You are on page 1of 21

Received: 27 November 2018 Revised: 15 March 2019 Accepted: 5 May 2019

DOI: 10.1002/dac.4025

RESEARCH ARTICLE

Power domain cyclic spread multiple access: An


interference‐resistant mixed NOMA strategy

R. Baby Shalini | S. Lenty Stuwart

University College of Engineering,


Summary
Nagercoil, Anna University, Tamil Nadu,
India The next generation wireless access technology highly relies on nonorthogonal
multiple access (NOMA) technique. This paper proposes a novel power domain
Correspondence
S. Lenty Stuwart, University College of
cyclic spread multiple access (PDCSMA) scheme for the design of NOMA sys-
Engineering, Nagercoil, Anna University, tem with power domain superposition coding (SC) and cyclic spreading at the
Tamil Nadu, India. transmitter concurrent with symbol level successive interference cancellation
Email: lenty_s@yahoo.com
(SL‐SIC) at the receiver. Based on acceptable difference in channel gain, the
Funding information users are grouped together to form PDCSMA clusters, and the unique power
Science and Engineering Research Board
is allotted to each user in a cluster. The user with good channel condition is
(SERB), Department of Science and Tech-
nology (DST), Government of India, allotted less power, and the user with poor channel condition is allotted more
Grant/Award Number: ECR/2017/001188 power. Each PDCSMA cluster has its own spreading code, and the data of
every user in a cluster are cyclic spread with the same code. Each cluster sup-
ports the number of multipath components equivalent to the length of the
spreading code. The use of cyclic spreading makes the signal suffered by mul-
tipath fading less prone to intra cluster interference. The user signal is decoded
by minimum mean square error‐frequency domain equalization (MMSE‐FDE)
or maximal ratio combining (MRC)–based receiver in which weak user is
detected with hard decision, and strong user is detected with SIC. Compared
with conventional power domain NOMA (PDNOMA) and interleaved NOMA,
the proposed PDCSMA achieves better bit error rate (BER) performance and
assures guaranteed detection.

KEYWORDS
intersymbol interference, nonorthogonal multiple access, superposition coding, successive
interference cancellation

1 | INTRODUCTION

Massive multiinput multioutput (MIMO) connectivity, lower latency, and diverse quality of service are the minimum
requirements needed for the upcoming 5G and beyond 5G wireless technologies, and these characteristics are promised
by nonorthogonal multiple access (NOMA) schemes. In recent years, the concept and key ideas behind NOMA are
widely studied.1 Multiple access techniques are classified as orthogonal multiple access (OMA) and NOMA. In OMA,
in order to mitigate multiple access interference, every user is served on exclusively allocated frequency channel, time
slot, or code. However, when two users are contending to access two different resource blocks, the user with good chan-
nel is served. But the service is denied to the user with poor channel because of very low received signal strength. As a
result, in such scenario, the user fairness is not guaranteed in OMA. In contrast to OMA, NOMA proves its user fairness
Int J Commun Syst. 2019;e4025. wileyonlinelibrary.com/journal/dac © 2019 John Wiley & Sons, Ltd. 1 of 21
https://doi.org/10.1002/dac.4025
2 of 21 SHALINI AND STUWART

by providing service to both good and bad channel users in the same resource block at the same time. Here, the received
signal strength at the weak user is improved by allocating more power compared with that of the strong user. The ser-
vice to the weak user with poor channel condition is assured while the strong user with good channel condition shares
the same resource block with weak user. Hence, the overall system throughput is improved in NOMA due to the reuse
of the resource block.2 The major categories of NOMA are scrambling‐based NOMA, code‐based NOMA, interleaving‐
based NOMA, spreading‐based NOMA, and mixed NOMA.
Scrambling‐based NOMA uses different scrambling code for each user and adopts repetition coding or channel cod-
ing for multiuser signal decoding. A special case of scrambling‐based NOMA called power domain NOMA (PDNOMA)
scheme with all “1” scrambling code is investigated in literature.3 The power difference between the near and far users
due to near far effect is utilized to share the same resource block by multiple users. Superposition coding at the trans-
mitter and successive interference cancellation (SIC) at the receiver are used in the PDNOMA. In superposition coding,
higher transmit power is allocated to the weak user, and lower transmit power is allocated to the strong user. Hence,
weak user is less affected by strong user signal and can use simple decoding when compared with strong user. On
the other hand, weak user signal is stronger than that of the strong user signal at the strong user. Therefore, strong user
first detects the weak user signal, removes it from the received signal, and then decodes its own data. Resource spread
multiple access (RSMA) scheme is proposed for massive machine‐type communication (mMTC) scenario in which low
cross‐correlation long pseudorandom sequences are used.4 Although the scheme promises for increased connection den-
sity, the latency and complexity are high.
Coding‐based NOMA schemes are developed in which sparse code book overlapped in space, time, code, or fre-
quency is used. Pattern division multiple access (PDMA) for 5G communication is investigated in Dai et al5 where mul-
tiple users sharing the same resource are differentiated by designing the pattern. A low‐density signature‐signature
vector extension (LDS‐SVE) scheme derived from LDS code division multiple access (LDS‐CDMA) is proposed for the
uplink systems where signature vector is generated by spreading each modulated symbol onto multiple subcarriers.6
Coded data bits are mapped to multidimensional complex codewords from predefined code sets in time and frequency
domain. Serial SCMA scheme for overloaded and constellations with smaller Euclidean distance is presented in Han
et al.7 Lower detection complexity and better link performance are obtained.
In interleaving‐based NOMA system, different channel interleavers are used to overlap multiple users, and low rate
channel coding is used for user decoding. Interleaved division multiple access (IDMA)–based NOMA scheme separates
users by allocating different random interleave pattern to each user. Interleaved domain multiuser detection (MUD)
architecture for the IDMA receiver with low latency is proposed.8 In order to decrease the latency and complexity,
MUD is performed directly without deinterleaving the received frame. Cyclic interleaving‐assisted repetition division
multiple access (RDMA)–based NOMA separates different users and achieves both frequency and time diversity by
the use of repetition pattern based on cyclic shift.9 Low code rate spreading (LCRS)–based NOMA scheme is introduced
in which different bit‐level interleaving is introduced for each user.10
Spreading‐based NOMA uses nonorthogonal, nonsparse, short, and low cross‐correlation spreading sequences for
distinguishing multiple users. Multiuser‐shared access (MUSA) scheme is introduced for NOMA system in which
complex‐valued spreading sequences are used to transmit the user symbols in time and frequency resource block.11
Nonorthogonal coded access (NOCA)–based NOMA utilizes the cyclic‐shifted base sequences as spreading sequences.
NOCA scheme utilizing the low cross‐correlation Zadoff‐Chu (ZC) sequences is investigated in Wang et al.12 User
overloading and higher spectral efficiency are achieved at the expense of increased receiver complexity.
In mixed NOMA, two or more basic NOMA schemes are combined to reap the benefit out of these schemes. Inter-
leave grid multiple access (IGMA) is basically a combination of coding‐cum‐interleaving–based NOMA schemes. Bit‐
level interleaving and symbol level grid mapping process are carried out for each user to provide very good block error
rate (BLER) performance and to provide user multiplexing capability.13,14 A scrambling‐cum‐coding–based power
domain sparse code multiple‐access (PSMA) scheme is introduced in Moltafet et al.15 In this scheme, multiple users
share the same sparse code book and they are differentiated by assigning different power level to each user.
Despite the various NOMA and mixed NOMA schemes, intersymbol interference (ISI) in the NOMA clusters
degrades the performance of both weak and strong users irrespective of their user combining strategies. The present
study tackles this situation by developing a scrambling‐cum‐spreading–based NOMA scheme called power domain
cyclic spread multiple access (PDCSMA) in which cyclic spreading is introduced to each user in a PDNOMA cluster.
User pairing in each cluster is performed by exploiting the difference in the multipath propagation mechanism among
the users. The user pairing policy is presented to achieve optimum performance in the proposed scheme. The main con-
tributions of this paper are outlined as follows:
SHALINI AND STUWART 3 of 21

The brief review of NOMA system and the description of cyclic spreading are presented. Due to the complex nature
of the multipath channel, cyclic spreading is introduced in PDNOMA system to enhance the performance in the pres-
ence of ISI. The significance of the proposed PDCSMA over conventional PDNOMA is illustrated with the aid of bit
error rate (BER) analysis. The user pairing with respect to the Ricean factor is investigated. The performance of different
user pairing strategy associated with Rayleigh and Ricean channel is evaluated, and guidelines related to user grouping
problem are presented.
The rest of this paper is organized as follows: System model is presented in Section 2. Section 3 explains the simula-
tion results and analysis. Conclusions are given in Section 4.

2 | SYSTEM MODEL

Let us consider K users in a cluster of a PDCSMA system as shown in Figure 1. Let d(0),d(1), … ,d(K − 1) be the data vec-
tors of the users 0,1, … ,K − 1, respectively. The data vector of user k with Q data symbols is in the form
h i
ðkÞ ðkÞ ðkÞ ðkÞ
dðkÞ ¼ d0 d1 d2 … dQ−1 . The spreading code available for the PDCSMA cluster is in the form C0 = [c0 c1 c2 … cN − 1]
which is an m‐sequence of length N. M multipath components are considered for the proposed system with maximum
number of multipath components supported that is N. The unique power allotted to user k in the cluster is Pk = αkPT,
where αk and PT are fractional power applied to user k and total transmitted power, respectively. Let hk be the channel
gain of user k, and user 0 is considered as a weak user with poor channel gain, where |h0| < |h1| < … < |hK − 1|. Before
combining the multiple user signal, based on the superposition principle, the powers in descending order are assigned to
users in ascending order of their channel gain, ie, P0 > P1 > … > PK − 1. Then, the superposed signal is cyclic spread
which is a combination of spreading and interleaving with m‐sequence. Although the proposed PDCSMA looks similar
to the IDMA‐based NOMA, the former method assigns the same interleaving to different users whereas the latter
method adopts different interleaving strategy for each user.
The process of cyclic spreading, illustrated in Figure 2, makes the signal less prone to interference in the PDCSMA
cluster and can combat with multipath fading environment. Before transmitting the cyclic spread data over additive
white Gaussian noise (AWGN) channel, adequate numbers of cyclic prefix symbol chips are added at the start of the
data frame, and they can be later exploited by the receiver to reconstruct the signal from the faded received signal. It
is assumed that the transmitted signal undergoes multipath propagation due to the channel length M.
The weak user signal decoding in the PDCSMA cluster is performed with the following procedure. Signal from each
received path consists of cyclic spread data of addition of all user data multiplied with their fractional power. The cyclic
prefix is removed from the received signal, and deinterleaving is performed. The cyclic spreading at the transmitter
makes the signal received from different paths resolvable, and hence, K user signals from each path are despread with
different cyclic spreading codes derived from C0. That is, data from path 0 is obtained by despreading the deinterleaved
data with C0, and data from path 1 is obtained by despreading with C1 and so on. After despreading, the superposed
signal from all paths are decorrelated with inverse cross‐correlation matrix in order to conquer the errors caused by
the nonzero cross‐correlation of spreading codes. Interestingly, the superposed data from all the M despreaders are in
cyclic‐shifted forms in such a way that the data from path 0 are not subjected to any cyclic shift, data from path 1
are cyclic shifted by one data symbol, and so on. Since the data in each path experienced right cyclic shift with respect
to one another, the corresponding left cyclic shifts are applied to compensate them. Now, the user data from all the

FIGURE 1 Transceiver structure of the


proposed PDCSMA scheme for
P0 > P1 > … > PK − 1
4 of 21 SHALINI AND STUWART

FIGURE 2 Data format of proposed PDCSMA‐MRC scheme for K = 2, Q = 3, and M = 2 with P0 > P1

paths become resolvable, and hence, they are added together using maximal ratio combining (MRC) technique. Finally,
the weak user data are detected using hard decision.
Likewise, the strong user data in the PDCSMA cluster are detected with the following procedure. The cyclic prefix–
removed symbol chips undergo minimum mean square error frequency domain equalization (MMSE‐FDE) followed by
deinterleaving. The multipath fading equalized output signal is then despread with spreading code C0. Based on αk and
M, the output from either MRC receiver or MMSE‐FDE receiver is subjected to SIC to decode the weak user signal.
The concept of proposed MRC receiver is well explained with Figure 2 which illustrates the systematic procedure of
cyclic spreading and also describes the entire detection process under two‐path fading channel. Two users namely, the
h i
ð0Þ ð0Þ ð0Þ
weak and strong user are considered based on the channel gain, ie, h0 < h1. Furthermore, d0 d1 d2 and
h i
ð1Þ ð1Þ ð1Þ
d0 d1 d2 are the user data corresponding to weak and strong users, respectively. The low channel gain user is given
more power, and the high channel gain user is given relatively less power, ie, P0 > P1.
The superposed data of two users are spread with a spreading code C0. After spreading, the symbol chips are subjected
to cyclic interleaving which involves two steps: Firstly, the symbol chip structure is flip up‐down, and the next is cyclic
shift. In the flip up‐down process, the data structure is up‐down reversed. In the cyclic shifting, the chips of data symbol 0
are down shifted by one chip, the chips of data symbol 1 are two chips cyclic shifted, and the chips corresponding to sym-
bol 2 are three chips cyclic shifted. Now, the data structure is reshaped into row‐wise serial form in order to make the
signal suitable for serial transmission. Before transmission, one symbol chip is cyclic prefixed in order to combat with
multipath fading with M = 2. The transmitted data consist of four blocks with index l which corresponds to the spreading
code length and three chips in each block with index q which corresponds to the number of data symbols transmitted.
SHALINI AND STUWART 5 of 21

At the receiver after the cyclic prefix is removed, the data are reshaped in such a way that it consists of three blocks
with index q and four chips in each block with index l. After reshaping, the data from path 0 and path 1 are obtained by
despreading with spreading code C0 and C1, respectively. From path 0, the addition of two user data is obtained, and
from path 1, the addition of two user data with one symbol right cyclic shift is obtained. Hence, one symbol left cyclic
shift is applied to the data from path 1. Next, data symbols from both the paths are combined by MRC, after which the
weak user data is obtained by hard decision and the strong user data is obtained by SIC.
Further, without the loss of generality, the description of the proposed PDCSMA scheme is presented with symbol
chip representation in which the range of chip index q and block index l correspond to 0 ≤ q ≤ Q − 1 and 0 ≤ l ≤ N − 1,
respectively. After cyclic spreading, the data frame of K user superposed signal contains a series of symbol chips in
which the qth symbol chip of lth block can be represented as,
pffiffiffiffiffiffiffiffiffiffi
dq ½l ¼ ∑k¼0 dðqkÞ αk PT cðqþQlÞ modN
K−1
(1)

The waveform of transmitted data frame appended with cyclic prefixed symbol chips is in the following form:
Q−1 N−1 Q−1
x ðt Þ ¼ ∑q¼0 dq ½N − 1βðt − qT c Þ þ ∑l¼0 ∑q¼0 dq ½lβðt − ðlQ þ q þ M ÞT c Þ (2)

Substituting (1) in (2) yields,


pffiffiffiffiffiffiffiffiffiffi
xðt Þ ¼ ∑q¼0 ∑k¼0 dðqkÞ
Q−1 K−1
αk PT cðqþQðN−1ÞÞ modN βðt − qT c Þ
pffiffiffiffiffiffiffiffiffiffi
þ ∑l¼0 ∑q¼0 ∑k¼0 dðqkÞ αk PT cðqþQlÞ modN βðt − ðlQ þ q þ M ÞT c Þ
N−1 Q−1 K−1
(3)

The first term in (3) represents cyclic prefix, and the second term represents the cyclic spread data of K user superposed
signal. A rectangular function β(t) is used to represent the pulse shaping characteristics of the symbol chips. The chan-
nel considered here is the Ricean/Rayleigh fading channel with AWGN. In mathematical form, the channel impulse
response is represented as,
M−1
hðt Þ ¼ ∑m¼0 hm δðt − τ m Þ (4)

where hm and τm represent the mth path normalized instantaneous complex gain and the time delay, respectively. δ(t) is
the Dirac delta function
Each PDCSMA cluster is formed to accommodate two users namely, the weak user and the strong user. Based on the
user pairing in each cluster, three different scenarios are considered. The detection at the cluster 1 is considered as sce-
nario 1 which is depicted in Figure 3. Detection of cluster 2 and cluster 3 is considered as scenario 2 and scenario 3,
respectively.
In scenario 1, user equipment 1 (UE1) is the strong user with good channel condition of Ricean factor 20 dB and UE0
is the weak user with poor channel condition of Ricean factor −20 dB. Hence, more power (P0) is given to the transmit-
ted data of UE0, and relatively less power (P1) is applied to UE1 data. The base station allows both the signals to be
transmitted over the same spectrum resource using the principle of superposition. When the signal reaches the weak
user over the Rayleigh channel, the strong user (user 1) is considered to be fully vanished. Hence, UE0 receives only
its signal, and it decodes its data by hard decision. At the same time, the strong user receives the superposed signal
of both the users and with the successive interference cancellation, the weak user signal is first decoded and removed

FIGURE 3 Scenarios considered for the proposed scheme


6 of 21 SHALINI AND STUWART

from the received signal, and then, the strong user signal is decoded. Similar to scenario 1, in scenario 2, Ricean and
Rayleigh users are grouped together to share the system resources with Ricean factor 20 and 0 dB. In the third scenario,
both the users possess a Rayleigh channel but having different Ricean factors of 0 and −20 dB.
The received signal at the UE in a PDCSMA cluster is the linear convolution of the transmitted signal x(t) and the
channel impulse response h(t) with the addition of WGN component n(t), which is in the following form:

yðt Þ ¼ x ðt Þ⨂hðt Þ þ nðt Þ (5)

Assuming quasistatic channel, the symbol chip representation of qth symbol chip of lth block after cyclic chip removal
is given as,

K−1 ðkÞ pffiffiffiffiffiffiffiffiffiffi K−1 M−1 ðkÞ pffiffiffiffiffiffiffiffiffiffi


yq ½l ¼ ∑k¼0 dðqþQÞ modQ αk PT cðqþQlÞ modN h0 þ ∑k¼0 ∑m¼1 dðqþQ−mÞ modQ αk PT cðqþQlþQmÞ modN hm þ nq ðlÞ (6)

The first term in (6) represents the line of sight (LOS) component (ie, m = 0), the second term represents the scattered
components, and the third term is the corresponding AWGN component. Based on the channel state information (CSI)
and grouping of users, the receiver selection is done. The proposed PDCSMA consists of two receivers which are MRC
and MMSE‐FDE; the output of either of two is chosen by the receiver selection block based on αk and M. Signal detec-
tion based on MRC and MMSE is discussed in the following subsections.

2.1 | MRC receiver

In MRC receiver, the cyclic prefix removed signal yq[l] in (6) is subjected to deinterleaving so that the data are arranged
in some particular order which is vital for despreading. After deinterleaving l becomes (l − q)/Q, the resultant data
frame has block index q with 0 ≤ q ≤ N − 1 and chip index l with 0 ≤ l ≤ Q − 1. The symbol chip representation of
deinterleaved data is described as,

K−1 ðkÞ pffiffiffiffiffiffiffiffiffiffi K−1 M−1 ðk Þ pffiffiffiffiffiffiffiffiffiffi


wl ½q ¼ ∑k¼0 dðqþQÞ modQ αk PT cðqþQððl−qÞ=QÞÞ modN h0 þ ∑k¼0 ∑m¼1 dðqþQ−mÞ modQ αk PT cðqþQððl−qÞ=QÞþQmÞ modN hm
þ nq ððl − qÞ=QÞ
K−1 ðkÞ pffiffiffiffiffiffiffiffiffiffi M−1 K−1 ðkÞ pffiffiffiffiffiffiffiffiffiffi
¼ ∑k¼0 dðqþQÞ modQ αk PT cðlÞ modN h0 þ ∑m¼1 ∑k¼0 dðqþQ−mÞ modQ αk PT cðlþQmÞ modN hm þ nl ðqÞ (7)

With proper rearrangement made in (7) of received signal, the data can be reconstructed in such a way that the added
data of all the users in path 0 are despread with spreading code C0 and the added data of all the users in the path 1 is
despread with spreading code C1 and so on. The qth symbol corresponding to jth path after despreading the
deinterleaved data with spreading code Cj is described as

N−1
Aq; j ¼ ∑l¼0 wl ½qcðlþQjÞ mod N (8)

Substituting (7) in (8) yields,

K−1 ðkÞ pffiffiffiffiffiffiffiffiffiffi N−1


Aq; j ¼ ∑k¼0 dðqþQÞ modQαk PT ∑l¼0 cðlÞ modN cðlþQjÞ modN h0
K−1 M−1 ðkÞ pffiffiffiffiffiffiffiffiffiffi N−1
þ ∑k¼0 ∑m¼1 dðqþQ−mÞ modQ αk PT ∑l¼0 cðlþQmÞ modN cðlþQjÞ modN hm þ nq; j (9)

The LOS and scattered components terms in (9) are in the form of the added symbols of all the users multiplied by their
associated power levels, channel coefficients, and correlation coefficients of spreading codes. Let
N−1
ρm; j ¼ ∑l¼0 cðlþQmÞ modN cðlþQjÞ modN be the correlation coefficient arises when m th path data is despread with the spread-
ing code Cj. Using ρm,j, the despread symbol in (9) can be rewritten as,

K−1 ðkÞ pffiffiffiffiffiffiffiffiffiffi K−1 M−1 ðkÞ pffiffiffiffiffiffiffiffiffiffi


Aq; j ¼ ∑k¼0 dðqþQÞ modQ αk PT h0 ρ0; j þ ∑k¼0 ∑m¼1 dðqþQ−mÞ modQ αk PT hm ρm; j þ nq; j (10)

In its equivalent vector form, (10) can be rewritten as,

Aq ¼ RHBq þ nq (11)
SHALINI AND STUWART 7 of 21

where R is the correlation matrix of spreading codes which is given by,

2 3
ρ0;0 ρ0;1 ⋯ ρ0;M−1
6 7
6 ρ1;0 ρ1;1 … ρ1;M−1 7
R¼6
6 ⋮
7
7 (12)
4 ⋮ ⋱ ⋮ 5
ρM−1;0 ρM−1;1 ⋯ ρM−1;M−1

The channel matrix is given by,

2 3
h0 0 ⋯ 0
6 0 … 0 7
6 h1 7
H¼6 7 (13)
4 ⋮ ⋮ ⋱ ⋮ 5
0 0 ⋯ hM−1

The data vector is written as,

h iT
K−1 pffiffiffiffiffiffiffiffiffiffi ðkÞ K−1 pffiffiffiffiffiffiffiffiffiffi ðkÞ K−1 pffiffiffiffiffiffiffiffiffiffi ðk Þ
Bq ¼ ∑k¼0 αk PT dðqþQÞ modQ ; ∑k¼0 αk PT dðqþQ−1Þ modQ ; …; ∑k¼0 αk PT dðqþQ−ðM−1ÞÞ modQ (14)

The noise vector is,


h iT
ðkÞ ðkÞ ðkÞ
nq ¼ nq;0 ; nq;1 ; …; nq;M−1 (15)

In order to remove the correlation coefficients, the inverse of correlation matrix is multiplied on both sides of (11),
which results in,

Dq ¼ R−1 Aq ¼ R−1 RHBq þ R−1 nq ¼ HBq þ R−1 nq (16)

Using (10), the jth element in (16) corresponding to the jth path qth symbol is described as,

K−1 ðkÞ pffiffiffiffiffiffiffiffiffiffi ðρ Þ


Dq; j ¼ ∑k¼0 dðqþQ− jÞ modQ αk PT hj þ nq; j (17)

Further, each element associated with scattered component in (17) is cyclic shifted by replacing q = q+j and are com-
bined together with MRC. Hence using (17), the qth output symbol of MRC is written as,

pffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffi M−1  2


Dq;MRC ¼ ∑j¼0 Dqþ j; j h*j ¼ ∑j¼0 ∑k¼0 dðqkÞ αk PT hj h*j þ nqþ j; j ¼ ∑k¼0 dðqkÞ αk PT ∑j¼0 hj  þ nq; j
M−1 M−1 K−1 K−1 ðMRCÞ
(18)

The resultant signal at arbitrary user equipment k is written as,


pffiffiffiffiffiffiffiffiffiffi M−1  2
Dq;MRC ¼ ∑k¼0 dðqkÞ αk PT ∑j¼0 hj;k  þ nq; j;k
ðkÞ K−1 ðMRCÞ
(19)

ðMRCÞ
where hj,k is the kth user, jth path channel coefficient and nq; j;k is the noise component. From (19), it is clear that MRC
output is the sum of K user data, and the output is free from ISI due to multipath components in the PDCSMA cluster.
Further, the user data are detected by applying either hard decision or SL‐SIC. Assuming two users in each PDCSMA
cluster, the detection of weak and strong users is described in the following subsections.

2.1.1 | Weak user detection

The strong user transmission power from the base station is low when compared with weak user counterpart. Hence, at
the weak user, the strong user signal strength is very poor and insignificant. The weak user with poor channel is
8 of 21 SHALINI AND STUWART

decoded with hard decision by dividing the total received signal with associated weak user power. Thus from (19), the
weak user (user 0) is decoded as,

Dq;MRC  K−1 ðkÞ pffiffiffiffiffiffiffiffiffiffi M−1  2  pffiffiffiffiffiffiffiffiffiffi


ð0Þ
Dð0Þ ¼ p ffiffiffiffiffiffiffiffiffiffi ¼ ∑ d α P ∑
k T j¼0
 hj;0
 þ n
ðMRCÞ
= α0 P T (20)
q
α0 P T k¼0 q q; j;0

h i
ð0Þ ð0Þ ð0Þ
where the detected data in vector form are represented as Dð0Þ ¼ D0 ; D1 …Dðq0Þ ; … DQ−1 . Hence, the qth data symbol
from weak user is detected as follows,
n h io n h pffiffiffiffiffiffiffiffiffiffi    pffiffiffiffiffiffiffiffiffiffiio
eð0Þ ¼ sgn real Dð0Þ ¼ sgn real ∑K−1 dðkÞ αk PT ∑M−1 hj;0 2 þ nðMRCÞ = α0 PT
d (21)
q q k¼0 q j¼0 q; j;0

2.1.2 | Strong user detection

The weak user transmission power from the base station is high when compared with that of strong user. This power
disparity forces the strong user to decode its signal by SIC. The general SIC procedure is as follows:

Step 1: The weak user is first decoded with hard decision as explained in Section 2.1.1.
Step 2: Decoded weak user signal is subtracted from the total received signal.
Step 3: The resultant signal is subjected to hard decision to decode the strong user.

Considering two users in the PDCSMA cluster, using (19) and (21), the strong user (user 1) decoded signal using SIC
is written as follows.
 ð0Þ pffiffiffiffiffiffiffiffiffiffi
ð1Þ
Dq;MRC − de α0 P T
q
Dðq1Þ ¼ pffiffiffiffiffiffiffiffiffiffi (22)
α1 P T

On substituting (19) and (21), (22) is written as,


 pffiffiffiffiffiffiffiffiffiffi M−1  2  pffiffiffiffiffiffiffiffiffiffi
∑k¼0 dðqkÞ αk PT ∑j¼0 hj;1  þ nq; j;1
ðMRCÞ
Dð1Þ
K−1
q ¼ = α1 P T
 n h p ffiffiffiffiffiffiffiffiffiffi    pffiffiffiffiffiffiffiffiffiffiiopffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffi (23)
− sgn real ∑k¼0 dðqkÞ αk PT ∑j¼0 hj;0  þ nq; j;0 = α0 PT
K−1 M−1 2 ðMRCÞ
α0 PT = α1 PT

Hence, using (23), the detected qth data symbol from strong user is written as,
n h io n h  pffiffiffiffiffiffiffiffiffiffi
eð1Þ ¼ sgn real Dð1Þ ¼ sgn real K−1 ðk Þ pffiffiffiffiffiffiffiffiffiffi M−1  2
 þ nðMRCÞ = α1 PT
d q q ∑k¼0 qd α k P T ∑ j¼0 h j;1 q; j;1
 n h  pffiffiffiffiffiffiffiffiffiffiiopffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiio (24)
K−1 ðkÞ pffiffiffiffiffiffiffiffiffiffi M−1  2
sgn real ∑k¼0 dq αk PT ∑j¼0 hj;0  þ nq; j;0 = α0 PT
ð MRC Þ
− α0 P T = α1 P T

In general, the decoding of arbitrary user i excluding the weak user is done by,
  pffiffiffiffiffiffiffiffiffiffi
i−1 eðk Þ pffiffiffiffiffiffiffiffiffiffi
DðqiÞ ¼ DðiÞ q;MRC − ∑k¼0 d q α k P T = αi PT (25)

Using (19) and (21), (25) is rewritten as,


 pffiffiffiffiffiffiffiffiffiffi M−1  2  pffiffiffiffiffiffiffiffiffiffi
∑k¼0 dðqkÞ αk PT ∑j¼0 hj;i  þ nq; j;i = αi PT
ðMRC Þ
DðqiÞ ¼
K−1

  n h pffiffiffiffiffiffiffiffiffiffi M−1  2  pffiffiffiffiffiffiffiffiffiffiiopffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffi (26)


− ∑k¼0 sgn real ∑k¼0 dðqkÞ αk PT ∑j¼0 hj;k  þ nq; j;k = αk PT
i−1 K−1 ðMRCÞ
αk PT = αi PT

Hence, using (26), the qth data symbol from user i is detected in MRC‐based receiver as follows:
SHALINI AND STUWART 9 of 21

n h io n h pffiffiffiffiffiffiffiffiffiffi M−1  2  pffiffiffiffiffiffiffiffiffiffi


eðiÞ ¼ sgn real DðiÞ ¼ sgn real
d ∑k¼0 dðqkÞ αk PT ∑j¼0 hj;i  þ nq; j;i
K−1 ðMRCÞ
= αi P T
q q
  n h p ffiffiffiffiffiffiffiffiffiffi    pffiffiffiffiffiffiffiffiffiffiiopffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiio (27)
− ∑k¼0 sgn real ∑k¼0 dðqkÞ αk PT ∑j¼0 hj;k  þ nq; j;k = αk PT
i−1 K−1 M−1 2 ðMRCÞ
αk PT = αi PT

2.2 | MMSE‐FDE receiver

MMSE criterion minimizes the mean square error between the transmitted signal and the equalized output. The MMSE
equalization is implemented in frequency domain wherein fast Fourier transforms (FFTs) of the cyclic prefix removed
received signal and channel impulse response are utilized. The received signal y(t) represented in (5) is subjected to
cyclic removal thereby obtaining u(t), and hence, U( f ) is written as

U ð f Þ ¼ FFT fuðt Þg (28)

Using (4), the channel transfer function is written as


n o
M−1
H ð f Þ ¼ FFT fhðt Þg ¼ FFT ∑m¼0 hm δðt − τ m Þ (29)

The MMSE equalization requires the knowledge about the channel coefficient and the variance of noise σ2. The equal-
ization coefficient based on the MMSE criterion results in H*( f )/(|H( f )|2+σ2), and the MMSE equalized signal is repre-
sented as

H*ð f Þ U ð f Þ
Y mmse ð f Þ ¼ (30)
j H ð f Þj 2 þ σ 2

where H*( f ) represents the complex conjugate of channel transfer function H( f ). Subsequently, the signal is brought
back to the time domain by taking IFFT on (30) which results in
!
H*ð f Þ U ð f Þ
ymmse ðtÞ ¼ IFFTðY mmse ð f ÞÞ ¼ IFFT (31)
j H ð f Þj 2 þ σ 2

Hence using (6), the qth symbol chip of lth block in the MMSE equalized signal can be represented as

K−1 ðkÞ pffiffiffiffiffiffiffiffiffiffi


yq;mmse ½l ¼ ∑k¼0 dðqþQÞ modQ αk PT cðqþQlÞ modN þ nq ðlÞ (32)

where nq(l) is the noise component in the lth block of qth chip. yq,mmse[l] is subjected to deinterleaving; thereafter, l
becomes (l − q)/Q, and the resultant data frame comprises block index q with 0 ≤ q ≤ N − 1 and chip index l
with 0 ≤ l ≤ Q − 1. The symbol chip representation of deinterleaved data is described as

K−1 ðkÞ pffiffiffiffiffiffiffiffiffiffi


wl;mmse ½q ¼ ∑k¼0 dðqþQÞ modQ αk PT cðlÞ þ nq;l (33)

where nq,l is the noise component. By despreading the deinterleaved signal with the spreading code C0, (33) becomes

N−1
Dq;mmse ¼ ∑l¼0 wl;mmse ½qcðlÞ (34)

By substituting (33) in (34) we get,


 pffiffiffiffiffiffiffiffiffiffi  pffiffiffiffiffiffiffiffiffiffi N−1
K−1 ðkÞ
Dq;mmse ¼ ∑l¼0 ∑k¼0 dðqþQÞ modQ αk PT cðlÞ þ nq;l cðlÞ ¼ ∑k¼0 dðqkÞ αk PT ∑l¼0 c2ðlÞ þ ∑l¼0 nq;l cðlÞ
N−1 K−1 N−1

K−1 ðk Þ pffiffiffiffiffiffiffiffiffiffi
¼ ∑k¼0 dðqÞ αk PT þ nq;mmse (35)

N−1
where ∑l¼0 c2ðlÞ ¼ 1 and nq,mmse is the noise component. From (35), it is obvious that the ISI component disappears after
10 of 21 SHALINI AND STUWART

despreading. Further, the detection of strong and weak user data in a two‐user PDCSMA cluster are presented in the
following subsections.

2.2.1 | Weak user detection

Since the strong user signal strength is very poor at the weak user, the weak user with poor channel is decoded with
hard decision by dividing the total received signal with associated weak user power. Thus, the weak user (user 0) is
decoded using (35) as

Dq;mmse  K−1 ðkÞ pffiffiffiffiffiffiffiffiffiffi  pffiffiffiffiffiffiffiffiffiffi


Dðq0Þ ¼ pffiffiffiffiffiffiffiffiffiffi ¼ ∑k¼0 dðqÞ αk PT þ nq;mmse = α0 PT (36)
α0 PT

Hence, using (36), the qth data symbol from weak user is detected as follows,
n h io n h pffiffiffiffiffiffiffiffiffiffi  pffiffiffiffiffiffiffiffiffiffiio
eð0Þ ¼ sgn real Dð0Þ ¼ sgn real ∑K−1 dðkÞ αk PT þ nq;mmse = α0 PT
d (37)
q q k¼0 ðqÞ

2.2.2 | Strong user detection

The weak user transmission power is high when compared with that of strong user, and hence, the strong user
with good channel is decoded by SIC. Considering two users in a PDCSMA cluster, using (35) and (37), the strong user
(user 1) decoded signal is written as
 ð0Þ pffiffiffiffiffiffiffiffiffiffi
Dq;mmse − d e α0 PT
q
Dðq1Þ ¼ pffiffiffiffiffiffiffiffiffiffi
 α1 PT  pffiffiffiffiffiffiffiffiffiffi  n h  pffiffiffiffiffiffiffiffiffiffiiopffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffi
K−1 ðk Þ pffiffiffiffiffiffiffiffiffiffi K−1 ðkÞ pffiffiffiffiffiffiffiffiffiffi
¼ ∑k¼0 dðqÞ αk PT þ nq;mmse = α1 PT − sgn real ∑k¼0 dðqÞ αk PT þ nq;mmse = α0 PT α0 PT = α1 PT
(38)

Hence using (38), the qth data symbol detected from strong user is written as
n h io n h  pffiffiffiffiffiffiffiffiffiffi
eð1Þ ¼ sgn real Dð1Þ ¼ sgn real K−1 ðkÞ pffiffiffiffiffiffiffiffiffiffi
d q q ∑ d
k¼0 ðqÞ α k P T þ nq;mmse = α1 P T
 n h p ffiffiffiffiffiffiffiffiffiffi  pffiffiffiffiffiffiffiffiffiffiiopffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiio (39)
K−1 ðk Þ
− sgn real ∑k¼0 dðqÞ αk PT þ nq;mmse = α0 PT α0 PT = α1 PT

In general, the decoding of arbitrary user i excluding the weak user is written as
  pffiffiffiffiffiffiffiffiffiffi
i−1 eðk Þ pffiffiffiffiffiffiffiffiffiffi
DðqiÞ ¼ Dq;mmse − ∑k¼0 d q α k P T = αi PT (40)

Substitute (35) and (37) in (40), we get,


  pffiffiffiffiffiffiffiffiffiffi  h n  pffiffiffiffiffiffiffiffiffiffioipffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffi
K−1 ðk Þ pffiffiffiffiffiffiffiffiffiffi K−1 ðkÞ pffiffiffiffiffiffiffiffiffiffi
DðqiÞ ¼
i−1
∑k¼0 dðqÞ αk PT þ nq;mmse = αi PT − ∑k¼0 sgn real ∑k¼0 dðqÞ αk PT þ nq;mmse = αk PT αk PT = αi PT
(41)

Hence using (41), the qth data symbol detected from user i in MMSE‐FDE‐based receiver is written as
n h io n h  pffiffiffiffiffiffiffiffiffiffi
eðiÞ ¼ sgn real DðiÞ ¼ sgn real K−1 ðkÞ pffiffiffiffiffiffiffiffiffiffi
d q q ∑k¼0 dðqÞ αk PT þ nq;mmse = αi PT
 h n  pffiffiffiffiffiffiffiffiffiffioipffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiio (42)
i−1 K−1 ðkÞ pffiffiffiffiffiffiffiffiffiffi
− ∑k¼0 sgn real ∑k¼0 dðqÞ αk PT þ nq;mmse = αk PT αk PT = αi PT

On comparing (27) and (42), it is clear that the MRC receiver gives an additional diversity due to the presence of the
M − 1 2
term ∑ hj;i  which indicates that the multipath components are resolvable. Conversely, the absence of such term
j¼0
SHALINI AND STUWART 11 of 21

in (42) may also lead to performance improvement in MMSE‐FDE receiver over MRC receiver when the multipath com-
ponents are insignificant compared with LOS. This argument is studied in detail with the help of simulations and pre-
sented in the following section.

3 | S I M U L A TI O N R E S U L T S

The parameters considered for simulation are given in Table 1. The simulation result gives the BER comparison among
proposed PDCSMA‐MRC, PDCSMA‐FDE, and conventional PDNOMA‐FDE with three different scenarios as given in
Table 2.
The cyclic behavior of the channel with respect to the spreading code used in the proposed PDCSMA is illustrated in
Figure 4. Let the spreading code in the first row be C0 = [c0 c1 c2 c3]. The user with three bits of information is consid-
ered. Hence, the spreading code is repeated for three times in the second row. The cyclic interleaved codes are repre-
sented in the third row. The next set of three rows represents the multipath components at the receiver. Path 0
represents the LOS component with spreading code C0. Similarly, path 1 and path 2 are cyclic‐shifted versions of path
0. That is, the spreading codes C0, C1, and C2 are present in path 0, path 1, and path 2, respectively, which makes the
multipath components orthogonal and also resolvable.
Scenario 1: Figures 5 and 6 compare the two users' BER performance of the proposed PDCSMA‐MRC, PDCSMA‐
FDE, and conventional PDNOMA‐FDE for 2 and 15 multipaths, respectively. As given in Table 2, scenario 1 considers
the weak and strong users with Ricean factors Kr = − 20 dB and Kr = 20 dB, respectively. The two‐path amplitude delay
profiles are [0.1 1] and [1 0.1] for Kr = − 20 dB and Kr = 20 dB, respectively. The power allocation for weak and strong
user are P0 = 0.8 and P1 = 0.2, respectively.
Figure 5 shows the plot of BER versus Eb/N0 for the proposed PDCSMA‐MRC, PDCSMA‐FDE, and conventional
PDNOMA‐FDE for two users with two‐path channel. For strong user at Kr = 20 dB, very strong LOS is present. All
the three systems show similar BER performance at low Eb/N0; however, marginal performance gap is observed at high
Eb/N0. Specifically, at high Eb/N0, superior performance is attained for the proposed PDCSMA‐MRC because the mul-
tipath components become resolvable. But, performance of PDCSMA‐FDE is slightly inferior to PDCSMA‐MRC due
to weak scattered component. For weak user at Kr = − 20 dB, the LOS is weak. Hence, large performance gap is per-
ceived between strong and weak users. Moreover, the performance gap between the strong and weak users increases
nonlinearly from low Eb/N0 to high Eb/N0. This shows the influence of relatively small noise variance in the perfor-
mance of weak user. The performance of PDNOMA‐FDE and proposed PDCSMA‐FDE are one and the same for the
entire Eb/N0 range. Although the proposed PDCSMA‐MRC shows no performance difference at low Eb/N0, the superior
performance difference is observed at high Eb/N0 due to resolvable multipath components.
Figure 6 depicts the BER performance of the proposed PDCSMA‐MRC, PDCSMA‐FDE, and conventional PDNOMA‐
FDE for two users with 15‐path channel. At Kr = 20 dB, the BER of strong user becomes almost same for both proposed

TABLE 1 Transmission parameters for BER simulation

Parameter Weak User (User 0) Strong User (User 1)

Transmission power 0.8 0.2


Modulation BPSK BPSK
Channel Rayleigh Rayleigh/Ricean
Ricean factor −20 or 0 dB 0/20 dB

TABLE 2 Channel and power allocation for various scenarios

Ricean Factor Kr, dB Power Allocation


Scenario Weak User Strong User Weak User (P0) Strong User (P1)

1 −20 (Rayleigh) 20 (Ricean) 0.8 0.2


2 0 (Rayleigh) 20 (Ricean) 0.8 0.2
3 −20 (Rayleigh) 0 (Rayleigh) 0.8 0.2
12 of 21 SHALINI AND STUWART

FIGURE 4 Illustration of cyclic behavior of the channel for the proposed PDCSMA frame with spreading codes C0 = [c0 c1 c2 c3],
C1 = [c3 c0 c1 c2], and C2 = [c2 c3 c0 c1.]

FIGURE 5 BER comparison of PDCSMA‐MRC, PDCSMA‐FDE, and PDNOMA‐FDE of scenario 1 with P0 = 0.8, P1 = 0.2, and M = 2 for
various Ricean factors Kr

PDCSMA‐FDE and PDNOMA‐FDE. When the BER approaches 10−5, both have an improved Eb/N0 of 3 dB over pro-
posed PDCSMA‐MRC. The drastic change in the performance of proposed PDCSMA‐MRC for 15‐path channel when
compared with 2‐path channel is due to the fact that the energy in each of the scattered components is relatively low
and become insignificant at the combiner. At Kr = − 20 dB, the weak user BER performance in proposed PDCSMA‐
MRC is superior when compared with that of the proposed PDCSMA‐FDE and conventional PDNOMA‐FDE. This is
due to the diversity achieved from significant scattered components. That is, when the number of paths is increased
SHALINI AND STUWART 13 of 21

FIGURE 6 BER comparison of PDCSMA‐MRC, PDCSMA‐FDE, and PDNOMA‐FDE of scenario 1 with P0 = 0.8, P1 = 0.2, and M = 15 for
various Ricean factors Kr

from 2 to 15 paths, better diversity gain is achieved by the weak user. Interestingly, the performance at very low Eb/N0 is
even better compared with systems with two‐path channel, and at moderate Eb/N0, still the performance is better com-
pared with that of PDCSMA‐MRC with two‐path channel. But, such dominance weakens at very high Eb/N0 because the
system reduces the role of noise variance on the performance. Likewise, the performance of PDCSMA‐FDE and conven-
tional PDNOMA‐FDE is always inferior to proposed PDCSMA‐MRC. This performance degradation is as a result of no
diversity by FDE. Furthermore, the performance gap between proposed PDCSMA‐FDE and conventional PDNOMA‐
FDE is regarded as cyclic spread gain in the proposed system.
Scenario 2: Figures 7 and 8 compare the two users' BER performance of the proposed PDCSMA‐MRC, PDCSMA‐
FDE, and conventional PDNOMA‐FDE for 2 and 15 multipaths, respectively. As given in Table 2, scenario 2 considers
the weak and strong users with Ricean factors Kr = 0 dB and Kr = 20 dB, respectively. The two‐path amplitude delay

FIGURE 7 BER comparison of PDCSMA‐MRC, PDCSMA‐FDE, and PDNOMA‐FDE of scenario 2 with P0 = 0.8, P1 = 0.2, and M = 2 for
various Ricean factors Kr
14 of 21 SHALINI AND STUWART

FIGURE 8 BER comparison of PDCSMA‐MRC, PDCSMA‐FDE, and PDNOMA‐FDE of scenario 2 with P0 = 0.8, P1 = 0.2, and M = 15 for
various Ricean factors Kr

profile are [1 1] and [1 0.1] for Kr = 0 dB and Kr = 20 dB, respectively. The power allocation for weak and strong user are
P0 = 0.8 and P1 = 0.2, respectively.
Figure 7 exhibits the BER performance of the proposed PDCSMA‐MRC, PDCSMA‐FDE, and conventional
PDNOMA‐FDE for two users with two‐path channel. At Kr = 20 dB, the performance of strong user is similar to the
one discussed in Figure 5 of scenario 1. The identical performance is obtained because the same user power ratio is
maintained in scenario 1 and scenario 2. At Kr = 0 dB, the weak user performance in the proposed PDCSMA‐MRC little
bit deviates from its strong user counterpart. This close performance achieved for the weak user is due to the fact that
the LOS and scattered component are made equally strong at Kr = 0 dB. Additionally, the performance of PDCSMA‐
MRC is superior to that of PDCSMA‐FDE and PDNOMA‐FDE because the multipath components are resolvable and
diversity is achieved in the MRC. Furthermore, the performance gap of the PDCSMA‐FDE and PDNOMA‐FDE is min-
imal at low Eb/N0, and the gap is large at high Eb/N0. The reason behind this is that the scattered component introduces
strong interference to LOS component. Hence, the decrease in noise variance shows only slight improvement in the
BER performance of PDNOMA‐FDE.
Figure 8 shows the BER performance of the proposed PDCSMA‐MRC, PDCSMA‐FDE, and conventional PDNOMA‐
FDE for two users with 15‐path channel. At Kr = 20 dB, when the number of paths is increased to 15 for strong user,
both the PDCSMA‐FDE and conventional PDNOMA‐FDE retain the same result as that of 2 paths and the BER perfor-
mance of PDCSMA‐MRC is degraded by Eb/N0 of 2 dB at BER=10−5. That is, the performance of strong user is similar to
the one discussed in Figure 6 of scenario 1. The identical performance is obtained because the same user power ratio is
maintained in scenario 1 and scenario 2.
At Kr = 0 dB, the weak user performance in the proposed PDCSMA‐MRC little bit deviates from its strong user coun-
terpart. This close performance achieved for the weak user is due to the fact that the LOS and scattered component are
made equally strong at Kr = 0 dB. However, the performance is poor when compared with the corresponding perfor-
mance at Kr = − 20 dB shown in Figure 6. Because, despite the energy in LOS component is equal to total energy
in scattered component, the less energy in the individual scattered component leads to the performance degradation.
Additionally, the performance of PDCSMA‐MRC is superior to that of PDCSMA‐FDE and PDNOMA‐FDE because
the multipath components are resolvable and diversity is achieved in the MRC. Furthermore, the performance gap of
the PDCSMA‐FDE and PDNOMA‐FDE is minimal at low Eb/N0 and the gap is large at high Eb/N0. The reason behind
this is that 14 scattered components introduce strong interference to LOS component. Hence, the decrease in noise var-
iance shows only slight improvement in the BER performance of PDNOMA‐FDE.
Scenario 3: Figures 9 and 10 compare the two users' BER performance of the proposed PDCSMA‐MRC, PDCSMA‐
FDE, and conventional PDNOMA‐FDE for 2 and 15 multipaths, respectively. As given in Table 2, scenario 3 considers
the weak and strong users with Ricean factors Kr = − 20 dB and Kr = 0 dB, respectively. To be exact, both the users
SHALINI AND STUWART 15 of 21

FIGURE 9 BER comparison of PDCSMA‐MRC, PDCSMA‐FDE, and PDNOMA‐FDE of scenario 3 with P0 = 0.8, P1 = 0.2, and M = 2 for
various Ricean factors Kr

FIGURE 10 BER comparison of PDCSMA‐MRC, PDCSMA‐FDE, and PDNOMA‐FDE of scenario 3 with P0 = 0.8, P1 = 0.2, and M = 15 for
various Ricean factors Kr

experience Rayleigh channel. The two‐path amplitude delay profile are [0.1 1] and [1 1] for Kr = − 20 dB and Kr = 0 dB,
respectively. The power allocation for weak and strong user are P0 = 0.8 and P1 = 0.2, respectively.
Figure 9 shows the BER performance of the proposed PDCSMA‐MRC, PDCSMA‐FDE, and conventional PDNOMA‐
FDE for two users with two‐path channel. At Kr = − 20 dB, the performance of weak user is similar to the one
discussed in Figure 5 of scenario 1. The identical performance is obtained because the same weak user power is main-
tained in scenario 1 and scenario 3. At Kr = 0 dB, the proposed PDCSMA‐FDE for strong user exhibits superior perfor-
mance when compared with PDCSMA‐MRC and conventional PDNOMA‐FDE. Although the strong user transmission
power is low, the FDE is able to equalize the effect of multipath interference. On the other hand, the proposed
PDCSMA‐MRC shows constant error floor for the entire Eb/N0 due to very weak signal strength of strong user. The per-
formance of PDNOMA‐FDE is linear at low Eb/N0, and error becomes irreducible at high Eb/N0, and also, the perfor-
mance lies between PDCSMA‐MRC and PDCSMA‐FDE.
16 of 21 SHALINI AND STUWART

Figure 10 displays the BER performance of the proposed PDCSMA‐MRC, PDCSMA‐FDE, and conventional
PDNOMA‐FDE for two users with 15‐path channel. At Kr = − 20 dB, the performance of weak user is similar to the
one discussed in Figure 6 of scenario 1. The identical performance is obtained because the same weak user power is
maintained in scenario 1 and scenario 3. At Kr = 0 dB, the proposed PDCSMA‐FDE for strong user exhibits superior
performance when compared with PDCSMA‐MRC and conventional PDNOMA‐FDE. Although the strong user trans-
mission power is low, the FDE is able to equalize the effect of multipath interference. On the other hand, the proposed
PDCSMA‐MRC shows constant error floor for the entire Eb/N0 due to poor strong user signal. Specifically, the poor
transmitted signal strength of the strong user makes the LOS and scattered components insignificant for the MRC.
The performance of PDNOMA‐FDE is linear at low Eb/N0, and error becomes constant at high Eb/N0, and also, the per-
formance lies between PDCSMA‐MRC and PDCSMA‐FDE. Interestingly, the strong user performance of all the three

FIGURE 11 BER performance of randomly located user in a cell with variable channel length, Ricean factor, and power allocation

FIGURE 12 Spider plot representation of BER performance for different power allocation strategies of strong and weak users with M = 2
and 15 with random channel conditions
SHALINI AND STUWART 17 of 21

systems is inferior to the weak user performance of PDCSMA‐MRC. The reason for the surprise result is that multipath
diversity gain is achieved in the MRC due to the proposed cyclic spreading.
Figure 11 depicts the Eb/N0 versus BER curves for ideal channel bound, interleaved NOMA, PDNOMA‐FDE, and
proposed PDCSMA for both strong and weak users. The ideal bound for strong and weak users are obtained with unity
channel gain and arbitrary power allocation. The strong user implements SIC, and weak user performs hard decision.
Under ideal channel conditions, the strong user attains the lower bound BER of 10−5 at Eb/N0 of 10 dB, and the weak
user brings about the lower bound of 10−5 BER at Eb/N0 ≈ 15 dB. When two users, namely, the strong and weak users
are randomly grouped together to form a NOMA cluster with random channel conditions, at all times, the performance
of strong user surpasses the weak user performance. Furthermore, considering the BER performance of weak users of

FIGURE 13 Spider plot of scenarios 1,2, and 3 with P0 = 0.9, P1 = 0.1, and M = 2

FIGURE 14 Spider plot of scenarios 1,2, and 3 with P0 = 0.9, P1 = 0.1, and M = 15
18 of 21 SHALINI AND STUWART

different methods, it is noted that always the proposed PDCSMA outperforms the other two methods. This is because of
the diversity combining technique used in the proposed method.
The power allocation for users in each NOMA cluster plays an important role in achieving the desired BER perfor-
mance. The synthesis of the BER performance of conventional PDNOMA and proposed PDCSMA at Eb/N0 = 14 dB is
presented by means of spider plot. Figure 12 portrays the BER performance based on different power allocation strate-
gies between strong and weak users with ratios 1 : 9, 1 : 4, 3 : 7,and 4 : 6. The comparisons are made for both 2 and 15
paths with random channel conditions. From the spider plot, it is noted that PDCSMA outperforms PDNOMA in all the
cases. In addition, in the proposed PDCSMA, the power ratios 1 : 9 and 1 : 4 achieve improved BER performance over
other power ratios. For 2 paths, the power ratio 1 : 4 offers the outstanding BER performance, while the power ratio 1 : 9
yields the best result for 15 paths. Unfortunately, the power ratios 3 : 7 and 4 : 6 are not able to provide the favorable

FIGURE 15 Spider plot of scenarios 1,2, and 3 with P0 = 0.8, P1 = 0.2, and M = 2

FIGURE 16 Spider plot of scenarios 1,2, and 3 with P0 = 0.8, P1 = 0.2, and M = 15
SHALINI AND STUWART 19 of 21

BER performances. Hence, proper attention is needed in allocating user powers for achieving significant BER
performance.
Figures 13 and 14 are the spider plot for comparing the BER performance of the proposed PDCSMA‐MRC, PDCSMA‐
FDE, and conventional PDNOMA‐FDE for 2‐path and 15‐path channels, respectively, with transmission power P0 = 0.9
and P1 = 0.1. The BER characteristics of all the three scenarios stated in Table 2 are compared for the chosen Eb/
N0 = 14 dB. Likewise, the BER characteristics of the three systems for three scenarios with transmission power
P0 = 0.8 and P1 = 0.2 are presented in Figures 15 and 16 for 2 paths and 15 paths, respectively.
Consolidating spider plots, Table 3 lists the receiver selection for various NOMA schemes in different wireless scenar-
ios. The best receiver for specific channel type, user type, and power ratio is marked with a tick symbol. Furthermore,
the suitable receiver for each scenario is encircled. In general, the user power ratio plays major role in the receiver selec-
tion. The proposed PDCSMA‐MRC and PDCSMA‐FDE always outperform PDNOMA‐FDE for all the three scenarios.
Based on the user power ratio and the number of multipath components, the superiority switches between
PDCSMA‐MRC and PDCSMA‐FDE. Precisely, for two‐path channel, PDCSMA‐MRC is a dominant receiver for both
weak and strong users in scenario 1 and scenario 2 for the user power ratio of 1 : 9 and 1 : 4. On the other hand, for
two‐path channel, PDCSMA‐FDE is emerging as superior receiver for both weak and strong users in scenario 3 for
the user power ratio of 1 : 9 and 1 : 4. Nevertheless, in all the three scenarios with both the user power ratios, 15‐path
channel necessitates the weak and strong users to use different receivers. That is, PDCSMA‐MRC is preferred for weak
user whereas PDCSMA‐FDE favors the strong user.

TABLE 3 Receiver selection for different user grouping

Power (P1 : P0) System User Scenario 1 Scenario 2 Scenario 3

1:9 2 paths Prop‐MRC Strong ✓

Weak ✓

Prop‐FDE Strong ✓
Weak ✓
PDNOMA‐FDE Strong
Weak
15 paths Prop‐MRC Strong
Weak ✓ ✓

Prop‐FDE Strong ✓ ✓

Weak
PDNOMA‐FDE Strong
Weak
1:4 2 paths Prop‐MRC Strong ✓

Weak ✓

Prop‐FDE Strong

Weak

PDNOMA‐FDE Strong
Weak
15 paths Prop‐MRC Strong

Weak ✓ ✓ ✓
Prop‐FDE Strong ✓

Weak
PDNOMA‐FDE Strong
Weak
20 of 21 SHALINI AND STUWART

4 | CONCLUSIONS

This paper proposed a mixed NOMA strategy based on cyclic spreading. The proposed PDCSMA encompasses cyclic
spreading at the transmitter and allows user to choose the MRC or FDE built receiver based on the channel condition.
The proposed scheme is compared with conventional PDNOMA and interleaved NOMA, and the proposed scheme is
tested for three different scenarios considering both Ricean and Rayleigh channels. The significant results attained
put forward the grouping of users with different channel conditions for the optimum BER performance and recom-
mend the receiver for the strong and weak users under different wireless environment. The inherent ability of the pro-
posed PDCSMA to adapt the user power and channel condition envisages the scheme suitable for the next generation
wireless systems. The proposed scheme is envisioned to give attention to weak user and to achieve reasonable BER per-
formance. However, we believe that the studies initiated here can be extended with codeword‐level SIC or joint mod-
ulation which may provide significant improvement in the strong user performance with reasonable burden on the
proposed receiver.

ACK NO WLE DGE MEN TS


The authors would like to express their gratitude to the anonymous reviewers for their detailed reviews, valuable
suggestions, and insightful recommendations made on this paper which helped a lot to improve the quality of this
paper.
This work was supported by Science and Engineering Research Board (SERB), a statutory body of Department of Sci-
ence and Technology (DST), Government of India, under Early Career Research (ECR) Award (Grant no. ECR/2017/
0011888).

ORCID
S. Lenty Stuwart https://orcid.org/0000-0002-9988-5624

R EF E RE N C E S
1. Dai L, Wang B, Yuan Y, Han S, Chih‐Lin I, Wang Z. Non‐orthogonal multiple access for 5G: solutions, challenges, opportunities, and
future research trends. IEEE Commun Mag. 2015;53(9):74‐81.
2. Ding Z, Lei X, Karagiannidis GK, Schober R, Yuan J, Bhargava VK. A survey on non‐orthogonal multiple access for 5G networks:
research challenges and future trends. IEEE J Sel Areas Commn. 2017;35(10):2181‐2195.
3. Moltafet M, Yamchi NM, Javan MR, Azmi P. Comparison study between PD‐NOMA and SCMA. IEEE Trans Veh Technol.
2018;67(2):1830‐1834.
4. Cao Y, Sun H, Soriaga J, Ji T. Resource spread multiple access—a novel transmission scheme for 5G uplink, in Proc. IEEE 86th Veh.
Technol. Conf. (VTC‐Fall), pp. 1‐5, 2017.
5. Dai X, Zhang Z, Bai B, Chen S, Sun S. Pattern division multiple access: a new multiple access technology for 5G. IEEE Wireless Commun.
2018;25(2):54‐60.
6. Zhang J, Wang X, Yang X, Zhou H. Low density spreading signature vector extension (LDS‐SVE) for uplink multiple access, in Proc. IEEE
86th Veh. Technol. Conf. (VTC‐Fall), pp. 1‐5, 2017.
7. Han Y, Zhou W, Zhao M, Zhou S. Enabling high order SCMA systems in downlink scenarios with a serial coding scheme. IEEE Access.
2018;6:33796‐33809.
8. Tran Thi TN, Leonardo L, Shingo Y, Hiroshi O. Low latency IDMA with interleaved domain architecture for 5G communications. IEEE J
of Emer and Sel Top in Cir and Sys (JETCAS). 2017;7(4):582‐593.
9. New uplink non‐orthogonal multiple access schemes for NR, document R1‐167535, Mediatek, 3GPP, Aug. 2016.
10. Multiple access schemes for new radio interface, document R1‐162385, Intel, 3GPP, Apr. 2016.
11. Contention‐based non‐orthogonal multiple access for UL mMTC, document R1‐164269, ZTE, 3GPP, May 2016.
12. Wang Q, Zhao Z, Miao D, Zhang Y, Sun J, Liu M, Zhong Z. Non‐orthogonal coded access for contention‐based transmission in 5G, in
Proc. IEEE Veh. Technol. Conf. (VTC‐Fall), pp. 1‐6, 2017.
13. Xiong Q, Qian C, Yu B, Sun C. Advanced NOMA scheme for 5G cellular network: interleave‐grid multiple access, in Proc. IEEE
Globecom Workshops, pp. 1‐5, 2017.
SHALINI AND STUWART 21 of 21

14. Su H, Bin Y, Qian C, Xiao Y, Xiong Q, Sun C, Gao Y. Non‐orthogonal interleave‐grid multiple access scheme for industrial internet‐of‐
things in 5G network, IEEE Trans on Industrial Informatics, (Early Access), pp. 1 – 11, 2018.
15. Moltafet M, Mokari N, Javan MR, Saeedi H, Pishro‐Nik H. A new multiple access technique for 5G: power domain sparse code multiple
access (PSMA). IEEE Access. 2018;6:747‐759.

How to cite this article: Shalini RB, Stuwart SL. Power domain cyclic spread multiple access: An interference‐
resistant mixed NOMA strategy. Int J Commun Syst. 2019;e4025. https://doi.org/10.1002/dac.4025

You might also like