NON-ORTHOGONAL MULTIPLE ACCESS FOR 5G
Pattern Division Multiple Access:
A New Multiple Access Technology for 5G
Xiaoming Dai, Zhenyu Zhang, Baoming Bai, Shanzhi Chen, and Shaohui Sun
Abstract
The anticipated 1000-fold increase in mobile
data traffic over the next decade and the explo-
sion of new services and applications pose great
challenges for the current orthogonal multiple
access (OMA)-based 4G systems. A promising
solution to address these challenges is to shift
from the currently predominant OMA to non-or-
thogonal multiple access (NOMA). This article first
introduces the principle of the complexity-con-
strained capacity-achieving NOMA design. Then
a non-orthogonal pattern division multiple access
(PDMA) scheme is proposed to meet the expo-
nentially growing demand of mobile users for
computing and information application services.
The key feature of the PDMA scheme is a joint
design of transmitter and receiver, which allows
low-complexity successive interference cancella-
tion (SIC)-based multi-user detection with substan-
tially improved performance over conventional
OMA schemes. More specifically, the patterns of
multiple users are judiciously designed so that the
data symbols of different users are of appropriate
diversity disparity at the symbol level and power
disparity at the resource element level. The appro-
priate disparity in diversity and power can be
effectively exploited by the low-complexity SIC-
based detector to realize the near-perfect cancel-
lation of multi-user interference. Moreover, the
PDMA system parameters can be flexibly adjusted
to provide different levels of overload, rendering it
suitable to meet the diverse traffic requirements in
future 5G systems. Link-level simulations illustrate
that PDMA is capable of accommodating a 300
percent overload, while it still enjoys transmission
reliability close to conventional OMA schemes.
The results demonstrated in this article indicate
that PDMA can be a promising multiple access
technology with low signaling overhead, low
latency, and massive connectivity support for 5G.
Introduction
The unprecedented increase of mobile data traf-
fic brought about by the wide proliferation of
smartphones and tablet computers is driving the
wireless communications industry to undergo an
unprecedented paradigm shift [1]. In addition, the
advent of the Internet of Things (IoT) will enable
new ways to monitor, assist, secure, and control
smart homes, smart factories, and so on, which
opens up a broad range of diverse applications
Digital Object Identifier: Xiaoming Dai and Zhenyu Zhang are with the University of Science and Technology Beijing; Baoming Bai is with Xidian University;
10.1109/MWC.2018.1700084 Shanzhi Chen and Shaohui Sun are with the China Academy of Telecommunications Technology.
54 1536-1284/18/$25.00 © 2018 IEEE
ranging from mission-critical services to massive
deployment of autonomous devices. These new
services may require the fifth generation (5G) net-
works to support massive connectivity of users
and/or devices to meet the demand for low laten-
cy, low-cost devices, and diverse service types.
Fast and efficient multiple access is the key tech-
nology to handle the massive number of sporad-
ic traffic-generating devices, such as the devices
which are inactive most of the time but regularly
access the network for minor updates without
human interaction.
The current wireless communication systems
have predominantly adopted orthogonal multiple
access (OMA) schemes, where users are allocat-
ed orthogonal physical resources in the time, fre-
quency, or space domain. Existing OMA schemes
efficiently eliminate multi-user interference and
thus allow relatively simple transceiver implemen-
tations. However, it is shown that OMA schemes
achieve strictly lower capacity than non-orthog-
onal multiple access (NOMA) schemes in the
downlink broadcast channel (BC) [2]. Such inef-
ficiency of OMA schemes is even exacerbated in
the uplink scenario [3]. Dimensioning the channel
access based on existing OMA paradigms may
lead to a severe waste of physical resources or
even fail to work in massive connectivity scenari-
os, such as the IoT applications.
To support the daunting task of massive spo-
radic connections, the wireless research commu-
nity is exploring different technical approaches,
such as novel cellular network architectures,
massive multiple-input multiple-output (MIMO)
techniques, spectrum utilizations at untapped mil-
limeter-wave frequency bands, new waveform
designs, and novel multiple access technologies.
Among these potential solutions, the NOMA
approach is especially suitable for meeting the
requirement of massive connectivity, and it is
also efficient in reducing transmission latency and
improving energy efficiency [4–8]. It has been
proven that NOMA is optimal in achieving the
entire capacity region of the BC [2] and exhibits
higher spectral and energy efficiency than OMA
for delay-sensitive applications in the multiple
access channel (MAC) [3].
However, the theoretically predicted gains of
NOMA over OMA rely on proper multi-user signal
separation at the receiver. To reap the full benefits
of NOMA, the maximum a posteriori probability
(MAP) multi-user detection (MUD) technique can
IEEE Wireless Communications • April 2018
be utilized to achieve the desired performance.
The computational complexity of the MAP MUD
scales exponentially with the number of users and
imposes a formidable challenge to practical hard-
ware implementations. As an alternative to the
optimal MAP detector, the low-complexity succes-
sive interference cancellation (SIC)-based detec-
tor with single-user decoding is able to achieve
the Shannon capacity region boundaries in both
the BC and MAC scenarios [9, 10]. Nonetheless,
one main disadvantage of SIC-based detectors is
that errors occurring in detection of transmitted
symbols will propagate further into subsequent
symbols due to interference subtraction. Such
error propagation may severely degrade the sys-
tem performance, especially when the number of
users is large.
In this article, we first introduce the complex-
ity-constrained capacity-achieving NOMA design
principle, which was not addressed in [7, 8]. Then
we propose a non-orthogonal pattern division
multiple access (PDMA) scheme based on a joint
design of the transmitter and an SIC-based detec-
tor at the receiver for the uncorrelated and cor-
related channel scenarios. The latter is an extension
of [7, 8]. The patterns of different users are judi-
ciously designed to exhibit appropriate diversity
disparity at the symbol level and power disparity at
the physical resource element level. Such diversi-
ty disparity and power disparity among users can
be effectively exploited by the SIC-based detector
to achieve near-perfect cancellation of multi-us-
er interference. Furthermore, the PDMA system
parameters can be flexibly adjusted to support a
wide range of overload to accommodate diverse
applications. The analysis based on the constel-
lation-constrained (CC) capacity shows that the
PDMA scheme outperforms conventional OMA
schemes with affordable computational complexi-
ty. In addition, an iterative detection and decoding
(IDD)-based receiver [11] structure is elaborated to
improve the performance of the PDMA scheme.
Link-level simulations show that the PDMA scheme
is able to support up to 300 percent overload
and achieves significant performance gains over
conventional OMA schemes. The superior perfor-
mance on massive connectivity support is also veri-
fied by system-level simulations.
Fundamentals of
Pattern Division Multiple Access
Basics of the SIC-Based Detector
The SIC-based detector [2] iteratively decodes
symbols by subtracting the detected symbols of
strong users first to facilitate the following detec-
tion of weak users. The decoded data of the early
detected symbol is re-encoded, and by using
accurate channel knowledge, it can be recon-
structed to closely resemble the real transmitted
signal. However, the error propagation resulting
from low diversity of early SIC detection stages
may severely degrade the system performance. It
is generally accepted that, for a system equipped
with an SIC-based detector, the performance is
highly dependent on the first-step detection accu-
racy. The low-complexity belief propagation (BP)
algorithm and its variant SIC-BP [12] are shown
to be able to achieve a close approximation of
the MAP MUD. The SIC-BP algorithm solves
IEEE Wireless Communications • April 2018
inference problems, exactly or approximately, via The SIC-BP algorithm
probabilistic graphical models [12]. The SIC-BP obtains a posteriori
algorithm obtains a posteriori estimates of the sys- estimates of the system
tem unknowns by iteratively passing locally calcu- unknowns by itera-
lated conditional probabilities between variable
and function nodes [12]. Similar to SIC-based tively passing locally
detectors, the performance of the SIC-BP algo- calculated conditional
rithm is also determined by the initial inference probabilities between
accuracy of the transmitted symbols involved with variable and function
the iterative detection process. nodes. Similar to SIC-
This observation suggests that enhancing the
first-step inference accuracy is of paramount based detectors, the
importance for improving the overall perfor- performance of the
mance of non-orthogonal systems employing an SIC-BP algorithm also
SIC-based detector, such as the BP algorithm. is determined by the
initial inference accu-
Pattern Division Multiple Access racy of the transmitted
We first introduce some notations for the PDMA
scheme, where K users can non-orthogonally share symbols involved with
N(N < K) orthogonal radio resource elements, a the iterative detection
chip for the code-division multiple access (CDMA) process.
system, and a subcarrier for the orthogonal fre-
quency-division multiple access (OFDMA) system.
The overload factor, which is the ratio of the num-
ber of users to the total number of utilized physi-
cal resource elements, is defined as a = K/N. The
pattern matrix of the PDMA is defined as S = [s1,
s2, ,sK], where sk = [s1k, s2k, sNk]T denotes the
pattern for user k. The set of positions of non-zero
elements in the nth row of the pattern matrix S
denotes the set of users that contribute their data
at the physical resource element. In addition, the
pattern matrix S consists of groups of user pat-
terns with the same number of non-zero entries.
The design philosophy of the PDMA scheme is
that user signals are judiciously allocated in a spe-
cific physical resource space (frequency, code, or
spatial domain) at the transmitter, which can be
effectively exploited to enhance the performance
of SIC-based detectors at the receiver. More spe-
cifically, the data of different users should exhibit
appropriate diversity disparity at the symbol level
and power disparity at the physical resource ele-
ment level. Such disparities are expected to intro-
duce a convergence-amenable characteristic that
can be fully exploited by the SIC-based detector
in eliminating multi-user interference as well as
retrieving transmit diversity at the receiver.
Inspired by the properties of the SIC-based
detector discussed earlier, we present a non-or-
thogonal PDMA scheme where the correspond-
ing pattern matrix has the following three features:
1. The number of groups having different num-
bers of non-zero elements in the pattern matrix
is maximized.
2. The interference among the user patterns in the
same type group is minimized.
3. The size of each group is maximized to the
degree allowable by the computational com-
plexity constraints (further detailed earlier).
The maximum number of supported users K
for the PDMA scheme with N orthogonal physical
resource elements is given by K = CN1 + CN2 + … +
CNN = 2N – 1, where CNn denotes the number of all
n-combinations of a set N.
Depending on whether the user’s data is sent
consecutively or in a distributed manner, we pro-
pose a distributed-mapping-based PDMA and a
localized-mapping-based PDMA, respectively, as
below.
55
The design philosophy
of the PDMA scheme
is that user signals are
judiciously allocated
in a specific physi-
cal resource space
(frequency, code, or
spatial domain) at the
transmitter, which can
be effectively exploit-
ed to enhance the
performance of SIC-
based detectors at the
receiver.
1 The illustration of PDMA
with N = 2 and N =3 in the
manuscript is mainly due
to their easy adaptability
to one physical resource
block (PRB) occupying 12
subcarriers by 7 orthogonal
frequency domain multiplex-
ing (OFDM) symbols, which
is specified in the 4G system.
The extension of the PDMA
scheme for N larger than 3 is
straightforward.
56
Distributed-Mapping-Based PDMA: First highest diversity order of 3, so the preliminary
we design a distributed-mapping-based PDMA inference of x1 is the most reliable among all user
scheme. We illustrate these three design features symbols. Then the accuracy of inference estima-
of the PDMA as presented above using the fol- tion of x1 will propagate in the iterative SIC-based
lowing two PDMA matrices in the frequency detection process. As a result, the error propaga-
domain for N = 2 and N = 3 as follows:1 tion can be significantly alleviated in the iterative
detection of symbols of other users. The symbols
⎡ ⎤
⎢ ⎥ x 2, x 3, and x 4 with a lower diversity order can
⎢ frequency
⎢
user1 user2 user3 ⎥
⎥
benefit from accurate inference of the previously
S(2×3)
dm = ⎢ fi 1 2 0 ⎥ detected symbol x1 with a higher diversity order.
⎢ ⎥
⎢
fi+d 1#
!" 0 #"###
!## 2 $ ⎥ The resulting reliable estimates of the correctly
⎢ group1 with size of 1(=C22 )
group 2 with size of 2 =C21 ( ) ⎥ detected symbols can, in turn, aid in the detection
⎢⎣ ⎥⎦
of previously detected symbols, thus leading to
more accurate detection performance and faster
⎡
⎢ frequency user1
user2 user3 user4 user5 user6 user7 ⎤
⎥
convergence of the iterative algorithm.
⎢
⎢ fi
1 3/2 3/2 0 3 0 0 ⎥
⎥ Similar phenomena can also be observed for
S(3×7)
dm = ⎢
⎢ fi+d
1 3/2 0 3/2 0 3 0 ⎥
⎥ the localized-mapping-based PDMA with pattern
1# 0 #3"
/ 2####
3 /$
2 0 #"0###
3
(3×7)
matrix Slm .
⎢ fi+2d !" !### !## $ ⎥
⎢
⎣
group1 with size of 1(=C33 )
( )
group 2 with size of 3 =C32 ( )
group 3 with size of 3 =C31 ⎥
⎦
Remark 1: The structural irregularity embodied
where fk represents the subcarrier k, and d is the in an appropriate diversity disparity at the sym-
subcarrier spacing between subcarriers where bol level and power disparity at the resource ele-
two PDMA encoded symbols are sent. When ment level can facilitate the convergence for the
d is a sufficiently large value, say d = 512 for low-complexity SIC-based receiver. The diversity
the 4G system with 15 kHz subcarrier spacing gains obtained in the iterative SIC process can
over the extended pedestrian A (EPA) channel, be leveraged to increase the transmission rate for
S(2×3)
dm and S(3×7)
dm correspond to the distribut- NOMA, so as to achieve higher spectral efficiency
ed-mapping-based design. When d = 1, S(2×3) dm than conventional OMA schemes.
and S(3×7)
dm degenerate into the localized-map-
ping-based design. Receiver Design for PDMA
The PDMA scheme is designed so that each BP-Based MUD: In this section, we describe
group (which is composed of users patterns the BP-based algorithm [12], which can effectively
with the same number of non-zero elements exploit the SIC-amenable structure of the PDMA
in the pattern matrix S) has a different number scheme to obtain near-optimal MUD. BP is an
of non-zero entries (within each group); that is, efficient iterative message passing algorithm for
the diversity orders of users’ data in different computing the marginal a posteriori distributions,
groups are different. Taking the pattern matrix which is designed on the factor graph (FG) of the
S(3×7)
dm as an example, the users defined by the underlying Bayesian inference networks [12]. Fig-
pattern matrix S(3×7)
dm can be categorized into ure 1 illustrates the FG of the PDMA scheme with
three groups, and the users belonging to different pattern matrix S(3×7)dm , where the FG is a bipar-
groups, each having a different diversity order. tite graph containing two types of nodes: variable
Specifically, group 1 consisting of user 1 has the nodes (VNs) and function nodes (FNs). In Fig. 1,
highest diversity order of 3, while group 2 with each VN xk (representing a user) is denoted by
users 2–4 has a 2-fold diversity, and users 4–7 in a circle, while each FN yn (representing a phys-
group 3 all have the lowest diversity order of 1. ical resource element) is illustrated by a square,
The overload factors for the PDMA schemes and d fn denotes the number of connected VNs
with S(2×3)
dm and S(3×7)
dm are 150 and 233 percent, for FN yn, e.g., d fn = 4, ∀n of S(3×7)
dm . The messag-
respectively. es are updated by iteratively exchanging them
Localized-Mapping-Based PDMA: For the between FNs and VNs along the respective edges
localized mapping approach, we can exploit (representing the non-zero element of the PDMA
the correlation between adjacent physical radio matrix). When the FG contains no loops, the BP
resource elements and design the PDMA matrix algorithm can be used to perform exact inference
with quasi-orthogonal property to mitigate the for each symbol after a sufficient number of itera-
multi-user interference. Based on this observa- tions [12].
tion, we design a localized-mapping-based PDMA Operating on the FG of the PDMA scheme,
scheme with N = 3 as follows: the BP algorithm iteratively approximates the glob-
⎡
user1 user2 user3 user4 user5 user6 user7 ⎤ al MAP detection by factorizing it into a prod-
uct of simpler local observations. When the FG
⎢ frequency ⎥
⎢ 1 3/2 3/2 0 3 0 0 ⎥
⎢ fi ⎥
S(3×7)
lm = ⎢
⎢ fi+1
1
1#
− 3/2 0 − 3/2 0 3 0 ⎥
⎥
contains cycles, it may lead the BP algorithm to
0 − "
3####
/ 2 3 /$
2 0 0 ### 3$ ⎥
converge to imprecise conditional distributions
!"
⎢ fi+2 !#### !### "
⎢ group 1 with size of 1( )=C33
( )
group 2 with size of 3 =C32 group 3 with size of 3( =C31 ) ⎥
⎣ ⎦
or, more critically, to diverge. PDMA consists of
groups of users with different diversity orders at
As can be seen from this example, user 1 is the symbol level and different power levels at the
orthogonal to users 2–4, and it also exhibits low resource element level. The structural irregularity
correlation with users 5–7. Users 2–4 have an of the PDMA pattern matrix is beneficial for initi-
overall higher correlation than user 1. Users 5–7 ating the convergence of the iterative detection,
experience the largest average interference level especially for the most difficult equi-powered case.
among all users. As shown in [12], the computational complex-
max
For the PDMA scheme with the distributed ity of the BP-based MUD is O(Xdf (S)), where
mapping based pattern matrix S(3×7)
dm , the sym- d max n
f (S) = max d f (S)
bol x1 (xk denotes the symbol of user k) has the 1<n≤N
IEEE Wireless Communications • April 2018
denotes the maximum row weight of S and |X|
Variable nodes
represents the size of the modulation order,
which is considerably lower than O( X K) of the x1 x2 x3 x4 x5 x6 x7
optimal MAP MUD.
Turbo BP for MUD Enhancement: We can fur-
ther enhance the performance of the BP-based
MUD by combing the BP detector with the chan-
nel decoder to form an outer-loop turbo BP receiv-
er structure. Figure 2 illustrates that two outer-loop
iterations (labeled Outer-it) can enhance the d1f
link-level performance for a PDMA system with
300 percent overload by about 2.1 dB at block
error rate (BLER) of 10–2 over the BP-based MUD y1 y2 y3
without outer-loop iteration (i.e., Outer-it = 0).
Function nodes
Constellation-Constrained Capacity Analysis for PDMA FIGURE 1. Factor graph representation of the PDMA scheme with pattern
We analyze the CC capacity to illustrate the matrix S(3×7)
dm .
achievable sum-rate for PDMA in the MAC. For a
K-user MAC, the effective received symbol vector
y at the base station is given by y = H ⊙ Sx + n = out loss of generality. As an illustrative example,
Hefx + n, where H = [h1, h2, …, hK]NK denotes we utilize two PDMA pattern matrices S(2×3)dm and
the channel matrix for all users, the entry hn,k, ∀n S(4×6)
dm (shown below) with the same overload
 {1, 2, … , N}, ∀k  {1, 2, … , K} is assumed to ratio of 150 percent to explain the design of a
be independent and identically distributed (i.i.d.) PDMA scheme.
complex Gaussian random variable with zero
mean and unit variance, Hef = H ⊙ S denotes the ⎡ ⎤
effective channel matrix for all K users, x = [x1, x2, ⎢ user1 user2 user3 ⎥
user4 user5 user6
…, xK]T refers to the transmitted symbol vector of ⎢ ⎥
all K users with normalized power E[ xK 2] = 1, ⎢ 4/3 0 4/3 2 2 0 ⎥
⎢ ⎥
⊙ denotes the element-wise Hadamard product (4×6) ⎢
S dm = 4/3 4/3 0 2 0 2 ⎥
of two matrices, and n ∼ CN (0, s 2I) is the noise ⎢ ⎥
vector. For illustrative purposes, we write the ⎢ 4/3 4/3 4/3 0 2 0 ⎥
⎢ ⎥
received signal vector y for the PDMA scheme 0
⎢ !### 4"/ 3###
4$/3 !0## #" 0###$2 ⎥
with pattern matrix S(2×3)
dm as follows: ⎢⎣ group1 group1 ⎥⎦

⎡ x
1
⎤ We compute the CC sum-rates of pattern
⎡ h11 h12 h13 ⎤ ⎡ 1 2 0 ⎤⎢ ⎥ ⎡ n1 ⎤
y=⎢ ⎥⊙⎢ ⎥ ⎢ x2 ⎥+⎢ ⎥. matrices S(2×3)
dm and S(4×6)
dm as shown in Fig.
⎢⎣ h21 h22 h23 ⎥ ⎢ 1
⎦ ⎣ 0 2 ⎥⎦ ⎢ ⎥ ⎢⎣ n2 ⎥⎦ 3, where the CC sum capacity of OMA is also
⎢⎣ x3 ⎥⎦
depicted for comparison. We can ascertain that
(1)
⎡ h x + h 2x ⎤ ⎡ n ⎤ all PDMA matrices can achieve a maximum sum-
⎢ 11 1 12
=Utilizing
2 ⎥
+⎢
1
⎥.
⎢ h x + hthe2 xchain
⎥ ⎢ rule
n2 ⎥ from the information rate of 3 b/s/Hz in the high signal-to-noise ratio
21 1 23 3 ⎣
theory,
⎣ we can express⎦ the⎦ sum of the CC capac- (SNR) regime, indicating a 50 percent throughput
ity as follows: increase compared with the conventional OMA (2×3)
scheme. The average CC sum-rate of the S dm
I (x, y) = I(x1 : y) + I(x2 : yx1) + I(x3 : yx1, x2), (2) is only 3 percent less than that of the S(4×6) in
dm
the low- and intermediate-SNR range due to less
where I (x, y) = H (y) − H (y x1 ), H (y) average diversity order of users.
1 For PDMA with pattern matrices S(2×3)dm and
=− ∫ p(y)log2 d y, p(y) = 3
Xk
∑ p(y x), S(4×6)
dm , we have d f = 2 and d f = 4, respective-
∏ k=1 x ly. The computational complexity of PDMA with
S(2×3)
dm is O(X2) compared with O(X4) of the
and  XK denotes the size of modulation order PDMA with S(4×6)dm . Although the PDMA with
of the kth user, which is assumed to be 4 (i.e., S(4×6)
dm achieves slightly higher CC sum-rate than
quadrature phase shift keying [QPSK] constella- the PDMA with S(2×3)dm
, its computational com-
tion is considered) for all users in this article with- plexity is about 16 times higher for MUD when

Hard
decision
LLRs LLRs
y in out
BP based Channel
MUD decoder 10-1
BLER

A priori
LLR -
Constellation + Outer-it = 0
probability Outer-it = 1
calculator 10-2 Outer-it = 2
Extrinsic LLR: log-likelihood ratio Outer-it = 3
LLRs
2 4 6 8 10 12
SNR (dB)

FIGURE 2. Turbo BP receiver structure for the PDMA and its performance.

IEEE Wireless Communications • April 2018 57

The structural
3
irregularity of the
PDMA pattern matrix 2.5

CC sum-rate (b/s/Hz)
is beneficial for initi­ 10-1
2
ating the convergence

BLER
of the iterative detec- 1.5 10-2 PDMA, 150%
(46)
tion, especially for the PDMA with Sdm PDMA, 233%
PDMA with Sdm(23) PDMA, 300%
most difficult 1 OFDMA
Conventional OMA 10-3
equi-powered 0 5 10 15 20 25
-2 0 2 4 6
case. SNR (dB) SNR (dB)

FIGURE 3. Comparison of CC sum-rate of the FIGURE 4. BLER performance of the PDMA with
PDMA scheme and the OMA scheme. different overload factors in the uplink.

QPSK is employed; thus, it is less attractive for
practical applications. Therefore, a PDMA scheme
with pattern matrix S(2×3)
dm achieves an attractive
performance-complexity tradeoff to realize the
required overload of 150 percent.
C -C
omplexity onstrained apacity chieving
NOMA D
esign rinciple
Assuming Gaussian MAC (GMAC), the sum-rate
of the NOMA system can be expressed as
1 ⎡ ⎛ 1 H⎞
⎤ narios, QPSK was employed, and the maximum
Rsum = log ⎢det ⎜ I + 2 SS ⎟⎥.
N ⎣ ⎝ σ ⎠⎦
R sum achieves the upper bound of the GMAC
when SS H = KI. The BP-based receiver can
achieve near-perfect interference cancellation for
properly designed S and the associated computa-
tional complexity is
⎛ d max (s) ⎞
O⎜ X f ⎟
⎝ ⎠
[12]. Therefore, it is highly desirable to design that the PDMA approaches that of the 150 percent
fulfills: 1) SSH = KI and 2) exhibits low dfmax(s) as case, indicating a near-perfect interference cancel-
much as possible. We can thus rigorously summa-
rize these two design targets as follows:
min d max
f (S)
s (3) over OFDMA is explained as follows: The PDMA
2
s.t. SS H = KI, d max
f (S) < K, s k = N.
The PDMA matrix provided in work exhibits
good performance-complexity tradeoff as demon-
strated later. However, the approach to the com-
plexity-constrained capacity-achieving NOMA
design defined in Eq. 3 still remains an open prob-
lem.
PDMA for Massive Connectivity Applications
A key performance indicator (KPI) for 5G is the
ability to support massive connectivity with a large
number of devices such as smartphones, tablet
computers, and IoT devices. In this section, we
first provide the link-level simulation of a PDMA
system with different overload factors. We then
carry out the system-level simulation to illustrate
2 We applied localized its advantages over conventional OMA schemes
mapping for OFDMA since for massive connectivity applications.
it closely resembles that
of the localized mapping
DFT-spread OFDM. The
S F
upport of lexible and arge verload
distributed-mapping of the For a PDMA system, the maximum overload fac-
PDMA thus exhibits better tor increases as the length of pattern N increases.
frequency diversity. For the number of orthogonal radio resource ele-
ments N = 2, 3, and 4, the maximum overload
factor can be 150, 233, and 375 percent, respec-
tively. Therefore, by varying N, we can accommo-
date a flexible overload for versatile applications.
We evaluated the performance of the PDMA
via a link-level simulation. The ITU Urban Macro
C -A
(UMa) channel with a 2 GHz carrier frequency
P was adopted. The maximum Doppler frequency,
f d, was set to 5.55 Hz, which corresponds to 3
km/h at the carrier frequency of 2 GHz. A perfect
channel state information is assumed. In all sce-
throughput of each user was assumed to be the
same. Figure 4 illustrates the curves of average
BLER of all users of PDMA with the overload fac-
tors 150, 233, and 300 percent (the first 12 col-
umns of S(4×15)
dm ), respectively. The BLER results of
the OMA-based 4G system were also provided as
a baseline.
It is shown from Fig. 4 that the BLER perfor-
mance of the PDMA degrades as the overload
factor increases in the intermediate SNR region.
As the SNR increases, the BLER performance of
lation for PDMA even with a high overload. The
maximum network throughput can be increased
by 200 percent (this is not shown due to space
constraints). The better performance of PDMA
users exhibit either better diversity order (for
those with df>1) or better frequency diversity than
that of OFDMA.2
R
obustness in attern ollision ases P C C
Machine type communications (MTC) are nor-
mally battery powered, and low power con-
sumption is essential for its implementation. The
excessive transmission delay and large signaling
overhead in the current scheduling-based grant
access mechanism are too expensive for low-cost
MTC equipment. To address this issue, the con-
tention-based grant-free access can be applied
to substantially reduce the transmission latency
and signaling overhead by eliminating the conven-
tional “request-and-grant” procedure. However,
such an “arrive-and-send” mechanism will inevi-
tably introduce collisions among users. It is thus
essential to design a multiple access scheme with
tolerance to multi-user collisions, which can fortu-
L O
nately be realized by the PDMA scheme due to
its convergence-amenable property.
In Fig. 5, we evaluate the link-level perfor-
mance of PDMA with user pattern collision. The

58 IEEE Wireless Communications • April 2018


Collision
No collision
10-1
BLER
10-2
-2 -1 0 1 2 3 4 5
SNR (dB)
FIGURE 5. Performance of user collision.
ideal case is no user collision, while user collision
happens when any two users randomly select the
same user pattern. It is shown in Fig. 5 that the
performance degradation due to pattern collision
is about 0.25 dB at the BLER of 10 -2, which is
acceptable for practical applications.
Support of Massive Connectivity in
Contention-Based Scenarios
In this subsection, we present a system-level sim-
ulation of the potential gains of PDMA over the
OFDMA scheme (currently used by 4G) in con-
tention-based scenarios. We consider an applica-
tion scenario for small packet transmission with
tight latency constraints. We employ a 19-hex-
agonal macrocell model with 3 sectors per cell.
The cell radius of each macrocell is set to be
500 m. The locations of the users are randomly
assigned with uniform distribution. The system
bandwidth is set to 10 MHz, and the transmis-
sion power of the macrocell is 46 dBm. The
antenna gains at the macrocell and user equip-
ment (UE) are 17 dBi and 0 dBi, respectively.
The contention region is set to be 6 resource
block (RB) pairs. Uplink traffic for each user fol-
lows Poisson distribution with a mean packet
inter-arrival time of 120 ms per user. The turbo
BP receiver is employed for PDMA, while a lin-
ear MMSE receiver is employed for OFDMA.
The system performance is evaluated in terms
of outage probability, where the system outage
is defined as the user’s packet drop rate being
larger than 1 or 5 percent.
Figure 6 illustrates that PDMA can support 116
users for system outage of 5 percent, while con-
ventional OFDMA can only accommodate 46
users; that is, about 1.5 times more users can be
supported by PDMA. For a system outage of 1
percent, PDMA is able to achieve a more prom-
inent performance gain; that is, about 2.2 times
more users can be accommodated by PDMA.
Thus, PDMA demonstrates an obvious advantage
over OFDMA in terms of supported number of
users while achieving the same system outage
performance.
Conclusions
In this article, we first introduce the complexi-
ty-constrained capacity-achieving NOMA design
principle. Then a new NOMA scheme named
PDMA was devised. The PDMA scheme is based
on a joint transmitter and receiver design that
facilitates low-complexity SIC-based MUD with
substantially improved performance over con-
ventional OMA schemes. The PDMA scheme is
flexibly designed to accommodate various over-
IEEE Wireless Communications • April 2018
Numerical results from
120
OFDMA link-level and sys-
PDMA
Number of supported users
100
tem-level simulations
80 illustrate that PDMA is
60 a promising candidate
40 technique for 5G mul-
20
tiple access due to it
being able to triple the
0
1% outage 5% outage overall system through-
System outage
put while keeping a
FIGURE 6. Number of supported users by PDMA link performance
and OFDMA under different outage probability close to orthogonal
constraints. transmissions.
loads and is thus suitable for diverse applications.
Furthermore, PDMA exhibits robust collision tol-
erance and is amenable to grant-free scenarios,
which is essential for IoT applications. Numerical
results from link-level and system-level simulations
illustrate that PDMA is a promising candidate
technique for 5G multiple access due to it being
able to triple the overall system throughput while
keeping a link performance close to orthogonal
transmissions.
Acknowledgment
This research is supported by the China
Mobile Research Institute with grant number
R170001240.
References
[1] L. Dai et al., “Non-Orthogonal Multiple Access for 5G: Solu-
tions, Challenges, Opportunities, and Future Research Trends,”
IEEE Commun. Mag., vol. 53, no. 9, Sept. 2015, pp. 74-81.
[2] T. Cover, “Broadcast Channels,” IEEE Trans. Info. Theory, vol.
18, no. 1, Jan. 1972, pp. 2–14.
[3] P. Wang, J. Xiao, and P. Li, “Comparison of Orthogonal and
Non-Orthogonal Approaches to Future Wireless Cellular Sys-
tems,” IEEE Vehic. Tech. Mag., vol. 1, no. 3, Sept. 2006, pp. 4–11.
[4] S. Yuya et al., “Non-Orthogonal Multiple Access (NOMA)
for Cellular Future Radio Access,” Proc. IEEE VTC-Spring
2013, May, 2013, pp. 1–5.
[5] K. Higuchi and A. Benjebbour, “Non-Orthogonal Multiple
Access (NOMA) with Successive Interference Cancellation
for Future Radio Access,” IEICE Trans. Commun., vol. E98-B,
no. 3, Mar. 2015, pp. 403–14.
[6] B. Wang et al., “Spectrum and Energy Efficient Beamspace
MIMO-NOMA for Millimeter- Wave Communications Using
lens Antenna Array,” IEEE JSAC, vol. 35, no. 10, Oct. 2017,
pp. 2370–82.
[7] X. Dai et al. “Successive Interference Cancelation Amenable
Multiple Access (SAMA) for Future Wireless Communica-
tions,” Proc. IEEE ICCS, Nov. 12–15, 2014, pp. 222–26.
[8] S. Chen et al., “Pattern Division Multiple Access-A Novel
Nonorthogonal Multiple Access for Fifth-Generation Radio
Networks,” IEEE Trans. Vehic. Tech., vol. 66, no. 4, Apr.
2017, pp. 3185–96.
[9] A. J. Viterbi, “Very Low Rate Convolutional Codes for Max-
imum Theoretical Performance of Spread-Spectrum Multi-
ple-Access Channels,” IEEE JSAC, vol. 8, no. 3, Mar. 1990,
pp. 641–649.
[10] N. Jindal and A. Goldsmith, “Dirty Paper Coding vs. TDMA
for MIMO Broadcast Channels,” IEEE Trans. Info. Theory, vol.
51, no. 7, May 2004, pp. 1783–94.
[11] X. Wang and H. V. Poor, “Iterative (Turbo) Soft-Interference
Cancellation and Decoding for Coded CDMA,” IEEE Trans.
Commun., vol. 47, no. 7, July 1999, pp. 1046–61.
[12] F. R. Kschischang and B. J. Frey, “Iterative Decoding of
Compound Codes by Probability Propagation in Graphical
Models,” IEEE JSAC, vol. 16, no. 2, Feb. 1998, pp. 219–30.
[13] G. Caire and S. Shamai (Shitz), “On the Achievable
Throughput of A Multi-Antenna Gaussian Broadcast Chan-
nel,” IEEE Trans. Info. Theory, vol. 49, no. 7, July 2003, pp.
1691–1706.
[14] Z. Ding, P. Fan, and H. V. Poor, “Impact of User Pairing
on 5G Non-Orthogonal Multiple Access Downlink Trans-
missions,” IEEE Trans. Vehic. Tech., vol. 19, no. 8, 2015, pp.
1462–65.
59
 [15] F. Kschischang, B. Frey, and H. Loeliger, “Factor Graphs
and the Sum-Product Algorithm,” IEEE Trans. Info. Theory,
vol. 47, no. 2, Feb. 2001, pp. 498–519.
Biographies
Xiaoming Dai is currently a professor in the Department of Tele-
communications at the University of Science and Technology
Beijing (USTB), China. His research interests expand on modula-
tion and coding, space-time coding, signal processing, and code
designs. His research activity has led to numerous publications
in leading international journals and to fruitful industrial applica-
tions, most notably the pattern division multiple access (PDMA)
scheme.
Zhenyu Zhang is a Ph. D. student, majoring in information and
communication engineering at USTB. His research interests are
signal detection in massive MIMO systems and non-orthogonal
multiple access.
60
Baoming Bai is a professor with the State Key Laboratory of
Integrated Services Networks, School of Telecommunication
Engineering, Xidian University, China. His research interests
include information theory and channel coding, wireless com-
munication, and quantum communication.
S hanzhi C hen is the director of the State Key Laboratory of
Wireless Mobile Communications and is a board member of
Semiconductor Manufacturing International Corporation. He
has devoted his work to the research and development of
TD-SCDMA third-generation industrialization and TD-LTE-Ad-
vanced fourth-generation standardization.
Shaohui Sun received his Ph.D. degree in communication and
information systems from Xidian University in 2003. Since January
2011, he has been the chief technical officer of Datang Wireless
Mobile Innovation Center, Datang Telecom Technology and Indus-
try Group. His current research areas of interest include multiple
antenna technology, heterogeneous wireless networks, and relays.
IEEE Wireless Communications • April 2018