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

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2596438, IEEE
Transactions on Vehicular Technology

Pattern Division Multiple Access (PDMA)

-A Novel Non-orthogonal Multiple Access for 5G
Radio Networks
Shanzhi Chen, Senior Member, IEEE, Bin Ren, Qiubin Gao, Shaoli Kang, Shaohui Sun, and Kai Niu

Abstract—In this paper, pattern division multiple access The fifth generation (5G) needs to support much higher
(PDMA), a novel non-orthogonal multiple access scheme, is pro- capacity and also provide much larger number of connected
posed for 5G radio networks. PDMA pattern defines the mapping users [2]. These requirements are rather challenging and it is
of transmitted data to a resource group which can consist of time,
frequency, and spatial resources or any combination of these difficult to satisfy the requirements by using OMA schemes.
resources. The pattern is introduced to differentiate signals of Non-orthogonal multiple access (NOMA) has thus been con-
users sharing the same resources, and the pattern is designed with sidered as a promising candidate to meet the requirements
disparate diversity order and sparsity so that PDMA can take the in connection numbers and system capacity of 5G. Signal
advantage of joint design of transmitter and receiver to improve of multiple users are superposed and advanced detection
system performance while maintaining detection complexity to
a reasonable level. System level simulation results show that algorithm are employed to separate the superposed signal.
PDMA can support 6 times simultaneous connections than that Theoretically, NOMA is optimal in terms of achieving the
of conventional and at least 30% improvement in spectrum capacity boundary [1].
efficiency over OFDMA. Actually NOMA has been used in previous wireless systems
Index Terms—Non-orthogonal Multiple Access, PDMA, SIC, such as 3G WCDMA system. For example, in the uplink of
5G. WCDMA, data symbols are spread by long spreading codes,
and multiple users transmit their spread signals on the same
frequency and time resources. Since long spreading codes are
I. I NTRODUCTION used, only linear detection algorithm is possible at the receiver
Multiple access scheme was regarded as the landmark of due to the complexity of non-linear detection algorithm. As a
each generation of mobile communication systems. Frequen- result, non-orthogonal transmission of this type is demonstrat-
cy division multiple access (FDMA) was used in the first ed to be inefficient in terms spectrum efficiency [3].
generation (1G). Time division multiple access (TDMA) and In recent years, a number of NOMA techniques are studied
code division multiple access (CDMA) were introduced in in both industry and academy, such as interleave division
the second generation (2G) and the third generation (3G), multiple access (IDMA) [4], bit division multiplexing (BDM)
respectively. Orthogonal frequency division multiple access [5], sparse code multiple access (SCMA) [6], and multi-user
(OFDMA) is the key component of the fourth generation (4G). sharing access (MUSA) [7], etc. With IDMA, different users
As already known, above multiple access schemes separate are distinguished by user-specific interleavers. BDM was first
users in orthogonal resources such as frequency, time, or proposed to provide differentiated services for broadcasting
code domain resources. These orthogonal multiple access system. It extends the multiplexing from symbol level to bit
(OMA) schemes make it possible to build the system with low level. Both SCMA and MUSA are based on code domain
complexity, however they are unable to achieve the capacity superposition [8]. To separate users being multiplexed on
boundary of a multi-user channel [1]. the same resource, nonlinear detection algorithms such as
successive interference cancellation (SIC), maximum a pos-
Copyright (c) 2015 IEEE. Personal use of this material is permitted. teriori (MAP) or maximum likelihood (ML) are suggested
However, permission to use this material for any other purposes must be for receiving. MAP and ML are so complex that it is very
obtained from the IEEE by sending a request to [email protected].
Shanzhi Chen and Shaohui Sun are with the State Key Laboratory of difficult to be implemented. SIC receiver reaches a good
Wireless Mobile Communications, China Academy of Telecommunications tradeoff between performance and complexity. However, SIC
Technology (CATT), and Datang Telecom Technology & Industry Group. suffers from the error propagation problem and degrades the
(e-mail: [email protected]; [email protected])
Bin Ren is with the School of Information and Communication Engineering, performance of NOMA transmission. Specifically, if a user
Beijing University of Posts & Telecommunications, China. happens to be not correctly detected, the following user will
(e-mail: [email protected]) be likely to be erroneously detected, i.e., the performance of
Qiubin Gao and Shaoli Kang are with the State Key Laboratory of Wireless
Mobile Communications, CATT. SIC relies heavily on the correctness of precedent detected
(e-mail: [email protected]; [email protected]) users.
Kai Niu is with the Key Lab. of Universal Wireless Communications This paper proposes a novel NOMA scheme based on code
(Ministry of Education), Beijing University of Posts and Telecommunications,
China. pattern, called Pattern Division Multiple Access (PDMA).
(email: [email protected]) Joint optimization of transmitting and receiving is considered

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2596438, IEEE
Transactions on Vehicular Technology

with SIC amenable pattern design at the transmitter side and estimation, the reconstructed signal will also be distorted.
SIC based detection at the receiver side. PDMA pattern are Even though the user’s packet is detected correctly, it still
designed to offer different order of transmission diversity, so has adverse effect on the following users’ detection.
that disparate diversity order between multiple users could Error propagation is a crushing blow for multi-user detection
be introduced to alleviate the error propagation problem of and it will deteriorate the performance of SIC based multi-
SIC receiver. PDMA pattern is also required to be sparse to user system. In general, two approaches can be considered
facilitate advanced detection algorithm such as belief prop- to alleviate the error propagation problem. The first is to
agation (BP). Iterations between BP and channel decoding enhance the reliability of those early-decoded users, either by
could further boost system performance. PDMA pattern is also selecting users with good channel condition or by designing
extended to include power scaling and phase shifting to harvest transmission parameters such that the early-decoded users have
additional constellation shaping gain. It is demonstrated by higher reliability and better channel condition.
simulation results that PDMA could provide significant gain Analytical results of MIMO detection from [10], [11] show
over OFDMA at both link level and system level. that the ith detected layers of SIC receiver could achieve
The main contribution of this paper is giving a complete diversity order
reporting about PDMA, including the principle, framework,
key technologies, application analysis, and performance eval- Ndiv (i) = NR − NT + i (1)
uation. The rest of this paper is organized as following: In
where NR is the receiving antenna number, NT is number of
section II, principle of PDMA as well as system model for
data layers.
both uplink and downlink are introduced. In section III, PDMA
The diversity order increases with the detection order.
pattern design at transmitter is discussed, including principle of
PDMA design is inspired by above result [12]-[14]. Multi-
PDMA pattern design and extension of PDMA pattern matrix.
user channel can be viewed as a virtual MIMO channel and the
In section IV, detection algorithms at receiver are discussed
above result could be generalized to multi-user non-orthogonal
and analyzed. PDMA application scenarios and system design
transmission. For non-orthogonal transmission employing SIC
aspects are described in details in section V. Performance
receiver, diversity order of each user varies with the order
evaluation as results are provided for PDMA and OFDMA
of detection. The first detected user has the lowest diversity
in section VI. Finally, in section VII, conclusion remarks are
order, and the last detected user has the highest diversity
given.
order. In a fading channel, diversity order affects transmission
reliability significantly. Increasing the diversity order typically
II. S YSTEM M ODEL leads to more reliable transmission. With SIC receiver, the first
A. Principle of PDMA detected user actually determines the overall detection perfor-
mance, but unfortunately its diversity order is the lowest. To
According to theoretical results of multi-user channel [9],
optimize system performance, it is desirable to have identical
superposition coding, at a transmitter and SIC at a receiver, are
pro-detection diversity order for each user.
able to achieve capacity boundary of multiple access channels
Diversity could be obtained from transmission or reception,
(MAC) or degraded broadcast channels (BC) when T&R are
or from both. Assuming that transmission diversity order of
working together. From theoretical perspective it is rational to
the ith detected user is DT (i), the diversity order after SIC
use SIC to achieve channel capacity, since the packet error
receiver can be expressed as
rate tends to be zero with the increased code length as long
as a user’s transmission rate is below the channel capacity. Ndiv (i) = DT (i) − K + i (2)
However, in a real system, detection error is inevitable due
to various non-ideal conditions, such as, limited code length, where K is the number of users. By joint design from
channel fading, and glitches, etc. transmitter and receiver, PDMA deliberately selects DT (i)
For SIC receiver, if a former user’s packet is detected so that the diversity order after SIC receiver is as close as
erroneously, it is very unlikely that the following users’ possible.
packet could be detected correctly. This is the so-called error The definition of transmission diversity means that multiple
propagation problem. Since multiple users are detected one by copies of a signal are transmitted from independent resources
one in serial order, the detection order of all users is usually to avoid transmission error due to deep fading on one resource.
arranged according to their signal strength. That is, the signal The resources could be time, frequency or spatial resource.
of the first detected user is the strongest, the signal of the PDMA maps transmitted data onto a group of resources
second detected user is weaker, and so on. For the first detected according to PDMA pattern to realize disparate transmission
user, it is recovered directly from the original receiving. diversity order. A PDMA pattern defines the mapping from
While for the following detected users, they are recovered transmitted data to a resource group. A resource group can
respectively from related cancellation receiving which should consist of time resource, frequency resource, spatial resource
cancel those former detected users from the original receiving or any combination of these resources. The number of mapped
by user reconstruction. If a user is not correctly detected, its resources in a group determines the order of transmission
reconstruction is impossible to be correct. In addition, the diversity. Data of multiple users can be multiplexed onto the
accuracy of reconstruction also impacts on the performance same resource group with different PDMA pattern. In this way,
of following users. For example, based on distorted channel non-orthogonal transmission is realized. By assigning PDMA

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2596438, IEEE
Transactions on Vehicular Technology

RE1 where gk is an N × 1 binary vector with elements “0” or “1”.

RE2 “1” means that the user’s data are mapped to the corresponding
RE3 RE, otherwise not. The K users’ PDMA patterns on N REs
[N,K]
RE4 construct PDMA pattern matrix GP DM A with the dimensions
User 1 User 2 User 3 User 4 User 5 User 6 of N × K:
[N,K]
GP DM A = [g1 , g2 , · · · , gK ] . (4)
Fig. 1. PDMA pattern for 6 users on 4 REs.
The received signal y at the base station (BS) can be
expressed as:
∑K
pattern with different diversity order, disparate transmission y= diag(hk )vk + n (5)
diversity order among users could be achieved. k=1
The second approach to alleviate error propagation problem
where n represents noise and interference at receiver; hk is
is to adopt more advanced and complex detection algorithm
the uplink channel response of the k th user; y, n and hk are
such as ML or MAP. It is anticipated that PDMA with
vectors with length N . diag(hk ) represents a diagonal matrix
advanced detection algorithm can alleviate error propagation
with elements from hk .
effect to a substantial degree. However, ML or MAP algorithm
Considering (3) and (4), (5) can be written in a compact
incurs tremendous detection complexity and it is hard to
form:
implement.
y = Hx + n (6)
Fortunately, the detection complexity could be reduced
T
significantly by making the PDMA pattern sparse. That is, data where x = [x1 x2 · · · xK ] , H is the PDMA equivalent
are only mapped to a small part of the resources in the resource channel response matrix of K users multiplexed on N REs
group. This draws on the idea of sparse coding in low density and is given by
parity check (LDPC) coding. Sparsity makes it possible to use [N,K]
low complexity BP algorithm to approach the MAP detection. H = HCH • GP DM A (7)
In addition, convergence of BP algorithm could be speed up HCH = [h1 , h2 , · · · , hK ] (8)
by disparate transmission diversity of PDMA.
In summary, PDMA uses PDMA pattern to define sparse where the (n, k) element of HCH is the channel response
mapping from data to a group of resources. PDMA pattern from the k th user to the base station on the nth RE, and •
could be represented by a binary vector. The dimension of the indicates element-wise product of two matrices.
vector equals to number of resource in a group. Each element Overload factor is defined as the ratio between the number
in the vector corresponds to a resource in a resource group. of users and the number of REs in a resource group. It
A “1” means that data shall be mapped to the corresponding reflects the multiplexing times of PDMA relative to orthogonal
resource. Actually, number of “1” in the PDMA pattern is scheme. Taking N = 3, K = 6 as an example, the overload
defined as its transmission diversity order. factor is then α = K/N = 200%, which means that PDMA
Fig. 1 shows an example of resource mapping according supports two times user compared with OMA. In an example,
to PDMA pattern. Six users are multiplexed on four resource the PDMA pattern matrix is expressed as
elements (REs). A PDMA pattern is assigned to a user. User1’s  
1 1 0 1 0 0
data are mapped to all four REs in the group, and user2’s GP DM A =  1 0 1 0 1 0 
[3,6]
(9)
data are mapped to the first three REs, etc. The order of 0 1 1 0 0 1
transmission diversity of the six users is 4, 3, 2, 2, 1, and
1 respectively. The received signal is then given by
 
x1
    x2   
y1 h1,1 h1,2 0 h1,4 0 0   n1
 x3 
B. System Model of Uplink  y2  =  h2,1 0 h2,3 0 h2,5 0   +  n2  . (10)
 x4 
y3 0 h3,2 h3,3 0 0 h3,6   n3
Without loss of generality, we assume both transmitter | {z } x5 
H x6
and receiver have single antenna, and an example of uplink
PDMA system is shown in Fig. 2(a). At the transmitter, As a further extension, the constellation mapping and
a PDMA encoder maps the modulation symbols xk onto PDMA encoding can be realized by mapping the user’s data
resources and generates PDMA modulation vector vk . At the bits directly to a PDMA modulation vector. This is a joint
receiver, PDMA multi-user detector is used to detect data of operation between PDMA encoding and constellation mapping
multiplexed users. - joint PDMA modulation. Specifically, the candidate set of
The uplink PDMA system consists of K users and their data PDMA modulation vector for user k, called codebook, shall
are mapped onto N REs by using distinguished PDMA pat- be designed according to the PDMA pattern of user k. The
terns. The PDMA modulation vector vk of user k is obtained codebook is designed offline and stored in a receiver and
by spreading the user’s modulation symbol xk according to transmitter pair. Once the user’s PDMA pattern is decided,
the PDMA pattern gk : the associated codebook is therefore determined. Joint PDMA
modulation is to choose a PDMA modulation vector from the
vk = gk xk , 1 ≤ k ≤ K (3) codebook based on the user’s data bits. The vectors in the

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2596438, IEEE
Transactions on Vehicular Technology

b1 Channel c1 Constellation
x1 PDMA v1 OFDM y1 bˆ1
User 1 Channel
encoding mapping encoding modulation User 1
decoding
OFDM PDMA
de- multi-user
modulation
yK detection
Channel bˆK
User K
bK Channel cK Constellation
xK PDMA vK OFDM decoding
User K encoding mapping encoding modulation

Receiver
Transmitter (a)

b1 c1 x1 v1
User 1 Channel Constellation PDMA
OFDM y1 PDMA c1 Channel bˆ1
encoding mapping encoding User 1
de-modulation multi-user detection decoding
z OFDM
S modulation

bK Channel
cK Constellation xK PDMA vK OFDM yK PDMA cK Channel bˆK
User K User K
encoding mapping encoding de-modulation multi-user detection decoding

Transmitter (b) Receiver

Fig. 2. PDMA system model. (a) Uplink. (b) Downlink.

codebook and the PDMA pattern associated with it have the 2 × 3 pattern matrix, i.e., 3 users are multiplexed on two REs.
same sparsity property, i.e., the zero elements appear at the The pattern matrix is:
same position. Actually, non-zero elements in the vector is [ ]
[2,3] 1 1 0
a multi-dimensional constellation in high dimensional space. GP DM A =
1 0 1
Joint PDMA modulation completes the PDMA encoding and
the constellation mapping simultaneously in high dimensional and another design for 150% overload is 4 × 6 pattern matrix:
space.  
1 0 1 1 1 0
 1 1 0 1 0 1 
GP DM A =  .
[4,6]
C. System Model of Downlink  1 1 1 0 1 0 
In a downlink system shown in Fig. 2(b), we assume that 0 1 1 0 0 1
each user is assigned a PDMA pattern. After PDMA encoding, Though both pattern matrices have the same overload factor,
multiple data streams are superimposed at the base station and [4,6]
GP DM A can achieve better performance while the cost of
transmitted simultaneously. The received signal yk at user k [2,3]
detection complexity is higher comparing to that of GP DM A .
can be expressed as PDMA pattern matrices with different dimensions are able
∑
K to achieve a given overload factor. With a higher dimension,
yk = diag(hk ) gi xi + nk detection complexity is also higher, and better performance is
i=1
(11) expected. Given overload factor, dimension of pattern matrix
[N,K]
= (diag(hk )GP DM A )x + nk shall be selected to reach a tradeoff between complexity and
= Hk x + nk performance.
If N is the size of resource group (row number of PDMA
[N,K]
Hk = diag(hk )GP DM A (12) pattern matrix), there are 2N − 1 possible binary vectors for a
pattern matrix. Assuming K is the column number determined
where nk represents noise and interference at receiver; hk
based on overload factor, we can thus choose K patterns out
is the downlink channel response of the k th user; yk , nk
from 2N − 1 candidates to construct PDMA pattern matrix.
and hk are vectors with length N . Hk is the PDMA e-
Selection of patterns also gives impacts on performance and
quivalent channel response matrix of user k on N REs.
T complexity:
x = [x1 x2 · · · xK ] , where xk is the k th user’s modulation
symbol. 1) A pattern with heavier weight (number of “1” elements
in the pattern) provides higher diversity order. More reli-
able data transmission can be anticipated, and detection
III. PDMA PATTERN D ESIGN
complexity is also increased. If the system can conduct
A. Pattern Matrix Design complex computation, patterns with heavy weight will
Users multiplexed PDMA patterns, on the same resource be preferable; otherwise, light weight patterns have to
group, can construct a PDMA pattern matrix. Each column of be selected, aiming at sparse PDMA pattern matrix.
the PDMA pattern matrix represents a PDMA pattern. Prop- 2) According to design principle of PDMA, it is desirable
erties of PDMA pattern matrix such as dimension and level of to have different diversity orders in the pattern matrix
sparsity contribute both complexity and system performance. to alleviate error propagation problem of SIC receiver
Given a certain overload factor, there are a number of pattern or fasten convergence of BP receiver. Thus the selected
matrices available, as long as K and N are selected properly. patterns shall have as many different diversity orders as
For example, overload factor of 150% could be achieved by a possible.

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2596438, IEEE
Transactions on Vehicular Technology

3) For patterns with identical diversity order, smaller inner PDMA pattern matrix with dimension 3 × 6 is expressed as
product between the patterns leads to less interference  
1 1 1 0 1 0
against each other. Small inner product means that
GP DM A =  1 1 0 1 0 1  .
[3,6]
the two patterns have less “1” elements in common
1 0 1 1 0 0
positions. That is, the number of REs shared by the two
patterns is low. Data of two users are multiplexed on And the finally selected PDMA pattern matrix with dimension
only few REs. For example, if two patterns have inner 4 × 8 is expressed as
product of 0, the two patterns actually maps data onto  
1 1 1 1 0 0 0 0
a different set of REs, hence there is not interference  1 1 0 0 1 1 0 0 
GP DM A =  
[4,8]
between the two patterns. For a given diversity order,  1 0 1 0 1 0 1 0 .
the selected patterns shall minimize the maximum inner
0 0 0 1 0 1 1 1
product between any two patterns. Of course this rule is
also applied to patterns with different diversity order.
The design of pattern matrix shall take overload factor, B. Extension of Pattern Matrix
[2,3]
diversity order and detection complexity into account. A good Taking PDMA pattern matrix GP DM A as an example, data
pattern matrix can reach good trade-off among these aspects. of 3 users are mapped onto two REs. The transmission signal
The criteria of maximum constellation constrained capacity on these REs can be expressed as:
(CC-Capacity) [15] can be used to design PDMA pattern  
[ ] [ ] x
matrix especially for uplink application. That is, with the v1 1 1 0  1 
= x2 (15)
input information on matrix dimension and its row weight, v2 1 0 1
x3
all candidate sets of PDMA pattern matrix are calculated by
CC-capacity, then the PDMA pattern matrix with maximum where vj is the transmission signal on the j th RE, and xk
CC-capacity is selected [16]. is the modulation symbol of the k th user. Unlike orthogonal
[N,K] transmission, transmission signal on each RE is linear combi-
Gopt = arg max{C(N, K, Ω)∥G[N,K] ⊂ G[N,M ] }, nation of multiple modulation symbols:
st.∥G[N,K] (:, k)∥2 = 1(k = 1, 2, · · · , K), x ∈ ΩK×1 (13)
v1 = x1 + x2
(16)
where C(N, K, Ω) denotes CC-Capacity for the parameter N , v2 = x1 + x3 .
K and Ω, Ω denotes constellation set, M = 2N − 1 and x is
defined in (6). This combination may alter the characteristics of transmis-
As an example, assuming the matrix dimension is 4 × 6, the sion signal on each RE. For example, if all three users adopt
calculation of C(N, K, Ω) is as follows. BPSK modulation, the modulation symbol of user 1, user 2 and
user 3 is either +1 or −1. The combined transmission signal
C(4, 6, ΩQP SK ) takes value from {−2, 0, +2}. Assuming a noiseless channel,
= I(v1 ; y) + I(v2 ; y|v1 ) + I(v3 ; y|v1 , v2 ) if the receiver receive −2, or +2 on a RE, then the receiver
(14) can infer that the transmitted symbols on the RE is [−1, −1]
+ I(v4 ; y|v1 , v2 , v3 ) + I(v5 ; y|v1 , v2 , v3 , v4 )
or [+1, +1]. But if 0 is received, it is impossible for the
+ I(v6 ; y|v1 , v2 , v3 , v4 , v5 ) receiver to recover the transmitted symbols, as both [+1, −1]
where I(.) represents the mutual information between the and [−1, +1] resulting in the same output. Furthermore, if
input QPSK symbol and the output of a Gaussian channel, each user adopts QPSK or 16QAM modulation, the combined
and I(v2 ; y|v1 ) denotes the conditional mutual information constellation consists of 9 or 49 constellation points.
between v2 and y with the given value of v1 , where vk From the above discussions, we can see that the combined
and y are defined in (3) and (5), respectively. I(.) and constellation has non-uniform distribution and it is no longer
I(v2 ; y|v1 ) in (14) can be calculated by adopting Monte Carlo one-to-one map between a constellation point and an input
integration method. Considering different row weights, the user data, i.e., the combination leads to ambiguity.
selected PDMA pattern matrix are expressed as following To resolve the ambiguity, power scaling and phase shifting
  can be introduced in PDMA pattern matrix. Specifically,
1 0 1 0 0 0 before two users’ symbols are mixed, a power scaling factor
 1 0 0 1 0 0 
row weight 2: GP DM A =   and a phase shifting factor shall be applied:
[4,6]
 0 1 0 0 1 0  √ √
0 1 0 0 0 1 v = βx1 ejφ + 1 − βx2 (17)
  where β is power scaling factor and φ is phase shifting factor.
1 1 1 0 0 0
 1 0 0 1 1 0  As an example, by setting φ = π/4 and β = 0.5, i.e.,
row weight 3: GP DM A =  
[4,6]
 0 1 0 1 0 1  only phase shifting difference is introduced between users, the
0 0 1 0 1 1 . combined constellation is shown in Fig. 3. It can be observed
that by adding a phase shifting factor, the ambiguity is
As another example, in case of overload factor 200%, resolved. Moreover, the distribution of combined constellation
assuming selecting the row weight 4, the finally selected is closer to Gaussian distribution. It is known that, Gaussian

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2596438, IEEE
Transactions on Vehicular Technology

QPSK 16QAM
2
1.5
SIC receiver
1.5
1 Received signal
1
User 1 detection User 1 data
0.5
0.5

+ -
0 0
Å Reconstruction
−0.5
−0.5

−1
User 2 detection User 2 data
−1
−1.5

+ -
−1.5
−1.5 −1 −0.5 0 0.5 1 1.5
−2
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2 Å Reconstruction
(a) (b)

...
...
...

...
Fig. 3. Combined constellation from two users by phase shifting φ = π/4. + -
Å
(a) QPSK. (b) 16QAM.
User K detection User K data
QPSK 16QAM
1.5 1.5

1 1
Fig. 5. SIC receiver.
0.5 0.5

0 0

−0.5 −0.5
A. SIC
−1 −1

As shown in Fig. 5, the basic idea of SIC receiver is

−1.5 −1.5
−1.5 −1 −0.5 0 0.5 1 1.5 −1.5 −1 −0.5 0 0.5 1 1.5
to reconstruct a user’s signal and then subtract it from the
(a) (b) received signal. The construction could be carried out either at
Fig. 4. Combined constellation from two users by power scaling and phase symbol level or codeword level. For the symbol level SIC (SL-
shifting β = 0.8, φ = π/4. (a) QPSK. (b) 16QAM. SIC), the construction is made from the demodulated symbols.
Instead, the codeword level SIC (CW-SIC) is based on signal
construction from decoded data bits. As channel decoding is
distribution is the capacity maximizing input distribution for able to correct most errors, CW-SIC is expected to perform
AWGN channel. better than SL-SIC.
That is, by introducing a phase shifting factor, channel
capacity gain is reaped. The gain is also called shaping gain.
When both power scaling and phase shifting are introduced, B. BP
the effect is exemplified in Fig. 4. As expected, the shape of BP algorithm has been demonstrated to be able to approach
constellation is changed, and it also approaches a Gaussian MAP detection asymptotically [17], [18]. Furthermore, the
distribution. sparsity of PDMA pattern reduces the complexity of BP
For PDMA, the power scaling and phase shifting can be algorithm, making it suitable for PDMA system, and disparate
naturally incorporated into PDMA pattern matrix. That is, transmission diversity of PDMA can speed up the convergence
the value “1” in PDMA pattern matrix is substituted with of BP.
a complex value reflecting both power scaling and phase Given the received signal vector y and the PDMA equiva-
shifting, forming an extended PDMA pattern matrix. For the lent channel response matrix H in (7), the optimal detection
[2,3] of x is a joint MAP detection
PDMA pattern matrix GP DM A , the extended PDMA pattern
matrix is given as x̂ = arg max p(x|y, H) (19)
[ ] x∈ℵK
[2,3] α11 e−jφ11 α21 e−jφ21 0
GE−P DM A = (18)
α12 e−jφ12 0 α32 e−jφ32 where ℵK represents constellation alphabet of K users.
where αkj and φkj is the power scaling and phase shifting Equation (19) can be approximated by a local MAP solution
factor of the k th user on the j th RE. based on Bayesian formula
The optimal value of power scaling and phase shifting ∑ ∏
xˆk = arg max P (x) p(yn |x) (20)
depend on the number of users and the shape of input s∈ℵ
x∈ℵK ,xk =s n∈Nv (k)
constellation.
where ℵ and Nv (k) represent constellation alphabet and the
IV. D ETECTION A LGORITHM AT R ECEIVER RE index set corresponding to the PDMA pattern of the k th
Detection algorithm at receiver is the key to reap perfor- user. According to [17], [18], the problem can be solved by
mance gain of PDMA in uplink and downlink. This section applying BP algorithm on the underlying factor graph.
[N,K]
is dedicated to the details of the three algorithms suitable for A PDMA system with pattern matrix GP DM A can be
PDMA: SIC, BP, and BP-IDD. By arranging diversity order represented by a factor graph consisting of channel observation
of PDMA patterns, error propagation problem of SIC receiver nodes (CND) and user nodes (UND). A factor graph with
[3,6]
could be alleviated to certain extent. Sparsity of PDMA pattern PDMA pattern matrix GP DM A is shown in Fig. 6. The
greatly reduces the complexity of BP and BP-IDD algorithm, k UND corresponds to the k th user’s data symbol, and
th

making them suitable for PDMA system. In addition, PDMA the j th CND represents the received signal on the j th RE
pattern could be designed to speed up convergence of BP and yj (1 ≤ j ≤ N ). If there is an edge between the k th UND and
[N,K]
BP-IDD. the j th CND (i.e., GP DM A (j, k) ̸= 0), the received signal

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2596438, IEEE
Transactions on Vehicular Technology

x1 x2 x3 x4 x5 x6 c 1 ,1 c1, m c 6 ,1 c 6 ,m
VND
UND Turbo Decoder

d (×) Converter of symbol-LLR & bit-LLR d (×)

x1 x2 x3 x4 x5 x6
UND

CND
y1 y2 y3

[3,6]
Fig. 6. Factor graph of PDMA with GP DM A .
CND
De- Turbo Hard y1 y2 y3
+ interleaver decoder decision
- BP Detector
-
Interleaver + [3,6]
Multi-
Fig. 8. Factor graph of PDMA with GP DM A and Turbo decoder.
y
user
detector TABLE I
De- Turbo Hard C OMPUTATION C OMPLEXITY PER M ODULATION S YMBOL
+ interleaver decoder decision
-
-
Algorithm Number of multiplications Number of additions
Interleaver +
SIC O(KN 3 ) O(KN 3 )
df
BP O(df N M ) O(Tin df N Qm M df )
Fig. 7. Structure of the BP-IDD receiver for PDMA uplink system.
df
BP-IDD O(df N M ) O(Tout Tin df N Qm M df )

on the j th RE includes contributions from the k th user. Detail

procedure about the BP algorithm can be referred to [17], [18]. and BP would be better than SIC.
Table I summarizes computation complexity of the above
three detection algorithms [20], [21]. In Table I, M denotes
C. BP-IDD the size of modulation constellation, Qm = log2 (M ), Tin ,
The basic principle of BP based iterative detection and de- Tout and df represent BP-IDD inner iteration number, outer
coding (BP-IDD) algorithm is that the decoded bit-LLR is fed iteration number, and maximum row weight of PDMA pattern
back from Turbo decoder and converted to symbol-LLR as the matrix. The number of additions of BP-IDD receiver is about
a priori information of the BP detector. There are two iterative Tout times of BP receiver, yet they have the same number of
processes in the BP-IDD receiver: inner iteration processing of multiplications as the value of |yj − fj x|2 in [21, eq. (17)] is
BP detector and outer iteration processing between BP detector calculated only once and reused for the following inner and
and Turbo decoder. As shown in Fig. 7, besides a traditional outer iteration in BP-IDD. It should also be noted that the
BP multi-user detector, the BP-IDD receiver includes multiple computation of Turbo decoding is not accounted in Table I.
parallel iterative processes, each of which is composed by As the processing ability of a base station is usually more
modules of deinterleaver, Turbo decoder and interleaver. Here powerful compared with that of a user terminal, BP-IDD and
soft information is transferred between the multi-user detector BP could be used in the PDMA uplink. Detection at user
and the Turbo decoder, in the form of log-likelihood ratio terminal in the PDMA downlink can choose either BP or SIC
(LLR). depending on its processing ability.
A factor graph of BP-IDD algorithm based on PDMA
[3,6] V. PDMA IN 5G S YSTEM
pattern matrix GP DM A is shown in Fig. 8. VND denotes
the variable node. Let xk (k = 1, . . . , K) be data symbols A. Application Scenarios
corresponding to the UND of the kth user and associated to the PDMA can be applied in typical scenarios of 5G, for
VND ck,i (i = 1, . . . , m), where m represents the modulation example, in enhanced mobile broadband (eMBB) or massive
order of k th user. The connection between UND and VND machine type communication (MTC).
shall satisfy certain condition imposed by channel encoding. In eMBB scenario, the main challenge is to enhance the
According to Fig. 8, the iterative process between UND and transmission rate. Massive MIMO, ultra dense networking,
CND, described in the previous section BP detection, is called and high frequency communication are candidate technologies.
inner iteration, and the iterative process between UND and PDMA can be used in conjunction with these technologies.
VND is named outer iteration. Detail procedure about the BP- In massive MTC scenario, a base station needs to provide
IDD algorithm can be referred to [18], [19]. connection to a huge number of low cost terminals. The main
challenge for massive MTC scenario is how to effectively deal
D. Comparison of SIC, BP and BP-IDD with massive connection with power constraint.
Among the detection algorithms described above, it is ex- In 4G LTE system, to transmit data, a user shall first issue
pected that BP-IDD is the best of all in terms of performance, scheduling request (SR) on periodically occurred resources

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2596438, IEEE
Transactions on Vehicular Technology

TABLE II
which is configured by base station. Base station then makes PDMA L INK - LEVEL S IMULATION A SSUMPTIONS
scheduling decisions and sends uplink grant to the user in-
dicating the resources on which the user can transmit data. Parameter Value
Generally, the procedure may take 10ms or more. For some Carrier 2 GHz
massive MTC application, such a long latency is unacceptable. System bandwidth 10 MHz
Moreover, the uplink grant is carried by downlink control sig- Channel model UMA-NLOS [22]
naling, and with massive number of connections the downlink Modulation coding rate QPSK 1/2; LTE Turbo
control channel may become a bottleneck. In such situation, Antenna configuration 1 Tx 2 Rx
grant-free transmission is a viable option. Channel estimation Perfect
By means of grant-free transmission, a user autonomously HARQ No
selects resource for transmission without SR and scheduling of Uplink overload factor 150%; 200%; 300%
base station. To avoid interfering with other traffic scheduled Uplink average signal to noise ratio (SNR) The same of all users
by base station, resource for grant-free transmission shall be
confined within a certain set of resources. The resource set
is called resource pool. For orthogonal transmission, resource Link adaptation for downlink PDMA is based on user
pool consists of resource in time and frequency domains. A reporting of the channel quality information (CQI). However,
user selects a resource from the pool for transmission. the multi-user pairing nature of PDMA makes it difficult
As there is no coordination between users sharing the same for a user to predict CQI without knowing its paring users
resource pool, it is likely that two users select the same and their PDMA patterns. Power domain optimization may
resource. When a collision happens, it may lead to failure further complicate the problem, as the user will not know
in detection. The probability of collision is proportional to what the transmission power is going to be before making the
the number of users sharing the resource pool, and is inverse scheduling decision. A flexible CQI calculating and reporting
proportional to the number of resources in the pool. That is, mechanism is needed.
enlarging the resource pool could reduce collision probability. As discussed above, the grant-free transmission is able
As a non-orthogonal transmission scheme, PDMA could to reduce data latency and to control overhead. Though by
naturally be incorporated into grant-free transmission to reduce introduction of PDMA, a resource pool could be extended, it
the collision probability. That is, PDMA provides another is still possible that two users collide on the same resource and
dimension for resource sharing - PDMA pattern. A traditional PDMA pattern especially when the system is heavy loaded.
resource pool could be extended to include PDMA pattern. To facilitate grant-free transmission, resource selection method
Specifically, each resource group in the pool is associated and mechanism to resolve conflicts play a fundamental role.
with a PDMA pattern matrix. A UE selects a time-frequency Radio resource management
resource as well as a PDMA pattern from the pattern matrix Traditionally, the radio resource management deals with
for transmission. Even though two users may select the same the allocations of time, frequency, and spatial resources to
time-frequency resource, as long as their PDMA patterns are make full use of the wireless resources. PDMA introduces
different, the receiver is able to decode the two users data another dimension - PDMA pattern. The optimization problem
successfully. becomes really challenging, as more optimization variables
The resource pool is α − 1 times larger than a traditional are involved. Low complexity radio resource management
resource pool where α is the overload factor of the PDMA algorithm, that could achieve near-optimal performance, is
pattern matrix. worthy to seek.

B. System Design Aspects VI. P ERFORMANCE E VALUATION

To enable PDMA in practice, a number of aspects in system A. Comparison of PDMA and OMA
design shall be considered. Taking LTE orthogonal system as a reference, uplink and
Air interface and process design downlink performances of PDMA are evaluated and results
PDMA enables large number of users to transmit on the are provided in this section. The assumptions of the link-level
same resource, especially when PDMA is used jointly with simulations are shown in Table II, and the results are given in
massive MIMO. The need for reference signal will be multi- Fig. 9 and Fig. 10. The system-level simulation assumptions
plied accordingly. Reference signal shall be designed carefully are shown in Table III, and the simulation results are provided
to meet the requirements of detecting PDMA signal and in Fig. 11 and Fig. 12.
keeping the overhead incurs by the reference signal to a Fig. 9 shows the total spectrum efficiency (SE) of uplink
reasonable level. PDMA under the overload factors of 150%, 200%, and 300%
Multi-user transmission of PDMA also leads to demanding as well as SE of OFDMA. For fair comparison, same number
requirements on control channel, as each users data shall of source bits is assumed for PDMA and OFDMA. For the
be accompanied by a control channel to provide necessary given overload factors of 150%, 200% and 300%, SE gains
information for detection. Techniques such as multi-subframe of 50%, 100% and 200% can be achieved by PDMA over
scheduling, group control signal design and the control sig- OFDMA when SNR is high enough.
naling content design could serve as a starting point to cope In a downlink system, the SNR differences between users
with the problem. have tremendous influence on the performance of PDMA.

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2596438, IEEE
Transactions on Vehicular Technology

TABLE III
PDMA S YSTEM - LEVEL S IMULATION A SSUMPTIONS 1
PDMA
OFDMA
Parameters Value
0.9
Hexagonal homogeneous network;
Topology

Total Spectral Efficiency(bps/Hz)

19sites/57sectors
0.8
Number of users
10, 20, or 30
per cell
Carrier 2GHz 0.7

Uplink: 5MHz
Bandwidth
Downlink: 10MHz 0.6
ISD 500m
Channel model ITU UMa [22] 0.5
Uplink: open-loop power control,
Power control
alpha = 1, P0 = −95dBm 0.4
−4 −2 0 2 4 6 8 10 12 14
The number of Uplink: 1Tx2Rx;
SNR (dB)
antenna Downlink: 2Tx2Rx
Uplink: User vertical polarization; BS ±45◦
Fig. 10. PDMA downlink performance.
Antenna cross polarization
configuration Downlink: User ±45◦ cross polarization; BS
1.2
±45◦ cross polarization 1.02
Channel estimation Perfect 1

per cell @1MHz˅

Uplink: Grant-free 0.8
Scheduler

(packet/ms
Downlink: PF schedule
0.6
Uplink: 160 bits @ 1PRB
MCS Downlink: Adaptive (Based on LTE downlink 0.4
MCS definition) 0.17
0.2
Maximum HARQ Uplink: 0
transmission times Downlink: 3 0
OFDMA PDMA
Uplink: Bursty traffic with small packet
Traffic model
Downlink: Full buffer traffic
PDMA uplink
Fig.system
11. performance
PDMA uplink(Supported packet arrival
system performance rate whenpacket
(Supported packetarrival
drop rate
rate= 1%)
Linear MMSE receiver for OFDMA, BP-IDD
Receiver when packet drop rate = 1%).
for PDMA

than 10dB.
The uplink grant-free OFDMA and PDMA transmissions
3 PDMA−150% are evaluated under traffics with small burst packet and the
PDMA−200%
latency is required to be not more than 1ms. Gain of 500%
Total Spectral Efficiency(bps/Hz)

PDMA−300%
2.5 in terms of the number of supported users under the given
OFDMA
system packet drop rate of 1% is observed in Fig. 11. The gain
2
comes from two facts. First, PDMA provides a larger resource
pool than OFDMA does, so that the collision probability of
1.5
PDMA is lower than that of OFDMA. Second, the BP-IDD
receiver employed by PDMA is more capable of dealing with
1
interference when collision occurs.
From the results of downlink PDMA shown in Fig. 12,
0.5
PDMA can get about 30% gain compared with OFDMA in
terms of both SE at cell edge and cell average SE. The gain
−2 0 2 4 6 8 10 increases with the user number in a cell increasing, because
SNR(dB)
with more users it is easier to find suitable users for pairing
in PDMA.
Fig. 9. PDMA uplink performance.

B. Comparison of Different Pattern Matrix

The performance gain of PDMA over OFDMA gets more Taking overload factor 150% and 200% as examples, uplink
remarkable when the difference gets larger. As shown in performances of PDMA with different pattern matrices are
Fig. 10, when the SNR difference of two users is 12dB, the SE provided in this section. For the given pattern matrix 4 × 6,
gain is 14% when the SNR is −4dB; reaches the maximum of 3 × 6 and 4 × 8 described in section III.A, Fig. 13 shows their
50% when SNR is 0dB; and vanishes when the SNR is higher performance comparison.

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2596438, IEEE
Transactions on Vehicular Technology

10

TABLE IV
35% 30.85% 31.44% R ECEIVER C OMPLEXITY OF D IFFERENT PDMA PATTERN M ATRIX
28.76%
Spectrum efficiency gain

30% 26.15% 26.85% Pattern matrix Computation complexity

25.14%
25% 4 × 6 row weight df = 2 Nmul ≈ 8 × 42

20% 4×6 row weight df = 3 Nmul ≈ 12 × 43

3×6 row weight df = 4 Nmul ≈ 12 × 44
15%
4×8 row weight df = 4 Nmul ≈ 16 × 44
10%
TABLE V
5%
C OMPLEXITY OF D IFFERENT R ECEIVERS
0%
10 UE 20 UE 30 UE Link type Receiver Computation complexity
BP-IDD Nmul ≈ 64, Nadd ≈ 1152
cell average gain cell edge gain Uplink [2, 3]
BP Nmul ≈ 64, Nadd ≈ 384
BP Nmul ≈ 64, Nadd ≈ 384
Downlink [2, 3]
Fig. 12. PDMA downlink system performance. SIC Nmul ≈ 24, Nadd ≈ 24
Note: Nmul denotes the number of multiplications, Nadd denotes
the number of additions, and ≈ means approximately equal to.

[4,6]−df=2
[4,6]−df=3
[3,6]−df=4
[4,8]−df=4
−1
10
BLER

−1
10
BLER

150%−BP
150%−BP−IDD
300%−BP
−2
−2 10 300%−BP−IDD
10

0 1 2 3 4 5 6
SNR(dB)
0 2 4 6 8 10 12
Fig. 13. Performance of different PDMA pattern matrix. SNR(dB)

Fig. 14. Performance of uplink PDMA with BP or BP-IDD at receiver.

As shown in Fig. 13, higher row weight (df ) can get better
performance. For PDMA pattern matrix 4×6, row weight df =
0
3 has 1dB gain compared with row weight df = 2. In addition, 10

PDMA pattern matrix with higher dimension can get better

performance. For the same overload 200%, pattern matrix 4×8
has 0.8dB gain compared with pattern matrix 3 × 6. However,
a higher row weight or a higher dimension in construction
PDMA pattern matrix means a higher complexity in detection.
BLER

Table IV gives a comparison of the computation complexity −1

10
about these pattern matrices. For the same dimension 4 × 6,
the complexity of pattern matrix with row weight df = 3 is 6 BP
times of pattern matrix with row weight df = 2. For the same SL−SIC
overload 200%, the complexity of pattern matrix 4 × 8 is 1.33
times of pattern matrix 3 × 6. Therefore, a tradeoff between
performance and complexity should be considered in pattern −2
10
matrix design. −2 0 2 4 6 8 10
SNR of far UE (dB)

C. Comparison of Different Receivers Fig. 15. Performance of downlink PDMA with BP or SIC at receiver.
For those receivers described in section IV, uplink perfor-
mances of PDMA with BP or BP-IDD receivers are com-
pared, and the results are shown in Fig. 14. Also, downlink the results are shown in Fig. 15. Table V gives comparisons
performances of PDMA with BP or SIC are compared, and of the computation complexity.

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2596438, IEEE
Transactions on Vehicular Technology

11

As shown in Fig. 14, BP-IDD receiver has gain compared [3] 5th Generation Non-orthogonal Waveforms for Asynchronous Signalling,
with BP receiver. Especially, the gain is higher with overload [Online]. Available: http://www.5gnow.eu.
[4] P. Li, L. Liu, K. Wu; W.Leung, “Interleave Division Multiple Access,”
increasing, e.g., the gain is 0.8dB for overload 150% with IEEE Trans. Wireless Commun., vol. 5, no. 4, pp. 938-947, Apr. 2006.
pattern matrix 2 ×3, and the gain is 1.6 dB for overload 300% [5] H. Jin, K. Peng, J. Song, “Bit Division Multiplexing for Broadcasting,”
with pattern matrix 4×12. This is because that the performance IEEE Trans. on Broadcast., vol. 59, no. 3, pp. 539-547, Apr. 2013.
[6] L. Lu, Y. Chen, W. Guo, et al. “Prototype for 5G new air interface tech-
of BP algorithm converges more slowly for overload factor of nology SCMA and performance evaluation,” China Commun. (Suppl.),
300% compared with 150% and the frequency diversity degree pp.38-48, Dec. 2015.
is higher for 300% than 150%. Whereas, from computation [7] Z.Yuan, G. Yu, W.Li, Y.Yuan, et al. “Multi-User Shared Access for
Internet of Things,” IEEE 83th Veh. Tech. Conf. (VTC Spring), 2016.
complexity, addition in BP-IDD receiver is 3 times of that in [8] L. Dai, et al, “Non-orthogonal Multiple Access for 5G: Solutions, Chal-
BP receiver. lenges, Opportunities and Future Research Trends,” IEEE Commun.Mag.,
As shown in Fig. 15, BP receiver has a gain of 2 dB vol. 53, no. 9, pp. 74-81, Sep. 2015.
[9] M. C. Thomas, A. T. Joy, Elements of Information Theory (2nd ed.), John
compared with SIC receiver in downlink where MCS of the Wiley & Sons, Inc. 2006.
near user and far user is QPSK 1/2, and power factor is [10] S. Loyka, F. Gagnon. “Performance Analysis of the V-BLAST Algo-
0.2 and 0.8. Whereas, the computation complexity of BP rithm: an Analytical Rpproach,” IEEE Trans. Wireless Commun., vol. 3,
no. 4, pp. 1326-1337, Feb. 2002.
receiver is about 1.67 and 15 times higher than SIC receiver [11] D. Truhachev, “Universal Multiple Access via Spatially Coupling Data
in multiplication and addition, respectively Transmission,” in Proc. Int. Symp. Wireless Pers. Information Theory
Proceedings (ISIT), pp. 1884-1888. Jul. 2013.
[12] X. Dai, S. Sun, and Y. Wang. “Successive Interference Cancellation
VII. C ONCLUDING R EMARKS Amenable Space-Time Codes with Good Multiplexing-Diversity Trade-
offs,” Wireless Pers. Commun., vol. 55, no. 4, pp.645-654, Dec. 2010.
Pattern division multiple access (PDMA) has been in- [13] Dai X. M., Sun S. H. and Wang, Y. M. “Reducing Complexity of Quasi-
corporated into ITU-R Report “Future technology trends of Maximum-Likelihood Detectors Through Companding for Coded MIMO
Systems,” IEEE Trans. Veh. Tech., vol. 61, no. 3, pp. 1109-1123, Mar.
terrestrial IMT systems” [23] by ITU organization since 2014. 2012.
It is a novel NOMA scheme based on successive interfer- [14] S. Sun, “Pattern Division Multiple Access (PDMA),” Future-Taiwan 5G
ence cancellation (SIC) amenable multiple access (SAMA) Workshop, Nov. 2014.
[15] R. Hoshyar, F. Wathan, and R. Tafazolli, “Novel Low-Density Signature
[24], [25]. To achieve an integral optimization for multi- for Synchronous CDMA Systems over AWGN Channel,” IEEE Trans.
user communication system, PDMA considers joint design of Signal Processing, vol.56, no.4, pp. 1616-1626, Jan. 2008.
transmitting and receiving, with SIC amenable pattern design [16] B. Ren, Y. Wang, X. Dai, etc, “Pattern Matrix Design based on PDMA
for 5G UL Application”, Submitted for publication. China Commun, May
at the transmitter and SIC based multi-user detection at the 2016.
receiver. [17] X. Dai, S. Sun, Y. Wang, “Reduced-Complexity Performance-Lossless
This paper provides a whole picture of PDMA, including (Quasi-)Maximum-Likelihood Detectors for S-QAM Modulated MIMO
Systems,” Elect. Lett., vol. 49, no. 11, pp. 724-725, May 2013.
system model, PDMA pattern design, PDMA detection algo- [18] X. Dai, Z. Zhang, K. Long, S. Sun, and Y. Wang, “Unequal error
rithm, application in 5G system, and performance evaluation. correcting capability aware iterative receiver for (parallel) Turbo coded
System-level simulation results show that PDMA can support communications,” IEEE Trans. Veh. Tech., vol. 63, no. 7, pp. 3446-3451,
Sep. 2014.
up to 6 times connected users and at least 30% improvement in [19] J. Xu, X. Dai, W. Ma, Y. Wang, “A component-Level Soft Interference
spectrum efficiency over OFDMA. Further research directions Cancellation Based Iterative Detection Algorithm for Coded MIMO
of PDMA include combination of PDMA encoding and mod- systems,” IEEE 80th Veh. Tech. Conf. (VTC Fall), September 2014.
[20] D. N. Liu and M. P. Fitz, “Low Complexity Affine MMSE Detector
ulation, low complexity detection algorithm, and combining for Iterative Detection-Decoding MIMO-OFDM systems,” IEEE Trans.
with multiple antennas, etc. Commun., vol. 56, no. 1, pp. 150-158, Jan. 2008.
At present, with the support of the 863 project (Item [21] B. Ren, X. Yue, W. Tang, etc, “Advanced IDD Receiver for PDMA
Uplink,” IEEE ICCC 2016, to be published.
No.2015AA01A709) by Chinese government, PDMA testbed [22] Rep. ITU-R M.2135-1, “Guidelines for Evaluation of Radio Interface
is being developed including multiple transmitters and multiple Technologies for IMT-Advanced,” Dec. 2009, [Online]. Available:
receivers. The PDMA field tests in both eMBB scenario and http://www.itu.int/dms pub/itu-r/opb/rep/R-REP-M.2135-1-2009-PDF-
E.pdf.
massive MTC scenario are expected in 2016. [23] Report ITU-R M.2320-0, Future technology trends of terrestrial IMT
systems, Nov. 2014, [Online]. Available: http://www.itu.int/ITU-R/.
[24] X. Dai et al., “Successive Interference Cancelation Amenable Multiple
ACKNOWLEDGMENT Access (SAMA) for Future Wireless Communications,” Proc. IEEE ICCS
This work was partly supported by the National High Sci- 2014, pp. 222-226, Nov. 2014.
[25] S. Kang, X. Dai, B. Ren, “Pattern Division Multiple Access for 5G,”
ence & Technology Plan (863 Plan, No. 2015AA01A709) and Telecommun. Network Technol., pp. 43-47, May 2015.
the National Science Fund for Distinguished Young Scholars
(Grant No. 61425012) in China. We would like to thank Dr.
Wang Yingmin from China Academy of Telecommunications
Technology and Prof. Dai Xiaoming from Beijing Science and
Technology University for their valuable comments.

R EFERENCES
[1] D. Tse, P. Viswanath. Fundamentals of Wireless Communication, Cam-
bridge Univ. Press, 2005.
[2] IMT-2020(5G) Promotion Group, Whitepapers, [Online]. Available:
http://www.imt-2020.org.cn/zh.

0018-9545 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2016.2596438, IEEE
Transactions on Vehicular Technology

12

Shanzhi Chen [SM’04] ([email protected]) Shaoli Kang ([email protected]) is the head expert
received his Ph.D. degree from Beijing University of 5G standardization in wireless innovation centre
of Posts and Telecommunications (BUPT), China, of China Academy of Telecommunication Technolo-
in 1997. He joined Datang Telecom Technology gy (CATT). From November 2000 to May 2005, Dr
& Industry Group in 1994, and has been serving Kang acted as a project manager in CATT, focusing
as CTO since 2008. He was a member of the on R&D of TD-SCDMA. Then she worked in the
steering expert group on information technology of communication centre of system research (CCSR)
the 863 Hi-Tech Research and Development Plan of University of Surrey as a research fellow, doing
of China from1999 to 2011. He is the director of research on projects from EPSRC and OFCOM and
State Key Laboratory of Wireless Mobile Commu- leading the antenna & propagation (AP) Club. From
nications, and the board member of Semiconductor September 2007, Dr Kang came back to CATT and
Manufacturing International Corporation (SMIC). He devoted his works to the acted as the vice chief engineer of TDD R&D product line, focusing on
research and development of TD-SCDMA 3G industrialization and TD-LTE- speeding up the standard and industrial progress of TDD technology. Since
advanced 4G standardization. He received 2001 and 2012 National Awards early 2011, Dr Kang has been working in the wireless innovation centre and
for Science and Technology Progress, China, the 2015 National Award for acted as the head expert, leading the 5G research in CATT. Till now she has
Technological Invention, China, and the 2014 Distinguished Young Scholar applied more than 50 patents, published more than 20 papers.
Award of National Natural Science Foundation, China. His current research
interests include network architecture, 5G mobile communication, and Internet
of Things (IoT). Shaohui Sun ([email protected]) received his
Ph.D. from Xidian University, Xi’an, China, in
2003. Form March 2003 to June 2006, he was a
postdoctoral research fellow at the Datang Telecom
technology and industry group, Beijing, China. From
Bin Ren ([email protected]) received the B.E. and June 2006 to December 2010, he worked at the
M.S. degrees from Beijing University of Posts and Datang mobile communications equipment company
Telecommunications, Beijing, China, in 2006 and Ltd., Beijing, where he had been deeply involved in
2009, respectively, and is currently studying for the development and standardization of the Third-
Ph.D. degree of Beijing University of Posts and Generation Partnership Project Long-Term Evolution
Telecommunications, Beijing, China. He joined the (3GPP LTE). Since January 2011, he has been the
Key Laboratory of Wireless Mobile Communication- Chief Technical Officer of Datang Wireless Mobile Innovation Center of
s, China Academy of Telecommunication Technol- the Datang Telecom technology and industry group. Now, his research area
ogy, Beijing, China in 2009. His current research of interest includes multiple antenna technology, heterogeneous wireless
interests include wireless communications theory networks, and relay.
and wireless communications systems.

Qiubin Gao ([email protected]) received his B.S. Kai Niu [S’00, M’04] ([email protected]) re-
and Ph.D. in control science and engineering from ceived the B. S. degree in Information Engineer-
Tsinghua University. He is currently a senior re- ing and Ph.D. degree in Signal and Information
search engineer at Datang Wireless Mobile Inno- Processing from Beijing University of Posts and T-
vation Center of China Academy of Telecommu- elecommunications (BUPT), China, in 1998 in 2003,
nications Technology (CATT). His current research respectively. Currently he is a Professor at the school
interests include physical layer design for mobile of information and communication engineering of
communication, multiple-antenna technology, CoM- BUPT. His research interests include coding theory
P, device to device communication, and system per- and its applications, space-time codes and broadband
formance evaluation. He is inventor/co-inventor of wireless communication. Since 2008 he is serving
more than 300 patents in wireless communications, as a senior member of the Chinese Institute of
and author/co-author of a number of journal and conference papers. He Electronics (CIE) and a committee member of the Information Theory Chapter
received URSI awards for Young Scientists in 2014. of CIE.

0018-9545 (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.