra-cell users demonstrates that C-NOMA can achieve the maximum diver-
are considered by each user. On that, the reported RI is suit- sity gain for all users. The cooperative NOMA scheme ensures
able for OMA. In case of NOMA, however, the transmission that the ith best user experiences a diversity of the order of
power is split between strong and weak users. Additionally, N conditioned on a specific power allocation ratio. However,
inter-user interference occurs in NOMA. Thus, both signal C-NOMA is expensive in terms of additional time slots, since
and interference powers experienced by NOMA users change, its cooperative phase requires message retransmissions from
which may result in a different rank from that of reported RI each user acting as a relay in a serial manner. To reduce
assuming full transmission power. Therefore, the rank feed- system complexity, C-NOMA performs user pairing based
back in NOMA will inherently place a limitation on the on distinctive channel gains. The performance of C-NOMA
achievable gain. A couple of rank optimization methods [21] can be further enhanced by adopting optimal power alloca-
could be deployed in order to overcome this limitation, thereby tion schemes [24], [25]. Direct derivation of the theoretical
enhancing both the outage probability and ergodic capacity achievable data rate in NOMA is quite difficult. However,
performance. Finally, coverage performances of NOMA in var- if the rate of conventional TDMA is compared with that of
ious wireless networks, including small cell networks, might non-cooperative NOMA, the performance difference is seen
be investigated for a deeper understanding of the impact of not as a function of power allocation coefficients but rather
path loss. as depending on how disparate two users’ channels are. And
a similar observation can be noted for C-NOMA.
NOMA for a multiple-antenna relay network has been
B. Cooperative NOMA (C-NOMA) studied in [26] and [27]. These studies analyzed the outage
In wireless networks, cooperative communications has behavior of the mobile users and derived closed-form expres-
gained a great deal of attention due to the ability to offer sions for the exact outage probability. If NOMA is combined
spatial diversity to mitigate fading, while resolving the difficul- with a multiple-antenna amplify-and-forward (AF) relay net-
ties of mounting multiple antennas on small communications work, where the base station and mobile users are equipped
terminals [22]. In cooperative communications, several relay with multiple antennas, the relay locations have a substantial
nodes are assigned to assist a source in forwarding informa- impact on the outage performance. When the relay loca-
tion to the respective destinations. Therefore, the integration of tion is close to the BS, NOMA outperforms conventional
cooperative communications with NOMA can further improve OMA. However, conventional OMA attains better outage
system efficiency in terms of capacity and reliability. The performance when the relay location is close to the users.
cooperative NOMA (C-NOMA) scheme [23] exploits prior In either case, NOMA offers better performance in terms of
information available in NOMA systems. In this scheme, SE and user fairness. Unlike C-NOMA systems, a coopera-
users with better channel conditions decode the messages tive relaying system (CRS) using NOMA [28] for spatially
for the others, and therefore, these users act as relays to multiplexed transmissions enhances SE. In this system, the
improve reception reliability for users with poor connections source transmits a superposed signal to the relay and the des-
to the base station. Cooperative communications for users tination during the first time slot. During the second time slot,
with better channel conditions than others can be imple- only the relay transmits the decoded symbol to the destination.
mented by using short-range communications techniques, such Note that the destination receives a single data symbol dur-
as ultra-wideband (UWB) and Bluetooth (BT). In particular, ing two time slots in conventional CRS (cooperative relaying
C-NOMA consists of two phases, namely, transmission phase without NOMA). With PSR and PRD being the average pow-
and cooperative phase. During the transmission phase, the BS ers for the channels of source-to-relay and relay-to-destination,
sends superposed messages (N users’ signal) to NOMA users. respectively, the average sum rate of CRS using NOMA can
At the end of this phase, successive detection will be carried be obtained by:
 
out by the users. The cooperative phase consists of (N − 1) 1 log2 e 1 1 β2
Ravg = log2 ρ − Ec − log2 + , (12)
time slots. At the ith time slot, 1 ≤ i ≤ (N−1), the (N−i+1)th 2 2 2 PSR PRD
728 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 19, NO. 2, SECOND QUARTER 2017

where Ec is the Euler constant. CRS using NOMA achieves by controlling the rate allocated to concurrent transmissions,
more SE than conventional CRS when the SNR is high which in turn can be achieved by using different scheduling
and the average channel power of the source-to-relay link is approaches [31].
superior to the source-to-destination and relay-to-destination
links. The concept of relaying can also be adopted to D. NOMA with MIMO and Beamforming
extend the cell coverage or overcome shadowing in case of
Beamforming (BF) is a signal processing technique used
NOMA transmission [29] where a BS can directly communi-
in various wireless systems for directional communications.
cate with a nearby user while communicating with a distant
Multi-user BF in MIMO systems is known as a capacity-
user only through a relay. In this system, the BS transmits
enhancing technology. In a multi-user BF system, each user is
a superposed signal to the relay and the nearby user during
supported by a single BF vector, orthogonal to the other users’
the first time slot. During the second time slot, while the relay
channels in order to eliminate interference from other users
transmits the decoded symbol for the distant user, the BS trans-
and thus maximize the achievable sum capacity. The integra-
mits the information for only the nearby user. Note that the
tion of NOMA with multi-user BF (NOMA-BF) thus has the
nearby user can remove the interference signal that originated
potential to capture the benefits of both NOMA and BF [25].
from the relay by using the side information (the signal of the
The NOMA-BF technique allows two users to share a sin-
distant user) obtained during the first time slot. The utilization
gle beamforming vector. To reduce the inter-beam interference
of NOMA in this sort of coordinated direct and relay transmis-
(from users of other beams) and intra-beam interference (from
sion provides considerable performance gain compared with
users sharing the same beamforming vector), NOMA-BF
NOMA in uncoordinated direct and relay transmission.
comes with a clustering and power allocation algorithm based
on correlation among users and channel gain difference,
C. Fairness in NOMA respectively. The NOMA-BF system improves the sum capac-
NOMA users experience unequal data rates, since this ity, compared to the conventional multi-user beamforming
access method is based on SIC decoding order [2], [3]. system. NOMA-BF also guarantees weak users’ capacity to
Although the decoding order is formulated based on users’ ensure user fairness. Suppose that two-user NOMA-BF con-
CSI, the NOMA service could be critical for some situations sists of N clusters, where each of them contains two users.
where strict fairness might be an issue. The power allocation In this scenario, a power allocation scheme for the nth cluster
problem from a fairness viewpoint can be investigated under maximizes the sum capacity while keeping the weak user’s
two assumptions: i) the BS has perfect CSI, and hence, users’ capacity at least equal to that of the conventional multi-
data rates adapt to the channel conditions; and ii) users have user beamforming system. It can be formulated as follows,
fixed targeted data rates under an average CSI. With these conditioned on (13a) and (13b):
assumptions, it is possible to come up with low-complexity
algorithms that yield globally optimal solutions. NOMA with β1n = arg max (R1 + R2 ), (15)
β1
n
a fairness constraint also outperforms OMA approaches by 1
significantly improving the performance of the users with the s.t. R2 ≥ R2,conv−BF , (15a)
2
worst channel conditions [30]. If instantaneous CSI is avail-
able at the BS, fairness among users can be ensured by where R1 and R2 are the capacities of the strong and weak
maximizing the minimum achievable data rate, i.e., users, respectively, whereas R2,conv−BF is the capacity of the
weak user if the weak user is supported by conventional beam-
max min Ri (β), (13) forming. Symbols β1n and β2n = 1 − β1n are the power fractions
β iN

N of the strong and weak users, respectively, in the nth cluster.
s.t. βj = 1, (13a) The optimal solution to (15) can be obtained by using the
j=1 Karush-Kuhn-Tucker (KKT) condition below:
0 ≤ βj , for jN. (13b) 1
β1n = 
 2 
Since the problem in (13) is not convex, it needs to first be 1 + h2,n  ρ
converted into a sequence of linear programming. Eventually,  
the optimal solution to (13) can be given by:  2  N  2 
 
1 + h2,n ρ − 1  
⎛ ⎞ i=1,i=n h2,i wi ρ + 1
2 − 1⎝
t N
−  .
βi = P|hi |2 βk + σn2 ⎠, i = N, N − 1, . . . , 1,  2   2 
P|hi |2 
ρ h2,i wi  
1 + h2,n ρ
k=i+1
(14) (16)
where t represents the minimum data rate. If the instantaneous The above clustering algorithm, based on channel corre-
CSI is not available, the outage probability should be opti- lation and a zero-forcing (ZF) precoding matrix, helps two
mized with average CSI. In this case, fairness among users users in the same cluster to achieve similar benefits from
can be ensured by minimizing the maximum outage probabil- the ZF precoding matrix. However, the performance of the
i (β) conditioned on (13a) and (13b).
ity as minβ maxiN Pout system becomes considerably degraded due to the increased
The fairness of a NOMA system can also be determined multi-user interference (MUI) when there is a shortage of user
ISLAM et al.: POWER-DOMAIN NOMA IN 5G SYSTEMS 729

sets with high channel correlation. In this regard, researchers

investigated the design of the precoding matrix for MU-MIMO
NOMA to achieve a higher sum-rate in the presence of said
interference [32].
As discussed above, random opportunistic
beamforming [25] for the MIMO NOMA system works
under the assumption of perfect CSI at the transmitter. It
becomes evident that with a relatively large number of users, Fig. 8. NOMA in a two-cell scenario (network NOMA).
the combination of NOMA and MIMO can achieve sufficient
throughput gain [33]. If perfect CSI at the transmitter is
unavailable due to limited feedback, statistical CSI can be rates to cell-edge users. The associated BSs for CoMP need
utilized for long-term power allocation to maximize the to allocate the same channel to a cell-edge user. As a result,
ergodic capacity of MIMO NOMA systems. The system- the SE of the system worsens as the number of cell-edge
atic implementation of MIMO NOMA has been reported users increases. In order to avoid this problem, a coordinated
in [21], [24], and [34] and demonstrates that the use of SC (CSC)-based NOMA scheme [38] was formulated by con-
MIMO NOMA can outperform conventional MIMO OMA. sidering SC for downlink transmissions to a group of cell-edge
Both optimal and low-complexity suboptimal power allocation users and users near the BS simultaneously with a common
schemes [35] can maximize ergodic capacity with a total access channel [17]. In other words, BSs transmit Alamouti
transmit power constraint. Ding et al. [36] proposed a new (space-time) [39] coded signals to user c (a cell-edge user),
design for precoding and detection matrices in a general while each BS also transmits signals to a user near the BS. The
NOMA downlink scenario with a fixed power allocation CSC-NOMA scheme with Alamouti code provides a cell-edge
scheme for all the participating users. The analytical and user with a reasonable transmission rate without diminishing
numerical results confirmed that MIMO NOMA can achieve the rates to nearby users, and increases SE. If Rc1 , Rc2 , and
better outage performance than conventional MIMO OMA, Rc are the rates to a user (U1 ) near BS 1, a user (U2 ) near
even for users who experience strong co-channel interference. BS 2, and a coordinated user (Uc ), respectively, the sum rate
Suppose a BS equipped with M antennas communicates with becomes Rc1 + Rc2 + Rc with
several users equipped with N antennas each. Also, the users ⎡ ⎛   ⎞⎤
h1,1 2 P1
are randomly grouped into M clusters with K users in each Rc1 = E⎣log2 ⎝1 +  2  ⎠⎦, (18)
cluster. Then, the outage probability, at a high SNR, for the E h1,2  P2 + σn2
kth ordered user in the ith cluster can be given by ⎡ ⎛   ⎞⎤
  h2,2 2 P2
N−M+1 K−k+1 Rc2 = E⎣log2 ⎝1 +  2  ⎠⎦,
K! ((N−M+1)!
∗) (19)
E h2,1  P1 + σn2
Pout ≈ , (17) !
(K − k)!(k − 1)(K − k + 1) Rc = min Z1, Z2 , Zc , (20)
where  ∗ represents the maximum required SINR to meet the where Z1 = E[log2 (1 + SINR1 )], Z2 = E[log2 (1 + SINR2 )],
QoS of users in the ith cluster. As can be seen from (17), and Zc = E[log2 (1 + SINRc )]. Let SINRi be the SINR for
MIMO NOMA can experience a diversity gain of (N − M + user Ui in decoding the signal of user Uc . Note that hi,j
1)(K−k+1). It is also intuitive that the extension of NOMA in denotes the channel coefficient from the BS j to user Ui .
massive MIMO systems can further enhance the SE. Unlike Because of the use of Alamouti code for CoMP communica-
the single-antenna system, where the strong user is responsible tions, the exchange of instantaneous CSI is not required. This
for successful SIC decoding, precoders in MIMO systems usu- is a significant advantage over coherent transmission schemes
ally affect the signal-to-interference-plus-noise ratio (SINR) of that require instantaneous CSI exchange, which results in an
both users. Thus, the weak user data rate will be limited by excessive backhaul overhead for high mobility cell-edge users.
not only the decoding performance of the weak user but also
by the decoding capability of the strong user. In this respect, F. Network NOMA
the same precoder for both users in each cluster may not be
Suppose there is a simple two-cell scenario of a cellular
sufficient. The above work on MIMO NOMA deals with the
system (see Fig. 8), where U3 and U4 are served by BS 1,
same precoder in each cluster. However, it is possible to fur-
while U1 and U2 are served by BS 2. Also, assume that
ther enhance the sum rate if the two users in a cluster can use
a two-user NOMA scheme is adopted, so that U3 is paired
different precoders [37].
with U4 , and U1 is paired with U2 . As a cell-edge user in
NOMA does not perform SIC before decoding its signal,
E. NOMA in a Coordinated System U4 and U1 are inherently unable to avoid the interference
In cellular systems, a cell-edge user usually experiences from U3 and U2 , respectively. In order to reduce the impact
a lower data rate compared to that experienced by a user near of this interference, BS 1 and BS 2 allocate more power
the BS. Coordinated multipoint (CoMP) transmission (and to U4 and U1 , respectively. As such, inter-cell interference
reception) techniques, where multiple BSs support cell-edge occurs in downlink transmissions and mutual interference
users together, are usually employed to increase transmission between U4 and U1 occurs in uplink transmissions [40]. To
730 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 19, NO. 2, SECOND QUARTER 2017

deal with this problem, associated with the employment of which users are grouped together. The impact of user pair-
NOMA in a multi-cell scenario, the straightforward application ing/grouping was investigated mathematically in [43]. This
of single-cell NOMA solutions is not appropriate; single-cell work demonstrates that the performance gain of NOMA with
NOMA needs to be extended to network NOMA. Like vari- fixed power allocation over conventional OMA can be fur-
ous inter-cell interference mitigation techniques employed in ther increased by selecting users whose channel conditions
OMA, some solutions are also required in NOMA to reduce are more distinctive. A two-step method for user pairing
the impact of interferences in the context of a multi-cell sce- based on proportional fairness (PF) was introduced in [44].
nario. One possible solution to mitigate the aforementioned In the first step, the power allocation for each candidate
interferences in network NOMA is to utilize joint precod- user set is optimized to find the highest scheduling metric.
ing of users’ signals across the neighboring cells. However, In a subsequent step, the optimal user pair, or single user
the design of an optimal precoder is difficult, since each with the maximum metric, is scheduled. Since this user-pair
BS should know all users’ data and CSI. The correlation- power allocation (UPPA) technique avoids unnecessary com-
based precoder design needs dynamic user selection for each parison of candidate user pairs by formulating the prerequisites
NOMA pair [25]. Moreover, the multi-user precoding appli- for user pairing, this method comes with considerably lower
cable to single-cell NOMA may not be realistic in network computation complexity compared to the tree-search–based
NOMA, since a beam generated via geographically separated transmission power allocation (TTPA). Suppose, the upper and
BSs does not support more than one spatially separated user lower bounds of the allocated power coefficient of U1 are β1u
for intra-beam NOMA. A low-complexity precoding scheme and β1l , respectively. Now, if any of the following conditions,
for network NOMA was introduced in [40] based on the fact termed the prerequisites, are not satisfied by a candidate user
that large-scale fading would be very disparate between the pair, the computation and comparison of their PF scheduling
links of different cells. Here, the joint precoder is applied only factor can be omitted. Thus, the compared number of user
to cell-edge users (e.g., U4 and U1 in Fig. 8) and the resulting pairs can be substantially reduced.
SINR of each user Ui of power Pi is found as:
β1u > 0 ⇐⇒ I2 > ρ0 > 0, (25)
 −1
 −1  β1 ≥ ρ0 I1−1  β1l , (26)
H41 (H41 )H P1
1,1
SINR1 = , (21) β1l < 1 ⇐⇒ I1 > ρ0 > 0, (27)
|h11 |2 P0 + |h12 |2 P2 + N0 B
|h22 |2 P2 where Ii is the channel quality indicator (CQI) of user Ui ,
SINR2 =   2  2  , (22) and ρ0 is the minimum required SINR threshold, while the
|h21 |2 P0 + w0,0  P1 + w0,1  P3 + N0 B modulation and coding scheme (MCS) with the lowest coding
|h30 |2 P3 rate is selected for an expected block error rate. Eventually, the
SINR3 =   2  2  , (23) solution to the optimal power ratio provided by the two-user
|h32 |2 P2 + w1,0  P1 + w1,1  P3 + N0 B proportional fairness-based NOMA can be given by:
 −1 ⎧ ∗
 −1  ∗
⎨ β1 β1 ≤ β1 ≤ β1
l u
H41 (H41 )H P4 ⎪
0,0 "1 = β u R1 < R2
β , (28)
SINR4 = , (24) ⎪ 1
|h41 |2 P0 + |h42 |2 P2 + N0 B ⎩ l
β1 others
where H41 = [h4 , h1 ]T with the channel vector of the ith user, where β1∗ is the solution to the modified scheduling factor [45].
hi = [hi1 , hi2 ], and hij (i ∈ {1, 2, 3, 4}, and j ∈ {1, 2}) being As (28) indicates, the available scheduling factor of every
the channel response between the jth BS and ith user. The zero- valid pair of users can be computed, and the largest one
forcing precoder, W, is the normalized pseudo-inverse of H41 , can be selected. A similar user set–selection algorithm is
−1
i.e., W = (H41 )H (H41 (H41 )H ) , B is the system bandwidth, formulated based on the mathematical characteristics of the
and N0 is the noise spectral density. In [41], two coordinated PF metric [46]. This method can also reduce the computa-
beamforming approaches have been proposed to deal with the tional complexity by judging whether a user set is worth
inter-cell interference in two-cell MIMO-NOMA networks. multiplexing based on a simple condition between the weakest
Then, in [42], this approach is extended to an arbitrary number two users within the set. Also, the impact of user pairing on
of cells; in addition, the maximum number of users sup- the performance gain of NOMA with a MIMO system was
ported by the proposed scheme in multi-cell MIMO networks investigated [36], demonstrating that the performance gain of
is characterized. MIMO NOMA over MIMO OMA increases as the number of
users in each group increases.
G. NOMA User Pairing
Since NOMA is an interference-limited system, it is prac- H. Energy-Efficient NOMA
tically unwise to ask all users in the system to perform NOMA employs some controllable interference via non-
NOMA jointly. In this regard, users can be divided into orthogonal resource allocation and realizes overloading at the
multiple groups, where NOMA is implemented within each cost of slightly increased receiver complexity. Consequently,
group, and different groups are allocated with OMA. Clearly, higher SE can be achieved by NOMA for 5G. Although SE
the performance of this hybrid MA scheme depends on shows how efficiently a limited spectrum resource is utilized,
ISLAM et al.: POWER-DOMAIN NOMA IN 5G SYSTEMS 731

it fails to provide any insight on how efficiently energy is J. NOMA with Link Adaptation
utilized. With the rise in desire for green communications Hybrid automatic repeat request (HARQ) protocol, an indis-
in recent years, reducing energy consumption has become of pensable part of link adaptation, is designed for reliable
prime importance for researchers, and 5G has also targeted EE communications by using retransmission diversity and chan-
as one of the major parameters to be achieved. Nonetheless, nel coding gain. Suppose, when multiple NOMA packets are
Shannon’s information capacity theorem illustrates that the transmitted, there is a collision, and a retransmission request
two objectives of minimizing consumed energy and maxi- is sent to the users in collision. Each packet can convey at
mizing SE are not achievable simultaneously, and calls for least a certain amount of information, even though the pack-
a trade-off. It can be noted that with circuit power under ets collided. All the retransmitted signals can be combined
consideration, there always exists an optimal point in the EE by using HARQ in order to improve the SE of NOMA [51].
versus SE (EE-SE) curve. An energy-efficient two-user single- Li et al. [52] provided a thorough investigation of HARQ
cell NOMA was studied in [40]. Under fixed total power design for NOMA with single user-MIMO (SU-MIMO). They
consumption, the EE-SE relationship was found to be linear found that the HARQ technique for a NOMA system is more
with a positive slope. Appropriate power allocation between challenging than for an OFDMA system. A good HARQ algo-
two users allows achieving any point in the EE-SE curve. For rithm should deal with two central problems in NOMA retrans-
a given SE for each user, maximum EE performance can be mission: when the HARQ combination should be directed,
achieved. The degree of efficiency can be adjusted by varying and how to conduct power assignment for retransmission.
the total power using power-control schemes. If the sum rate Based on channel responses to initial transmission and retrans-
capacity of the cell is Rsum with total power consumption Pcell , mission, it was found that the optimum time for conducting
EE can be written as ηE = Rsum /Pcell = BηS /Pcell , where ηS is an opportunistic HARQ combination by the distant user is
the spectrum efficiency. The EE optimization can be achieved when the SINR of the initial transmission is reasonably larger
in both single-input single-output and MIMO systems [47]. than the retransmission SINR. And one approach for conduct-
ing retransmission power assignment would be assigning the
I. NOMA in Visible Light Communications power in such way so that it maximizes the geometric mean
One of the major downsides to visible light communica- user throughput of all the users in retransmission. In sum, an
tions (VLC) systems is the narrow modulation BW of the advanced HARQ design can bring NOMA gain improvement
light sources, which results in a barrier to attaining compe- both in cell throughput and cell-edge throughput.
tent data rates. Like wireless communications, optical wire-
less communications also considers various signal-processing
K. Other NOMA Solutions
techniques, and multicarrier and multi-antenna systems for
achieving higher data rates in VLC systems. Since NOMA is 1) NOMA with Raptor Codes: For a given integer, k,
now a potential candidate for next-generation wireless com- and a real number , Raptor codes, first proposed by
munications, the feasibility of NOMA in VLC is also be Shokrollahi [53], encode a message of k symbols into a poten-
a subject of interest. It is viable to apply a NOMA scheme tially limitless sequence of encoding symbols such that any
to enhance the achievable throughput in high-rate VLC. In subset of k (1 + ) encoding symbols allows the message to
fact, studies reveal that NOMA is a promising MA scheme be recovered with high probability. Raptor codes were recently
for downlink in VLC networks [48], [49]. However, these found effective in several cooperative communications scenar-
works simply reflect a scenario with multiple light emitting ios. The integration of Raptor codes with NOMA has been
diodes (LEDs) in an indoor environment; real high-speed studied in [54], where an interfering channel with Raptor code
OFDM-based communications has not been considered. Other was added to an existing main non-orthogonal wireless chan-
research investigated optical OFDM transmission and com- nel. It was demonstrated that the coded interference does
pared the achievable capacities of NOMA and OFDMA [50]. not affect the performance of the main channel, while the
It also analyzed the impact of cancellation error in SIC interfering signal itself can successfully be decoded with high
receivers for VLC systems. The achievable data rates for probability.
U1 and U2 in two-user NOMA with DC-biased optical OFDM 2) NOMA with Network Coding: Random linear network
(DCO-OFDM), assuming a frequency-flat fading channel, can coding (RLNC) is a good encoding scheme that allows
be written as data retransmission. In RLNC, the source does not need
  to be aware of the packets lost by the intended receiver.
B γo2 Lo2 Pβ1,n G21
R1 = log2 1 +  , (29) To date, various RLNC techniques have been proposed to
2 o γ 2o Lo2 P2 1 − β1,n G21 + N0 B improve transmission efficiency in both multicast and broad-
 
B γo2 Lo2 Pβ2,n G22 cast services. The performance of multicast services in
R2 = log2 1 + , (30) downlink networks can be further enhanced by integrating
2 γo2 Lo2 P2 β2,n G22 + N0 B RLNC with NOMA. NOMA with RLNC was studied by
where the optical parameters (γo , Lo , o ) are the receiver Park and Cho in [55]. In conventional NOMA, the power
responsivity, number of LED chips, and the signal cancella- domain multiplexing of multiple receivers is considered
tion error, symbols G1 and G2 are channel gain for U1 and U2 , for unicast services, whereas NOMA-RLNC utilizes power-
respectively, and the symbol βi,n denotes the power coefficient domain multiplexing for multiple reception groups of receivers
for the nth subcarrier for Ui . for multicast services. It was found that NOMA-RLNC
732 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 19, NO. 2, SECOND QUARTER 2017

improves the packet success probability of providing multicast

services where a source superposes multiple coded packets
before transmitting.
3) Coexistence of NOMA and OMA: In terms of capacity
enhancement, which is a major goal of 5G, NOMA is a poten-
tial candidate for future radio access. That said, it does not
mean OMA schemes will be entirely replaced by NOMA. For
example, OMA might be preferred over NOMA for small cells
if the number of users is small and the near–far effect is not
important. It can be concluded that both OMA and NOMA will
coexist to fulfill varied requirements of different services and
applications in future 5G. As a matter of fact, the long-term
coexistence of different radio access technologies is, in gen-
eral, an important feature of 5G networks. Dai et al. [15]
discussed NOMA schemes for 5G and analyzed their basic
principles, key features, and receiver complexity. They con-
clude that the concept of software-defined multiple access Fig. 9. Outage performance of NOMA in 5G with random users in a cell.
can offer various services and applications with different
requirements.
4) SIC Receiver Variations: An ideal SIC receiver per- IV. P ERFORMANCE E VALUATIONS
fectly cancels the interference from the cell-edge user at
To illustrate the performance gains, this section provides
the receiver of a cell-center user. Until now, the seminal
numerical results for various NOMA setups. Two primary
works on NOMA adopt a SIC receiver at cell-center users,
performance metrics, namely, outage probability and achiev-
while cell-edge users do not take SIC detection into account.
able rate are considered for this purpose. Also, the section
However, it is possible to further improve the performance
provides some insights into SE and EE performances of
of NOMA by considering an advanced receiver at cell-edge
NOMA.
users [56]. By advanced receiver, this paper means a receiver
that can perform ideal SIC processing, since the improvement
in NOMA performance gain with non-ideal SIC is not satis- A. Outage Probability
factory. SIC where the cell-edge user’s signal is demodulated Based on (10), Fig. 9 compares the outage performance
and a hard decision is made without channel decoding is called of a NOMA scheme with that of an OMA scheme for
symbol-level SIC (SLSIC). In codeword-level SIC (CWSIC), a cellular network comprising randomly deployed users with
the signal of a cell-edge user is demodulated and decoded. N = 2, L = 10, αp = 2, and RD = 3 m. The users
The probability of successful cell-edge user signal recovery are uniformly located. The targeted data rates of 0.1 bits per
increases, compared with SLSIC, since channel decoding is channel use (BPCU) and 0.5 BPCU, respectively, for a weak
involved in signal detection. Consequently, the impact of error user and a strong user (user 2) were used. Since the con-
propagation can be reduced in CWSIC [57], [58]. However, ventional orthogonal scheme is considered for benchmarking,
it increases computational complexity. If the Gray map- its targeted rate is 0.6 BPCU (the addition of two users’
ping of superposed signals is achieved by jointly modulating data rates). Also, note that the numerical results are based
multi-users’ signals at the transmitter, the desired signal can on the normalized SNR model. As can be seen from this
be decoded with the application of a log-likelihood method at figure, NOMA outperforms the comparable scheme, and the
the receiver. Since the decoding can be performed without SIC diversity order of the users is a function of their channel condi-
processing, the complexity of CWSIC can be reduced [57]. tions. Note that in this case, the ratio of the power assigned to
Appropriate power control mechanism for NOMA employing a strong user to the power assigned to a weak user is 1:4. The
minimum mean squared error-based linear filtering followed outage probability given by (7) is basically valid at a high
by SIC (MMSE-SIC) can mitigate inter-cell interference and SNR condition. On that, in order to recognize the compar-
the fairness can be ensured by employing a PF-based multiuser ative outage performance, attention should be paid to high
scheduling [59]. SNR regions where both users outperform the OMA scheme.
5) NOMA with SWIPT: In addition to enhancing SE, Since the assigned power to the strong user is proportionally
another major goal of 5G networks is to maximize EE. Many lower, the outage performance at a low SNR region is poor.
researchers demonstrated that simultaneous wireless infor- However, as the SNR becomes high enough, the power-domain
mation and power transfer (SWIPT) is expected to provide multiplexing becomes dominant, and thereby shows the best
a viable solution to overcome EE issues. It is possible to performance with superior diversity order.
establish a cooperative communications protocol by combining With the same number of users and a power allocation
SWIPT and NOMA, where cell-center users act as energy- ratio used in Fig. 9, Fig. 10 presents the outage probability
harvesting relays to help cell-edge users. The use of SWIPT that is achieved by non-cooperative NOMA and cooperative
does not affect the diversity orders of both nearby and distant NOMA (discussed in Section III-B) as a function of SNR. It
users, compared to conventional NOMA [60]. shows that cooperative NOMA transcends the comparable
ISLAM et al.: POWER-DOMAIN NOMA IN 5G SYSTEMS 733

Fig. 10. Outage performance of a cooperative NOMA. Fig. 12. Sum rate performance of NOMA in 5G with random users in a cell.

Fig. 11. Outage performance of a MIMO NOMA system.

Fig. 13. Achievable rate performance of NOMA with relay.

scheme, since it ensures that the maximum diversity gain is

achievable by all users. This high diversity gain is explained
below. Under C-NOMA, users with the worst channel con- B. Achievable Rate
dition get assistance from the other N − 1 users, along with With the same system parameters used in Fig. 9,
their own direct links to the source. Although non-cooperative Fig. 12 presents the sum rate (the sum of the data rates of the
NOMA can attain only a diversity order of i for the ith ordered NOMA users) achieved by a two user NOMA as a function of
user, C-NOMA ensures that a diversity order of N is achievable SNR. The targeted data rates of the randomly deployed users in
for all users by exploiting user cooperation. the cell are determined opportunistically according to the chan-
Fig. 11 demonstrates the performance of a typical MIMO nel conditions. In this case, the data rate of user 2 has always
NOMA system (discussed in Section III-D) in terms of out- been considered higher than that of user 1, since the channel
age probability. This figure considers two clusters, where each gain of user 2 is always higher than that of user 1. With this
cluster contains two users equipped with three antennas. The consideration, the figure demonstrates that NOMA can attain
power allocation ratio is 1:4, and target data rates of the strong a higher sum rate than the OMA scheme across the entire
user and the weak user are 3 BPCU and 1.3 BPCU, respec- SNR range.
tively. The figure indicates that the MIMO NOMA system In Fig. 13, the rate performance of a cooperative relaying
outperforms MIMO with OMA, particularly at high SNR val- system using NOMA (discussed in Section III-B) with fixed
ues. By comparing the slopes of the performance curves, it power allocation is presented. The figure demonstrates that
can be concluded that the diversity gain of MIMO NOMA is CRS with NOMA achieves better average rate performance
the same as MIMO OMA. However, MIMO NOMA provides than the usual relaying system, particularly at a high SNR,
better SE, since this gain is achieved by allowing both users when the average power of the channel coefficient of source
from the same cluster to utilize the same BW. to relay is much higher than that of relay to destination.
734 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 19, NO. 2, SECOND QUARTER 2017

Fig. 14. Sum capacity of NOMA beamforming. Fig. 16. Performance gain in PF-based user pairing.

Fig. 15. Sum capacity of CSC-based NOMA. Fig. 17. Capacity regions of light communications with NOMA.

Therefore, the relays should be nominated from the cell-center loss exponent of 3. Note that non-CSC–based NOMA consid-
users in a NOMA-based relay network in order to get the ers only one BS (either one) to employ SC to serve a pair of
benefits of NOMA. cell-edge and nearby users simultaneously. It is observed that
Fig. 14 shows the sum capacities of NOMA-BF (discussed the sum rate of a CSC-based system exponentially increases
in Section III-D) and conventional multi-user beamform- with SNR, and is higher than a non-CSC–based system.
ing with a correlation threshold ρ = 0.75, system BW Fig. 16 compares the performance of user pairing with
of 4.32 MHz, maximum transmission power per cluster at proportional fairness–based power allocation (discussed in
43 dBm, and noise density of -169 dBm/Hz. As can be seen, Section III-G) to that of the TTPA scheme [61] in terms of per-
NOMA-BF improves the sum capacity. Here, the users are centage gain with respect to OFDMA. A better performance
randomly located with uniform distribution in a cell radius of gain is obtained across the entire range of numbers of users per
500 m. NOMA-BF is better in terms of sum capacity compared cell when user pairing is formed through PF-based schedul-
to conventional multi-user beamforming, since correlation- ing. This is because PF-based user pairing effectively preserves
based clustering with effective power allocation reduces the user diversity by its power allocation metric. Also, removing
inter-beam and intra-beam interference. In view of the fact invalid user pairs during prerequisite checking perhaps partly
that two users share a single beamforming vector, the num- contributes to this performance enhancement, since it ensures
ber of supportable users can easily be increased by utilizing distinct channel variations among candidate users, to some
NOMA-BF. extent.
Fig. 15 compares the sum rate performance of CSC-based Fig. 17 shows the achievable capacity regions of NOMA and
NOMA (discussed in Section III-E) with that of non-CSC- OFDMA-based VLC systems (discussed in Section III-I) with
based NOMA under symmetric channel conditions with a path the following numerical parameters: number of LED chips
ISLAM et al.: POWER-DOMAIN NOMA IN 5G SYSTEMS 735

Fig. 18. Spectral efficiency of network NOMA. Fig. 19. A trade-off between energy efficiency and spectral efficiency.

Lo = 100 with responsivity γo = 0.25 A/W, and system differences in the performance of these two methods become
BW B = 20 MHz. In NOMA, the capacity region has been even more significant in high SNR regions.
obtained by varying the power coefficients. In OFDMA, the
rate pair was found by varying the BW assigned to each V. NOMA C HALLENGES
user. As shown, NOMA-based light communications demon- Many researchers have worked on designing and imple-
strates higher rate pairs than OFDMA-based VLC. Therefore, menting NOMA techniques and on solving various technolog-
it can be concluded that power-domain multiplexing is also ical problems associated with those methods. The literature
promising in light communications. demonstrates that NOMA is compatible with cooperative
communications, relaying, and MIMO, and it significantly
C. Spectral Efficiency (SE) and Energy Efficiency (EE) enhances the performance gains. This paper has shown various
useful NOMA solutions to address the problems associated
Suppose the power gains of a 4 × 2 downlink channel
with the multi-cell network, wireless link adaptation, EE-
in the network NOMA scenario described in Fig. 8 are
SE trade-off, and user pairing. Researchers have introduced
[−77, −117; −97, −107; −77, −117; −97, −107] dB.
a technique where the low data rate and bad quality of ser-
The SE of this network NOMA scheme is presented in
vices experienced by cell-edge users can easily be improved
Fig. 18. The baseline to compare the performance of network
by adopting NOMA-based coordination. In addition to the
NOMA in terms of total SE is the performance of the entire
research concerns in the literature, there are several other chal-
network based on single-cell NOMA, where base stations do
lenges and open issues that need to be carefully addressed.
not cooperate with each other. This figure shows that net-
This section will try to provide some research directions for
work NOMA outperforms conventional NOMA. The reason
researchers interested in investigating NOMA on a larger scale.
is that severe inter-cell interference (mainly between cell-edge
users) causes performance degradation in conventional single-
cell NOMA, since there is no inter–base-station cooperation. A. Dynamic User Pairing
However, the precoder for network NOMA works in such As mentioned earlier in this paper, co-channel interference
a way that base stations transmit jointly to cell-edge users, is strong in NOMA systems, since multiple users share the
and the cell-center users first detect and subtract the signals same time, frequency, and spreading code. As a result, it
of cell-edge users to mitigate the mutual interference. is difficult to ask all the users in the system to perform
In Fig. 19, the EE-SE trade-off (discussed in Section III-H) NOMA jointly. Alternatively, the users in the system are
curve of a single-user MIMO NOMA scheme is compared divided into several groups, where NOMA is applied within
with that of single-user MIMO TDMA as an OMA approach to each group, and different groups are allocated with orthogonal
explaining what happens to EE performance when we change BW resources. The static case is usually considered, where the
spectrum efficiency, and vice-versa. We considered statistical mth user and the nth user (|m−n| is constant and considerably
CSI in both cases. The figure demonstrates that NOMA out- high) are paired to implement NOMA. Although it is diffi-
performs OMA in terms of better EE-SE trade-off. Both SE cult in practice, the design of dynamic user pairing/grouping
and EE corresponding to the peak operating point offered by schemes is necessary to achieve the maximum benefits offered
the NOMA scheme are higher values than those achieved by by NOMA. In this regard, the analytical insights obtained
the OMA scheme. The reason is that EE optimization maxi- by Ding et al. [43] for the case with two selected users
mizes EE under the constraints of total transmit power and the can be considered as criteria for the design of distributed
minimum rate of the weak user with the help of NOMA. The methods. It is known that the UPPA scheme [44] improves
736 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 19, NO. 2, SECOND QUARTER 2017

performance gains. However, only two users are permitted to So, resource management has to assist with effective utiliza-
be scheduled by NOMA. An encouraging extension of the tion. Wireless resource management is a series of processes
UPPA scheme would be a power allocation scheme in the required to determine the timing and amount of related
multi-user NOMA system where the scheduler allows more resources to be allocated to each user [64]. It also depends on
than two users. Also, it requires more research in order to the type of resources. According to the Shannon’s informa-
apply UPPA schemes and variations [46] to a practical system tion capacity theorem, BW is one of these wireless resources.
with a suitable modulation and coding scheme. As a part of the effective management of the system BW in
a communications system, the total BW is first divided into
B. Impact of Transmission Distortion several chunks. Each chunk is then assigned to a particular
The transmission of source information, such as voice user or a group of users, as in the case of NOMA. Also, the
and video over communications channels is generally consid- number of packets for each user varies over time. Therefore,
ered lossy. The transmitted data always experience distortion user-pairing and optimum power allocation among users in
while they propagate to the receiver. To deal with this lossy NOMA requires a sophisticated algorithm to provide the best
transmission, considerable theoretical attention in assessing performance with the usages of minimum resources. Resource
source fidelity over fading channels has been paid to date. allocation in NOMA can also be explored from a mathematical
Different source coding and channel coding have been framed optimization theory point of view. For example, Lei et al. [65]
to minimize the end-to-end distortion. However, source cod- considered joint power and channel allocation for NOMA in
ing diversity and channel coding diversity provide conflicting 5G networks. They take user power control and SIC implemen-
situations among the preferred amounts of distortion, cost, and tation into account to solve the power and channel allocation
complexity. Choudhury and Gibson [62] compared the source problem. Power allocation in NOMA-based cognitive radio
distortion for two definitions of channel capacity, namely, networks [66] is also an unexplored area of research.
ergodic capacity and outage capacity. Both information capac-
ity and distortion depend on outage probability. It is evident E. NOMA with Multiple Antenna
that the outage probability that maximizes the outage rate may The larger the rank of the MIMO channel matrix, the larger
not provide the minimum expected distortion [62], [63]. An the number of de-correlated channels obtained, and thus, bet-
investigation can be carried out to optimize the outage proba- ter system performance occurs. Therefore, the channel matrix
bility for which a NOMA scheme can provide the maximum rank plays a key role in evaluating MIMO NOMA. However,
outage rate with acceptable distortion. existing works in MIMO NOMA consider full-rank channel
matrices in order to investigate system performance. Under
C. Impact of Interference this constraint, for example, the analytical results developed by
Although interference analysis is a generic term in wire- Ding et al. [36] provide the upper bounds of outage probability
less communications, here this survey focuses on coopera- and capacity. It is therefore now essential to study the design
tive NOMA [23], which utilizes Bluetooth-like short-range of MIMO NOMA with consideration of rank-deficient channel
communications in the cooperative phase. However, the use matrices. Rank adjustment for NOMA can be adopted in order
of BT radio in cellular communications faces an extreme to improve MIMO NOMA performance [21]. A joint opti-
interference scenario from existing wireless personal area net- mization of rank and transmission power assignment can be
work (WPAN) operations. BT interference decreases the cov- investigated to further enhance performance. It is also interest-
erage and throughput, causes intermittent or complete loss ing to focus on existing MIMO NOMA techniques combined
of connectivity, and results in difficult pairing during the with other promising wireless techniques, for example, orthog-
user’s discovery phase. In fact, interference in the deployed onal frequency and code division multiplexing (OFCDM).
environment, payload size, and distance between cooperative Most of the MIMO NOMA algorithms (for example, the beam-
users all affect the deployment of channel allocation. Also, forming algorithm proposed by Kim et al. [32]) visited in this
a self-organizing scatternet to manage BT nodes needs to be paper have very high computational complexity. Therefore,
reformulated to make it functional with NOMA, since the there is an urgent need for research activities into complexity
users in NOMA are paired according to CSI. In addition, reduction.
a robust scatternet should offer valid routes between nodes
with high probability, even though users’ mobility causes com- F. Heterogeneous Networks
plete loss of some of the wireless links. Furthermore, due to the A heterogeneous network (HetNet) is a wireless network
mobility of users, interference becomes dynamic. Therefore, consisting of nodes with diverse transmission powers and
performance analysis of a cooperative NOMA scheme in this coverage sizes. The HetNet has potential enough for next-
dynamic interfering environment will be an interesting study. generation wireless networks in terms of capacity and coverage
with reduced energy consumption. The infrastructure featuring
D. Resource Allocation a high-density deployment of low-power nodes can also sig-
In order to accommodate a diverse set of traffic require- nificantly increase EE, compared to low-density deployment
ments, 5G systems should be capable of supporting high of fewer high-power nodes. There are several research works
data rates at very low latency and in reliable ways. However, on HetNets, for example, node cooperation, optimal load bal-
this is a very difficult task, since resources are limited. ancing, and enhanced inter-cell interference coordination [67].
ISLAM et al.: POWER-DOMAIN NOMA IN 5G SYSTEMS 737

A system framework of a cooperative HetNet for 5G was more significant if there are more local obstacles. Deriving
recently studied in [68] with the aims of both spectrum effi- a NOMA scheme that provides users (especially located at
ciency and EE. Since the objective of NOMA coincides with a distance greater than 150 m up to the cell boundary, in the
that of a HetNet, the specific utilization of NOMA in a partic- case of mmW cellular) uniform outage experiences would be
ular HetNet can offer extended benefits. Also, the non-uniform excellent work.
spatial distribution of mobile users will affect the performance
of NOMA. Therefore, investigation of outage performance,
ergodic capacity, and user fairness in NOMA schemes with J. NOMA with Antenna Selection
spatial user distribution can be worthy work.
The simultaneous use of multiple antennas at the trans-
mitter side requires a corresponding number of parallel radio
G. Outage Probability Analysis
frequency (RF) chains at the front end. As such, it increases
Outage analysis is fundamental to understanding the system complexity, power consumption, and cost with an
performance of any wireless system. As a matter of fact, increase in the number of antennas. To overcome such issues,
the achievable capacity depends upon the outage behavior transmit antenna selection (TAS) is often preferred to make
of users. Various researchers have investigated the outage the transmitter structure simple. An effective TAS technique
probability of the basic NOMA scheme, in general. For exam- opportunistically selects the best antenna out of multiple
ple, compared to OMA, less outage occurs in NOMA for antennas and uses a single RF chain for transmission. This
users randomly deployed in a cell [19]. That analysis con- eventually reduces system complexity, power consumption,
sidered the impact of path loss. Also, it was observed that and cost, as well as size, at the expense of acceptable
a NOMA-BF system improves the sum capacity, compared to performance loss. In this respect, research should be con-
the conventional multi-user BF system [25]. However, when ducted into finding a novel TAS-NOMA scheme for downlink
NOMA comes with beamforming, the outage probability of communications from a base station equipped with multiple
users will be changed. On that, outage performance anal- antennas to multiple users, each equipped with a single
ysis of NOMA-BF can be investigated. Similarly, network antenna. The aim of the scheme should ultimately improve
NOMA [40] primarily focuses on inter-cell interference miti- the sum rate, considering the targeted user rate is allocated
gation by designing a precoder. But it needs explicit analysis opportunistically based on channel conditions.
to understand the outage behaviors of cell-edge users. Other
NOMA works, for example, NOMA with VLC [50] and
NOMA with coding [54], [55], can be further investigated to K. Carrier Aggregation
analyze outage performance.
In order to increase the BW, and thereby, increase the
H. Practical Channel Model data rate of its users, LTE-Advanced utilizes the concept
of carrier aggregation (CA) [71]. The CA concept is that
To support the ever-growing amount of consumer data, users are allocated aggregated resources consisting of two or
next-generation wireless networks require not only an effi- more component carriers (CCs). A CC is nothing but each
cient radio access technique but also spectrum availability. aggregated carrier. The arrangement of aggregation would
For the time being, it is obvious that 5G will use spectrum be either contiguous allocation, where CCs are set adja-
allocations in the unused millimeter wave (mmW) frequency cent to each other, or non-contiguous allocation, where there
bands. Also, the backbone networks for 5G are expected is a gap in between. It is possible to integrate CA with
to move from copper and fiber to mmW wireless connec- NOMA in order to take the advantages offered by both.
tions, allowing rapid deployment and mesh-like connectivity. To do so, however, the user pairing will be different from
The mmW frequencies between 30 and 300 GHz are a new that of basic NOMA. It is well known that a NOMA user
frontier for cellular networks that offers a huge amount of is paired with another NOMA user in a basic two-user
BW. Understanding the challenges of mmW cellular commu- NOMA scheme; each carrier is usually assigned to two users
nications, in general, and channel behavior in particular, is based on channel conditions. The user paring in CA-enabled
therefore extremely important, and is a fundamental require- NOMA can be explained as follows. A CC (say, fAN1 ) assigned
ment to developing 5G mobile systems, as well as back- to a NOMA user (UAN1 ) will also be a CC of another
haul techniques [69]. The existing studies on NOMA assume NOMA user (UAN2 ). The situation is that the user UAN1 is
the wireless links between transmitter and receiver exhibit paired with UAN2 with respect to the CC fAN1 . At the same
a Rayleigh fading channel with AWGN. A more realistic anal- time, it is also possible that the user UAN1 is paired with
ysis would be revealed if the measured path loss and delay UAN3 with respect to the CC fAN2 . On that, a NOMA user
spread values [70] could be considered to reflect the radio might be pared with multiple different users at the same time
channel in mmW band. based on the number of CCs if CA is integrated with basic
NOMA. To the best of authors’ knowledge, answering the
I. Uniform Fairness question of what CA type is appropriate for NOMA solu-
In mmW cellular, at a distance greater than 175 m, most tions represents an open issue, and thus, investigation of
locations experience a signal outage [69]. Since outage is different types of CA in NOMA is an interesting research
highly dependent on the environment, actual outage may be direction.
738 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 19, NO. 2, SECOND QUARTER 2017

L. Other Challenges This group-wise SC and SIC operation basically provides

There are also some other challenges that need to be a tradeoff between performance gain and implementation
addressed before NOMA becomes a part of 5G in the future. In complexity.
a downlink scenario, for example, the transmitter allocates the
power to the users based on their respective CSI. Therefore, B. Error Propagation
a proper mechanism for CSI feedback, a suitable channel It is intuitive that once an error occurs in SIC, all other
estimation scheme with proper reference signal design, is user information will likely be decoded erroneously. However,
important for achieving robust performance. In addition, the the effect of error propagation can be compensated by using
benefit of NOMA is proven only under the ideal condi- a stronger code (e.g., increasing the block length) when the
tion of perfect acquisition of CSI data at the transmitter number of users is reasonably small [3]. In case of degrada-
side. One possible solution might be the use of a limited tion in the performance for some users, nonlinear detection
feedback channel to acquire CSI [72]. However, it requires techniques can also be considered to suppress the error propa-
additional BW to convey various channel quality indicators. gation. Based on computer simulations, Benjebbour et al. [2]
From this perspective, proper user selection and power alloca- showed that the error propagation can have a marginal impact
tion schemes are necessary when the transmitter has to work on the NOMA performance. The reason is that a user with
under imperfect CSI and/or with limited feedback. In multicar- bad channel gain is assigned to another user with good channel
rier communications, the peak-to-average-power ratio (PAPR) gain during NOMA scheduling. This finding has been reported
can cause the transmitter’s power amplifier (PA) to run within under the conditions of a worst-case model which assumes
a non-linear operating region. This causes significant signal that the decoding of the strong user at the second stage is
distortion in PA output. The effect of PAPR is thus critical always unsuccessful whenever the decoding of the weak user
to determining what techniques to use to achieve the best is unsuccessful at the first stage of the strong user receiver.
NOMA performance. The concept of relaying [29] to extend Although there exist works that analytically study the SIC
cell coverage can be applied in non-orthogonal coordinated error propagation in basic MIMO systems [75], [76], there is
transmissions between multiple small cells and a macro cell no prominent research that provides a mathematical under-
in order to achieve more capacity gain. A hybrid scheme standing of the effect of imperfect SIC on NOMA schemes.
proposed by Kalokidou et al. [73] combines the principles Therefore, a mathematical analysis of the impact of imper-
of topological interference management and NOMA schemes fect SIC on NOMA performances represents an interesting
to achieve better performance in terms of sum rate. Further research direction.
work on the MIMO form of this hybrid method can be investi-
gated. However, the challenge, especially for a dense network,
C. Quantization Error
is how to determine a fair power allocation method. To adopt
NOMA in 5G, NOMA should also be made robust in terms of When the received signal strengths of users are disparate,
system scalability, since 5G must support heterogeneous traf- the analog-to-digital (A/D) converter needs to support a very
fic and diverse radio environments. To date, most of the work large full-scale input voltage range, and requires high reso-
has been dedicated to verifying the performance of NOMA in lution to accurately quantize a weak signal, since the more
theory, but not in practice. Although Xiong et al. [74] intro- levels the ADC uses for quantization, the lower its quantization
duced a software-defined radio (SDR) NOMA as a prototype, noise power. However, there is a limitation placed on arbi-
more over-the-air experiments are required for the purpose of trarily high–resolution ADC due to its cost, conversion time,
demonstrating NOMA’s potential in 5G mobile networks. and hardware complexity. This constraint eventually leads to
a trade-off between the quantization error and SIC gain.
VI. I MPLEMENTATION I SSUES
NOMA implementation includes designing and operating D. Power Allocation Complexity
the system. A lot of what will be attempted at this stage will The achievable throughput of a user is affected by the trans-
rely upon a particular application and environment. In this mit power allocation to that user. This particular power alloca-
section, this survey tries to give some generic points, including tion also affects the achievable capacity of other users, since
some on computational complexity and error propagation that the basis of NOMA is power-domain user multiplexing. To
can be critical to assessing NOMA performance. achieve the best throughput performance in NOMA, a brute-
force search over the possible user pairs with dynamic power
A. Decoding Complexity allocation is required. However, this kind of exhaustive search
is computationally expensive.
Signal decoding by using SIC requires additional implemen-
tation complexity compared to orthogonal schemes, since the
receiver has to decode other users’ information prior to decod- E. Residual Timing Offset
ing its own information [3]. Also, this complexity increases Synchronous transmission has been considered in
as the number of users in the cell of interest increases. NOMA research, and this consideration is reasonable
However, users can be clustered into a number of groups, for the downlink scenario, because the BS controls trans-
where each cluster contains a small number of users with bad mission for all users. However, perfect synchronization
channels. SC/SIC can then be executed within each group. among NOMA users is impractical on uplink, since users
ISLAM et al.: POWER-DOMAIN NOMA IN 5G SYSTEMS 739

are spatially distributed, and the mobile communications J. Standardization Status

channel is usually dynamic in nature. In asynchronous com- There exist some standardization efforts on NOMA tech-
munications, OFDM symbols from the superposition-coded nologies. Below is the summary of the current status of
users are time-misaligned. Thus, NOMA users’ performance ongoing standardization of NOMA. The 3rd Generation
substantially depends on the relative time offset between Partnership Project (3GPP) LTE Release 13 approved the
interfering users [77]. On that, it is important to further study item (SI) of downlink multiuser superposition transmis-
investigate NOMA in asynchronous communications. In sion (MUST) [81]. The primary objective of MUST was to
practice, the asynchronous scheme requires information on identify promising enhancements of downlink multiuser trans-
multiple symbols for detection and interference cancellation. mission schemes within one cell. To achieve this objective,
Otherwise, performance degrades when complete information the SI focused on the evaluation of system level gain and
on interfering signals is unknown. complexity-performance trade-off under practical deployment
scenarios and traffic models [82]–[87]. The study concludes
F. NOMA or IDMA that NOMA can increase system capacity and improve user
Both NOMA and interleave division multiple experiences. A new work item (WI) of downlink multiuser
access (IDMA) are prospective multiple access schemes superposition transmission for LTE has recently been approved
for equal rate transmission without link adaptation. The by 3GPP LTE Release 14 [88]. The core objective of this WI
question may arise as to which scheme should be adopted in is to identify necessary techniques to enable LTE to support
practical system implementation. Chen et al. [78] concluded downlink intra-cell multiuser superposition transmission for
that IDMA is able to deal with the user power–balanced the physical downlink shared channel.
scenario and still offer robust performance with higher com-
putational complexity. On the other hand, NOMA is effective VII. C ONCLUSION
in the power-imbalanced scenario, smartly performing the This paper provides a comprehensive overview of
trade-off between performance and complexity. the present and emerging power-domain SC-based
NOMA research into 5G, and discusses NOMA performance
G. Signaling and Processing Overhead with numerical results. It is clear that NOMA is a candidate
There are several sources of additional signaling and pro- multiple access technology for next-generation radio access.
cessing overhead in NOMA compared to its orthogonal Its diversity gain originates from the power domain of the sig-
counterparts. For example, to collect the CSI from different nals to be transmitted in a superposed fashion. Many research
receivers and to inform the receivers of the SIC order, some results have been found in favor of NOMA in terms of outage
time slots need to lapse. This causes data rate degradation probability, achievable capacity, weak users’ rate guarantees,
in NOMA. Also, with dynamic power allocation, and encod- and cell-edge user experiences. In addition to perfect SC at
ing and decoding for SC and SIC, NOMA signal processing the transmitter and error-free SIC at the receiver, optimum
requires additional energy overhead. power allocation, QoS-oriented user fairness, appropriate user
pairing, and good link adaptation are also required to obtain
the maximum benefits offered by NOMA. In addition, this
H. Limited Number of User Pairs paper discusses how NOMA works with various standard
NOMA exploits power-domain multiplexing to offer its ben- wireless technologies, including cooperative communications
efits. More specifically, about 8 dB difference in propagation and MIMO. For a deeper understanding of NOMA, this
loss is required to pair a cell-edge user with a cell-center paper provides a discussion on how inter-cell interference
user [79]. Therefore, the possible number of pairs of users in in a network can be mitigated, and explains how a trade-off
a typical NOMA scheme is limited, which eventually reduces between energy efficiency and spectral efficiency can be
the capacity gain of NOMA. Finding the appropriate signal achieved. The discussions of several important issues, such
detection and decoding strategy that increases the number of as dynamic user pairing, distortion analysis, interference
pairs of users is an important issue. analysis, resource allocation, heterogeneous networks, carrier
aggregation, and transmit antenna selection, are expected
I. NOMA Deploying Environment to facilitate, and provide a basis for further research on
Since the power domain is exploited, and channel gains NOMA in 5G. This paper offers a general view of some
are naturally adequately large in macrocells, researchers com- implementation issues, including computational complexity,
monly consider macrocell deployment. In addition to macro- error propagation, deployment environments, and standard-
cells, small cells are also becoming important and are being ization status. In sum, the results of this survey are expected
studied for future 5G networks. Fortunately, the NOMA gain to be useful to researchers working in the area of wireless
is still achievable, even in a small cell where NOMA offers communications and NOMA.
a higher performance gain compared to OMA [80]. It is
also important to note that NOMA gain in terms of cell ACKNOWLEDGEMENT
throughput in small cells is even larger than in macrocells. The authors would like to thank the anonymous reviewers
Therefore, NOMA can be equally deployed in both macrocell for their valuable comments and suggestions to improve the
and small-cell environments. quality of the paper.
740 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 19, NO. 2, SECOND QUARTER 2017

R EFERENCES [24] Y. Hayashi, Y. Kishiyama, and K. Higuchi, “Investigations on power

allocation among beams in non-orthogonal access with random beam-
[1] P. Wang, J. Xiao, and L. Ping, “Comparison of orthogonal and non- forming and intra-beam SIC for cellular MIMO downlink,” in Proc.
orthogonal approaches to future wireless cellular systems,” IEEE Veh. IEEE Veh. Technol. Conf., Las Vegas, NV, USA, Sep. 2013, pp. 1–5.
Technol. Mag., vol. 1, no. 3, pp. 4–11, Sep. 2006. [25] B. Kim et al., “Non-orthogonal multiple access in a downlink multiuser
[2] A. Benjebbour et al., “Concept and practical considerations of non- beamforming system,” in Proc. IEEE Milit. Commun. Conf., San Diego,
orthogonal multiple access (NOMA) for future radio access,” in Proc. CA, USA, Nov. 2013, pp. 1278–1283.
Int. Symp. Intell. Signal Process. Commun. Syst. (ISPACS), Naha, Japan, [26] J. Men and J. Ge, “Non-orthogonal multiple access for multiple-
Nov. 2013. pp. 770–774. antenna relaying networks,” IEEE Commun. Lett., vol. 19, no. 10,
[3] D. Tse and P. Viswanath, Fundamentals of Wireless Communication. pp. 1686–1689, Oct. 2015.
Cambridge, U.K.: Cambridge Univ. Press, 2005. [27] J. Men and J. Ge, “Performance analysis of non-orthogonal multiple
[4] R. Hoshyar, F. P. Wathan, and R. Tafazolli, “Novel low-density signature access in downlink cooperative network,” IET Commun., vol. 9, no. 18,
for synchronous CDMA systems over AWGN channel,” IEEE Trans. pp. 2267–2273, Dec. 2015.
Signal Process., vol. 56, no. 4, pp. 1616–1626, Apr. 2008. [28] J.-B. Kim and I.-H. Lee, “Capacity analysis of cooperative relaying
[5] R. Razavi, R. Hoshyar, M. A. Imran, and Y. Wang, “Information the- systems using non-orthogonal multiple access,” IEEE Commun. Lett.,
oretic analysis of LDS scheme,” IEEE Commun. Lett., vol. 15, no. 8, vol. 19, no. 11, pp. 1949–1952, Nov. 2015.
pp. 798–800, Aug. 2011. [29] J.-B. Kim and I.-H. Lee, “Non-orthogonal multiple access in coordinated
[6] M. Al-Imari, P. Xiao, M. A. Imran, and R. Tafazolli, “Uplink non- direct and relay transmission,” IEEE Commun. Lett., vol. 19, no. 11,
orthogonal multiple access for 5G wireless networks,” in Proc. Int. Symp. pp. 2037–2040, Nov. 2015.
Wireless Commun. Syst. (ISWCS), Aug. 2014, pp. 781–785. [30] S. Timotheou and I. Krikidis, “Fairness for non-orthogonal multiple
[7] M. Al-Imari, M. A. Imran, and R. Tafazolli, “Low density spreading for access in 5G systems,” IEEE Signal Process. Lett., vol. 22, no. 10,
next generation multicarrier cellular systems,” in Proc. IEEE Int. Conf. pp. 1647–1651, Oct. 2015.
Future Commun. Netw. (ICFCN), Baghdad, Iraq, Apr. 2012, pp. 52–57. [31] M. Mollanoori and M. Ghaderi, “Uplink scheduling in wireless networks
[8] H. Nikopour and H. Baligh, “Sparse code multiple access,” in Proc. with successive interference cancellation,” IEEE Trans. Mobile Comput.,
IEEE Int. Symp. Pers. Indoor Mobile Radio Commun. (PIMRC), London, vol. 13, no. 5, pp. 1132–1144, May 2014.
U.K., Sep. 2013, pp. 332–336. [32] J. Kim, J. Koh, J. Kang, K. Lee, and J. Kang, “Design of user clustering
[9] H. Nikopour et al., “SCMA for downlink multiple access of 5G wire- and precoding for downlink non-orthogonal multiple access (NOMA),”
less networks,” in Proc. IEEE Glob. Telecommun. Conf. (GLOBECOM), in Proc. IEEE Milit. Commun. Conf. (MILCOM), Tampa, FL, USA,
Austin, TX, USA, Dec. 2014, pp. 3940–3945. Oct. 2015, pp. 1170–1175.
[10] S. Kang, X. Dai, and B. Ren, “Pattern division multiple access for 5G,” [33] K. Higuchi and Y. Kishiyama, “Non-orthogonal access with random
Telecommun. Netw. Technol., vol. 5, no. 5, pp. 43–47, May 2015. beamforming and intra-beam SIC for cellular MIMO downlink,” in Proc.
[11] L. Hanzo, M. Münster, B. J. Choi, and T. Keller, OFDM and IEEE Veh. Technol. Conf., Las Vegas, NV, USA, Sep. 2013, pp. 1–5.
MC-CDMA for Broadband Multi-User Communications, WLANs and [34] Y. Lan, A. Benjebboiu, X. Chen, A. Li, and H. Jiang, “Considerations
Broadcasting. Piscataway, NJ, USA: IEEE Press, 2003. on downlink non-orthogonal multiple access (NOMA) combined with
[12] T. Abe and T. Matsumoto, “Space-time turbo equalization in frequency- closed-loop SU-MIMO,” in Proc. IEEE Int. Conf. Signal Process.
selective MIMO channels,” IEEE Trans. Veh. Technol., vol. 52, no. 3, Commun. Syst., Gold Coast, QLD, Australia, Dec. 2014, pp. 1–5.
pp. 469–475, May 2003. [35] Q. Sun, S. Han, C.-L. I, and Z. Pan, “On the ergodic capacity of
[13] M. Y. Alias, S. Chen, and L. Hanzo, “Multiple-antenna-aided OFDM MIMO NOMA systems,” IEEE Wireless Commun. Lett., vol. 4, no. 4,
employing genetic-algorithm-assisted minimum bit error rate multiuser pp. 405–408, Aug. 2015.
detection,” IEEE Trans. Veh. Technol., vol. 54, no. 5, pp. 1713–1721, [36] Z. Ding, F. Adachi, and H. V. Poor, “The application of MIMO to non-
Sep. 2005. orthogonal multiple access,” IEEE Trans. Wireless Commun., vol. 15,
[14] L. Hanzo, S. Chen, J. Zhang, and X. Mu, “Evolutionary-algorithm- no. 1, pp. 537–552, Jan. 2016.
assisted joint channel estimation and turbo multiuser detection/decoding [37] Q. Sun et al., “Sum rate optimization for MIMO non-orthogonal multiple
for OFDM/SDMA,” IEEE Trans. Veh. Technol., vol. 63, no. 3, access systems,” in Proc. Wireless Commun. Netw. Conf. (WCNC),
pp. 1204–1222, Mar. 2014. New Orleans, LA, USA, Mar. 2015, pp. 747–752.
[15] L. Dai et al., “Non-orthogonal multiple access for 5G: Solutions, chal- [38] J. Choi, “Non-orthogonal multiple access in downlink coordinated two-
lenges, opportunities, and future research trends,” IEEE Commun. Mag., point systems,” IEEE Commun. Lett., vol. 18, no. 2, pp. 313–316,
vol. 53, no. 9, pp. 74–81, Sep. 2015. Feb. 2014.
[16] T. M. Cover, “Broadcast channels,” IEEE Trans. Inf. Theory, vol. 18, [39] S. M. Alamouti, “A simple transmit diversity technique for wire-
no. 1, pp. 2–14, Jan. 1972. less communications,” IEEE J. Sel. Areas Commun., vol. 16, no. 8,
[17] S. Vanka et al., “Superposition coding strategies: Design and exper- pp. 1451–1458, Oct. 1998.
imental evaluation,” IEEE Trans. Wireless Commun., vol. 11, no. 7, [40] S. Han, C.-L. I, Z. Xu, and Q. Sun, “Energy efficiency and spectrum
pp. 2628–2639, Jul. 2012. efficiency co-design: From NOMA to network NOMA,” IEEE Commun.
[18] N. I. Miridakis and D. D. Vergados, “A survey on the successive Soc. MMTC E-Lett., vol. 9, no. 5, pp. 21–24, Sep. 2014.
interference cancellation performance for single-antenna and multiple- [41] W. Shin et al., “Coordinated beamforming for multi-cell
antenna OFDM systems,” IEEE Commun. Surveys Tuts., vol. 15, no. 1, MIMO-NOMA,” IEEE Commun. Lett., to be published,
pp. 312–335, 1st Quart., 2013. doi: 10.1109/LCOMM.2016.2615097.
[19] K. Higuchi and A. Benjebbour, “Non-orthogonal multiple [42] W. Shin, M. Vaezi, J. Lee, and H. V. Poor, “On the number of users
access (NOMA) with successive interference cancellation,” IEICE served in MIMO-NOMA cellular networks,” in Proc. Int. Symp. Wireless
Trans. Commun., vol. E98-B, no. 3, pp. 403–414, Mar. 2015. Commun. Syst. (ISWCS), Poznań, Poland, Sep. 2016, pp. 638–642.
[20] Z. Ding, Z. Yang, P. Fan, and H. V. Poor, “On the performance of [43] Z. Ding, P. Fan, and H. V. Poor, “Impact of user pairing on 5G
non-orthogonal multiple access in 5G systems with randomly deployed nonorthogonal multiple-access downlink transmissions,” IEEE Trans.
users,” IEEE Signal Process. Lett., vol. 21, no. 12, pp. 1501–1505, Veh. Technol., vol. 65, no. 8, pp. 6010–6023, Aug. 2016.
Dec. 2014. [44] F. Liu, P. Mähönen, and M. Petrova, “Proportional fairness-based
[21] X. Chen, A. Benjebbour, Y. Lan, A. Li, and H. Jiang, “Impact of rank user pairing and power allocation for non-orthogonal multiple access,”
optimization on downlink non-orthogonal multiple access (NOMA) with in Proc. IEEE 26th Annu. Int. Symp. Pers. Indoor Mobile Radio
SU-MIMO,” in Proc. IEEE Int. Conf. Commun. Syst. (ICCS), Nov. 2014, Commun. (PIMRC), Hong Kong, Aug./Sep. 2015, pp. 1127–1131.
pp. 233–237. [45] N. Otao, Y. Kishiyama, and K. Higuchi, “Performance of non-orthogonal
[22] A. S. Ibrahim, A. K. Sadek, W. Su, and K. J. R. Liu, “Cooperative access with SIC in cellular downlink using proportional fair-based
communications with relay-selection: When to cooperate and whom resource allocation,” in Proc. Int. Symp. Wireless Commun. Syst., Paris,
to cooperate with?” IEEE Trans. Wireless Commun., vol. 7, no. 7, France, 2012, pp. 476–480.
pp. 2814–2827, Jul. 2008. [46] T. Seyama, T. Dateki, and H. Seki, “Efficient selection of user sets for
[23] Z. Ding, M. Peng, and H. V. Poor, “Cooperative non-orthogonal downlink non-orthogonal multiple access,” in Proc. IEEE 26th Annu.
multiple access in 5G systems,” IEEE Commun. Lett., vol. 19, no. 8, Int. Symp. Pers. Indoor Mobile Radio Commun. (PIMRC), Hong Kong,
pp. 1462–1465, Aug. 2015. Sep. 2015, pp. 1062–1066.
ISLAM et al.: POWER-DOMAIN NOMA IN 5G SYSTEMS 741

[47] Q. Sun, S. Han, C.-L. I, and Z. Pan, “Energy efficiency optimization for [70] T. S. Rappaport et al., “Millimeter wave mobile communications for 5G
fading MIMO non-orthogonal multiple access systems,” in Proc. IEEE cellular: It will work!” IEEE Access, vol. 1, pp. 335–349, 2013.
Int. Conf. Commun. (ICC), London, U.K., Jun. 2015, pp. 2668–2673. [71] G. Yuan, X. Zhang, W. Wang, and Y. Yang, “Carrier aggregation for
[48] H. Marshoud, V. M. Kapinas, G. K. Karagiannidis, and S. Muhaidat, LTE-advanced mobile communication systems,” IEEE Commun. Mag.,
“Non-orthogonal multiple access for visible light communications,” vol. 48, no. 2, pp. 88–93, Feb. 2010.
IEEE Photon. Technol. Lett., vol. 28, no. 1, pp. 51–54, Jan. 1, 2016. [72] S. Liu and C. Zhang, “Downlink non-orthogonal multiple access system
[49] L. Yin, X. Wu, and H. Haas, “On the performance of non-orthogonal with limited feedback channel,” in Proc. IEEE 7th Int. Conf. Wireless
multiple access in visible light communication,” in Proc. IEEE 26th Commun. Signal Process. (WCSP), Nanjing, China, Oct. 2015, pp. 1–5.
Annu. Int. Symp. Pers. Indoor Mobile Radio Commun. (PIMRC), [73] V. Kalokidou, O. Johnson, and R. Piechocki, “A hybrid TIM-
Hong Kong, Sep. 2015, pp. 1354–1359. NOMA scheme for the SISO broadcast channel,” in Proc. IEEE
[50] R. C. Kizilirmak, C. R. Rowell, and M. Uysal, “Non-orthogonal multiple Int. Conf. Commun. Workshop (ICCW), London, U.K., Jun. 2015,
access (NOMA) for indoor visible light communications,” in Proc. pp. 387–392.
4th Int. Workshop Opt. Wireless Commun. (IWOW), Istanbul, Turkey, [74] X. Xiong, W. Xiang, K. Zheng, H. Shen, and X. Wei, “An open source
Sep. 2015, pp. 98–101. SDR-based NOMA system for 5G networks,” IEEE Wireless Commun.,
[51] J. Choi, “H-ARQ based non-orthogonal multiple access with succes- vol. 22, no. 6, pp. 24–32, Dec. 2015.
sive interference cancellation,” in Proc. IEEE Glob. Telecommun. Conf., [75] P. Li, R. C. de Lamare, and R. Fa, “Multiple feedback successive
New Orleans, LA, USA, Nov./Dec. 2008, pp. 1–5. interference cancellation detection for multiuser MIMO systems,” IEEE
[52] A. Li, A. Benjebbour, X. Chen, H. Jiang, and H. Kayama, “Investigation Trans. Wireless Commun., vol. 10, no. 8, pp. 2434–2439, Aug. 2011.
on hybrid automatic repeat request (HARQ) design for NOMA with [76] J. Kim, D. Kim, and S. K. Yun, “Mitigating error propagation in suc-
SU-MIMO,” in Proc. IEEE 26th Annu. Int. Symp. Pers. Indoor Mobile cessive interference cancellation,” IEICE Trans. Commun., vol. E89-B,
Radio Commun. (PIMRC), Hong Kong, Aug./Sep. 2015, pp. 590–594. no. 10, pp. 2956–2960, Oct. 2006.
[53] A. Shokrollahi, “Raptor codes,” IEEE Trans. Inf. Theory, vol. 52, no. 6, [77] H. Haci and H. Zhu, “Performance of non-orthogonal multiple access
pp. 2551–2567, Jun. 2006. with a novel interference cancellation method,” in Proc. IEEE Int. Conf.
[54] M. J. Hagh and M. R. Soleymani, “Raptor coding for non-orthogonal Commun. (ICC), London, U.K., Jun. 2015, pp. 2912–2917.
multiple access channels,” in Proc. IEEE Int. Conf. Commun. (ICC), [78] Y. Chen, J. Schaepperle, and T. Wild, “Comparing IDMA and
Jun. 2011, pp. 1–6. NOMA with superimposed pilots based channel estimation in uplink,”
[55] S. Park and D.-H. Cho, “Random linear network coding based on non- in Proc. IEEE 26th Annu. Int. Symp. Pers. Indoor Mobile Radio
orthogonal multiple access in wireless networks,” IEEE Commun. Lett., Commun. (PIMRC), Hong Kong, Sep. 2015, pp. 89–94.
vol. 19, no. 7, pp. 1273–1276, Jul. 2015. [79] T. Yazaki and Y. Sanada, “Effect of joint detection and decoding in non-
[56] X. Chen, A. Beiijebbour, A. Li, H. Jiang, and H. Kayama, orthogonal multiple access,” in Proc. Int. Symp. Intell. Signal Process.
“Consideration on successive interference canceller (SIC) receiver at Commun. Syst. (ISPACS), Kuching, Malaysia, Dec. 2014, pp. 245–250.
cell-edge users for non-orthogonal multiple access (NOMA) with [80] Y. Saito, A. Benjebbour, Y. Kishiyama, and T. Nakamura, “System-
SU-MIMO,” in Proc. IEEE 26th Annu. Int. Symp. Pers. Indoor. Mobile level performance of downlink non-orthogonal multiple access (NOMA)
Radio Commun. (PIMRC), Hong Kong, Aug./Sep. 2015, pp. 522–526. under various environments,” in Proc. IEEE 81st Veh. Technol.
[57] C. Yan et al., “Receiver design for downlink non-orthogonal multiple Conf. (VTC), Glasgow, U.K., May 2015, pp. 1–5.
access (NOMA),” in Proc. IEEE 81st Veh. Technol. Conf. (VTC), [81] MediaTek Inc., “Study on downlink multiuser superposition trans-
Glasgow, U.K., May 2015, pp. 1–6. mission for LTE,” 3GPP TSG RAN Meeting #68, Malmö, Sweden,
[58] K. Saito, A. Benjebbour, Y. Kishiyama, Y. Okumura, and T. Nakamura, Tech. Rep. RP-151100, Jun. 2015.
“Performance and design of SIC receiver for downlink NOMA with [82] A. Benjebbour et al., “NOMA: From concept to standardization,” in
open-loop SU-MIMO,” in Proc. IEEE Int. Conf. Commun. Proc. IEEE Conf. Standards Commun. Netw. (CSCN), Tokyo, Japan,
Workshop (ICCW), London, U.K., Jun. 2015, pp. 1161–1165. May 2015, pp. 18–23.
[59] Y. Endo, Y. Kishiyama, and K. Higuchi, “Uplink non-orthogonal access [83] NTT DOCOMO, “Deployment scenarios for downlink multiuser super-
with MMSE-SIC in the presence of inter-cell interference,” in Proc. position transmissions,” 3GPP TSG RAN WG1 Meeting #80bis,
Int. Symp. Wireless Commun. Syst. (ISWCS), Paris, France, Aug. 2012, Belgrade, Serbia, Tech. Rep. R1-152062, Apr. 2015.
pp. 261–265. [84] NTT DOCOMO, “Evaluation methodologies for downlink multiuser
[60] Y. Liu, Z. Ding, M. Eikashlan, and H. V. Poor, “Cooperative non- superposition transmissions,” 3GPP TSG RAN WG1 Meeting #81,
orthogonal multiple access in 5G systems with SWIPT,” in Proc. 23rd Fukuoka, Japan, Tech. Rep. R1-153332, May 2015.
Eur. Signal Process. Conf. (EUSIPCO), Nice, France, Aug./Sep. 2015, [85] MediaTek Inc., “Candidate non-orthogonal multiple access,” 3GPP TSG
pp. 1999–2003. RAN WG1 Meeting #81, Fukuoka, Japan, Tech. Rep. R1-153335, May
[61] A. Li, A. Harada, and H. Kayama, “A novel low computational com- 2015.
plexity power allocation method for non-orthogonal multiple access [86] NTT DOCOMO, “System-level evaluation results for downlink
systems,” IEICE Trans. Fundamentals Electron. Commun. Comput. Sci., multiuser superposition transmissions,” 3GPP TSG RAN WG1 Meeting
vol. E97-A, no. 1, pp. 57–67, Jan. 2014. #82, Beijing, China, Tech. Rep. R1-154536, Aug. 2015.
[62] S. Choudhury and J. D. Gibson, “Information transmission over fad- [87] NTT DOCOMO, “Link-level evaluation results for downlink multiuser
ing channels,” in Proc. IEEE Glob. Telecommun. Conf. (GLOBECOM), superposition transmissions,” 3GPP TSG RAN WG1 Meeting #82,
Washington, DC, USA, Nov. 2007, pp. 3316–3321. Beijing, China, Tech. Rep. R1-154537, Aug. 2015.
[63] S. M. R. Islam and K. S. Kwak, “Outage capacity and source distortion [88] MediaTek Inc. and CMCC, “Downlink multiuser superposition transmis-
analysis for NOMA users in 5G systems,” Electron. Lett., vol. 52, no. 15, sions for LTE,” 3GPP TSG RAN Meeting #71, Gothenburg, Sweden,
pp. 1344–1345, Jul. 2016. Tech. Rep. RP-160680, Mar. 2016.
[64] B. G. Lee, D. Park, and H. Seo, Wireless Communications Resource
Management. Hoboken, NJ, USA: Wiley, 2009. S. M. Riazul Islam (M’10) received the B.S. and
[65] L. Lei, D. Yuan, C. K. Ho, and S. Sun, “Joint optimization of power M.S. degrees in applied physics and electronics
and channel allocation with non-orthogonal multiple access for 5G cel- from the University of Dhaka, Bangladesh, in 2003
lular systems,” in Proc. IEEE Glob. Telecommun. Conf. (GLOBECOM), and 2005, respectively, and the Ph.D. degree in
San Diego, CA, USA, Dec. 2015, pp. 1–6. information and communication engineering from
[66] M. Zeng, G. I. Tsiropoulos, O. A. Dobre, and M. H. Ahmed, Inha University, South Korea, in 2012. Since
“Power allocation for cognitive radio networks employing non- 2014, he has been with the UWB Wireless
orthogonal multiple access,” in Proc. IEEE Glob. Telecommun. Communications Research Center, Inha University,
Conf. (GLOBECOM), Washington, DC, USA, Dec. 2016. as a Research Professor. He served the University
[67] R. Q. Hu and Y. Qian, Heterogeneous Cellular Networks. Chichester, of Dhaka as an Assistant Professor and a Lecturer
U.K.: Wiley, 2013. with the Department of Electrical and Electronic
[68] R. Q. Hu and Y. Qian, “An energy efficient and spectrum efficient wire- Engineering (formerly Department of Applied Physics, Electronics and
less heterogeneous network framework for 5G systems,” IEEE Commun. Communication Engineering) from 2005 to 2014. In 2014, Dr. Islam was
Mag., vol. 52, no. 5, pp. 94–101, May 2014. with Samsung Research and Development Institute Bangladesh as a Chief
[69] S. Rangan, T. S. Rappaport, and E. Erkip, “Millimeter-wave cellular Engineer with the Department of Solution Laboratory for six months. His
wireless networks: Potentials and challenges,” Proc. IEEE, vol. 102, research interests include wireless communications, signal processing for
no. 3, pp. 366–385, Mar. 2014. communications, and enabling technologies for 5G and beyond.
742 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 19, NO. 2, SECOND QUARTER 2017

Nurilla Avazov received the B.Sc. and M.Sc. Kyung-Sup Kwak (M’81) received the Ph.D.
degrees in telecommunications from the Tashkent degree from the University of California at San
University of Information Technologies, Uzbekistan, Diego in 1988. From 1988 to 1989, he was
in 2005 and 2007, respectively, and the Ph.D. with Hughes Network Systems, San Diego,
degree in telecommunications engineering from CA, USA. From 1989 to 1990, he was with
the University of Agder, Grimstad, Norway, in the IBM Network Analysis Center, Research
2015. Since 2016, he has been a Research Triangle Park, NC, USA. Since then, he has
Professor with the UWB Wireless Communications been with the School of Information and
Research Center, Inha University. His current Communication Engineering, Inha University,
research interests include wireless communications, South Korea, as a Professor, where he had been
5G, multiple-input multiple-output systems, vehicle- the Dean of the Graduate School of Information
to-vehicle communications, Internet of Things, and channel modeling for Technology and Telecommunications from 2001 to 2002. He has been the
UWB communication systems. Director of the UWB Wireless Communications Research Center (formerly
Key National IT Research Center), South Korea, since 2003. In 2006, he
served as the President of Korean Institute of Communication Sciences,
and in 2009, the President of Korea Institute of Intelligent Transport
Systems. In 2008, he had been selected for Inha Fellow Professor and now
Octavia A. Dobre (M’05–SM’07) received the for Inha Hanlim Fellow Professor. Dr. Kwak published more than 200
Engineering Diploma and Ph.D. degrees from peer-reviewed journal papers and served as TPC/Track chairs/organizing
Politehnica University of Bucharest (formerly chairs for several IEEE related conferences. His research interests include
Polytechnic Institute of Bucharest), Romania, in wireless communications, UWB systems, sensor networks, WBAN, and nano
1991 and 2000, respectively. Between 2002 and communications. He was a recipient of the number of awards, including
2005, she was with Politehnica University of the Engineering College Achievement Award from Inha University, the LG
Bucharest and New Jersey Institute of Technology, Paper Award, the Motorola Paper Award, the Haedong Prize of research, and
USA. In 2005, she joined Memorial University, various government awards from the Ministry of ICT, the President, and the
Canada, where she is currently a Full Professor Prime Minister of Korea, for his excellent research performances.
and a Research Chair. She was a Visiting Professor
with the Université de Bretagne Occidentale, France,
and the Massachusetts Institute of Technology, USA, in 2013. Her research
interests include blind signal identification and parameter estimation tech-
niques, 5G technologies, cognitive radio systems, transceiver optimization
algorithms for wireless communications, and optical and underwater com-
munications. She has co-authored over 190 journal and conference papers
in the above areas and gave over 40 invited and keynote talks to indus-
try and academia. Her research has been supported by the Natural Sciences
and Engineering Research Council of Canada, Mathematics of Information
Technology and Complex Systems, Canada Foundation for Innovation,
Research and Development Corporation, Atlantic Canada Opportunities
Agency, Defense and Research Development Canada, Communications
Research Centre Canada, Altera Corporation, DTA Systems, ThinkRF, and
Agilent Technologies. Dr. Dobre was a recipient of the Royal Society
Scholarship at Westminster University, U.K., in 2000, and the Fulbright
Fellowship at the Stevens Institute of Technology, USA, in 2001. She serves
as the Editor-in-Chief of the IEEE C OMMUNICATIONS L ETTERS, as well
as an Editor for the IEEE Communications Surveys and Tutorials. She was
an Editor and a Senior Editor for the IEEE C OMMUNICATIONS L ETTERS,
an Editor for the IEEE T RANSACTIONS ON W IRELESS C OMMUNICATIONS,
and a Guest Editor for other prestigious journals. She served as the General
Chair of CWIT, and a Technical Co-Chair of symposia at numerous confer-
ences, such as IEEE GLOBECOM and ICC. She is the Chair of Women in
the Communications Engineering ComSoc Standing Committee, and a mem-
ber of the IEEE Instrumentation and Measurement Society Administrative
Committee.