ACCEPTED FROM OPEN CALL

Orthogonal Time-Frequency Space Modulation:

A Promising Next-Generation Waveform
Zhiqiang Wei, Weijie Yuan, Shuangyang Li, Jinhong Yuan, Ganesh Bharatula, Ronny Hadani, and Lajos Hanzo

Abstract Nevertheless, recently an increasing amount of

research attention has been dedicated to design-
Sixth-generation (6G) wireless networks are ing new modulation waveform and schemes for
envisioned to provide global coverage for the high-mobility communications of next-generation
intelligent digital society of the near future, rang- wireless networks.
ing from traditional terrestrial to non-terrestri- High-mobility communications operating at
al networks, where reliable communications in high carrier frequencies suffer from severe Dop-
high-mobility scenarios at high carrier frequencies pler spreads, mainly caused by the relative motion
would play a vital role. In such scenarios, the con- between the transmitter, receiver, and scatterers.
ventional orthogonal frequency division multiplex- Conventional orthogonal frequency-division mul-
ing (OFDM) modulation, that has been widely tiplexing (OFDM) modulation, which has been
used in both the fourth-generation (4G) and the widely used in both the fourth-generation (4G) and
emerging fifth-generation (5G) cellular systems as the emerging fifth-generation (5G) cellular systems
well as in WiFi networks, is vulnerable to severe as well as in WiFi networks, suffers in high-mobility
Doppler spread. In this context, this article aims scenarios. OFDM waveform is impaired by severe
to introduce a recently proposed two-dimension- inter-carrier interference (ICI), which is aggravated
al modulation scheme referred to as orthogonal by the fact that the highest and lowest subcarriers
time-frequency space (OTFS) modulation, which exhibit rather different normalized Doppler. Hence,
conveniently accommodates the channel dynamics synchronization is also a challenge. Recently, a
via modulating information in the delay-Doppler new two-dimensional (2D) modulation scheme,
domain. This article provides an easy-reading over- namely orthogonal time-frequency space (OTFS)
view of OTFS, highlighting its underlying motiva- [2], has been proposed as a promising candidate
tion and specific features. The critical challenges for high-mobility communications.
of OTFS and our preliminary results are present- OTFS modulates information in the delay-Dop-
ed. We also discuss a range of promising research pler (DD) domain rather than in the time-frequency
opportunities and potential applications of OTFS in (TF) domain of classic OFDM modulation, pro-
6G wireless networks. viding a strong delay-resilience and Doppler-resil-
ience, while enjoying the potential of full diversity
Introduction [2], which is the key for supporting reliable com-
Sixth-generation (6G) wireless networks are munications. Additionally, OTFS modulation can
expected to support ubiquitous connectivity to transform a time-variant channel into a 2D qua-
a wide range of mobile terminals, spanning from si-time-invariant channel in the DD domain, where
autonomous cars to unmanned aerial vehicles its attractive properties can be exploited. Given
(UAV), low-earth-orbit (LEO) satellites, and high that most of the existing wireless system designs
speed-trains, etc. One of the critical challenges for have been conceived for low-mobility and low-car-
these services is to provide reliable communica- rier scenarios, OTFS introduces new critical chal-
tions in high-mobility environments. Additionally, lenges in transceiver architecture and algorithmic
the spectrum congestion under 6 GHz creates designs. To unleash the full potential of OTFS, chal-
a fundamental bottleneck for capacity improve- lenging fundamental research problems have to be
ment and sustainable system evolution. The ultra- addressed, including channel estimation, detection,
high data rate requirements of panoramic and as well as multi-antenna and multi-user designs.
holographic video streaming push mobile pro- In contrast to existing tutorial papers on OTFS
viders to utilize higher frequency bands, such as [2, 3], this article portrays OTFS modulation con-
the millimeter wave (mmWave) bands, where a ceived for communications over high-mobility envi-
huge chunk of unused spectrum is available. In ronments by providing an easy-reading overview
general, wireless communications in high-mobility of its fundamental concepts, highlighting the chal-
scenarios at high carrier frequencies are extreme- lenges and potential solutions as well as explor-
ly challenging due to the hostile channel varia- ing new promising areas for future research. The
tions. Relying on adaptive coherent/non-coherent rest of this article is organized as follows. The next
detection [1] is beneficial for attaining a certain section introduces the fundamentals of OTFS. The
degree of robustness against channel variations. potential applications and research opportunities

Zhiqiang Wei, Shuangyang Li, and Jinhong Yuan are with the University of New South Wales;
Digital Object Identifier: Weijie Yuan (corresponding author) is with Southern University of Science and Technology; Ganesh Bharatula is with Telstra Corp Ltd.;
10.1109/MWC.001.2000408 Ronny Hadani is with Cohere Technologies, Inc. and University of Texas at Austin; Lajos Hanzo is with University of Southampton.

136 1536-1284/20/$25.00 © 2020 IEEE IEEE Wireless Communications • August 2021

HANZO_LAYOUT.indd 136 8/31/21 12:48 AM

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on November 01,2022 at 10:50:07 UTC from IEEE Xplore. Restrictions apply.
 of OTFS are then presented. Then, before conclud- Given that most of the
Dispersive time-invariant Time-variant LoS channel
ing, we demonstrate the most important design multipath channel existing wireless system
challenges of OTFS and their potential solutions. designs have been con-
FundAmentAls oF otFs ceived for low-mobility
and low-carrier scenar-
A fundamental question to answer for motivating
the research community and industry to investi- ios, OTFS introduces
gate OTFS is why we shall perform modulation new critical challenges
in the DD domain. Hence, commencing from in transceiver architec-
the channel characteristics of high-mobility com- Frequency-selective ture and algorithmic
Time-selective
munications, we present the basic concepts and
properties of the DD domain channels, the DD designs. To unleash the
domain multiplexing, the OTFS transceiver archi- Doubly-selective full potential of OTFS,
tecture and signal waveform, as well as the OTFS challenging fundamen-
system design principle. tal research problems
have to be addressed,
From tIme-InvArIAnt to tIme-vArIAnt chAnnels including channel esti-
Wireless channels can be modeled by a linear
time-invariant (LTI) system, provided that the mation, detection, as
channel impulse response (CIR) is time-invariant well as multi-antenna
or has a long coherence time, cf. Fig. 1. In the Time-variant dispersive multipath channel
and multi-user designs.
presence of multiple scatterers, the dispersive FIGURE 1. An illustration of frequency-selective,
LTI channel’s output is a temporal-smeared ver- time-selective, and doubly-selective channel
sion of the transmitted signal, but again, the CIR models.
is time-invariant. In this case, a one-dimensional
(1D) CIR in the delay domain h(t) is sufficient for
characterizing the time-dispersive channel. The and frequency f. Due to the limited coherence
Fourier transform (FT) of this CIR is a frequen- time and coherence bandwidth (coherence
cy-selective channel transfer function (CTF). With region in Fig. 2) of LTV channels, channel acqui-
increasing the CIR delay spread, the selectivity sition in the TF domain would be challenging and
becomes more severe since the separation of fre- would impose a significant signaling overhead.
quency-domain (FD) fades is increasingly propor- For instance, for an OFDM system having a car-
tional to the CIR-length. rier frequency of f c = 3.5 GHz and a subcarri-
However, the assumption of having LTI CIRs may er spacing of Df = 15 kHz supporting a relative
no longer hold in the face of increased user mobility velocity of v = 300 km/h, the maximum Doppler
and carrier frequency. Therefore, the linear time-vari- shift is nmax = 972.22 Hz and the OFDM symbol
ant (LTV) channel model [4] has attracted consid- duration including a 20 percent cyclic prefix (CP)
erable research attention in high-mobility scenarios. is 80 s. Assuming that the channel’s coherence
LTV channels give rise to frequency shifts due to the time is 1/(4nmax) = 257.14 s, one channel coher-
Doppler effect, yielding a spectral-smeared version ence interval can only accommodate at most
of the transmitted signal, that is, frequency-dispersive. three OFDM symbols.
Frequency-dispersive channels are time-selective and Applying the FT to h(t, t) w.r.t. t yields the
the separation of the channel’s time-domain (TD) DD domain channel (spreading function), h(t,
fades is increasingly proportional to the Doppler n). The DD domain channel h(t, n) characteriz-
spread. In practice, LTV channels of high-mobility es the intensity of scatterers having a propagation
scenarios are often doubly-dispersive due to the joint delay of t and Doppler frequency shift of n, which
presence of multipath propagation and Doppler directly captures the underlying physics of radio
effects, cf. Fig. 1. The transmitted signals suffer from propagation in high-mobility environments. More
dispersion both in the TD and FD. In such scenari- importantly, the LTV channel in the DD domain
os, each tap of the CIR function is time-dependent, exhibits beneficial features of separability, stability,
fluctuating according to the rate of /(2n) between compactness, and possibly sparsity, as illustrated in
consecutive TD fades, cf. Fig. 2, where  denotes Fig. 2 and detailed below, which can be exploited
the wavelength and v is the relative speed between to facilitate efficient channel estimation and data
the transmitter and receiver. Hence, this results in detection.
a 2D CIR function h(t, t) in the time-delay domain. Separability: Additionally introducing the Dop-
In contrast to the traditional way of treating TD and pler domain of wireless channels allows us to
FD dispersion as undesired channel impairments, separate the propagation paths experiencing an
we can beneficially exploit the additional degrees identical delay and thus fully discloses the available
of freedom (DoF) of doubly-dispersive channels for channel DoF.
achieving reliable diversity-aided communications in Stability: Only the drastic change of propaga-
high-mobility scenarios. tion path lengths and moving speeds may cause
channel variations in the DD domain. Consequent-
ltv chAnnels In tF And dd domAIns ly, DD domain channels fluctuate much slower
Apart from the time-delay domain, LTV chan- than time-delay domain or TF domain channels.
nels can be equivalently described in either the Compactness: It is worth noting that in typi-
TF or DD domain, cf. Fig. 2. To emphasize the cal wireless channels, we have 4t maxn max≤ 1 [4],
TF selectivity, the TF domain channel, h(t, f), can where tmax indicates the maximum delay. Hence,
be obtained by the FT of h(t, t) with respect to h(t, n) has a compact DD domain support within
(w.r.t.) delay t. Note that h(t, f) can be interpreted the intervals [0, t max] and [–nmax, nmax] along the
as the complex CTF coefficient at time instant t delay and Doppler dimensions, respectively.

IEEE Wireless Communications • August 2021 137

HANZO_LAYOUT.indd 137 8/31/21 12:48 AM

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on November 01,2022 at 10:50:07 UTC from IEEE Xplore. Restrictions apply.
 In contrast to the
Fourier transform
existing 1D TD or FD
modulation techniques,
OTFS is a 2D modula-
tion technique, where
the data symbols are
multiplexed in the
DD domain and each
symbol is spread right
across the entire TF
domain. This property Fourier transform
Symplectic Fourier transform
is desirable for attaining
FIGURE 2. LTV channels in the time-delay, TF, and DD domains.
the maximum achiev-
able diversity for trans-
Potential Sparsity: When an open-space rural es independent fading. The maximum attainable
mission over doubly propagation environment having few moving scat- diversity order is determined by the number of
dispersive channels, terers is considered, the DD domain channel exhib- independently fading resolvable paths in the DD
provided that each TD its a sparse response[5]. domain. To fully exploit the maximum attainable
and FD sample expe- Note that the above discussions are mainly relat- diversity order, advanced detection techniques are
riences independent ed to the deterministic description of LTV channels. required that can efficiently combine the informa-
More details on the stochastic characterization of tion conveyed by the different propagation paths
fading. LTV channels can be found in [4]. Parameterizing to the OTFS receiver. Although the asymptotic
channel with the aid of delay and Doppler is not diversity gain of the uncoded OTFS system can be
new, which has been widely adopted in the areas as low as one in some special cases as revealed in
of radar and sonar. The main benefit of the OTFS [6], the probability of occurrence of these cases
waveform is the DD domain multiplexing, which tends to zero for a moderate DD domain grid size
will be detailed in the next section. [7]. For example, it becomes as low as 10(–150) for
a DD domain grid of M = N =16 [7]. Furthermore,
From TF to DD Domain Multiplexing channel coding is capable of avoiding the occur-
The classic modulation techniques typically multi- rence of these cases and thus OTFS can achieve
plex data in the TD or FD, which are briefly intro- the full diversity order almost surely.
duced as follows. The information symbols are carried by the DD
Time-Division Multiplexing (TDM): TDM carries domain pulses, which are 2D Sinc functions [2,
information (QAM) symbols in unique, user-specific 8], cf. Fig. 3. The width of the DD domain basis
time-slots, within the same frequency band. function along the delay dimension is inversely
Frequency-Division Multiplexing (FDM): FDM proportional to the system bandwidth MDf, and
multiplexes streams or users in dedicated FD slots its width along the Doppler dimension is inverse-
occupied at the same time. ly proportional to time duration NT occupied by
Code-Division Multiple Access (CDMA): The an OTFS frame [8]. When M → ∞, and N → ∞,
information symbols (of different users) are car- the DD domain basis function tends to a 2D Dirac
ried over unique (user-specific) single-carrier TD, impulse, which corresponds to a 2D localized
multi-carrier FD or multi-carrier time-frequency-do- pulse in the DD domain, cf. Fig. 3. This pulse local-
main spreading sequences. ization capability experienced in the DD domain
OFDM: OFDM transmits its information sym- stands in sharp contrast to the properties of its dual
bols by overlapping Sinc-shaped orthogonal sub- pair, namely the TF domain, where the Heisenberg
carriers in parallel. uncertainty principle prevents corresponding pulse
Index Modulation (IM): IM relies on the gen- localization.
eralized ON/OFF keying principle to map the Moreover, adopting 2D localized pulses for
information bits to the indices of spatial-, time- and conveying the information in the DD domain is
frequency-domain resources. beneficial for exploiting the underlying physical
However, these traditional modulation tech- propagation properties. We note that convention-
niques may not work well in the face of severe al wireless communication designs treat channel
Doppler spread. For instance, the popular OFDM fading as an inevitable deleterious effect, aiming
modulation efficiently transforms a frequency-se- for combating or exploiting it while ignoring its
lective fading channel to multiple parallel frequen- basic underlying causes. In more detail, the chan-
cy-flat subchannels, which allows a low-complexity nel impairments of propagation delay and Doppler
single-tap FD equalization for transmission over LTI frequency shifts are modeled as a pair of opera-
channels. However, the orthogonality of OFDM tions imposed by wireless channels on the trans-
waveform erodes in high-mobility scenarios when mitted waveform. Furthermore, fading is viewed
conventional wireless transceivers are adopted. as a phenomenon that manifests itself at the chan-
In contrast to the existing 1D TD or FD modu- nel’s output imposed by the combined destructive
lation techniques, OTFS is a 2D modulation tech- effect of this pair of fundamental channel-induced
nique, where the data symbols are multiplexed in phenomena. In contrast, OTFS generates a com-
the DD domain and each symbol is spread right plete orthogonal family of waveforms, which is
across the entire TF domain. This property is desir- closed under arbitrary combinations of the time
able for attaining the maximum achievable diversity delay and frequency shift. In other words, upon
for transmission over doubly dispersive channels, transmitting a signal in this family, the received sig-
provided that each TD and FD sample experienc- nal will remain within the family under arbitrary

138 IEEE Wireless Communications • August 2021

HANZO_LAYOUT.indd 138 8/31/21 12:48 AM

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on November 01,2022 at 10:50:07 UTC from IEEE Xplore. Restrictions apply.
 FIGURE 3. OTFS transceiver architecture, the concept of DD domain data multiplexing and their coupling with DD domain channels.

IEEE Wireless Communications • August 2021 139

HANZO_LAYOUT.indd 139 8/31/21 12:48 AM

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on November 01,2022 at 10:50:07 UTC from IEEE Xplore. Restrictions apply.
 We can observe that channel impairments. This unique characteristic a benefit of its Doppler-resilience, OTFS is more
the OTFS transceiver of the OTFS waveform is due to the quasi-period- robust against the carrier frequency offset than
can be implemented icity property [9] of the DD domain basis func- OFDM. These advantages render OTFS eminently
based on the conven- tions, cf. Fig. 3. This property gives rise to a 2D suitable for high-mobility and high-carrier scenarios.
(quasi-) circular localized inter-symbol interference
tional OFDM architec- (ISI) pattern in the DD domain, representing the OTFS System Design Principle
ture by adding some channel impairments, cf. Fig. 3. Since the time- and According to Fig. 3, the data rate of OTFS sys-
pre-processing and frequency-shifts are separated in the DD domain, tems is determined by the amount of data sym-
post-processing blocks, the destructive effects of fading are substantially bols accommodated within a single OTFS frame,
thus making OTFS mitigated. with its largest possible value being MN. Since a
single OTFS frame occupies a time duration of
attractive from the per- OTFS Transceiver Architecture and Waveform NT and a bandwidth of MDf, the spectral efficien-
spective of implemen- The OTFS transceiver is shown in Fig. 3, where cy of OTFS systems is (1 – h)Rclog2M bit/s/Hz,
tation. Nevertheless, the modulated (pilot) symbols are first mapped to where Rc is the code rate, M is the modulation
as a block modulation the DD domain. Then, an orthogonal 2D precod- order, and h captures the training overhead as
scheme, OTFS systems ing, such as the inverse symplectic finite Fourier well as the guard interval overhead. The guard
transform (ISFFT) and Walsh-Hadamard transform interval in the TD has to be longer than the chan-
suffer from a higher [10], transplants the DD domain signal into the TF nel’s delay spread to avoid interference between
latency than classic domain. Then, a multi-carrier modulator, such as OTFS frames. Furthermore, the code rate Rc and
OFDM systems. OFDM or filterbank multicarrier (FBMC) modula- the modulation order M have to be carefully
tor, is employed in each time slot for further trans- selected to guarantee communication reliability.
forming the TF domain signal to the TD before As such, a key technique of improving the spec-
being transmitted over the channel. At the receiv- tral efficiency of OTFS systems is that of reduc-
er side, a cascaded combination of multi-carri- ing the training overhead, as shown in Fig. 4. On
er demodulation and 2D orthogonal decoding the other hand, the OTFS system parameters M,
transforms the received signal back into the DD N, Df, and T have to be chosen appropriately for
domain and retrieves the transmitted symbols in its practical implementations. In particular, the
the DD domain using an appropriate channel esti- OTFS frame duration NT has to be smaller than
mator [11] and equalizer [12]. We can observe the tolerable application latency. Furthermore,
that the OTFS transceiver can be implemented as the computational complexity and PAPR of
based on the conventional OFDM architecture OTFS modulation are proportional to N [12], we
by adding some pre-processing and post-process- prefer to choose a smaller N with a longer slot
ing blocks, thus making OTFS attractive from the duration T. Meanwhile, the available spectrum is
perspective of implementation. Nevertheless, as granulated into more subcarriers, that is, a higher
a block modulation scheme, OTFS systems suffer M is associated with a smaller subcarrier spacing
from a higher latency than classic OFDM systems. Df = 1/T.
As illustrated in Fig. 3, a single pulse represent-
ing a symbol at (t, n) in the DD domain will be Potential Applications and Opportunities
spread across the whole TF domain. The resultant OTFS modulation is envisioned to have diversified
TD waveform is the fluctuating pulse train seen applications in next-generation wireless networks,
in Fig. 3, where the fluctuation rate is determined as discussed in the following.
by the Doppler frequency n, while the pulse loca-
tion within each time slot is determined by the Vehicular Networks
delay t. Consequently, the TD waveform of OTFS Vehicular networks allow various vehicles to wire-
behaves locally like TDMA (localized pulses in the lessly exchange information with each other or
TD), globally like OFDM (localized pulses in the with roadside units (RSUs) to provide a variety
Doppler domain) and spreading like CDMA (2D of benefits, including cooperative traffic manage-
spreading in the TF domain), thus inheriting their ment, road-safety improvements and the support
beneficial properties. For example, OTFS exhib- of autonomous driving. OTFS can play a key role
its resilience to narrow-band interference and it is in future vehicular networks owing to its intrinsic
eminently suitable for multi-user scenarios. Addi- advantages for communications over high-mobil-
tionally, OTFS transforms the violently fluctuat- ity channels. The current standards of vehicular
ing TF domain channel into a quasi-time-invariant communications, such as IEEE 802.11bd and 5G
channel in the DD domain, which can potential- NR V2X, mainly consider OFDM-based wave-
ly be exploited for striking a compelling trade-off forms, where the impact of channel variations is
between the performance, computational com- mitigated either by inserting a midamble or by
plexity and signaling overhead. increasing the subcarrier spacing. By contrast,
Apart from its potential of exploiting full diver- the potential channel stability of OTFS exhibit-
sity and Doppler-resilience, OTFS also has some ed in the DD domain enables prompt initial link
further benefits over conventional modulation setup, agile sidelink scheduling, as well as predic-
techniques. For example, despite its multicarrier tive resource scheduling. Furthermore, the OTFS
nature, the peak-to-average power ratio (PAPR) waveform enjoys a lower PAPR than OFDM,
of OTFS is much lower than that of both OFDM allowing a better communication coverage for
and generalized frequency division multiplexing vehicular networks.
(GFDM), which is particularly beneficial for pow-
er-limited systems, such as the Internet-of-Things Millimeter-Wave Communications
(IoT). Additionally, the guard intervals are only The millimeter-wave frequency band possesses a
required between consecutive OTFS frames, rather large amount of under-utilized spectrum and has
than between time slots, and thus the associated the potential of offering giga-bit-per-second com-
idle time is significantly reduced. Furthermore, as munication services in future wireless networks.

140 IEEE Wireless Communications • August 2021

HANZO_LAYOUT.indd 140 8/31/21 12:48 AM

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on November 01,2022 at 10:50:07 UTC from IEEE Xplore. Restrictions apply.
 The Doppler effect becomes more severe upon
increasing the carrier frequency even at a low/ 100
medium velocity. Although increasing the subcar-
rier spacing to mitigate the resultant ICI is feasi-
ble, the TD symbol duration will be shorter and
inserting a CP for guarding against ISI will intro-
duce a significant overhead. The excessive phase
noise associated with high-frequency oscillators 10-1
also results in a time-varying composite channel. Performance gain of the
OTFS provides a strong immunity to the oscillator proposed DDCE

MSE
phase noise, which is crucial for mmWave com-
munications.

non-terrestrIAl netWorks 10-2

Non-terrestrial networks (NTN) provide a new Rect. - 12.19% overhead
telecommunication infrastructure based on air- DC - 12.19% overhead
borne or spaceborne vehicles, such as satellites, DC - 0.1% overhead
UAVs or high altitude platforms (HAPs). They Performance DC - 0.1% overhead (DDCE)
are capable of supporting the terrestrial 5G net- gain of window TF domain 33.33% overhead
works in the provision of global coverage and design
mobility, as well as ubiquitous connectivity and 10-3
enhanced network reliability. Since the airborne 10 15 20 25 30 35 40 45 50
and spaceborne vehicles usually move fast, the SNR (dB)
high Doppler spread experienced by the NTN
imposes new challenges on its air interface FIGURE 4. Performance comparison of the DD and TF domain channel esti-
design. OTFS modulation has rich potential in mation with different training overheads. The number of paths in the DD
the NTN owing to its prominent capability of domain is 5, N = 32, M = 32, nmax = 2/(NT), and tmax = 4/(MDf).
handling the Doppler effect. Additionally, air-
borne and spaceborne vehicles have limited
on-board power supply and computing capa- chAnnel estImAtIon
bility, hence the low PAPR and low-complexity The channel envelope fluctuates violently even in
of OTFS are of pivotal importance. Moreover, a short time period in high-mobility environments.
the corresponding NTN communication links Accurately estimating the CIR in OTFS systems
spanning to the ground terminals usually exhibit is a challenging but vital requirement for reliable
spatial channel sparsity in the DD domain, which detection. Thanks to the DD domain channel
allows OTFS to strike an attractive performance sparsity and quasi-stationarity, channel acquisi-
vs. complexity trade-off. tion in the DD domain is more convenient than
that in the TF domain, even for a lower training
underWAter AcoustIc communIcAtIons overhead, cf. Fig. 4. However, the DD domain
Underwater acoustic (UWA) channels are channel may not always be sparse [13], particu-
regarded as one of the most challenging wire- larly in the presence of fractional Doppler [12].
less channels, due to their high delay spread, For example, when the exact Doppler frequency
limited bandwidth, and rapid time variations. Sin- straddles a pair of finite-resolution bins in Fig. 3,
gle-carrier modulation using decision feedback rather than falling exactly into the 1-th bin, the
equalizers (DFE), OFDM and orthogonal signal DD domain channel is spread across all the Dop-
division multiplexing (OSDM) are the most popu- pler indices. Due to this channel spreading, a
lar schemes for UWA communications. However, much larger guard space is needed around the
they all transmit information in the TF domain, pilot symbols to avoid the interference caused by
where both the ISI and ICI equalization become unknown data symbols for channel estimation,
tedious tasks. which imposes a significant training overhead.
By contrast, OTFS is Doppler-resilient, hence A promising solution to address this issue is
transmitting information in the DD domain may enhancing the channel sparsity via designing a
outperform these TF domain modulation schemes bespoke TF domain window. In particular, we pro-
in UWA channels. Furthermore, UWA channels posed to apply a Dolph-Chebyshev (DC) window
tend to be sparse in the DD domain, where the at the OTFS transmitter or receiver [13] to suppress
equalization might be easier than that in the TF the channel spreading. Due to the enhanced chan-
domain. Moreover, UWA communications are nel sparsity, the DC windowing achieves a much
usually considered as wideband systems, since the improved channel estimation accuracy over the
ratio of acoustic signal bandwidth over the carri- conventional rectangular window [11], cf. Fig. 4.
er frequency is typically much higher than that in To further reduce the training overhead, we pro-
terrestrial communications. Hence, the potential pose to use decision-directed channel estimation
multipath-scale diversity of the DD domain channel (DDCE), where the reliably detected data symbols
[4] can be beneficially exploited. can be used for refining channel estimates based
on the classic decision-directed principle, rather
chAllenges And solutIons than purely relying on the known pilot symbols.
As a fledgling waveform, OTFS modulation unveils The refined channel estimates can again be used
new opportunities but also has its own challenges. for OTFS symbol detection, which can in turn
In this section, we introduce three fundamental improve the channel estimation accuracy. Figure
research problems of OTFS and their potential 4 demonstrates that the proposed DDCE scheme
solutions. without guard space consumes only 0.1 percent

IEEE Wireless Communications • August 2021 141

HANZO_LAYOUT.indd 141 8/31/21 12:48 AM

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on November 01,2022 at 10:50:07 UTC from IEEE Xplore. Restrictions apply.
 domain, while the extrinsic information is iterative-
100 ly exchanged between the time domain and DD
VB 150km/h VB 300km/h
domain via the corresponding unitary transforma-
MPA 150km/h MPA 300km/h
tion. The proposed CDID detector is also capable
MAP 150km/h MAP 300km/h
10-1 of exploiting the time domain channel sparsity and
CDID 150km/h CDID 300km/h
OFDM MMSE 150km/h OFDM MMSE 300km/h
the DD domain symbol constellation constraints,
which can achieve a close-to-optimal performance
10-2 at a much reduced computational complexity, cf.
Fig. 5. For all these detectors, the OTFS perfor-
mance remains similar upon increasing the velocity
BER

10-3 from 150 km/h to 300 km/h. In contrast, the detec-

tion performance of the MMSE OFDM detector
remains poor due to the excessive ICI.
10-4
coded otFs system
While OTFS has the potential of attaining the max-
OTFS imum achievable diversity gain, the channel codes
10-5
have to be carefully designed for OTFS modula-
tion. Moreover, perfect detection at near-capacity
signal-to-noise ratios (SNRs) may not be attained
10-6 for practical OTFS systems due to the associat-
10 12 14 16 18 20
ed poor channel conditions. Hence, the channel
Eb/N0 (dB) decoder has to cope with the OTFS detector’s
residual errors, which would require iterative
FIGURE 5. Performance comparison of OTFS with different detectors, moving OTFS receivers. However, how to design such a
speeds, and waveforms. The number of paths in the DD domain is 4, N = 8, receiver and how to choose the coding parame-
M = 16, nmax = 3/(NT) for the velocity of 150 km/h, nmax = 6/(NT) for the ters for near-capacity joint detection and decoding
velocity of 300 km/h, and tmax = 3/(MDf). remains an interesting open issue. Recent research
has unveiled a fundamental trade-off between the
training overhead while it can achieve an order of diversity gain and coding gain, which may shed
magnitude reduction in channel estimation error light on the code design of OTFS systems [15].
than that of its conventional counterpart. In particular, the diversity gain of OTFS systems
improves with the number of independent resolv-
eFFIcIent dd domAIn dAtA detectIon able channel paths in the DD domain, while the
The output signal in the DD domain can be coding gain declines, cf. Fig. 6. This is not unexpect-
regarded as a 2D circular convolution of the input ed as the transmitted signal energy is distributed
data symbols and the effective aggregate chan- to multiple paths, as discussed in [15]. Moreover,
nel, cf. Fig 3, which results in a rather specific both the coded and uncoded OTFS systems out-
interference pattern, where a pair of symbols far perform the corresponding OFDM systems, which
from each other in the DD domain may interfere clearly shows the advantage of OTFS modulation.
with each other. Mitigating this peculiar interfer-
ence requires a bespoke receiver. Adopting the outlook For otFs
optimal maximum a posteriori (MAP) detector OTFS has great potential in providing reliable
would indeed perfectly mitigate the interference communications for high-mobility applications
between symbols, but at an excessive complexity, and its open facets are expected to stimulate new
precluding its deployment in practical systems. research, as discussed below.
Hence, most OTFS detectors are focused on
complexity reduction, based on the classic mes- Fdd or tdd?
sage passing algorithm (MPA) and its variants. The As demonstrated by Cohere [2], the DD domain
main problem of MPA-based detection is its poor channels are capable of reaching a coherence
convergence behavior in the face of short cycles, time of 100 ms and a coherence bandwidth of
which may lead to performance degradation. 100 MHz. Hence, accurate channel reciprocity
A potent solution is to adopt the variational holds in the DD domain and thus time-division
framework of [14], which can adaptively construct duplexing (TDD) is a good option for OTFS.
the distributions of OTFS symbols according to their Besides, due to the significantly increased coher-
interference patterns. By appropriately constructing ence bandwidth of the DD domain channel, we
the distributions of OTFS symbols for variational can potentially infer the downlink channel from
purposes, we can design rapidly converging OTFS the uplink channel even for frequency-division
detection. Owing to its better convergence, the duplexing (FDD) OTFS systems, which significant-
variational Bayes (VB) OTFS detector can achieve a ly reduces the channel feedback overhead. How-
modest performance gain over the MPA detector, ever, the key is to find an accurate deterministic
cf. Fig. 5. The performance of MAP detection is or statistical mapping between the uplink and
also provided as the baseline, which has the best downlink channels. Furthermore, it is also import-
performance, albeit at the cost of excessive com- ant to conduct a systematic comparative study of
plexity. Another potential solution is the so-called TDD and FDD-aided OTFS systems.
cross-domain iterative detection (CDID), where
a conventional linear minimum mean squared scAlAble multIple Access schemes For otFs
error (L-MMSE) estimator is adopted for equal- How to support a multiplicity of users in
ization in the time domain and low-complexity high-mobility environments is a very challenging
symbol-by-symbol detection is utilized in the DD issue. OTFS offers the opportunity to accommo-

142 IEEE Wireless Communications • August 2021

HANZO_LAYOUT.indd 142 8/31/21 12:48 AM

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on November 01,2022 at 10:50:07 UTC from IEEE Xplore. Restrictions apply.
 date multiple users in the DD domain, where
employing carefully designed user scheduling 100
and guard spaces has the potential of avoid-
ing multi-user interference. However, how to
scale the systems for accommodating a large
number of users without a significant over- 10-1
head is an interesting open research problem.
The coexistence of promising multiple access
schemes and OTFS, such as non-orthogonal
multiple access, spatial-division multiple access,

FER
and interleave-division multiple access, is worth 10-2
exploring further.

Integrated Sensing and Communications Using OTFS Uncoded OTFS, Two Paths
Since the DD domain channel directly exploits the -3 Coded OTFS, Two paths
physics of propagation, relying on the distance, 10 Uncoded OTFS, Four paths
speed and scattering intensity, OTFS is eminently Coded OTFS, Four paths Coding gain
suitable for integrating sensing and communica- Uncoded OFDM, Four paths
tions solutions in a single platform. Efficient sens-
Coded OFDM, Four paths
ing algorithms to exploit the OTFS signal structure
10-4
are still unknown. Finding the optimal trade-off 0 5 10 15 20
between the sensing and communication perfor-
Eb /N0 (dB)
mances remains an interesting open question.
Moreover, as location and velocity can serve as
FIGURE 6. Performance comparison of coded/uncoded OTFS and OFDM
beneficial side information for improving commu-
modulations, N = 8, M = 16, nmax = 3/(NT), and tmax = 5/MDf.
nication performance, sensing-based communica-
tions relying on OTFS is an exciting open topic to
investigate. Conclusions
MIMO-OTFS In a nutshell, OTFS constitutes a promising
Applying OTFS in multi-antenna systems provides next-generation candidate. We commenced with
additional hitherto unexploited spatial DoF for an overview of the fundamental concept of OTFS,
multiplexing. In contrast to TF domain channels, including the main features of the DD domain
which may fluctuate dynamically for different channel, the DD domain multiplexing and OTFS
antennas in different time slots and subcarriers, transceiver architecture. The critical challenges of
the DD domain channels tend to remain quasi-sta- OTFS, such as channel estimation, efficient data
tionary both in the time and antenna domains, detection and coding/decoding problems were
which may result in an efficient channel estima- highlighted and pertinent preliminary results were
tion and multi-input multiple-input multiple-output provided. The potential applications of OTFS and
(MIMO) detection. How to design sophisticated several promising research directions were intro-
beamforming/precoding to fully exploit all the duced. It is hoped that this article will help inspire
available spatial DoFs and how to perform future research in this exciting new area and pave
low-complexity detection for MIMO-OTFS consti- the way for designing next-generation networks.
tute intriguing problems. Moreover, the analytical
framework of MIMO-OTFS system performance Acknowledgments
versus the number of antennas is also unexplored This work was supported in part by the Australian
in the open literature. Research Council (ARC) Discovery Projects under
Grant DP190101363, and in part by the Linkage
Index Modulation for OTFS Project under Grant LP170101196. L. Hanzo
OTFS modulation maps the classic modulated would like to acknowledge the financial support
symbols to the DD domain and spreads them of the European Research Council’s Advanced
across the whole TF domain grid for transmis- Fellow Grant QuantCom (Grant No. 789028).
sion. Therefore, IM can be carried out in the The authors would like to thank the support from
TF domain for improving the spectral efficiency, Telstra Corporation Ltd., particularly Dr. Paul G.
while slightly sacrificing the transmission diversity Fitzpatrick, Dr. Taka Sakurai, and Mr. Paul Spor-
order. In particular, the additional information ton for valuable discussions during this work.
bits can be mapped to the ON/OFF states of the
TF grid indices. Additionally, we can also extend References
[1] C. Xu et al., “Adaptive Coherent/Non-Coherent Spatial Mod-
this to the spatial dimension, such as antenna ulation Aided Unmanned Aircraft Systems,” IEEE Wireless
indices or propagation paths, for incorporating Commun., vol. 26, no. 4, 2019, pp. 170–77.
IM into OTFS systems, where the information [2] R. Hadani and A. Monk, “OTFS: A New Generation of Mod-
bits are transmitted by active antenna or path ulation Addressing the Challenges of 5G,” arXiv preprint
arXiv:1802.02623, 2018.
indices. The intriguing error rate performance [3] M. Ramachandran, G. Surabhi, and A. Chockalingam,
vs. complexity and spectral efficiency trade-offs “OTFS: A New Modulation Scheme for High-Mobility Use
of IM-OTFS systems remain to be investigated. Cases,” J. Indian Institute of Science, vol. 100, no. 2, 2020,
Furthermore, the design of efficient detectors for pp. 315–36.
[4] F. Hlawatsch and G. Matz, Wireless Communications over
jointly demodulating the additional information Rapidly Time-varying Channels, Academic Press, 2011.
bits and conventional constellation symbols of [5] W. Shen et al., “Channel Estimation for Orthogonal Time
IM-OTFS systems is also a critical open research Frequency Space (OTFS) Massive MIMO,” IEEE Trans. Signal
challenge. Process., vol. 67, no. 16, 2019, pp. 4204–17.

IEEE Wireless Communications • August 2021 143

HANZO_LAYOUT.indd 143 8/31/21 12:48 AM

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on November 01,2022 at 10:50:07 UTC from IEEE Xplore. Restrictions apply.
 [6] G. D. Surabhi, R. M. Augustine, and A. Chockalingam, “On Australia, in 2019. He is currently an assistant professor with the
the Diversity of Uncoded OTFS Modulation in Doubly-Dis- Department of Electrical and Electronic Engineering, the South-
persive Channels,” IEEE Trans. Wireless Commun., vol. 18, ern University of Science and Technology.
no. 6, June 2019, pp. 3049–63.
[7] P. Raviteja et al., “Effective Diversity of OTFS Modulation,” Shuangyang Li [S’18] ([email protected]) received
IEEE Commun. Lett., vol. 9, no. 2, 2020, pp. 249–53. the B.S. and M.S. degrees from Xidian University, China, in
[8] S. K. Mohammed, “Derivation of OTFS Modulation from 2013 and 2016, respectively. He is currently pursuing the Ph.D.
First Principles,” IEEE Trans. Veh. Technol., early access, 2021. degree at Xidian University and the University of New South
[9] A. J. Janssen, “The Zak Transform: A Signal Transform for Wales. His research interests include signal processing, channel
Sampled Time-continuous Signals,” Philips J. of Research, vol. coding and their applications to communication systems.
43, no. 1, 1988, pp. 23–69.
[10] T. Zemen et al., “Iterative Detection for Orthogonal Pre- Jinhong Yuan [F’16] ([email protected]) received the B.E.
coding in Doubly Selective Channels,” Proc. IEEE Personal, and Ph.D. degrees from the Beijing Institute of Technology, Bei-
Indoor and Mobile Radio Commun. Sympos., Sept. 2018, jing, China, in 1991 and 1997, respectively. In 2000, he joined
pp. 1–7. the School of Electrical Engineering and Telecommunications,
[11] P. Raviteja, K. T. Phan, and Y. Hong, “Embedded Pilot-Aid- University of New South Wales, Sydney, Australia, where he
ed Channel Estimation for OTFS in Delay-Doppler Chan- is currently a professor and Head of the Telecommunication
nels,” IEEE Trans. Veh. Technol., vol. 68, no. 5, May 2019, Group with the School.
pp. 4906–17.
[12] P. Raviteja et al., “Interference Cancellation and Iterative G anesh B haratula ([email protected])
Detection for Orthogonal Time Frequency Space Modula- received the master’s degree in microwave communications and
tion,” IEEE Trans. Wireless Commun., vol. 17, no. 10, Oct. electronics and another master’s degree in solid state physics
2018, pp. 6501–15. from Delhi University, India. He has been with Telstra Corpora-
[13] Z. Wei et al., “Transmitter and Receiver Window Designs tion Ltd., since 1984, starting first with IT Research Laboratories
for Orthogonal Time-Frequency Space Modulation,” IEEE and then joining its CTO Department and Wireless Engineering.
Trans. Commun., vol. 69, no. 4, 2021, pp. 2207–23. He is currently a networks principal with the Wireless Access
[14] W. Yuan et al., “A Simple Variational Bayes Detector for Group and looks after assessment and development of emerg-
Orthogonal Time Frequency Space (OTFS) Modulation,” ing wireless technologies.
IEEE Trans. Veh. Technol., vol. 69, no. 7, 2020, pp. 7976–80.
[15] S. Li et al., “Performance Analysis of Coded OTFS Systems Ronny Hadani ([email protected]) obtained a master
and Code Design,” IEEE Trans. Wireless Commun., early degree in applied mathematics from the Weizmann Institute,
access, 2021. Rehovot, Israel, in 2000. He obtained a Ph.D. in mathematics
from Tel Aviv-University, Israel, in 2006. He had a post-doc
Biographies position at the University of Chicago during 2006-2009, and
Zhiqiang Wei [M’19] ([email protected]) received the B.E. joined the faculty of mathematics at the University of Texas–
degree from Northwestern Polytechnical University (NPU), Austin in August 2009, where he is currently an associate pro-
Xi’an, China, in 2012, and the Ph.D. degree from the University fessor.
of New South Wales, Sydney, Australia, in 2019. He is currently
a Humboldt Postdoctoral Research Fellow with the Friedrich-Al- Lajos Hanzo [F’04] ([email protected]) received his degree
exander University Erlangen-Nuremberg. He received the Best in electronics in 1976 and his doctorate in 1983. He holds an
Paper Award from the IEEE International Conference on Com- honorary doctorate from the Technical University of Budapest
munications (ICC) in 2018. (2009) and from the University of Edinburgh (2015). He is a
member of the Hungarian Academy of Sciences and a former
Weijie Yuan [M’19] ([email protected]) received the Ph.D. Editor-in-Chief of IEEE Press. He is a Governor of both IEEE Com-
degree from the Beijing Institute of Technology, China, and Soc and of VTS, and has published 1900+ contributions at IEEE
the Ph.D. degree from the University of Technology Sydney, Xplore.

144 IEEE Wireless Communications • August 2021

HANZO_LAYOUT.indd 144 8/31/21 12:48 AM

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on November 01,2022 at 10:50:07 UTC from IEEE Xplore. Restrictions apply.