Received February 8, 2017, accepted March 6, 2017, date of publication March 17, 2017, date of current version April

24, 2017.
Digital Object Identifier 10.1109/ACCESS.2017.2683531

Performance Analysis of Cooperative Pattern

Division Multiple Access (Co-PDMA)
in Uplink Network
WANWEI TANG1,2,4 , SHAOLI KANG2 , AND BIN REN2,3
1 School of Electronic and Information Engineering, Beihang University, Beijing 100191, China
2 State Key Laboratory of Wireless Mobile Communications, China Academy of Telecommunications Technology, Beijing 100191, China
3 School of Information and Communication Engineering, Beijing University of Posts and Telecommunications, Beijing 100876, China
4 Intelligence and Information Engineering College, Tangshan University, Tangshan 063000, China

Corresponding author: W. Tang (tww1008@163.com)

This work was supported by the National High Technology Research and Development Program of China (863 Program)
under Grant 2015AA01A709.

ABSTRACT As a promising candidate non-orthogonal multiple access scheme for the fifth generation (5G)
wireless communication system, pattern division multiple access (PDMA) has received considerable atten-
tion recently. Using pattern matrix (G[2,3]
PDMA , as an example, is used in this paper), PDMA directly maps
the information bits of different users to radio resources, such as code, power, time and frequency, and
space resource. PDMA can meet the requirements of massive connectivity and higher spectral efficiency
for the 5G mobile network. In order to further improve transmission reliability and enhance the coverage,
an uplink cooperative PDMA (Co-PDMA) scheme with half-duplex decode and forward relay is proposed.
The analytical expressions of outage probability (OP) and sum data rate are derived to characterize the
performance of the proposed scheme. The results show that the proposed Co-PDMA scheme achieves
superior outage performance, with gains of 8 and 12 dB over the non- Co-PDMA and cooperative orthogonal
multiple access (Co-OMA), respectively, @ OP = 0.1. Besides, the scheme is found to outperform two other
schemes in terms of higher sum data rate and achieve an almost maximum 50% gain over Co-OMA when
every user has the same target data rate.

INDEX TERMS PDMA, DF relaying, outage probability, uplink.

I. INTRODUCTION wireless transmission difficult, especially for the users at

As a novel multiple access scheme, pattern division multiple
access (PDMA) is considered as a promising candidate for
for the fifth generation (5G) wireless communication system
due to its superior spectral efficiency [1]. In [2], basic con-
ception, framework, key technologies, system design and per-
formance evaluation were presented for PDMA. Advanced
receivers and pattern design methods were studied for PDMA
in [3] and [4], respectively. In addition, some researchers
conduct PDMA theoretical analysis and performance eval-
uation [5], [6] and the simulation results show that PDMA
can significantly improve spectral efficiency compared
with orthogonal frequency division multiplexing access
(OFDMA). At present, PDMA is being widely discussed at
Release 14 of 5G new radio (NR) in 3rd Generation Partner-
ship Projec (3GPP) [7], [8].
In 5G, close to 100% reliability is proposed. Time-varying
and frequency-selective fading phenomena make reliable
the edge of cell. Transmit diversity based on Multiple-Input
Multiple-Output (MIMO) is an effective technique to combat
the fading effect. Unfortunately, the use of multiple antennas
in uplink communications is restricted due to the limitation
of size and complexity of the mobile equipment. Relay-based
cooperative communication can expand the cell coverage area
and increase the network capacity without adding complexity
of multiple antennas [9], [10]. Wireless relaying techniques
have been received a considerable attention for improving
coverage and performance. There are mainly two different
cooperative protocols: amplify and forward (AF) and decode
and forward (DF). AF relay is easy to implement but can-
not achieve high performance gain since it just amplifies
the incoming signal and forwards it to the destination with-
out decoding. Therefore, the fading and white noise in the
channel are also amplified by AF relay. Compared with AF,
DF decodes the incoming signal, re-encodes it, and then

2169-3536
2017 IEEE. Translations and content mining are permitted for academic research only.
3860 Personal use is also permitted, but republication/redistribution requires IEEE permission. VOLUME 5, 2017
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
W. Tang et al.: Performance Analysis of Co-PDMA in Uplink Network

retransmits it to the destination, which can further improve

outage performance and reliability [11].

A. MOTIVATION AND CONTRIBUTIONS

The aforementioned literatures have made a lot of research
works on PDMA, however, to the best of our knowledge,
few contributions on cooperative PDMA are available yet.
In this paper, we propose an uplink cooperative pattern
division multiple access (Co-PDMA) scheme with half-
duplex DF relay. The proposed cooperative diversity tech-
nique and the diversity gain for all transmit signals from
the source using only one half-duplex relay. Data streams
are transmitted simultaneously through both direct and relay
channels. At the destination, a successive interference can-
cellation (SIC) algorithm without an ordering process is used.
Simulation results show that the proposed Co-PDMA scheme
can achieve superior outage performance compared with the
FIGURE 1. Cooperative PDMA uplink network with DF relay.
non-cooperative PDMA and cooperative orthogonal multiple
access (Co-OMA). The primary contributions of this paper
are summarized as follows:
1) A Co-PDMA scheme model is proposed and the as follows:
closed-form expressions of outage probability (OP) 1
ρ = , (1)
for the users in Co-PDMA are derived. In order to α1 /α2 + φ1 + α1 /α3
make comparison, the corresponding OP expressions
for non-cooperative PDMA and Co-OMA are also (φ1 + α1 /α3 )(α1 /α2 + φ1 )
β = ln , (2)
given in this paper. From the analysis results, we can (α1 /α2 )(α1 /α3 )
observe that the proposed Co-PDMA can obtain signif-
a = (α3 + α2 ) α1 γ 2 − α3 α2 γ 2 φ1
icant gain in terms of OP performance compared with
the other two schemes. b = 2α1 γ − (α2 + α3 ) φ1 γ
2) Additionally, we analyze the sum data rates of the three
schemes, i.e., Co-PDMA, non-cooperative PDMA and c = −φ1
Co-OMA, analysis results indicate that the proposed
1 = b2 − 4ac
Co-PDMA can achieve higher sum data rate than the √ √
other two schemes. The correctness of all the theoretic −b − 1 −b + 1
x1 = , x2 = . (3)
analysis is verified by Monte Carlo simulation. 2a 2a

B. ORGANIZATION II. SYSTEM MODEL

The rest of this paper is organized as follows. Section II
describes the Co-PDMA system model with a half-duplex
DF relay over Rayleigh fading channels. Section III derives
the exact OP closed-form expressions for the users in the
three schemes. Section IV provides the theoretical anal-
ysis and comparisons in terms of sum data rate for the
three schemes. Numerical results are presented in Section V
for verifying our analysis and followed by our conclusions
in Section VI.
Notation: Overall this paper, we use bold and capital
symbols to denote the matrix and use bold and lowercase
symbols to denote the vector; (•)H , (•)T and (•)−1 denote
the conjugate transpose, transpose and inversion of a matrix,
respectively; H is the channel matrix between transmitter
and receiver. I N is an N × N identity matrix. αi , P, and
Pi = αi P are the power control factor, target arrived power
and transmit power, respectively. γ = P/N0 denotes the
We consider a half-duplex dual-hop communication scenario
for PDMA uplink network in a fading environment with
multiple users, one base station (BS) and one DF relay,
where the direct link is considered. As depicted in Fig. 1, all
the users are divided into several groups by channel quality
indicator (CQI), timing advance, power or SINR information,
each group have users with similar CQI and allocate a PDMA
pattern randomly to each user. The number of users in differ-
ent groups can be changed by service requirement. The users
in different groups are allocated different time & frequency
resources, while the users in the same group will share the
same time & frequency resources. The channels from users
to relay and from relay to BS are frequency non-selective
channels that undergo independent Rayleigh fading. Thus,
the channel gains from user to BS, from relay to BS and
from user to relay can be denoted by hub ij
∼ CN (0, µub ) ,
hij ∼ CN (0, µrb ) and hij ∼ CN (0, µur ), respectively,
rb ur

maximum transmit SNR for each user. In order to simplify where i and j denote the ith resource element (RE) and the
the computing expressions, we do same variable substitution jth user.

VOLUME 5, 2017 3861

 W. Tang et al.: Performance Analysis of Co-PDMA in Uplink Network

For PDMA system, the received signal at the receiver from

the transmitter can be expressed as
K
X
y = H PDMA x + n = diag(hk )gk xk + n
k=1
X
= h̃k xk + h̃i xi + n, i, k = 1, 2, · · · K . (4) FIGURE 2. Three users multiplexing on two REs.
i6=k
| {z }
zk
signal-to-interference-plus- noise ratio (SINR) of the k th user
where is given by
,K ]
H PDMA = H CH  G[N
PDMA SIN Rk = Pk (h̃k )H (K zk )−1 h̃k . (8)
= [h1 h2 . . . hK ]  g1 g2 . . . gK
 
h i In order to simplify the progress of analysis, we take a
= h̃1 h̃2 . . . h̃K typical PDMA matrix,
 
h11 g11 h12 g12 · · · h1K g1K
 
1 1 0
G[2,3] ,
 
 h21 g21 h g · · · h2K g2K PDMA = g g g
1 2 3 = (9)
22 22
 1 0 1
= . . .. ..  (5)
 
 .. .. . .  as an example to derive the closed-form expression of SINR
hN 1 gN 1 hN 2 gN 2 · · · hNK gNK for each user. Fig. 2 shows the pattern matrix and correspond-
indicates PDMA equivalent channel response matrix from ing resource mapping: the information of the three users is
users to BS with dimension N × K ; y is an N × 1 received multiplexed on two REs, and the SOR is K /N = 3/2 =
signal vector on the N resources; x = [x1 x2 · · · xK ]T is 150%. Data from user 1 are mapped onto two REs while data
a K × 1 modulation symbol vector transmitted by K users, from user 2 and user 3 onto RE 2 and RE 3. The column
and xk is modulation symbol of the k th user; n indicates the weight of the three users is 2, 1 and 1, respectively.
All users in a cell are divided into T groups and each
white noise andmolded by n ∼ CN (0, N0 I N ×N ); H CH = group has 3 users. Each user picks one pattern from the
h1 , h2 , · · · , hK is the channel response matrix from users to
BS; ‘‘’’ indicates element-wise dot product of two matrices.
We assume that all the users are located in a homogeneous
environment and the signal from each user to the BS expe-
riences independent identically distributed (i.i.d) frequency
non-selective Rayleigh fading and additive white Gaussian
noise (AWGN). The N ×1 channel vector between the k th user
and the nth RE at the BS is denoted by hk = [h1k h2k · · · hNk ]
matrix randomly. The user picking g1 is named user 1 and
detected firstly, the user picking g2 is named user 2 and
detected secondly and the user picking g3 is named user 3
and detected lastly. PDMA equivalent channel response
matrix from users to BS, from users to relay and from relay
to BS are expressed as
h i 
H ub = H ub  G[2,3] ub ub ub

where, hik is modeled by hik ∼ CN (0, µ); diag(hk ) represents PDMA CH PDMA = h1 h2 h3  g1 g2 g3
a diagonal matrix with elements from hk ; h ub ub ub i  hub hub 0 
= h̃1 h̃2 h̃3 = 11 12 , (10)
  hub 0 hub
g11 g12 · · · g1K h 21
i23 
H ur = H ur  G[2,3] ub ub ub

[N ,K ]    g21 g22 · · · g2K  PDMA = h1 h2 h3  g1 g2 g3
= g1 g2 . . . gK =  .
 PDMA CH
GPDMA . .. .. 

 .. .. . .  h ur ur ur i  hur hur 0 
= h̃1 h̃2 h̃3 = 11 12 (11)
gN 1 gN 2 ··· gNK hur
21
0 hur 23
(6)
and
denotes a PDMA pattern matrix with dimensions of N × K ,
h i 
[2,3]
H rb rb rb rb rb

= H  G = h h h 3  g1 g2 g3
and K /N is defined as system overload rate (SOR); gk is PDMA CH PDMA 1 2
the PDMA pattern used by the k th user; The data of users h rb rb rb i  hrb hrb 0 
= h̃1 h̃2 h̃3 = 11 12 (12)
will be mapped on the resources defined on power, time & hrb
21
0 hrb
23
frequency or spatial domain with PDMA pattern matrix; zk
represents the noise plus interference faced by the k th user. respectively.
The covariance of zk can be explicitly calculated by During the first time slot, the users transmit the signal
with CP-inserted to both relay and BS. According to (4), the
K
X received signals at the BS and relay are expressed as
K zk = N0 I N + Pi h̃i (h̃i )H , (7)
ub X ub
i6=k yub = h̃k xk + h̃i xi + n, i, k = 1, 2, 3. (13)
i6=k
where Pi is the power associated with the ith user. | {z }
According to [12, eqs. (8-67)], the corresponding zub
k

3862 VOLUME 5, 2017

W. Tang et al.: Performance Analysis of Co-PDMA in Uplink Network

and k − 1 users have been successfully detected, so the OP of the

ur X ur k th user can be denoted as:
y ur
= h̃k xk + h̃i xi + n, i, k = 1, 2, 3. (14)
i6=k Pkout = 1 − Pr(Ā1 ∩ . . . ∩ Āk−1 ∩ Āk )
| {z }
k
zur
k
Y
= 1− Pr(Āj ). (20)
During the second time slot, the relay re-modulates the j=1
recovered data, yur , and transmits it to the BS after inserting
Theorem 1: During the first time slot, the information from
CP. At the BS, the received signal from relay is given by
users are transmitted to relay and BS. The OPs of the three
rb X rb
yrb = h̃k xk + h̃i xi + n, i, k = 1, 2, 3. (15) users in a pair group can be expressed as
i6=k 1 1 φ1
| {z } P1ub_out = 1 − exp(− )
zrb
k
1 + φ1 α3 /α1 µub α1 γ
α12 − α 1γ µφ1 ρ ρα3
 
According to (7), we can firstly calculate the covariances of − e 1 ub ρ β − 2
+
interferences (i.e., zub rb α2 α3 φ1 + α1 /α3 α1
k , and zk in (13) and (15), respectively),
then, according to (8), we can obtain each user’s SINRs α1 ρβ − α 1γ µφ1
− e 1 ub , (21)
experiencing different paths (e.g., from users to BS and from α2 α3 γ µub
φ2 φ2
relay to BS): − µ1 − µ1
P2ub_out = 1 − e ub α2 γ + P1ub_out · e ub α2 γ (22)
K
ub X ub ub ub
SIN Rub
k = Pk (h̃k )H (N0 I N + Pi h̃i (h̃i )H )−1 h̃k , and
φ2 φ3
− 1 − 1
i6=k
 
P3ub_out = 1 − 1 − P1ub_out · e µub α2 γ · e µub α3 γ . (23)
(16)
Relevant variable symbols are defined in Section I and
and
Section II.
K
rb X rb rb ub Proof: At the BS, the user 1 is detected firstly. According
SIN Rrb H
k = Pk (h̃k ) (N0 I N + Pi h̃i (h̃i )H )−1 h̃k . (17)
to (16), we have
i6=k
3
The BS processes the received signals from the users in ub X ub ub ub
SINRub H
1 = P1 (h̃1 ) [N0 I N + Pi h̃i (h̃i )H ]−1 h̃1
the first time slot and relay in the second time slot by using i=2
selection combining. Then an outage event occurs if neither  2  2
the direct transmission nor the relaying transmission suc- P1 hub
11
 P1 hub
21

=  2 +  2
ceeds. Therefore, the outage probability of the ith user can N0 + P2 hub  N0 + P3 hub 
12 23
be expressed as  ub 2  ub 2
α1 γ h11  α1 γ h21 
 2 .
 h i 
=  2 + (24)
Piout = Pr max SINRub i , SINRi < φi
rb
    1 + α2 γ hub12
 1 + α3 γ hub23

= Pr SINRub i < φ i P SINR rb
i < φ i , (18) According to (19), we have
ub
where φi = 2r̂i − 1, (i = 1, 2, 3) is the target SINR of the Pr(Ā1 ) = 1 − Pr(SINRub < φ1 )
 1
ith user, r̂i is the target data rate (TDR). We define ri as the

 
actual date rate of the ith user, correspondingly.

 

  ub 2  ub 2 
 α 1 γ h 

α1 γ  h  

 2 < φ1
11 21
= 1 − Pr  2 +
III. OUTAGE PROBABILITY 1+α2 γ h12 1+α3 γ hub

  ub   

23 
A. OUTAGE BEHAVIOR FROM USERS TO BS

| {z } | {z } 

 
Z1 Z2
Define Ak = {rk < r̂k } or Ak = {SIN Rk < φk } as
= 1 − Pr(Z1 + Z2 < φ1 ) (25)
the event that the BS cannot detect the data of the k th user.
Then the probability of the event that the k th user’s data are The cumulative distribution function (CDF) of Z1 is
successfully detected is expressed as expressed as
Pr(Āk ) = 1 − Pr(Ak ) FZ1 (z1 ) = Pr(Z1 < z1 )
z1 (1 + α2 γ x)
Z ∞  
= 1 − Pr(SIN Rk < φk ), (19)  2
= Pr(hub  < )f ub 2 (x)dx
0
11
α1 γ h12 
where Āk is the complementary set of Ak and Pr(Āk ) is the Z ∞ z (1+α γ x)

probability of the event Āk happens. − 1 1 2 1 − µ1 x
= 1 − e µub α1 γ e ub dx
Since the SIC receiver is used at the receiver end, for the 0 µub
k th user, the event that the BS can successfully detect the data 1 z1 1
= 1− exp(− ). (26)
of the k th user must be under the condition that all the front 1 + z1 α2 /α1 µub α1 γ

VOLUME 5, 2017 3863

 W. Tang et al.: Performance Analysis of Co-PDMA in Uplink Network

Similarly, the CDF of Z2 is expressed as and

n o
1 z2 1 ub
FZ2 (z2 ) = 1 − exp(− ). (27) Pr(Ā3 ) = 1 − Pr SINRub3 < φ3
1 + z2 α3 /α1 µub α1 γ   2 
= 1 − Pr α3 γ h23  < φ3
 ub 
Then, we have
Pr(Z1 + Z2 < φ1 ) − µ1
φ3

Z φ1 =e ub α3 γ . (34)
= Pr(Z1 < φ1 − z2 )fZ2 (z2 )dz2 According to (20), we can obtain the OPs of the three users
0
in the selected pair group
φ1   
(φ1 −z2 )
Z
− α 1γ
 
ub
= 1 − ψe 1 µub
fZ2 (z2 )dz2 P1ub_out = 1 − Pr Ā1
0
φ1 Z φ1   = J1 − J2 , (35)
(φ1 −z2 )
Z
− 1
   
ub ub
= fZ2 (z2 )dz2 − ψe α1 γ µub fZ2 (z2 )dz2 , P2ub_out = 1 − Pr Ā1 Pr Ā2
|0 } |0 φ2
− µ1
{z {z }
J1 J2 = 1 − (1 − (J1 − J2 )) · e ub α2 γ (36)
(28) and
     
and J2 in (28) can be further expressed as ub ub ub
P3ub_out = 1 − Pr Ā1 Pr Ā2 Pr Ā3
Z φ1 
(φ −z )

− α 1υ 1 µ 2 − µ1
φ2
− µ1
φ3
J2 = ψe 1 fZ2 (z2 )dz2 = 1 − (1 − (J1 − J2 )) · e ub α2 γ ·e ub α3 γ , (37)
0

α12 − α 1γ φ1

ρ ρα3
 respectively.
= e 1 µub
ρ β−
2
+ The proof is completed.
α2 α3 φ1 + α1 /α3 α1
α1 ρβ − α 1γ φ1
B. OUTAGE BEHAVIOR FROM RELAY TO BS
+ e 1 µub
, (29) During the second time slot, we assume that all users are
α2 α3 γ µub
detected successively at the relay, then the recovered data
where, ρ and β is defined in Section I. Then, (25) is further
relay re-modulates will experience same path to BS (i.e.,
derived as
ij = h , i = 1, 2; j = 1, 2, 3.) So (12) can be further
hrb rb
ub
Pr(Ā1 ) = 1 − Pr(Z1 + Z2 < φ1 ) = 1 − (J1 − J2 ). (30) expressed as
 rb
hrb 0

For user 2, according to (16) and (19), we have h
rb
H PDMA = , (38)
hrb 0 hrb
3
ub X ub ub ub
SINRub H
2 = P2 (h̃2 ) [N0 I N + Pi h̃i (h̃i )H ]−1 h̃2 and the users still pick the same pattern as the processing in
i=3 the first time slot.
H " #−1  Theorem 2: During the second time slot, at the BS, the OPs
hub hub
 
N0 0 of three users can be expressed as
= P2 12  2 12
0 0 N0 + P3 hub 0 
1, if a < 0, 1 < 0 or

23 

a < 0,  1 ≥ 0, x2 < 0
 2 

P2 hub   2 

12
= α2 γ hub

= (31) 1
  

12 
N0 x2 ,

 exp −
µrb



and

if a< 0, 1≥ 0, x2> 0 and  x1 < 0


n o 

ub
2 < φ2

Pr(Ā2 ) = 1 − Pr SINRub 1 1


1 − exp − x1 + exp − x2 ,


µrb µrb
  

 2 
= 1 − Pr α2 υ hub < φ

if a < 0, 1 ≥ 0, x1 > 0
  
2

12 
P1rb_out =
− µ1
φ2  0, if a > 0, 1 < 0 or
.

=e ub α2 γ (32)

a >0, 1 ≥0, x2 < 0




1

For user 3, according to (16) and (19), we have

x2 ,

 1 − exp −
µ


 ub H rb

ub 
SINRub [N0 I N ]−1 h̃3
if a > 0,  1 ≥ 0,  x2 > 0 and
 x1 < 0

3 = P 3 h̃3 



1 1
 2 

P3 hub x2 ,

  exp − x1 − exp −
µrb µrb
23

=



N0 

if a > 0, 1 ≥ 0, x1 > 0,
 2
= α3 γ hub (33) (39)
 
23


3864 VOLUME 5, 2017

W. Tang et al.: Performance Analysis of Co-PDMA in Uplink Network

   
rb rb
P2rb_out = 1 − Pr Ā1 Pr Ā2 and
φ φ
n o
rb
Pr(Ā2 ) = 1 − Pr SINRrb2 < φ2
− µ1 α 2γ − µ1 α 2γ
= 1−e rb 2 + P1rb_out ·e rb 2 (40)
  2 
= 1 − Pr α2 γ h  < φ2
and  rb 
     
rb rb rb
P3rb_out = 1 − Pr Ā1 Pr Ā2 Pr Ā3 φ2
− µ1
 
− 1 2
φ
− 1 3
φ =e rb α2 γ . (46)
= 1 − 1 − P1rb_out · e µrb α2 γ · e µrb α3 γ . (41)
For user 3, according to (17) and (19), we have
Relevant variable symbols are defined in Section I and  rb H rb
Section II. SINRrb
3 = P 3 h̃3 [N0 I N ]−1 h̃3
Proof: For user 1, according to (17) we have  2
3 P3 hrb   2
= = α3 γ hrb  (47)
 
rb X rb rb rb
SINRrb H
1 = P1 (h̃1 ) [N0 I N + Pi h̃i (h̃i )H ]−1 h̃1 N0
i=2
 rb 2  2 and
α1 γ h
  α1 γ hrb 
 2 .
n o
rb
3 < φ3
=  2 + (42) Pr(Ā3 ) = 1 − Pr SINRrb
1 + α2 γ hrb  1 + α3 γ hrb   
 2
According to (19), we have = 1 − Pr α3 γ hrb  < φ3
 
rb
Pr(Ā1 ) − µ1
φ3
(  2  2 ) =e rb α3 γ . (48)
α1 γ hrb  α1 γ hrb 
= 1 − Pr  2 +  2 < φ1 According to (20), we can obtain the outage probabilities
1 + α2 γ hrb  1 + α3 γ hrb 
(   4 ) of the three users in the pair group as follows:
(α3 + α2 ) α1 γ 2 − α3 α2 γ 2 φ1 hrb 
= 1 − Pr  
rb
P1rb_out = 1 − Pr Ā1 = Pr {$ },
 2
+ [2α1 γ − (α2 + α3 ) φ1 γ ] hrb  − φ1 < 0 (49)
= 1 − Pr{$ }. (43)    
rb rb
P2rb_out = 1 − Pr Ā1 Pr Ā2
After discussion under different conditions, we derive
φ2
− µ1


 1, if a < 0, 1 < 0 or = 1 − (1 − Pr {$ }) · e rb α2 γ

a < 0,  1 ≥ 0, x2 < 0

φ2 φ2

− µ1 − µ1

rb α2 γ + P1rb_out · e rb α2 γ



 1 = 1−e (50)
x2 ,

 exp −
µrb

and



if a< 0, 1≥ 0, x2> 0 and  x1 < 0

      
rb rb rb

P3rb_out = 1 − Pr Ā1 Pr Ā2 Pr Ā3


1 1


1 − exp − x1 + exp − x2 ,


µrb µrb φ2 φ3

 1 1
− −
= 1 − (1 − Pr {$ }) · e µrb α2 γ · e µrb α3 γ


if a < 0, 1 ≥ 0, x1 > 0


Pr {$ } = φ φ
− 1 2 − 1 3
 


 0, if a > 0, 1 < 0 or = 1 − 1 − P1rb_out · e µrb α2 γ · e µrb α3 γ . (51)
a >0, 1 ≥0, x2 < 0



The proof is completed.

1


x2 ,

 1 − exp −
µ


rb

C. OUTAGE BEHAVIOR AFTER SELECTION COMBINING

if a > 0,  1 ≥ 0, x2 > 0 and
 x1 < 0



According to the description in the section II, when selection


1 1


x2 , combining is used at BS, an outage event occurs if neither the

 exp − x1 − exp −
µrb µrb


direct transmission nor the relaying transmission succeeds.


if a > 0, 1 ≥ 0, x1 > 0


So we have
(44)  h i 
Pr (A1 ) = Pr max SINRub 1
, SINR rb
1
< φ 1
Similarly, for user 2, according to (17) and (19), we have    
3 = Pr SINR1 < φ1 Pr SINRrb
ub
1
< φ 1
rb X rb rb rb
SINRrb H
2 = P2 (h̃2 ) [N0 I N + Pi h̃i (h̃i )H ]−1 h̃2  ub 2  2
α1 υ h11 α1 υ hub
!
  
 2 < φ1
i=3 12
= Pr  2 +
hrb
 H 
N0 0
−1  rb 
h 1 + α2 υ hub21
 1 + α3 υ hub32

= P2  rb 2 !
0 0 N0 + P3 h
  0 α1 υ|hrb |2 α1 υ|hrb |2
· Pr + < φ1
1 + α2 υ|hrb |2 1 + α3 υ|hrb |2
 rb 2
P2 h   2
= = α2 γ hrb  (45)
 
N0 = (J1 − J2 ) · Pr {$ }, (52)

VOLUME 5, 2017 3865

 W. Tang et al.: Performance Analysis of Co-PDMA in Uplink Network

 h i 
Pr (A2 ) = Pr max SINRub 2
, SINRrb
2
< φ 2 Comparing (53), (54) and (59), we can see that the OP of
    Co-OMA has the similar expression with Pr (A2 ) and Pr (A3 )
= Pr SINR2 < φ2 Pr SINRrb
ub
2
< φ2 since that the patterns of user 2 and user 3 are orthogonal.
For user 1, its signal

T are repeated mapped onto two REs with
  2    2 
= Pr α2 γ h21  < φ2 Pr α2 γ h  < φ2
 ub   rb 
pattern g1 = 1 1 , which can bring diversity gain to obtain
 φ
 φ
 superior OP performance. When SIC receiver is used, the OPs
− µ1 α 2γ − µ1 α 2γ
= 1−e ub 2 1−e rb 2 (53) of user 2 and user 3 are derived as (56) and (57), in which
the user 1 with lower OP (i.e., better OP performance) can
and further reduce the OPs of user 2 and user 3. So we come
 h i 
Pr (A3 ) = Pr max SINRub , SINR rb
< φ into the conclusion that the average OP of the three users in
3 3 3
    Co-PDMA system is superior to that of Co-OMA, which is
= Pr SINR3 < φ3 Pr SINRrb
ub
3
< φ 3 verified by the following simulation.
  2    2 
= Pr α3 γ hub < φ α γ < φ
 rb  IV. SUM DATA RATE
Pr
 
32  3 3 h  3
From the OP expressions derived earlier in this article, we can
φ φ
  
− µ1 α 3γ − µ1 α 3γ see that the OPs of the multiple users are directly proportional
= 1−e ub 3 1−e rb 3 . (54)
to their own targeted SINRs φi , i = o, 1, 2, 3, respectively,
According to (20), after selection combining the outage and also directly proportional to the targeted data rate r̂i
probability of the three users can be expressed as because of φi = 2r̂i − 1, then the achievable sum data rate
for non-cooperative PDMA and Co-PDMA can be shown as
P1out = Pr (A1 )    
= 1 − (J1 − J2 ) · Pr {$ }, (55) Rd_sum = r̂1 1 − P1ub_out + r̂2 1 − P2ub_out
P2out = 1 − Pr Ā1 Pr Ā2


 
+ r̂3 1 − P3ub_out (60)
= 1 − (1 − Pr (A1 )) (1 − Pr (A2 )) (56)
and
and
Rr_sum = r̂1 1 − P1out + r̂2 1 − P2out + r̂3 1 − P3out




P3out




= 1 − Pr Ā1 Pr Ā2 Pr Ā3
(61)
= 1 − (1 − Pr(A1 )) (1 − Pr(A2 )) (1 − Pr (A3 )). (57)
respectively.
Then, the average OP of the three users can be expressed
For the PDMA pattern matrix G[2,3]
PDMA , the data from three
as
users are multiplied on two REs. In order to make fair com-
out
= P3out + P3out + P3out /3.


PPDMA_ave (58) parison between Co-OMA and Co-PDMA transmission, the
achievable sum data rate of Co-OMA on two REs is
In order to make comparison between Co-PDMA and
Ro_sum = 2 · r̂o 1 − Poout ,


conventional Co-OMA, we assure that the same REs carry (62)
the same target rate. For PDMA, there are three users mul-
tiplexing on two REs, so that for each user occupying one where r̂o is the TDR for the conventional Co-OMA.
RE in Co-OMA, its target date rate should satisfy r̂o = Rr_sum is the sum data rate of Co-PDMA and higher than
r̂1 + r̂2 + r̂3 /2. The OP of the user in Co-OMA is given

Rd_sum , the sum data rate of non-cooperative PDMA, since
as [13]: that after relay decoding and forwarding, the OP perfor-
 h i  mance will be further improved (P1out , P2out and P3out are
Poout = Pr max SINRub o , SINR rb
o < φ o lower than P1ub_out , P2ub_out and P3ub_out , respectively) Rr_sum
    is also higher than Ro_sum since that the average OP of
= Pr SINRo < φo Pr SINRrb
ub
o < φo Co-PDMA is lower than that of Co-OMA and the TDR
satisfies 2r̂o = r̂1 + r̂2 + r̂3 .
  2    2 
= Pr αo γ hub < φ α γ < φ
 rb 
Pr
 
o  o o h o  o
V. NUMERICAL RESULTS
− 1 φo − 1 φo
   
= 1 − e µub αo γ 1 − e µrb αo γ , (59) In this section, the performance evaluations of Co-PDMA,
non-cooperative PDMA and Co-OMA in terms of OP and
where φo = 2r̂o − 1 is the target SINR for conventional sum data rate are presented by using Monte Carlo simulation.
Co-OMA. Without loss of generality, we normalize the distance between
From the OP expressions of Co-PDMA and Co-OMA, we all users and the BS to unity and let d be the normalized
can see that the target SINR or TDR have a great impact distance between all users and the relay. As a result, µub = 1
on the OP performance: lower TDR r̂i , i = 0, 1, 2, 3, and µur = d −ι , where ι is the path loss exponent. In the
or lower target SINR φi , i = 0, 1, 2, 3, brings supe- following simulation, we further assume d = 0.5 and ι = 2
rior OP performance according to (55), (56), (57) and (59). (Free-space model is used here) [14].

3866 VOLUME 5, 2017

W. Tang et al.: Performance Analysis of Co-PDMA in Uplink Network
FIGURE 3. Outage performances of Co-PDMA and Co-OMA with different
TDR.
FIGURE 4. Outage performances achieved by three schemes.
The effects of TDR are firstly investigated. The curves for
the OP of Co-PDMA and Co-OMA with different TDRs are
demonstrated in Fig. 3. The exact analysis curves for the
OP of the Co-PDMA and Co-OMA are plotted according
to (58) and (59). Note that an excellent agreement between
the exact analytical results and Monte Carlo simulations is
observed. It is shown that lower target data rate can obtain bet-
ter performance which validate the derived analytical results.
The results in Fig. 3 can also be explained by an actual
communication system since lower target SINR can be more
easily achieved.
In Fig. 4, the average OPs achieved by Co-OMA scheme,
non-cooperative PDMA and Co-PDMA, are shown as a func-
tion of SNR. Both theoretical results and simulation results
are presented. As shown in the figure, Co-PDMA outper-
forms the other two schemes and obtains 8 dB and 12 dB gains
over non-cooperative PDMA and Co-OMA @ OP=0.1.
Fig. 5 compares the achievable sum data rate among afore-
mentioned three schemes. For the three users in PDMA,
each user’s TDR is 0.1. The TDR of every user in Co-OMA
is equivalent to or 1.5 times more than that for users in
Co-PDMA. The exact analysis curves for sum data rate of the
three schemes are plotted according to (60), (61) and (62).
VOLUME 5, 2017
FIGURE 5. Achievable sum data rates of three shcemes.
Clearly Co-PDMA transmission outperforms the Co-OMA
and non-cooperative PDMA. When the TDR of every
Co-OMA user is 1.5 times more, Co-PDMA achieves a sig-
nificant gain at lower SNR region (e.g., below 5 dB). When
the TDR of Co-OMA equals Co-PDMA, Co-PDMA achieves
almost 50% gain over Co-OMA at higher SNR region (e.g.,
above 8 dB) due to the diversity gain. The exact analysis
curves are also match well with the Monto Carlo simulation
curves. It should be noted that the performance improvement
Co-PDMA brings is at the expense of system complexity
increasement (multiple relay stations need to be set up)
and spectral efficiency degradation (duplicate transmission is
conducted).
VI. CONCLUSION
In this paper, we propose an uplink Co-PDMA scheme, in
which PDMA is designed with half-duplex DF relay. The
outage behaviors of the three paired users with PDMA pattern
matrix G[2,3]
PDMA are analyzed and closed-form expressions of
them are derived. In addition, the sum data rates of the three
users in the system are also analyzed. Simulation results show
that Co-PDMA can achieve 8 dB and 12 dB gain over non-
cooperative PDMA and Co-OMA @ OP=0.1 respectively
and achieve higher sum rate than non-cooperative PDMA
and Co-OMA. Under the assumption of same target data rate
for each user, Co-PDMA achieves almost 50% sum data rate
gain over Co-OMA at higher SNR region (e.g., above 8 dB).
It should be point out that the proposed scheme and analytical
method can be further extended to the scenarios for PDMA
[3,6]
pattern matrices with higher overload (e.g., GPDMA with SOR
200% or G[4,12]
PDMA with SOR 300%), where higher sum data rate
gain can be achieved.
REFERENCES
[1] ‘‘Future technology trends of terrestrial IMT systems,’’ ITU-R, Geneva,
Switzerland, Tech. Rep. 524 ITU-R M.2320-0, Nov. 2014. [Online]. Avail-
able: http://www.itu.int/pub/R-REP-M.2320-2014
[2] S. Chen, B. Ren, Q. Gao, S. Kang, S. Sun, and K. Niu, ‘‘Pat-
tern division multiple access (PDMA)—A novel non-orthogonal multi-
ple access for 5G radio networks,’’ IEEE Trans. Veh. Technol., to be
published.
3867
 W. Tang et al.: Performance Analysis of Co-PDMA in Uplink Network

[3] B. Ren et al., ‘‘Advanced IDD receiver for PDMA uplink system,’’ in Proc. SHAOLI KANG was a Project Manager with the
IEEE/CIC Int. Conf. Commun. China (ICCC) Chengdu, China, Jun. 2016, China Academy of Telecommunications Technol-
pp. 1–6. ogy (CATT), focusing on Research and Develop-
[4] B. Ren, Y. Wang, X. Dai, K. Niu, and W. Tang, ‘‘Pattern matrix design of ment of TD-SCDMA, from 2000 to 2005. Then,
PDMA for 5G UL applications,’’ China Commun, vol. 13, pp. 159–173, she was with the Communication Center of Sys-
Jan. 2017. tem Research, University of Surrey, as a Research
[5] J. Zeng et al., ‘‘Pattern division multiple access (PDMA) for cellular future Fellow, where she was involved in projects from
radio access,’’ in Proc. IEEE Int. Conf. Wireless Commun. Signal Process.,
EPSRC and OFCOM and leading the Antenna and
Oct. 2015, pp. 1–5.
Propagation Club. In 2007, she was with CATT
[6] B. Wang, K. Wang, Z. Lu, T. Xie, and J. Quan, ‘‘Comparison study of
non-orthogonal multiple access schemes for 5G,’’ in Proc. IEEE Int. Symp. and acted as the Vice-Chief Engineer of TDD
Broadband Multimedia Syst. Broadcast., Ghent, Belgium, Jun. 2015, Research and Development Product Line, focusing on speeding up the stan-
pp. 1–5. dard and industrial progress of TDD technology. Since 2011, she has been
[7] CATT, Usage Scenarios of Non-Orthogonal Multiple Access, document involved in the Wireless Innovation Center and acted as the Head Expert,
R1-166466, 3GPP, Aug. 2016. leading the 5G research with CATT. She has authored over 20 papers and
[8] CATT, Discussion on Scenarios and Use Cases for Multiple Access, doc- holds 50 patents.
ument R1-164246, 3GPP, May 2016.
[9] I. Y. Abualhaol and M. S. Bawa’aneh, "Capacity analysis of cooper-
ative relay-based communication system,’’ in Proc. IEEE GCC Conf.
Exhibit. (GCC), Dubai, United Arab Emirates, Feb. 2011, pp. 21–24.
[10] A. B. Saleh, S. Redana, J. Hämäläinen, and B. Raaf, ‘‘On the cover-
age extension and capacity enhancement of inband relay deployments in
LTE-advanced networks,’’ J. Elect. Comput. Eng., vol. 2010, no. 894846,
pp. 1–12, 2010.
[11] Y.-W. P. Hong, W.-J. Huang, and C.-C. J. Kuo, Cooperative Communica-
tions and Networking. Cambridge, U.K.: Cambridge Univ. Press, 2009.
[12] D. Tse and P. Viswanath, Fundamentals of Wireless Communication,
Cambridge, U.K.: Cambridge Univ. Press, 2005.
[13] N. Zhang, J. Wang, G. Kang, and Y. Liu, ‘‘Uplink nonorthogonal multiple
access in 5G systems,’’ IEEE Commun. Lett., vol. 20, no. 3, pp. 458–461,
Mar. 2016.
[14] A. Goldsmith, Wireless Communication, Cambridge, U.K.:
Cambridge Univ. Press, 2005.

WANWEI TANG received the B.E. degree from

Dalian Polytechnic University, in 2006, and the
M.S. degree from Dalian Maritime University, in
2008. He is currently pursuing the Ph.D. degree
with Beihang University, Beijing, China. He is cur-
rently with the Key Laboratory of Wireless Mobile
Communications, China Academy of Telecommu-
nications Technology, Beijing, China. He is also
with the Intelligence and Information Engineering
College, Tangshan University, Tangshan, China.
His current research interests include wireless communications theory, wire-
less communications systems, and cooperative networks.
BIN REN received the B.E. and M.S. degrees from
the Beijing University of Posts and Telecommuni-
cations, Beijing, China, in 2006 and 2009, respec-
tively, where he is currently pursuing the Ph.D.
degree. He joined the Key Laboratory of Wire-
less Mobile Communications, China Academy
of Telecommunications Technology, Beijing, in
2009. His current research interests include pat-
tern division multiple access, NOMA, and wireless
communications systems.

3868 VOLUME 5, 2017