This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE SYSTEMS JOURNAL 1

Nonorthogonal Multiple Access With Orthogonal

Time–Frequency Space Signal Transmission
Aritra Chatterjee , Vivek Rangamgari, Shashank Tiwari, and Suvra Sekhar Das, Member, IEEE

Abstract—Orthogonal time–frequency space (OTFS) is being is one of the most successful waveforms used is popular broad-
pursued in recent times as a suitable wireless transmission technol- band wireless communication systems namely DVB-T, DVB-A,
ogy for use in high-mobility scenarios. In this article, we propose DVB-S [8], [9], WiFi [10], and 4G-LTE [11]. However, it is well
nonorthogonal multiple access (NOMA) based OTFS which may be known that OFDM suffers from intercarrier interference (ICI)
called NOMA-OTFS system and evaluate its performance from a
due to high Doppler in such scenarios [12]. Although in the
system-level and link-level perspective. The challenge lies in the fact
that while OTFS transmission technology is known for its resilience
upcoming 5G new radio, the subcarrier bandwidth of OFDM
to high-mobility conditions, while NOMA is known to yield high is made flexible [13] to adapt to various channel conditions,
spectral efficiency (SE) in low-mobility scenarios in comparison to yet it is limited due to several other constraints as will be
orthogonal multiple access (OMA). We present a minimum mean discussed in later sections. On the other hand, orthogonal time–
square error (MMSE)-successive interference cancellation based frequency space (OTFS) [14], which places signal constellation
receiver for NOMA-OTFS, for which we derive expression for in delay-Doppler (De-Do) plane as opposed to time–frequency
symbol-wise postprocessing SINR in order to evaluate system sum (T–F) plane, is being explored with enthusiasm by researchers
SE. We develop power allocation schemes to maximize the sum across the globe [15]–[18], as it provides great improvements in
SE in the high-mobility version of NOMA. We further design a performance especially in such new high-mobility scenarios.
realizable codeword-level SIC (CWIC) receiver using low-density A radio access technology (RAT) comprises of transmission
parity check (LDPC) codes along with MMSE equalization for
evaluating link-level performance of such practical NOMA-OTFS
technology and multiple access (MA). In this article, we focus
system. The system-level and link-level performance of the pro- on multiple access for OTFS transmission technology.
posed NOMA-OTFS system are compared against benchmark The MA technique can be broadly classified in two categories,
OMA-OTFS, OMA-orthogonal frequency division multiplexing namely first, orthogonal multiple access (OMA) and, second,
(OMA-OFDM) and NOMA-OFDM schemes. From system-level nonorthogonal multiple access (NOMA). In OMA, resource
performance evaluation, we observe interestingly that NOMA- allocation orthogonality is maintained, i.e., one resource unit
OTFS provides higher system sum SE than OMA-OTFS. When is allocated to only one user. With reference to OTFS, two
compared to NOMA-OFDM, we find that outage SE of NOMA- types of OMA-OTFS are reported first, T–F MA OTFS [19]:
OTFS is improved at the cost of decrease in mean SE. The link-level where users are allocated different T–F resources and second,
results additionally show that the developed CWIC-based NOMA- De-Do MA-OTFS [14]: where users are allocated different D–D
OTFS receiver performs significantly better than NOMA-OFDM
in terms of block error rate, goodput, and throughout.
resources.
In contrast to OMA methodology, NOMA schemes allo-
Index Terms—Block error rate (BLER), low-density parity check cate more than one user in one resource unit. Power-domain
(LDPC), nonorthogonal multiple access (NOMA), orthogonal NOMA (PD-NOMA) schemes realized using superposition cod-
time–frequency space (OTFS), power allocation, spectral efficiency ing (SC) at the transmitter along with successive interference
(SE), successive interference cancellation (SIC). cancellation (SIC) at the receiver is known achieve the capac-
ity of Gaussian broadcast channel. PD-NOMA schemes are
found to significantly outperform OMA as well as code-division
I. INTRODUCTION NOMA schemes [20] in terms of sum spectral efficiency (SE)
A. Background and Motivation performance [21], [22].
Therefore, the main goal of this article is to investigate
E ARE experiencing new high-mobility scenarios such
W as high-speed railways [1]–[3], unmanned aerial vehicle
communications [4], vehicle-to-vehicle communications [5],
NOMA-OTFS, which has attracted only limited attention till
now. We aim to develop and investigate the NOMA-OTFS
and compare its performance against NOMA-OFDM and
[6], etc., where providing high-quality wireless communica- OMA-OTFS.
tion service using existing transmission technologies is a chal- Important aspects pertaining to the implementation of NOMA
lenge [7]. Orthogonal frequency division multiplexing (OFDM) are first, division of total available transmit power at BS among
users, second, user grouping and T–F resource allocation. Such
Manuscript received September 19, 2019; revised March 14, 2020 and May issues are addressed at length for both downlink and uplink
25, 2020; accepted May 30, 2020. (Corresponding author: Aritra Chatterjee.) directions in [23]–[28] and references therein. Such resource
The authors are with the G. S. Sanyal School of Telecommunica- allocation is done based on either full or partial channel state
tions, Indian Institute of Technology Kharagpur, Kharagpur 721302, India information (CSI) at transmitter. An overview on resource
(e-mail: [email protected]; [email protected]; shashankpbh@
gmail.com; [email protected]). allocation and performance analysis of power-domain NOMA
Digital Object Identifier 10.1109/JSYST.2020.2999470 systems, is available in [29] and [30].

1937-9234 © 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: University of Canberra. Downloaded on June 25,2020 at 09:25:41 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE SYSTEMS JOURNAL

As indicated above, in order to achieve optimal gains, NOMA 5) In [31] and [32], arbitrary De-Do channel is considered
transmitter must be made aware of the user’s instantaneous chan- for performance analysis, whereas we have evaluated per-
nel coefficients, which change rapidly in high-mobility scenarios formance of the proposed NOMA-OTFS in practical ITU
as considered in this article. The CSI fed back to transmitter De-Do channel model [33]. The results provide better es-
becomes outdated very fast which limits the achievable gain of timate of such NOMA-OTFS in future realistic scenarios
using T–F domain NOMA. for 5G and beyond.
Furthermore, because of OFDM’s limited capability to handle 6) The framework developed in this article is made flexible
high Doppler restricts the choice of NOMA-OFDM as a RAT in so as to handle OTFS and OFDM in an unified matrix
high-mobility scenarios. Since OTFS is resilient to high Doppler representation. It is also worth noting that the modified
in comparison to OFDM, we aim to investigate the use of OFDM framework we adopt in this article use block cyclic
NOMA with OTFS so that multiuser extension of OTFS can prefix (CP) along with MMSE equalizer. Accordingly, we
be achieved in such high-mobility conditions. Such investiga- analyze the performance of OFDM with block processing
tion is expected to pave the path for future research on meth- and ICI canceling receiver for comparison against OTFS.
ods for multiuser SE enhancement techniques in high-mobility The above discussions are for system-level performance eval-
scenarios. uation. While SE performance analysis gives us one perspective,
it is also vital to evaluate the link-level performance for such
NOMA-OTFS system in order to have a comprehensive view
B. Related Works and Contribution of the performance of such newly proposed system namely
The interplay of two futuristic technologies namely OTFS NOMA-OTFS. Accordingly, the following are included.
and NOMA has attracted the attention of researchers as reported 1) In order to get a link-level performance estimation, we
in [31] and [32]. In [31], the authors have proposed the use of need to develop a receiver for NOMA-OTFS system.
NOMA with OTFS in order to serve users having heterogeneous Accordingly, we have developed a codeword-level low-
mobility profiles for both downlink and uplink. In [32], beam- density parity check (LDPC)-SIC receiver which uses
forming aspect of “OTFS assisted NOMA” networks has been symbol-level log-likelihood ratio (LLR) values of the
explored (in presence of multiantenna base station) to maximize MMSE-based ICI canceling receiver (see Section V),
the low-mobility NOMA users’ data rate while maintaining which to the best of the authors’ knowledge first such
high-mobility OTFS user’s target data rate. attempt.
In light of the limited state-of-the-art available as indicated 2) The performance of such realistic LDPC enabled MMSE-
above, the major contributions of this article are outlined as SIC receiver is further compared with NOMA-OFDM
follows. and OMA-OTFS in terms of block error rate (BLER),
1) The system models in [31] and [32] assume only the user throughput (in bits/s/Hz) and goodput (in bits/s/Hz).
with highest velocity is served in De-Do plane (using 3) A comprehensive performance analysis taking into ac-
OTFS scheme). Whereas, the rest of low-mobility users count the system-level and link-level performance has
are served in same T–F plane (using OFDM modulation) been presented in this article (see Section VI).
which are multiplexed in PD-NOMA. Therefore, the user Notations: We use the following notations throughout the
with high mobility (served with OTFS) does not partic- article. We let x, x, and X represent scalars, vectors, and
ipate in De-Do NOMA transmission. In this article, we matrices, respectively. The superscripts (.)T and (.)† indicate
propose and develop a holistic framework to obtain the transpose and conjugate transpose operations, respectively. IN
De-Do PD-NOMA-OTFS where multiple high-mobility and WL represents identity matrix with order N and L-order
users are served by OTFS in the same De-Do resource normalized IDFT matrix, respectively. Kronecker product oper-
block, which is the first such proposal to the best of our ator is represented by ⊗. The Frobenius norm of any matrix X
knowledge. is denoted by ||X||F . diag[.] denotes a diagonal matrix whose
2) The pulse shape used in [31] and [32] is considered to diagonal elements are formed by the elements of the vector
be ideal in nature, which is not realizable in practice due inside. circ{.} denotes a circulant matrix whose first column
to time–frequency uncertainty principle [17]. Such ideal is given by the vector inside. The expectation parameter is
assumption simplifies the system equations which yields denoted by E[.]. Column-wise vectorization of matrix (.) is
to block circulant system matrices. In this article, we represented by vec{.}. The ceiling operator is denoted as ..
consider realizable time domain rectangular pulse which N[ab] √represents the set of natural numbers ranging from a to b.
does not offer such simplification. j = −1.
3) Furthermore, in [31, Section VII], NOMA users are allo-
cated fixed power without taking into account their chan-
nel condition. Such elementary power allocation restricts II. OTFS SIGNAL MODEL
NOMA gain. We evaluate the performance of NOMA- We consider a multicarrier and multi time-slot system with
OTFS with different dynamic power allocation strategies total Tf s duration and B Hz bandwidth. We have total M
suitably designed for high-mobility environments. number of subcarriers having Δf subcarrier bandwidth and N
4) In [31], SE results are obtained using Shannon’s expres- number of symbols having T symbol duration, thus B = M Δf
sion using ideal SIC at the receiver. In this article, we and Tf = N T . We consider the OTFS system to be critically
compute postprocessing symbol-level SINR for practical sampled such that T Δf = 1.
ICI canceling MMSE with SIC NOMA receiver, which For a user (termed as ith user henceforth), the QAM
renders the results more close to reality (in Sections III-A2 modulated Delay-Doppler data symbols, di (k, l) ∈ C, k ∈
and III-B2). N[0N − 1], l ∈ N[0M − 1], are arranged over Doppler-delay

Authorized licensed use limited to: University of Canberra. Downloaded on June 25,2020 at 09:25:41 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

CHATTERJEE et al.: NONORTHOGONAL MULTIPLE ACCESS WITH OTFS SIGNAL TRANSMISSION 3

lattice Λ = {( NkT , M Δf
l
)}. Data symbols di (k, l) is mapped (4) as
to time–frequency domain data Xi (n, m) on lattice Λ⊥ = 
K
{(nT, mΔf )}, n ∈ N[0N − 1] and m ∈ N[0M − 1] by using s=A βi Pdi . (5)
inverse symplectic fast Fourier transform. Thus, Xi (n, m) can i=1
be given as [16]
We consider linear time varying (LTV) channels for all the

N −1 M
 −1 users. Let, the ith user’s channel consists of Pi paths with
1 nk ml
Xi (n, m) = √ di (k, l)ej2π[ N − M ] . (1) hpi complex attenuations, τpi delays and νpi Doppler values
NM k=0 m=0 for pi th path where pi ∈ N[1Pi ]. Thus, Delay-Doppler channel
spreading function for the ith user can be given as
Next, a time–frequency modulator modulates Xi (n, m) to time
domain using Heisenberg transform as 
Pi
hi (τ, ν) = hpi δ(τ − τpi )δ(ν − νpi ), i = 1, . . . , K. (6)
 −1 M
 −1
N
√ pi =1
si (t) = ( PXi (n, m))g(t − nT )ej2πmΔf (t−nT )
n=0 m=0
The delay and Doppler values for pi th path is given as τpi =
i
(2) lp ki
M Δf and νpi = NpT , where lpi ∈ N[0M − 1] and kpi ∈
where g(t) is transmitter pulse of duration T and transmit
power is denoted by P. Furthermore, si (t) is sampled at the N[0N − 1] are the number of delay and Doppler bins on the
Doppler-delay lattice for pi th path. We assume that N and M
sampling interval of M T
. We collect samples of si (t) in si = are sufficiently large so that there is no effect of fractional delay
[si (0)si (1) · · · si (M N − 1)]. The QAM symbols di (k, l) are and Doppler on the performance. We also assume the perfect
arranged in M × N matrix as knowledge of (hpi , lpi , kpi ), pi ∈ N[0Pi − 1], at the receiver
Di = of ith user, as previously considered in [16] and [18]. One
i i
work on such estimation is given in [34]. Let τmax and νmax
⎡ ⎤
di (0, 0) di (1, 0) ··· di (N − 1, 0) be the maximum delay and Doppler spread for users. Channel
⎢ di (0, 1) di (1, 1) ··· di (N − 1, 1) ⎥ delay length αi = τmax i
M Δf  and channel Doppler length
⎢ ⎥ β = νmax N T . LCP = maxi=1,...,K (αi ).
i i
⎢ .. .. .. .. ⎥.
⎣ . . . . ⎦ At the ith user’s receiver, after removal of CP, the received
di (M − 1, 0) di (M − 1, 1) ··· di (N − 1, M − 1) signal can be written as [17]
(3) ri = Hi s + ni , i = 1, . . . , K. (7)
The transmitted signal can be written as matrix-vector multipli- where ni is white Gaussian noise vector of length M N with
cation as elemental variance σn2 and Hi is an M N × M N channel matrix
√ for ith user which is given by
si = A Pdi (4)

Pi
i i
where di = vec{Di }. Finally, AM N ×M N = WN ⊗ IM de- Hi = hip Πlp Δkp , i = 1, . . . , K (8)
notes the OTFS modulation matrix. A CP of length LCP ≥ pi =1
LCP − 1 is appended at the starting of the s, where LCP is the
channel’s maximum excess delay length. In order to implement with ΠM N ×M N = circ{[010 · · · 0]TM N ×1 } is a circulant de-
1 M N −1
OFDM in the same framework, the modulation matrix is modi- lay matrix and Δ = diag[1ej2π M N · · · ej2π M N ] is a diagonal
fied as A = IN ⊗ WM . Doppler matrix. Using the abovementioned signal model for the
De-Do PD-NOMA-OTFS in downlink, we proceed to evaluate
III. DELAY-DOPPLER POWER-DOMAIN NOMA-OTFS the corresponding SINR and SE experienced by the nonorthog-
onally multiplexed users.
In this section, we further extend the OTFS signal model 2) Receiver Processing, SINR and SE Analysis: In OTFS,
presented in Section II in order to develop the multiuser De-Do when the signal passes through the LTV channel, it encounters
PD-NOMA-OTFS signal models and derive the SINR and SE both ICI and intersymbol interference (ISI), thereby degrading
expressions of the same for both downlink and uplink. We its performance. In order to negate such induced ICI and ISI, we
consider K users with high velocity are multiplexed in power equalize the received signal through LMMSE equalizer, as done
domain all of which are served by OTFS (unlike [31] and [32]) in [35]. Furthermore, in the later stage, SIC has been performed
in both downlink and uplink transmissions. in order to mitigate the NOMA interference at the receiver, which
has been discussed subsequently.
A. De-Do PD-NOMA-OTFS Downlink The total effective noise at the ith receiver amounts to
1) Transmit Signal Model: Among the K high-mobility 
K

users multiplexed in power domain, we assume the ith user is ñiDL = βi PHi Adi + ni . (9)
i =1,i =i
allocated βi fraction of total power P. Clearly, K i=1 βi = 1.
Choice of βi is described in detail in Section IV. Following the Assuming the total effective noise following Gaussian distribu-
principle of superposition, the composite transmitted signal from tion, LMMSE equalization on the received signal ri in (7) results
the transmitter intended for all users can be written by modifying in estimated data vector for ith user as given in (10) shown at

Authorized licensed use limited to: University of Canberra. Downloaded on June 25,2020 at 09:25:41 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE SYSTEMS JOURNAL

the bottom of this page, where Γi denotes the average SNR of of the system in bps/Hz is given by
ith user. Rewriting (10) by using (7) and (5), we obtain

K
DL
Rsum = log2 (1 + ΥPost-D ). (14)

K i
d̂i = Bi (βi P)di + Bi (βi P)di i=1

i =1,i =i It is noteworthy that the SE performance presented here for
desired signal 
NOMA interference
downlink (and subsequently for uplink in Section III-B2) are
done for such realizable MMSE-SIC receiver only. SE cal-
+ Ci n i , i = 1, . . . , K (11) culation using log-determinant method2 of the delay-Doppler

noise component channel Hi is beyond the scope of the current work.

where,
√ for notational simplicity, we assign Ci = B. De-Do PD-NOMA-OTFS Uplink
βi (Hi A)† [(Hi A)(Hi A)† + Γ1i I]−1 and Bi = Ci Hi A.
At this point, without loss of generality, we consider 1) Transmit Signal Model: For uplink OTFS-NOMA, all the
that from the transmitting BS the distance of ith user is K users transmit data simultaneously to the base station in
higher than the (i + 1)th user for i = 1, . . . , (K − 1), thus delay-Doppler plane, thus making it a multiple-access channel.
in terms of received average SNR, it can be written as: For the sake of simplicity, perfect carrier and clock synchroniza-
Γ1 < Γ2 < · · · < Γi−1 < Γi < · · · < ΓK . Thus, we assume tion among the transmitting users has been assumed. It has also
that following the principle of NOMA, the ith user will not been assumed that both the users are implementing same OTFS
face any interference due to the signals intended for 1st, grid size (M, N ). It has also been assumed that the receiver
2nd,. . . , (i − 1)th users through perfect SIC.1 Using these BS has perfect knowledge about the channels from transmitting
assumptions and expanding (11), the symbol-wise pre- and users. The OTFS modulated transmitted vector from ith user is
post-SIC received SINR at any user can be formulated. For the given by
ith user, the downlink pre- and post-SIC SINR for jth symbol sui = A Piu dui (15)
(denoted as ΥPre-Dij and ΥPost-D
ij , respectively) can be given by
(12) and (13), shown at the bottom of this page, respectively, where Piu
and dui
denote the transmit power and vectorized
with i = 1, . . . , K and j = 1, . . . , M N . bipq and cipq denote the transmit data of the ith user, respectively. The uplink average
(p, q)th elements of Bi and Ci , respectively. SNR of the i-th user is given by Γui = Piu /σn2 . The aggre-
In OTFS, the SINR achieved in all symbols are nearly same gate received signal at the base station after removal of CP is
for large M and N values [38] and thus, henceforth, we drop given by
subscript j and represent the pre- and post-SIC SINR of ith user
as ΥPre-D
i and ΥPost-D
i , respectively. Thus, the downlink sum rate 
K
ru = Hui sui + n (16)
i=1

1 Consideration of imperfect SIC and subsequent error propagation can be 2 as usually done in conventional point-to-point multiple-input multiple-output
seen as a potential future work [36], [37] (MIMO) systems [39]

 
K −1  −1
1 1
d̂i = βi (Hi A)† βi (Hi A)(Hi A)† + βi (Hi A)(Hi A)† + I ri = βi (Hi A)† (Hi A)(Hi A)† + I ri
Γi Γi
i =1,i =i
(10)

βi P|bijj |2

desired power
ΥPre-D
ij = M N K   M N  (12)
MN
βi P |bi |2 + βi P |bijl |2 + |cijl |2 σn2
l=1,l=j jl i =1,i =i

l=1

l=1

ISI noise power
NOMA interference

βi P|bijj |2

desired power
ΥPost-D
ij = M N K     (13)
MN MN
βi P |bi |2 + βi P |bijl |2 + |cijl |2 σn2
l=1,l=j jl i =i+1

l=1

l=1

ISI noise power
NOMA interference

Authorized licensed use limited to: University of Canberra. Downloaded on June 25,2020 at 09:25:41 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

CHATTERJEE et al.: NONORTHOGONAL MULTIPLE ACCESS WITH OTFS SIGNAL TRANSMISSION 5

where Hui denotes the M N × M N delay-Doppler uplink chan- Using similar assumptions made for downlink direction, the
nel matrix from ith user to the BS. Similar to the downlink sum rate (in bps/Hz) in uplink direction is given by
scenario presented before, we further proceed to evaluate the
SINR and SE experienced by the PD-NOMA-OTFS users in 
K

uplink direction in the following section.

UL
Rsum = log2 (1 + ΥU
i ). (22)
i=1
2) Receiver Processing, SINR, and SE Analysis: During up-
link transmission in NOMA, the signal from the users with
higher SNR are sequentially decoded and successively can- IV. POWER ALLOCATION SCHEMES AMONG DOWNLINK
celed from the aggregate signal. For the same user ordering as NOMA-OTFS USERS
considered in downlink transmission (Γu1 < · · · < ΓuK ), while In this section, we briefly outline the various power allocation
decoding the ith user’s signal, the BS will consider the 1st, 2nd, schemes used for performance evaluation of NOMA-OTFS in
· · · , (i − 1)th users’ signals as noise. Thus, for ith user, the downlink direction.
effective noise can be denoted as It is noteworthy that in high-mobility scenarios, where OTFS
offers distinguishable advantages over OFDM, feeding back full

i−1
ñi = Piu Hui Adui + n. (17) instantaneous CSI to the BS becomes increasingly difficult due
i =1
to coherence time constraint of the channel. However, we also
describe the full CSI-based power allocation algorithm, which
Subsequently, the noise variance for the ith user is given by is used as a benchmark in comparative study.
It is also worth noting that in this article, while considering

i−1
power allocation for weighted sum rate maximization (discussed
σ̃n2 i = E[ñi ñH
i ]= Piu Hi HH 2
i + σn I. (18) in Section IV-C), we have considered only two users multiplexed
i =1 together following the principle of NOMA for analytical sim-
After processing the received signal (ru ) through LMMSE plicity. Extension to generalized K-user scenario can be seen as
equalizer (similar to (10) for downlink), the estimated data vector a potential future investigation, which is beyond the scope of the
for the ith user at the BS can be expressed by (19), shown at the current work.
bottom of this page.
For notational simplicity, we denote Cui = (Hui A)† A. Fixed Power Allocation
Γu
[(Hui A)(Hui A)† + i−1 i
i =1 Γu (Hui A)(Hui A)† + Γ1u I]−1 , Buii This power allocation scheme is simplistic in nature where
i i
= Ci Hi A and Bii = Ci Hi A. Thus, combining (15) and
u u u u u the fractions of transmit power for different users are deter-
(16), (19) can be rewritten as mined a priori. Such conventional scheme has been used in
NOMA performance analysis for simplicity and in order to have

i−1 a benchmark for other sophisticated power allocation strate-
dˆui = Buii Piu dui + Bu  Piu di gies [23], [40]. The power fractions are independent of user
 i =1 ii channel conditions and system performance. In order to maintain
desired signal 
NOMA interference fairness among users, it is a general practice to allocate more
power to the users with lower received average SNR. Thus, for
+ Cui ni , i = 1, . . . , K. (20) the SNR order mentioned in Section III-A2, the fixed transmit

noise component power fractions will be ordered as: β1 > β2 · · · > βK , with the
constraint K i=1 βi = 1.
Expanding (20), the uplink SINR for jth symbol of the ith user
can be formulated as
B. Fractional Transmit Power Allocation (FTPA)
2
Piu |buii
jj | In this dynamic power allocation scheme, the fraction of

desired power power allocated to any user is proportional to the inverse of
ij = 
ΥU M N i−1 
MN
 its channel gain so that the users with lower channel gain gets

Piu |buii |2 + P u |buii
jl |
2
+ greater transmit power in order to maintain system fairness [24,
l=1,l=j jl i =1 i l=1
  Section II-B]. Depending on the nature of information about the
ISI NOMA interference channel available at the BS, following two FTPA schemes have
M N  been investigated in this article.
2 2
|cui
jl | σn
1) Average SNR Based FTPA: In this scheme, it is assumed
l=1
 that the BS has the access to only slowly time-varying average
noise power received SNR values (Γi ) of the users (which can be fed back
(21) to the BS from user through feedback channel or measured
uii
with i = 1, . . . , K and j = 1, . . . , M N . buii
pq , b pq , and cui
pq in reverse channel). Therefore, we propose to use the average
denote the (p, q)th elements of Bii , Bii and Ci , respectively.
u u u
received SNR values to allocate the users’ power fractions using

 −1  −1

i−1
Piu σn2 
i−1 u
Γi u u † 1
ˆu u † u u †
di = (Hi A) (Hi A)(Hi A) + u u †
(H A)(Hi A) + u I u † u u†
ri = (Hi A) Hi Hi + H H  + u I ri

Piu i Pi 
Γui i i Γi
i =1 i =1
(19)

Authorized licensed use limited to: University of Canberra. Downloaded on June 25,2020 at 09:25:41 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE SYSTEMS JOURNAL

FTPA. The fraction of power allocated to the ith user is given jth symbol) derived in (13) as follows:
by
Maximize Rsum
Inst
= w1 log(1 + ΥPost-D
1j ) + w2 log(1 + ΥPost-D
2j )
Γ−1
βi = i
. (23) subject to β1 + β2 = 1, 0 ≤ β1 , β2 ≤ 1. (28)
K
i =1 Γ−1
i

2) Channel Norm Based FTPA: The base station has access Using the notations P|b1jj |2 = Γ1d , P M N 1 2
l=1,l=j |bjl | = Γ1ISI ,
to the partial CSI of all users in terms of the instantaneous P( M N 1 2
σn2 M N 1 2
P|b2jj |2 =
l=1 |bjl | ) = Γ1N , l=1 |cjl | = P1n ,
channel norms, it is assumed that we use those values to evaluate 2 2 2 MN 2 2
the users’ power fractions. Therefore, the fraction of power Γ1d , P M N
l=1,l=j |bjl | = Γ2ISI , and σn l=1 |cjl | = P2n ,
allocated to the ith user can be expressed as the instantaneous weighted sum rate in terms of β2 can be
expressed as
||Hi ||−1  
βi = F
(24) (1 − β2 )(Γ1d + Γ1ISI ) + β2 Γ1N + P1n
K
i =1 ||Hi ||−1
F Rsum = w1 log
Inst
(1 − β2 )Γ1ISI + β2 Γ1N + P1n
where Hi is defined in (8).  
β2 (Γ2d + Γ2ISI ) + P2n
+ w2 log . (29)
C. Power Allocation for Weighted Sum Rate Maximization β2 Γ2ISI + P2n
(WSRM)
As done in previous section, we use the differentiation method
Similar to the case for FTPA, we present two weighted sum in order to obtain a suboptimal solution of β2 [28, Section III-A].
rate maximization framework based on average SNR informa- Inst
Differentiating Rsum w.r.t. β2 and equating it to zero results in
tion and instantaneous channel information at the base station. (30), shown at the bottom of this page. By numerically solving
1) Average SNR-Based WSRM: In case the base station has (30) using available software tools, optimal value of β2 (β2Opt )
access to the average SNR information of the users, the opti- can be obtained. Similar to the earlier case, we finally assign
mization problem can be formulated based on the AWGN rates β2Opt = max(0, min(1, β2Opt )) and β1Opt = 1 − β2Opt .
as described below It is to be noted that judicious assignment of weights for the
 
β 1 Γ1 users has been addressed in literature considering proportional
Maximize Rsum AWGN
= w1 log 1 + fairness [28], [41]. However, this being an early investigation,
1 + β 2 Γ2
for simplicity we consider assignment of fixed weights as w1 =
+ w2 log(1 + β2 Γ2 ) 0.6, w2 = 0.4.
It is also important to note that in order to implement the power
subject to β1 + β2 = 1, 0 ≤ β1 , β2 ≤ 1 (25)
allocation schemes described in Sections IV-B2 and IV-C2,
where w1 and w2 are the weights assigned to the two users channel information (like channel norm for the first scheme and
in order to maintain fairness in power allocation. This being an B and C matrices for the second scheme) have to be either fed
early work, we obtain a suboptimal solution of the maximization back to the BS by an error-free feedback channel, or measured
problem by differentiating the cost function, as done in [28, at BS itself by exploiting uplink-downlink duality. Such channel
Section III-A], although concavity of such cost function in (25) measurement and feedback has to be done at least once in every
is not straightforward to be shown. Reducing the problem in delay-Doppler coherence time of the OTFS channel. It has been
AWGN
terms of only β2 , differentiating Rsum w.r.t. β2 and equating reported in literature that the delay-Doppler coherence time of
it to zero finally yields OTFS channel is significantly larger than the coherence time
in time–frequency domain for OFDM [15]. Thus, instantaneous
w 1 Γ2 w 2 Γ1 De-Do CSI based NOMA power allocation schemes in high
− = 0. (26)
1 + β 2 Γ2 1 + β 2 Γ1 Doppler scenarios are easily realizable in OTFS systems com-
pared to OFDM systems.
Solving the linear equation, the optimal value of β2 can be
obtained as
V. LINK-LEVEL PERFORMANCE ANALYSIS OF
w 2 Γ1 − w 1 Γ2
β2Opt = . (27) NOMA-OTFS SYSTEMS
(w1 − w2)Γ1 Γ2
It has been reported in literature that scheduling a high number
In order to impose the associated constraints stated in (25), we of users (more than 2–3) in power domain NOMA in same
assign β2Opt = max(0, min(1, β2Opt )). Clearly, β1Opt = 1 − β2Opt . resource block does not offer much gain despite prohibitively in-
2) Instantaneous Channel Information Based WSRM: If the creasing the complexity of transmit signal processing, signaling
base station has access to partial information about the instan- overhead as well as realizing the successive interference canceler
taneous channel of each user [in terms of Bi and Ci matrices receiver in polynomial or exponential order [42]. Therefore,
defined after (10)], the two-user optimization problem can be we limit this early investigating work to two user multiplexed
formulated using the exact post-SIC SINR expression (for the system as in [43]–[45].

w1 (Γ1N − Γ1ISI − Γ1d ) w1 (Γ1N − Γ1ISI ) w2 (Γ2d + Γ2ISI ) w2 Γ2ISI

− + − = 0 (30)
(1 − β2 )(Γ1d + Γ1ISI ) + β2 Γ1N + P1n (1 − β2 )Γ1ISI + β2 Γ1N + P1n β2 (Γ2d + Γ2ISI ) + P2n β2 Γ2ISI + P2n

Authorized licensed use limited to: University of Canberra. Downloaded on June 25,2020 at 09:25:41 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

CHATTERJEE et al.: NONORTHOGONAL MULTIPLE ACCESS WITH OTFS SIGNAL TRANSMISSION 7

which the bit bj = k. See the paragraph following (11) for the
definition of B1 and C1 matrices. These LLRs are then fed into
the LDPC decoder to decode first user’s data. Let L1 denotes a
matrix where L1 (η, j) = L(bj1η |d̂1 (η)) for η = 1, 2, . . . , M N
and j = 0, 1, . . . , Ki − 1. L1 is reshaped to Lcl × Ncw matrix
where Lcl and Ncw denote the LDPC codeword length and
number of codewords, respectively. Each column of L1 sub-
sequently regenerates codeword c1ι for ι = 1, 2 · · · , Ncw using
the Min-Sum Algorithm [46] employed by the LDPC decoder.
This algorithm iteratively updates the variable node and check
node equation as discussed below.
1) Variable Node Update
(l−1)
m(l)
μ,ν = Zμ + Σν  =ν mμ, ν (32)
where the Zμ is the channel LLR calculated from (31)
(l)
for the μth bit in the codeword and mν,μ is the message
received from the νth check node to the μth variable node
in the iteration l.
Fig. 1. Representative block diagram of two-user NOMA-OTFS system in
downlink. 2) Check Node Update
(l) (l)
m(l)
ν,μ = Πμ =μ sign(mμ, ν ) min

(|mμ, ν |) (33)
μ =μ

A. Downlink MMSE SIC Receiver With LDPC Coding where the product and the minimum operator is taken over
This section illustrates the practical realization of a two-user the set of neighboring variable nodes except the message
LDPC enabled codeword-level SIC OTFS-NOMA receiver for recipient itself.
downlink transmission as highlighted in Fig. 1. The base station 3) Decoding Decision
generates the data for both the users (denoted as b1 and b2 , re- Zμtot = Zμ + Σν m(l)
ν,μ . (34)
spectively), encode using the LDPC encoder and then modulate
the data using modulation supported by the user. The encoded The algorithm terminates when the termination condi-
signals for both the users are denoted as d1 and d2 , respectively, tions of LDPC termination are satisfied or iteration count
in Fig. 1. The modulated symbols are further superimposed with reaches the maximum number of iterations(Nimax ) and the
allocated power (βi ). The superimposed time-domain signal is decoded codeword bit cμ = 1 if Zμtot >= 0 and cμ = 0 if
further modulated for OTFS using SFFT and the Heisenberg Zμtot < 0.
transform. The resulting signal s (refer to (5) for its mathematical 2) Processing at the Second User: Since second user experi-
expression) is broadcast through the delay-Doppler channel to ences higher SNR, it performs the SIC in which it decodes first
both the users. user’s data and then uses it to cancel the interference to decode
Since in this article, we consider two users, we let K = 2 in its own data. The detected first user’s data at the second user is
(7). Both the users first perform LMMSE equalization in order given as
to mitigate the ISI and ICI. Additionally, as it is assumed that  −1
the SNR of the second user is higher than the first user, thus the 1
second user performs the SIC. d̂12 = β1 (H2 A)† (H2 A)(H2 A)† + I r2 . (35)
Γ2
1) Processing at the First User: The equalized data using
the MMSE equalizer can be described using (10) with i = 1. Corresponding LLR of the equalized data of fist user is
In order to decode the equalized data using LDPC decoder, the calculated as
channel LLR values are calculated from the equalized symbols  
j ||d̂12 (η) − s||2
as L(b12 η |d̂12 (η)) ≈ min0
  sSj σ122 (η)
j ||d̂1 (η) − s||2  
L(b1η |d̂1 (η)) ≈ min0 ||d̂12 (η) − s||2
sSj σ12 (η) − min1 (36)
  sSj σ122 (η)
||d̂1 (η) − s||2
− min1 (31) where σ122 (η) is the ηth element of σ122 = 1 2 †
+
sSj σ12 (η) β1 diag(σn C2 C2
β2 B2 B†2 ).The residual received signal at second user after
where di (η) is the ηth element of di mapped from the bits canceling the interference due to first user is given by
i −1
b0iη b1iη · · · bK
iη , Ki is the number of bits per symbol for user
r̃2 = r2 − β1 PH2 Ad̃12 (37)
i and σ12 (η) is the ηth element of σ12 = β11 diag(σn2 C1 C†1 +
β2 B1 B†1 ). The aggregate interference and noise is assumed where d̃12 is generated at second user after passing the LDPC
to follow Gaussian distribution, as previously stated in decoded codeword obtained from d̂12 through symbol mapper.
Section III-A2. Sjk denotes the set of constellation symbols in After doing MMSE equalization on the residual signal given in

Authorized licensed use limited to: University of Canberra. Downloaded on June 25,2020 at 09:25:41 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE SYSTEMS JOURNAL

TABLE I
KEY SYSTEM PARAMETERS

Fig. 2. Representative block diagram of two-user NOMA-OTFS system in

uplink.

(37), the detected second user’s data at the second user itself is
given by
 −1 where ηth element of dui , dui (η) is mapped from bits
† † 1
d̂2 = β2 (H2 A) β2 (H2 A)(H2 A) + I r̃2 . (38) 0 1 (K −1)
buiη buiη · · · buiη i , Ki is the number of bits per symbol for user
Γ2
i and σ22 (η) is the ηth element of σ22 = P12 diag(σn2 Cu2 Cu† 2 +
As done for first user, the bit-level LLRs for second user from
the symbols are calculated as P1 Bu21 Bu†
21 ). The matrices C2 and B21 are defined after (19).
u u

  The calculated LLR values are processed by LDPC decoder in

j ||d̂2 (η) − s||2 order to produces the message word. The obtained message is
L(b2η |d̂2 (η)) ≈ min0
sSj σ22 (η) again encoded and modulated to generate the recovered data d̃u2
  for second user, which is used to cancel the interference from
||d̂2 (η) − s||2 aggregate received signal to decode the first user’s data as
− min1 (39)
sSj σ22 (η) r̃u1 = ru − P2u Hu2 Ad̃u2 . (42)
where σ22 (η) is the ηth element of σ22 = β12 diag(σn2 C2 C†2 ) (see After doing MMSE equalization of residual signal at the BS
the paragraph following (11) for the definition of C2 matrix). given by (42), the detected first user’s data is given by
The LLRs are updated and the data of user two is generated by  −1
† † 1
the LDPC decoder to generate the data using Min-Sum algorithm d̂1 = (H1 A) (H1 A)(H1 A) +
u
I r̃u1 . (43)
as described in detail for user one in Section V-A1. Γ1
The equalized data is d̂u1 is used to calculate the LLR as follows:
B. Uplink MMSE SIC Receiver With LDPC Coding  
This section focuses on a realization of an OTFS-NOMA link j ||d̂u1 (η) − s||2
L(b1η |d̂1 (η)) ≈ min0
u
coupled with LDPC codes for a two user scenario in uplink sSj σ12 (η)
transmission as highlighted in Fig. 2. For two user case, the  
received signal can be expressed using (16) by putting K = 2. ||d̂u1 (η) − s||2
− min1 (44)
At the BS, since it is assumed Γu1 < Γu2 , we first decode second sSj σ12 (η)
user’s data as
 where assuming perfect SIC, d̂u1 (η) and σ22 (η) are the ηth
Γu
d̂u2 = (Hu2 A)† (Hu2 A)(Hu2 A)† + 1u (Hu1 A)(Hu1 A)† element of d̂ and σ12 = P1u diag(σn2 Cu1 Cu†
1 ). The matrix C1 is
u
Γ2 1
−1 also defined after (19). The calculated LLR values are then fed
1 to LDPC decoder to reproduce the data of user one.
+ uI ru . (40)
Γ2
VI. SIMULATION RESULTS AND DISCUSSION
LLR values of second user can be computed as
  In this section, we present detailed performance analysis in
uj u ||d̂u2 (η) − s||2 terms of system-level and link-level evaluation of the presented
L(b2η |d̂2 (η)) ≈ min0 NOMA-OTFS schemes in a high-speed scenario through exten-
sSj σ22 (η)
sive Monte Carlo simulation for both downlink and uplink. The
  important simulation parameters are listed in Table I. Doppler is
||d̂u2 (η) − s||2 generated using Jake’s formula, νp = νmax cos(θp ), where θp is
− min1 (41)
sSj σ22 (η) uniformly distributed over [−ππ]. The CP is chosen long enough

Authorized licensed use limited to: University of Canberra. Downloaded on June 25,2020 at 09:25:41 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

CHATTERJEE et al.: NONORTHOGONAL MULTIPLE ACCESS WITH OTFS SIGNAL TRANSMISSION 9

TABLE II
MEAN AND 5% OUTAGE SE (IN bps/Hz) FOR DOWNLINK NOMA IN OTFS AND OFDM FOR Γ1 = 15 dB, Γ2 = 25 dB FOR USER VELOCITY = 500 km/h

to accommodate the maximum excess delay of the channel. We

present results of an equivalent OFDM system with synchronous
CP length and block based signal structure as described in
Section II. The MMSE equalizer implemented in this work can
efficiently cancel the ICI at the receiver, unlike the single tap
equalizer used in traditional systems [47].
While 5G-NR has provision for variable subcarrier band-
width [12], [48], the EVA channel model restricts us to a subcar-
rier bandwidth of upto 60 kHz corresponding to numerology two
contrary to the maximum of 240 kHz with numerology four due
to the coherence bandwidth of about 56 kHz. However, it is worth
mentioning that from ICI perspective, numerology four with
maximum of 240 kHz is more desirable. It is important to note
that the receiver used here is MMSE with ideal channel estimates
thus serving the purpose to cancel inter subcarrier interference of
LTV channel which appear due to Doppler spread. The system
performance is then accordingly evaluated using a subcarrier
bandwidth of 15 kHz which is also valid for system design of
4 G systems [11], [49].
In Section VI-A, we present the system-level performance Fig. 3. CDF of downlink sum rate for various NOMA power allocation
of downlink and uplink in terms of sum SE results for the schemes under OTFS/OFDM for Γ1 = 15 dB, Γ2 = 25 dB for user velocity
MMSE-SIC receiver developed in Sections III-A2 and III-B2, = 500 km/h. Solid and dashed lines represent OTFS and OFDM results,
respectively. In Section VI-A1, we evaluate the performance respectively. Markers denote results corresponding to various power allocation
schemes described in Section IV.
results of NOMA-OTFS and OMA-OTFS in order to find the
feasibility of NOMA-based multiuser multiplexing in high-
mobility scenarios. We then conduct a comparative performance
analysis between OTFS-based and OFDM-based NOMA im- A. System-Level SE Results
plementation with an aim to explore relative gains that can be
1) Comparison Between NOMA/OMA-OTFS: In Fig. 3, the
obtained by such system design in high Doppler scenarios (in
cumulative distribution functions (CDF) of downlink sum rates
Section VI-A2). Since we evaluate different power allocation
achieved under various OMA and various NOMA power alloca-
for NOMA-OTFS in Section IV, we analyze their relative per-
tion schemes for OTFS and OFDM are shown. The exact values
formance in Section VI-A3. After system-level performance
of mean and 5% sum rate are given in Table II. From the CDF
evaluation, we delve into link-level performance verification
curves and tabulated values, we observe a significant increase
of LDPC coded NOMA-OTFS CWIC MMSE-SIC receiver in
in mean and outage sum rate in the NOMA-OTFS scheme
Section VI-B. We also compare the same against the equivalent
compared to the OMA-OTFS. We notice that there is more
NOMA-OFDM as well as OMA-OTFS systems as described
than 16% improvement in both mean and 5% outage sum SE,
earlier in this work. The performance is evaluated in terms of
respectively, in case of NOMA-OTFS with average SNR based
metrics like BLER (Pe ), throughput and goodput. The through-
FTPA power allocation with respect to OMA-OTFS. The gain
put of a link is defined as the number of bits transmitted per unit
is even higher for power allocation schemes like channel-norm
time and is given by
based FTPA and instantaneous CSI based WSRM NOMA-OTFS
Throughput = Σ2i=1 RKi bits/s/Hz (45) schemes.
2) Comparison Between OTFS and OFDM Performances:
where R and Ki denote code rate and bits per QAM symbol, From the CDF trends in Fig. 3 and tabulated values in Table II,
respectively. Whereas, the goodput of a link is defined as the it can be observed that the 5% outage sum SE shows significant
number of bits that are successfully received and expressed by improvement for OTFS with respect to OFDM. For example,
Goodput = Σ2i=1 RKi (1 − Pei )bits/s/Hz (46) in case of average SNR based FTPA and channel-norm based
FTPA schemes, an improvement of around 19.7% and 27.5%,
where Pei denotes BLER for ith user, respectively. respectively, is observed. The gain is even higher for weighted

Authorized licensed use limited to: University of Canberra. Downloaded on June 25,2020 at 09:25:41 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

10 IEEE SYSTEMS JOURNAL

are suboptimal. The optimistic results obtained here promotes

further investigation of such schemes which can be implemented
in high Doppler scenario.

B. Link-Level Performance of NOMA-OTFS

In this section, we discuss about the link-level performance
of LDPC coded codeword-level SIC NOMA-OTFS system and
compare it against OMA-OTFS and NOMA-OFDM system. We
first discuss about the downlink NOMA performance and then
the uplink. Tables III and IV show the goodput performance
along with BLER values of the users for downlink and uplink,
respectively. While generating such results, the average SNR
values are considered for choosing the modulation scheme in
order to guarantee that the experienced BLER remains below
the threshold 10−1 [43].
1) Performance of NOMA-OTFS in Downlink: In the down-
link direction, at base station, each user’s data is encoded using
LDPC with code rate R = 2/3. The encoded bit stream is
modulated using QPSK, 16QAM, or 64QAM (Ki = 2, 4, and
Fig. 4. CDF of uplink sum rate for NOMA users under OTFS/OFDM for
Γu u 6, respectively). In order to achieve BLER less than 0.1 with
1 = 10 dB, Γ2 = 30 dB for user velocity = 500 km/h.
LDPC code rate and length shown in Table I, SNR thresholds to
support QPSK, 16QAM, and 64QAM for OTFS are 9.5, 15, and
23.5 dB, respectively. For OFDM, the thresholds are 10.8, 18,
sum rate maximization schemes, reaching to nearly 26% and and 26 dB, respectively. The modulation schemes for both the
55%, respectively. Similar improvement (around 10%) has been users are chosen based on the average SINR(Υ̃) experienced by
observed for OMA scheme as well, highlighting the utility of the users, which are functions of Γ1 , Γ2 , β1 , and β2 . In downlink,
OTFS over OFDM even for orthogonal multiuser scenario. Exact the average SINR of user one are obtained assuming interference
values of gains are tabulated in last column of Table II. The as Gaussian noise is given by
outage improvement in OTFS over OFDM reflects the diversity  
β 1 Γ1
gain of OTFS with respect to OFDM. However, we note that Υ̃1 (in dB) = 10log10 . (47)
OFDM-based NOMA provides nominally higher mean SE (in β 2 Γ1 + 1
the order of 5%–14%) than OTFS-based NOMA. Detailed ana- The post SIC average SINR for user two assuming perfect SIC
lytical treatment in order to investigate this issue is a potential can be expressed as
future work.
Similar improvement in 5% outage sum rate (∼12%) and Υ̃2 (in dB) = 10log10 (β2 Γ2 ). (48)
reduction in mean sum rate (∼17%) in OTFS compared to OFDM
has also been reflected for uplink NOMA in Fig. 4 for the The modulated data of the users is transmitted using SC with
MMSE-SIC receiver designed in Section III-B2. β1 = 0.9 and β2 = 0.1 as described in Section III-A1. This
3) Comparison of Various NOMA Power Allocation choice of βi s results in Υ̃1 = 8.35 dB for Γ1 = 15 dB which
Schemes: The CDF curves in Fig. 3 and tabulated values in puts the system close to minimum operational range. Though
Table II of downlink sum rates for various NOMA power al- the SINR of 8.35 dB is insufficient to satisfy the BLER threshold
location schemes reveal important assessment of the schemes in as per previous discussion but it is observed that the system is
the considered system model. The average SNR based weighted able to support required BLER with these βi s. This observation
sum rate maximization scheme’s (described in Section IV-C1) suggests that the Gaussian assumption considered for evaluating
performance is seen to be the worst compared to the other SINR may not hold true. User one (weak user) decodes the signal
schemes. This is mainly due to the fact that strong user’s average using MMSE equalization as outlined in Section V-A1. On the
received SNR (Γ2 ) is significantly higher than the weak user’s other hand, as detailed in Section V-A2, user two (strong user)
average received SNR (Γ1 ), thus resulting in allocation of full experience higher SNR and thus perform SIC at the codeword
power to the weak user [see (27)], effectively turning the scheme level. Same SINR thresholds are used for uplink modulation
to OMA with only weak user. Judicious choice of weights scheme selection as in downlink. BLERs are evaluated using
(wi ) incorporating proportional fairness can be used to alleviate Monte Carlo link-level simulation for each user for downlink as
the issue. The average SNR based FTPA scheme (described in well as uplink.
Section IV-B1) also gets partially affected due to the same issue Based on the obtained BLER results, we compute throughput
and thus the scheme marginally outperforms the OMA scheme. (45) and goodput (46) for each user. Table III is generated
The channel norm based FTPA (described in Section IV-B2) keeping user one’s modulation as QPSK, as Γ1 = 15 dB. Γ2 is
and instantaneous CSI based weighted sum rate maximization varied such that the higher modulation schemes can be supported
(described in Section IV-C2) schemes have all achieved better by user two. Γ2 = 22, 30, 35 dB are considered. Corresponding
performance in terms of mean and outage SE which is signifi- to these Γ2 values, Υ̃2 obtained from (48) support modulation
cantly higher than the OMA scheme, which advocates for use schemes QPSK,16-QAM, and 64-QAM, respectively, following
of De-Do PD-NOMA for better system performance. It is worth the discussion made earlier about the SNR thresholds corre-
noting that the NOMA power allocation methods discussed here sponding to the modulation schemes. Though the SNR range

Authorized licensed use limited to: University of Canberra. Downloaded on June 25,2020 at 09:25:41 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

CHATTERJEE et al.: NONORTHOGONAL MULTIPLE ACCESS WITH OTFS SIGNAL TRANSMISSION 11

TABLE III
USERWISE BLER RESULTS FOR DOWNLINK NOMA IN OTFS AND OFDM

(β1 = 0.9 and β2 = 0.1, code rate= 2/3, UE1 using QPSK Modulation with SNR 15 dB resulting SINR 8.35 dB).

TABLE IV
USERWISE BLER RESULTS FOR UPLINK NOMA IN OTFS AND OFDM

(Code rate = 2/3, UE1 using QPSK Modulation with SNR 10 dB resulting SINR 10 dB).

for both users can be between −3 and 40 dB, only these com- is 5.33 bits/s/Hz. Thus, a gain of 50% in throughput can be
binations of SNRs are selected as representative values in order obtained in NOMA with respect to OMA.
to demonstrate system performance. An important aspect of
NOMA name user selection is dependent on achievable NOMA
gain, which in turn depends on supportable data rate. Here, VII. CONCLUSION
such considerations are made from the link-level perspective. In In this article, we have presented the performance analysis
operational system, EESM [50], [51] based mapping of user’s of a SC based De-Do domain PD-NOMA-OTFS system in
experienced SINR can be done to choose appropriate rate while high-mobility scenarios. In order to realize NOMA-OTFS, we
satisfying required BLER. have presented a linear MMSE-SIC receiver. Symbol-wise post-
For the SNR pair, Γ1 = 15 dB and Γ2 = 35 dB in the processing SINR is derived for both downlink and uplink for sub-
Table III, user one and user two are assigned QPSK (K = 2) sequent SE analysis of such a system. A few partial CSI-based
and 64-QAM (K=6), respectively, resulting in a throughput power allocation techniques have been implemented among
of 5.33 bits/s/Hz (which is evaluated from (45)) for OTFS downlink NOMA users for performance comparison. In order
while the goodput achieved is 5.10 bits/s/Hz, which is evaluated to investigate the practical applicability of such a system, we
by taking K1 = 2, K2 = 6, Pe1 = 5.6 × 10−2 , Pe2 = 5 × 10−3 have also develop a CWIC receiver with LDPC error-correcting
and R = 2/3 in (46). For the same scenario in OMA case, codes along with MMSE equalization for two-user NOMA case.
user one can support upto 16QAM while user two can support Results show that the De-Do domain two-user PD-NOMA-
upto 64QAM resulting in throughput of (4*2/3 + 6*2/3)/2 = OTFS, as presented in this article, is better than traditional OMA-
3.33 bits/s/Hz. Here, the percentage gain in throughput with OTFS by upto 16% in terms of both mean and outage sum SE
NOMA-OTFS over OMA-OTFS is 37.52%. performance.
When NOMA-OFDM is employed for the same conditions, We have also observed that NOMA-OTFS has upto 50% better
user one is assigned QPSK and user two is assigned 16QAM in outage sum SE when compared to NOMA-OFDM for partial-
order to satisfy BLER threshold resulting in a goodput of 3.99 CSI-based power allocation schemes. For full-CSI-based power
bits/s/Hz. Thus, NOMA-OTFS offers 21.76% gain in goodput allocation schemes, the gain is in the order of 27%. This also
over NOMA-OFDM. indicates that the OTFS gain over OFDM is not reduced by using
2) Performance of NOMA-OTFS in Uplink: In uplink direc- NOMA. However, we note that mean sum SE of appropriately
tion, the Table IV is generated by keeping Γu1 = 10 dB and modified NOMA-OFDM is better than NOMA-OTFS. Thus, we
varying the user two’s SNR, Γu2 = 25, 30, 40 dB thus varying the find that there is a tradeoff between mean and outage SE. The
user two’s modulation scheme as QPSK,16QAM, and 64QAM, improved outage sum SE indicates a more resilient system in
respectively. It can be observed that NOMA using OFDM is high-mobility scenario, which is highly desirable.
unable to support user one as Pe1 in above threshold. For the SNR The link-level performance obtained from the developed
pair, Γu1 = 10 dB and Γu2 = 40 dB, Pe2 = 3.6 × 10−2 , thus user codeword-level SIC receiver shows that the NOMA-OTFS sys-
two can be supported with modulation scheme 64QAM but user tem has upto 21.76% and 10.60% improved goodput in down-
one is unable to transmit even using QPSK due to the resulting link and uplink, respectively, compared to NOMA-OFDM. It
BLER of about 4.1 × 10−1 , as a result of error propagation. The also shows 37.52% and 50% better throughput for NOMA-
resulting NOMA-OFDM goodput is 4.64 bits/s/Hz compared to OTFS over OMA-OTFS system in downlink and uplink,
NOMA-OTFS goodput of 5.19 bits/s/Hz, thus, a gain of 10.60% respectively.
is shown for NOMA with OTFS over OFDM in uplink. If OMA Therefore, based on the system developed and presented
is employed for the same scenario, then user one and user two performance analysis we conclude that NOMA-OTFS has the
can support upto QPSK and 64QAM, respectively, resulting potential to improve the performance of regular OMA-OTFS
in a throughput of 2.67 bits/s/Hz, while NOMA throughput and NOMA-OFDM in high-mobility conditions.

Authorized licensed use limited to: University of Canberra. Downloaded on June 25,2020 at 09:25:41 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

12 IEEE SYSTEMS JOURNAL

REFERENCES [27] M. S. Ali, H. Tabassum, and E. Hossain, “Dynamic user clustering

and power allocation for uplink and downlink non-orthogonal multi-
[1] B. Ai et al., “Challenges toward wireless communications for high-speed ple access (NOMA) systems,” IEEE Access, vol. 4, pp. 6325–6343,
railway,” IEEE Trans. Intell. Transp. Syst., vol. 15, no. 5, pp. 2143–2158, 2016.
Oct. 2014. [28] G. Nain, S. S. Das, and A. Chatterjee, “Low complexity user selection with
[2] R. He et al., “High-speed railway communications: From GSM-R to LTE- optimal power allocation in downlink NOMA,” IEEE Wireless Commun.
R,” IEEE Veh. Technol. Mag., vol. 11, no. 3, pp. 49–58, Sep. 2016. Lett., vol. 7, no. 2, pp. 158–161, Apr. 2018.
[3] F. Hasegawa et al., “High-speed train communications standardization in [29] Z. Ding, X. Lei, G. K. Karagiannidis, R. Schober, J. Yuan, and V. K.
3GPP 5G NR,” IEEE Commun. Standards Mag., vol. 2, no. 1, pp. 44–52, Bhargava, “A survey on non-orthogonal multiple access for 5 g networks:
Mar. 2018. Research challenges and future trends,” IEEE J. Sel. Areas Commun.,
[4] S. Hayat, E. Yanmaz, and R. Muzaffar, “Survey on unmanned aerial vol. 35, no. 10, pp. 2181–2195, Oct. 2017.
vehicle networks for civil applications: A communications viewpoint,” [30] L. Dai, B. Wang, Z. Ding, Z. Wang, S. Chen, and L. Hanzo, “A survey of
IEEE Commun. Surv. Tut., vol. 18, no. 4, pp. 2624–2661, Oct.–Dec. 2016. non-orthogonal multiple access for 5G,” IEEE Commun. Surv. Tut., vol. 20,
[5] W. Viriyasitavat, M. Boban, H.-M. Tsai, and A. Vasilakos, “Vehicular com- no. 3, pp. 2294–2323, Jul.–Sep. 2018.
munications: Survey and challenges of channel and propagation models,” [31] Z. Ding, R. Schober, P. Fan, and H. V. Poor, “OTFS-NOMA: An efficient
IEEE Veh. Technol. Mag., vol. 10, no. 2, pp. 55–66, Jun. 2015. approach for exploiting heterogenous user mobility profiles,” IEEE Trans.
[6] L. Liang, H. Peng, G. Y. Li, and X. Shen, “Vehicular communications: Commun., vol. 67, no. 11, pp. 7950–7965, Nov. 2019.
A physical layer perspective,” IEEE Trans. Veh. Technol., vol. 66, no. 12, [32] Z. Ding, “Robust beamforming design for OTFS-NOMA,” IEEE Open J.
pp. 10 647–10 659, Dec. 2017. Commun. Soc., vol. 1, pp. 33–40, 2020.
[7] J. Wu and P. Fan, “A survey on high mobility wireless communications: [33] ITU, “Guidelines for evaluation of radio interface technologies for IMT-
Challenges, opportunities and solutions,” IEEE Access, vol. 4, pp. 450– advanced,” International Telecommunication Union, Geneva, Recommen-
476, 2016. dation M2135, Dec. 2009.
[8] R. V. Nee and R. Prasad, OFDM for Wireless Multimedia Communications. [34] P. Raviteja, K. T. Phan, Y. Hong, and E. Viterbo, “Embedded delay-Doppler
Norwood, MA, USA: Artech House, 2000. channel estimation for orthogonal time frequency space modulation,” in
[9] R. Prasad, OFDM for Wireless Communications Systems. Norwood, MA, Proc. IEEE 88th Veh. Technol. Conf., 2018, pp. 1–5.
USA: Artech House, 2004. [35] S. Tiwari, S. S. Das, and V. Rangamgari, “Low complexity LMMSE
[10] “IEEE standard for information technology– local and metropolitan area receiver for OTFS,” IEEE Commun. Lett., vol. 23, no. 12, pp. 2205–2209,
networks– specific requirements– part 11: Wireless LAN medium access Dec. 2019.
control (MAC) and physical layer (PHY) specifications amendment 5: En- [36] I. Abu Mahady, E. Bedeer, S. Ikki, and H. Yanikomeroglu, “Sum-rate
hancements for higher throughput,” IEEE Std 802.11n-2009 (Amendment maximization of NOMA systems under imperfect successive interfer-
to IEEE Std 802.11-2007 as amended by IEEE Std 802.11k-2008, IEEE ence cancellation,” IEEE Commun. Lett., vol. 23, no. 3, pp. 474–477,
Std 802.11r-2008, IEEE Std 802.11y-2008, and IEEE Std 802.11w-2009), Mar. 2019.
Oct. 2009, pp. 1–565. [37] X. Wang, R. Chen, Y. Xu, and Q. Meng, “Low-complexity power allocation
[11] E. Dahlman, S. Parkvall, and J. Skold, 4G: LTE/LTE-Advanced for Mobile in NOMA systems with imperfect SIC for maximizing weighted sum-rate,”
Broadband. Norwell, MA, USA: Academic, 2013. IEEE Access, vol. 7, pp. 94 238–94 253, 2019.
[12] S. S. Das, E. D. Carvalho, and R. Prasad, “Performance analysis of OFDM [38] R. Hadani et al., “Orthogonal time frequency space modulation,” 2018,
systems with adaptive sub carrier bandwidth,” IEEE Trans. Wireless Com- arXiv:1808.00519.
mun., vol. 7, no. 4, pp. 1117–1122, Apr. 2008. [39] A. Paulraj, R. Nabar, and D. Gore, Introduction to Space-Time Wireless
[13] S. S. Das and R. Prasad, Evolution of Air Interface Towards 5G: Radio Communications, 1st ed. New York, NY, USA: Cambridge Univ. Press,
Access Technology and Performance Analysis. Gistrup, Denmark: River 2008.
Publishers, 2018. [40] T. Hou, X. Sun, and Z. Song, “Outage performance for non-orthogonal
[14] R. Hadani and S. S. Rakib, “Multiple access in an orthogonal time multiple access with fixed power allocation over nakagami-m fading
frequency space communication system,” May 8, 2018, U.S. Patent channels,” IEEE Commun. Lett., vol. 22, no. 4, pp. 744–747, Apr. 2018.
9,967,758. [41] T. Seyama, T. Dateki, and H. Seki, “Efficient selection of user sets for
[15] A. Monk, R. Hadani, M. Tsatsanis, and S. Rakib, “OTFS-orthogonal time downlink non-orthogonal multiple access,” in Proc. IEEE 26th Annu. Int.
frequency space,” 2016, arXiv:1608.02993. Symp. Pers., Indoor, Mobile Radio Commun., 2015, pp. 1062–1066.
[16] R. Hadani et al., “Orthogonal time frequency space modulation,” in Proc. [42] Q. Wang, R. Zhang, L. Yang, and L. Hanzo, “Non-orthogonal multiple
IEEE Wireless Commun. Netw. Conf., 2017, pp. 1–6. access: A unified perspective,” IEEE Wireless Commun., vol. 25, no. 2,
[17] P. Raviteja, Y. Hong, E. Viterbo, and E. Biglieri, “Practical pulse-shaping pp. 10–16, Apr. 2018.
waveforms for reduced-cyclic-prefix OTFS,” IEEE Trans. Veh. Technol., [43] K. Saito, A. Benjebbour, Y. Kishiyama, Y. Okumura, and T. Nakamura,
vol. 68, no. 1, pp. 957–961, Oct. 2018. “Performance and design of SIC receiver for downlink NOMA with open-
[18] G. Surabhi, R. M. Augustine, and A. Chockalingam, “On the diversity of loop SU-MIMO,” in Proc. IEEE Int. Conf. Commun. Workshop, 2015,
uncoded OTFS modulation in doubly-dispersive channels,” IEEE Trans. pp. 1161–1165.
Wireless Commun., vol. 18, no. 6, pp. 3049–3063, Jun. 2019. [44] C. Yan, A. Harada, A. Benjebbour, Y. Lan, A. Li, and H. Jiang, “Receiver
[19] V. Khammammetti and S. K. Mohammed, “OTFS-based multiple-access design for downlink non-orthogonal multiple access (NOMA),” in Proc.
in high Doppler and delay spread wireless channels,” IEEE Wireless IEEE 81st Veh. Technol. Conf., 2015, pp. 1–6.
Commun. Lett., vol. 8, no. 2, pp. 528–531, Apr. 2019. [45] L. Yuan, J. Pan, N. Yang, Z. Ding, and J. Yuan, “Successive interference
[20] M. T. P. Le, G. C. Ferrante, T. Q. S. Quek, and M.-G. Di Benedetto, cancellation for LDPC coded nonorthogonal multiple access systems,”
“Fundamental limits of low-density spreading noma with fading,” IEEE IEEE Trans. Veh. Technol., vol. 67, no. 6, pp. 5460–5464, Jun. 2018.
Trans. Wireless Commun., vol. 17, no. 7, pp. 4648–4659, Jul. 2018. [46] J. Zhao, F. Zarkeshvari, and A. H. Banihashemi, “On implementation of
[21] D. Tse and P. Viswanath, Fundamentals of Wireless Communication. min-sum algorithm and its modifications for decoding low-density parity-
Cambridge, U.K.: Cambridge Univ. Press, 2005. check (LDPC) codes,” IEEE Trans. Commun., vol. 53, no. 4, pp. 549–554,
[22] T. M. Cover and J. A. Thomas, Elements of Information Theory. Hoboken, Apr. 2005.
NJ, USA: Wiley, 2012. [47] W. G. Jeon, K. H. Chang, and Y. S. Cho, “An equalization technique
[23] Y. Saito, Y. Kishiyama, A. Benjebbour, T. Nakamura, A. Li, and for orthogonal frequency-division multiplexing systems in time-variant
K. Higuchi, “Non-orthogonal multiple access (NOMA) for cellular future multipath channels,” IEEE Trans. Commun., vol. 47, no. 1, pp. 27–32,
radio access,” in Proc. IEEE 77th Veh. Technol. Conf., 2013, pp. 1–5. Jan. 1999.
[24] Y. Saito, Y. Kishiyama, A. Benjebbour, T. Nakamura, A. Li, and [48] A. A. Zaidi et al., “Waveform and numerology to support 5G services and
K. Higuchi, “System-level performance evaluation of downlink non- requirements,” IEEE Commun. Mag., vol. 54, no. 11, pp. 90–98, Nov. 2016.
orthogonal multiple access (NOMA),” in Proc. IEEE 24th Int. Symp. Pers. [49] S. Sesia, I. Toufik, and M. Baker, LTE-the UMTS Long Term Evolution:
Indoor Mobile Radio Commun., 2013, pp. 611–615. From Theory to Practice. Hoboken, NJ, USA: Wiley, 2011.
[25] P. Parida and S. S. Das, “Power allocation in OFDM based NOMA systems: [50] S. N. Donthi and N. B. Mehta, “An accurate model for EESM and its
A DC programming approach,” in Proc. IEEE Globecom Workshops, 2014, application to analysis of CQI feedback schemes and scheduling in LTE,”
pp. 1026–1031. IEEE Trans. Wireless Commun., vol. 10, no. 10, pp. 3436–3448, Oct. 2011.
[26] B. Di, L. Song, and Y. Li, “Sub-channel assignment, power allocation, [51] H. Song, R. Kwan, and J. Zhang, “Approximations of EESM effective
and user scheduling for non-orthogonal multiple access networks,” IEEE SNR distribution,” IEEE Trans. Commun., vol. 59, no. 2, pp. 603–612,
Trans. Wireless Commun., vol. 15, no. 11, pp. 7686–7698, Nov. 2016. Feb. 2011.

Authorized licensed use limited to: University of Canberra. Downloaded on June 25,2020 at 09:25:41 UTC from IEEE Xplore. Restrictions apply.