0% found this document useful (0 votes)
42 views6 pages

Orthogonal Time Frequency Space Modulation

Uploaded by

Belgacem
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
42 views6 pages

Orthogonal Time Frequency Space Modulation

Uploaded by

Belgacem
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

Orthogonal Time Frequency Space Modulation

R. Hadani⋄∗ , S. Rakib∗ , M. Tsatsanis∗ , A. Monk∗ , A. J. Goldsmith†∗ , A. F. Molisch‡∗ , and R. Calderbank♯∗

Abstract—A new two-dimensional modulation technique called time-varying multipath channel. OTFS can be implemented as
Orthogonal Time Frequency Space (OTFS) modulation designed a pre- and post-processing block to filtered OFDM systems,
in the delay-Doppler domain is introduced. Through this design, thus enabling architectural compatibility with LTE.
which exploits full diversity over time and frequency, OTFS cou-
pled with equalization converts the fading, time-varying wireless The main premise behind OTFS is to transform the time-
channel experienced by modulated signals such as OFDM into varying multipath channel into a two-dimensional channel
a time-independent channel with a complex channel gain that is in the delay-Doppler domain. Through this transformation,
roughly constant for all symbols. Thus, transmitter adaptation is coupled with equalization in this domain, all symbols over
not needed. This extraction of the full channel diversity allows a transmission frame experience the same channel gain. This
OTFS to greatly simplify system operation and significantly
improves performance, particular in systems with high Doppler, constant gain results from spreading information across the en-
short packets, and large antenna arrays. Simulation results tire delay-Doppler plane through the OTFS transform, whose
indicate at least several dB of block error rate performance full diversity can then be extracted by an efficient equalizer
improvement for OTFS over OFDM in all of these settings. In structure. While full diversity could also be extracted in
addition these results show that even at very high Dopplers (500 the time-frequency domain through an appropriately-designed
km/h), OTFS approaches channel capacity through linear scaling
of throughput with the MIMO order, whereas the performance of equalizer, the sparsity and lower variability of the channel
OFDM under typical design parameters breaks down completely. in the delay-Doppler domain makes our approach far more
practical and robust. The relatively constant channel gain over
all symbol transmissions obtained with our system greatly
I. I NTRODUCTION reduces the overhead and complexity associated with physical
The 5G air interface and associated waveform have to layer adaptation. It also presents the transport and application
support a number of diverse requirements and usage scenarios, layer with a robust fixed-rate channel, which is highly desir-
including terminal speeds of up to 300 km/h for vehicle-to- able when running over TCP/IP and for the delay-sensitive
vehicle and vehicle-to-infrastructure systems, and up to 500 applications envisioned for 5G. Moreover, full diversity en-
km/h for high-speed train applications. The 4G modulation of ables linear scaling of throughput with the number of antennas,
Orthogonal Frequency Division Multiplexing (OFDM) breaks regardless of channel Doppler. In addition to OTFS’s full diver-
down in this setting as channel estimation is no longer effec- sity benefits, since the delay-Doppler channel representation
tive, nor is the associated time-frequency adaptation to these is very compact, OTFS enables dense and flexible packing
channel estimates that makes OFDM near-capacity-achieving of reference signals, a key requirement to support the large
in lower-Doppler environments. These requirements motivate antenna arrays used in massive MIMO applications.
the need to examine new waveforms that can meet the desired The delay-Doppler channel model upon which OTFS is
performance requirements in the new 5G usage scenarios. based was originally derived in Bello’s landmark paper [2].
This paper proposes Orthogonal Time Frequency Space Several prior works have investigated optimal signaling for
(OTFS) modulation, a new modulation scheme whereby each Bello’s channel model to fully capture diversity across both
transmitted symbol experiences a near-constant channel gain the time and frequency domains. In particular, [3] established
even in channels with high Doppler, large antenna arrays (mas- a signal model that presents the received signal as a canonical
sive MIMO), or at high frequencies such as millimeter waves. decomposition into delay and Doppler shifted versions of a
OTFS is a generalization of both CDMA and OFDM and basis signal, and suggests a delay-Doppler RAKE receiver
hence inherits their most compelling attributes. In particular, that exploits the dispersion in both dimensions. In [4] a gen-
OTFS modulates each information symbol onto one of a set eral framework for communication over a doubly-dispersive
of two-dimensional (2D) orthogonal basis functions that span fading channel based on an orthogonal short-time Fourier
the bandwidth and time duration of the transmission burst or basis consisting of time-frequency shifts of a given pulse was
packet. OTFS reduces to CDMA when these basis functions developed. A different approach was taken in [5] which first
are used as (one-dimensional) spreading codes, and to OFDM demonstrates that a time-frequency RAKE receiver does not
when they are the subchannel carriers. However, in contrast obtain full diversity as it is not optimized on the transmit side.
to CDMA and OFDM, the modulation basis function set of The work then derives optimal precoders which, in conjunction
OTFS is specifically derived to combat the dynamics of the with a jointly designed RAKE receiver, obtain the full diversity
of the doubly dispersive channel. These works all differ from
∗ Cohere Technologies, Inc., Santa Clara, CA 95051 USA.
⋄ University of Texas, Austin, TX 78712
OTFS in that their modulation design is in the time-frequency
† Stanford University, Stanford, CA 94305
domain rather than the delay-Doppler domain.
‡ University of Southern California, Los Angeles, CA 90089 A separate body of prior work focuses on time-frequency
♯ Duke University, Durham, NC 27708 pulse shape design to minimize dispersion after transmission

Authorized licensed use limited to: UNIVERSITE DE GABES. Downloaded on May 16,2024 at 12:25:09 UTC from IEEE Xplore. Restrictions apply.
978-1-5090-4183-1/17/$31.00 ©2017 IEEE
through the channel. A special case of this body of work is
pulse-shaped OFDM. Various criteria for pulse shape opti-
mization have been considered in earlier works including sup-
pressing ISI and ICI [6], maximizing signal-to-interference-
plus-noise ratio (SINR) [7], or optimizing spectral efficiency
through the use of non-rectangular lattices in the time-
frequency domain [8]. These works all differ from OTFS in
that they attempt to mitigate or fully remove the ISI and Fig. 1. OTFS Modulation Block Diagram: Transmitter and Receiver
ICI through pulse shape design in the time domain. OTFS is
designed so that its information symbols experience minimal
cross-interference as well as full diversity in the delay-Doppler operates on the input s(t) to produce the output r(t):
domain through appropriate design of the modulation lattice Π
Πh (s) : s(t) −−→
h
r(t). (2)
and pulse shape design in that domain.
Having differentiated OTFS from prior work, the remainder In the mathematics literature, the representation of the rela-
of the paper will describe OTFS modulation design and tionship in (1) as an operator Πh parameterized by a function
performance. In Section II we describe the wireless channel h(τ, ν) and operating on a function s(t) as defined in (2) is
in terms of its delay-Doppler characteristics, for which OTFS called a Heisenberg transform [10]. As we will see below,
is designed. Section III develops the details of OTFS as OTFS modulation also utilizes a Heisenberg transform on
a modulation that matches wireless channel characteristics the transmitted symbols, hence the received signal becomes
through two processing steps; the transmitter first maps the a cascade of two Heisenberg transforms, one corresponding
two-dimensional delay-Doppler domain, where the informa- to the OTFS modulation, and the other corresponding to the
tion symbols reside, to the time-frequency domain through channel. As will be discussed in more detail in the next section,
a combination of the inverse symplectic Fourier transform applying an equalizer in the delay-Doppler domain to the
and windowing. It then applies the Heisenberg transform, a corresponding structure of the received signal extracts the full
generalization of the OFDM transform, to the time-frequency diversity of the channel, resulting in a near-constant gain on
modulated signal to convert it into the time domain for trans- each of the transmitted symbols as well as a particularly simple
mission. The receiver performs the reverse operations. We also mechanism to recover these symbols.
show in this section that the modulated symbols in OTFS, after
transmission through the channel, exhibit a constant channel III. OTFS M ODULATION
gain over each frame. The full time and frequency diversity of
OTFS modulation is comprised of a cascade of two two-
the channel can then be extracted by an equalizer. Performance
dimensional transforms at both the transmitter and the receiver,
results for OTFS coupled with equalization demonstrating its
as shown in Fig. 1. The transmitter first maps the informa-
advantages over OFDM in high Doppler channels, with short
tion symbols x[n, m] residing in the two-dimensional delay-
packets, and with MIMO arrays are presented in Section IV.
Doppler domain to symbols X[n, m] in the time-frequency
The paper concludes in Section V.
domain through a combination of the inverse symplectic
Fourier transform and windowing. We call this cascade of
II. T HE D ELAY-D OPPLER C HANNEL operations the OTFS transform. Next the Heisenberg transform
is applied to X[n, m] to convert the time-frequency modulated
The complex baseband channel impulse response h(τ, ν)
signal to the time domain signal s(t) for transmission over the
characterizes the channel response to an impulse with delay τ
channel. The reverse operations are performed in the receiver,
and Doppler ν [1]. The received signal due to an input signal
mapping the received time signal r(t) first to the time-
s(t) transmitted over this channel is given by
∫∫ frequency domain through the Wigner transform (the inverse
of the Heisenberg transform), and then to the delay-Doppler
r(t) = h(τ, ν)s(t − τ )ej2πν(t−τ ) dν dτ . (1)
domain for symbol demodulation. In the next subsections
we describe the components and general properties of time-
An important feature of the h(τ, ν) representation is its
frequency modulation, the specific time-frequency modulation
compactness: since typically there are only a small number
of OTFS, and the impact of the channel on the transmitted
of channel reflectors with associated Dopplers, far fewer
OTFS symbols.
parameters are needed for channel estimation in the delay-
Doppler domain than in the time-frequency domain. This
sparse representation for typical channel models, including A. Time-Frequency Modulation
those in LTE, has important implications for channel estima- We begin this section with a generic description of time-
tion, equalization and tracking. A more in-depth discussion of frequency modulation, which has the following components:
the channel sparsity in the delay-Doppler domain for typical • A lattice or grid Λ in the time-frequency domain that is
channels as well as the implications of this sparsity in receiver a sampling of the time and frequency axes at intervals T
processing is provided in [9]. and ∆f respectively:
Note that (1) can also be interpreted as a linear operator
Πh (·), parameterized by the impulse response h(τ, ν), that Λ = {(nT, m∆f ), n, m ∈ Z}. (3)

Authorized licensed use limited to: UNIVERSITE DE GABES. Downloaded on May 16,2024 at 12:25:09 UTC from IEEE Xplore. Restrictions apply.
• A packet burst with total duration N T seconds and total the discrete twisted convolution of X[n, m] and h(τ, ν):
bandwidth M ∆f Hz.
f (τ, ν) = h(τ, ν) ∗σ X[n, m] =
• A set of modulated symbols X[n, m], n = 0, . . . , N − 1,
M/2−1 N −1
m = 0, . . . , M − 1 that we wish to transmit over a given ∑ ∑
packet burst. X[n, m]h(τ − nT, ν − m∆f )ej2π(ν−m∆f )nT .
m=−M/2 n=0
• A transmit pulse gtx (t) (and associated receive pulse
grx (t)) whose inner product is bi-orthogonal with respect (10)
to translations by time T and frequency ∆f , i.e. With this result established we are now ready to examine the

∗ receiver processing steps.
gtx (t)grx (t − nT )ej2πm∆f (t−nT ) dt = δ(m)δ(n).
(4)
B. Reception of Time-Frequency Modulated Signals: Sufficient
Note that the bi-orthogonal property of the pulse shapes Statistics and Channel Distortion
described by (4) is what eliminates cross-symbol interference
The sufficient statistic for symbol detection based on the re-
in symbol reception, as will be shown in the next subsection.
ceived signal is obtained by matched filtering with the channel-
A time-frequency modulator with these components maps
distorted, information-carrying pulses (assuming that the ad-
the two-dimensional symbols X[n, m] on the lattice Λ to a
ditive channel noise is white and Gaussian). The matched
transmitted waveform s(t) via a superposition of delay-and-
filter first requires computation of the cross-ambiguity function
modulate operations on the pulse waveform gtx (t), namely
Agrx ,r (τ, ν), defined as follows.
M/2−1 N −1
∑ ∑ ∫
s(t) = X[n, m]gtx (t−nT )ej2πm∆f (t−nT ) . (5) Agrx ,r (τ, ν) , gr∗ (t − τ )r(t)e−j2πν(t−τ ) dt. (11)
m=−M/2 n=0
This function, when sampled on the lattice Λ, i.e., at τ = nT
We call the modulation (5) the Heisenberg transform of and at ν = m∆f , yields the matched filter output
X[n, m], which is shown in [9] to be a generalization of
the OFDM transform mapping modulated symbols in the Y [n, m] = Agrx ,r (τ, ν)|τ =nT,ν=m∆f . (12)
frequency domain (i.e. on each subcarrier) to the transmitted
The operation of (12) is called the Wigner transform. This
signal in the time domain. Similar to interpreting the channel
transform inverts the Heisenberg transform of (5) and, as
operation (1) as a Heisenberg operator (2) applied to the
shown in [9], is a generalization of the inverse OFDM
transmitted signal s(t), the modulation of (5) can also be
transform mapping the received OFDM signal to modulated
interpreted as a Heisenberg operator ΠX (·) with parameters
symbols.
X[n, m] that is applied to the pulse shape gtx (t) as
We now establish the relationship between the matched filter
s(t) = ΠX (gtx ). (6) output Y [n, m] and the transmitter input X[n, m]. We have
already established in (9) that the input to the matched filter
This interpretation is useful when we consider the received sig- r(t) can be expressed with respect to a Heisenberg operator
nal as a cascade of two Heisenberg operators, one associated Πf (gtx (t)) parameterized by the impulse response f (τ, ν) and
with the modulator and one associated with the channel, given operating on the pulse shape grx (t) (plus noise). The output
the following property of such a cascade, which is shown in of the matched filter then has two contributions:
Proposition 1 of [9].
Given a pair of Heisenberg operators parameterized by h1 Y (τ, ν) = Agrx ,Πf (gtx ) (τ, ν) + Agrx ,v (τ, ν). (13)
and h2 and applied in cascade to a waveform g(t), direct
The last term on the right side is the contribution of noise,
computation yields that
which we will denote as V (τ, ν) = Agrx ,v (τ, ν), while the first
Πh2 (Πh1 (g(t))) = Πh (g(t)), (7) term is the output of the matched filter corresponding to the
input in the absence of noise. Direct computation reveals that
where h(τ, ν) = h2 (τ, ν)∗σ h1 (τ, ν) is the twisted convolution this term can be expressed as the twisted convolution of the
of h1 (τ, ν) and h2 (τ, ν), defined as two-dimensional impulse response f (τ, ν) with the function
Agrx ,gtx (τ, ν):
h2 (τ, ν) ∗σ h1 (τ, ν) =
∫∫
′ ′ Agrx ,Πf (gtx ) (τ, ν) = f (τ, ν) ∗σ Agrx ,gtx (τ, ν). (14)
h2 (τ ′ , ν ′ )h1 (τ − τ ′ , ν − ν ′ )ej2πν (τ −τ ) dτ ′ dν ′ . (8)
Substituting (10) into (14) and using this expression for the
Applying the above result to the cascade of the modulation and first term in (13) yields the end-to-end channel description as
channel Heisenberg operators (6) and (1) yields the received follows:
signal
∫∫ Y (τ, ν) = h(τ, ν)∗σ X[n, m]∗σ Agr ,gtr (τ, ν)+V (τ, ν). (15)
r(t) = f (τ, ν)gtx (t − τ )ej2πν(t−τ ) dνdτ + v(t), (9) Evaluating Eq. (15) on the lattice Λ we obtain the matched
filter output estimate of the modulation symbols
where v(t) is additive noise at the receiver input, and f (τ, ν)
is the impulse response of the combined transform given by X̂[m, n] = Y [m, n] = Y (τ, ν)|τ =nT,ν=m∆f . (16)

Authorized licensed use limited to: UNIVERSITE DE GABES. Downloaded on May 16,2024 at 12:25:09 UTC from IEEE Xplore. Restrictions apply.
It is shown in [9] that in the case of an ideal channel, i.e. where ~ denotes two-dimensional circular convolution. This is
h(τ, ν) = δ(τ )δ(ν), this reduces to Y [n, m] = X[n, m] + similar to the convolution property of the conventional discrete
V [n, m] where V [n, m] is the additive white noise. Thus Fourier transform. We are now ready to define OTFS as a time-
the matched filter output perfectly recovers the transmitted frequency modulation with an additional pre-processing step.
symbols (plus noise) under ideal channel conditions. OTFS modulation: Consider a set of QAM information
Let us now consider the channel output for more general symbols arranged on a 2D grid x[k, l], k = 0, . . . , N − 1, l =
channels. In this case, it is shown in the second theorem of 0, . . . , M −1 that we wish to transmit. Further, assume a time-
[9] that if h(τ, ν) has finite support bounded by (τmax , νmax ) frequency modulation system defined by the lattice, packet
and if Agrx ,gtx (τ, ν) = 0 for τ ∈ (nT − τmax , nT + τmax ), ν ∈ burst, and bi-orthogonal transmit and receive pulses described
(m∆f −νmax , m∆f +νmax ), then Y [n, m] = H[n, m]X[n, m] in Section III-A. In addition to these components, OTFS
for incorporates a transmit windowing square summable function
∫∫
Wtx [n, m] that multiplies the modulation symbols in the time-
H[n, m] = h(τ, ν)ej2πνnT e−j2π(ν+m∆f )τ dνdτ. (17)
frequency domain. Given the above components, we define the
Note that for Y [n, m] = H[n, m]X[n, m] there is no cross- modulated symbols in OTFS as follows:
symbol interference affecting X[n, m] in either time n or
X[n, m] = Wtx [n, m] SFFT−1 (x[k, l]). (21)
frequency m. Hence, the received symbol X[n, m] is the
same as the transmitted symbol except for a complex scale The transmitted signal s(t) = ΠX (gtx (t)) is obtained from the
factor H[n, m], similar to OFDM transmitted through time- Heisenberg transform defined in (6). We call (21) the OTFS
invariant frequency-selective fading channels. Note that if the transform, which combines an inverse symplectic transform
ambiguity function is only approximately bi-orthogonal in the with a windowing operation. The second equation describes
neighborhood of (T, ∆f ) then there is some minimal cross- the Heisenberg transform of gtx (t) parameterized by the sym-
symbol interference. The bi-orthogonality and, in its absence, bols X[n, m] into the transmitted signal s(t). The composition
residual cross-symbol interference is a function of the transmit of these two transforms comprises OTFS modulation, as was
and receive pulses gtx and grx . The fading H[n, m] in (17) shown in the two transmitter blocks of Fig. 1.
that each symbol X[n, m] experiences has a complicated A different basis function representation useful in the OTFS
expression as a weighted superposition of exponentials. In demodulation process we discuss below is as follows:
the next section we will describe the specific transforms

N −1 M
∑ −1
associated with OTFS as a time-frequency modulation, and 1
how these transforms result in constant channel gains for each X[n, m] = Wtr [n, m] x[k, l]bk,l [n, m]
MN (22)
k=0 l=0
information symbol.
b [n, m] = e−j2π( M − N ) .
ml nk
k,l
C. OTFS Modulation and Demodulation
The interpretation of (22) is that each information symbol
With the background of the previous two sections, we can x[k, l] is modulated by a 2D basis function bk,l [n, m] in the
now define the specific transforms associated with OTFS mod- time-frequency domain.
ulation and demodulation that result in a near-constant channel
OTFS Demodulation: Let the receiver employ a receive
gain across symbols. The transforms utilize a variant of the
windowing square summable function Wrx [n, m]. Then the
standard Fourier transform called the Symplectic Finite Fourier
demodulation operation consists of the following steps:
Transform (SFFT). This transform is defined as follows. Let
Xp [n, m] denote the periodized version of X[n, m] with period 1) Take the Wigner transform of the received signal, which
(N, M ). The SFFT transform of Xp [n, m] is then defined as yields
xp (k, l) = SFFT(Xp [n, m]) for Y [n, m] = Agrx ,r (τ, ν)|τ =nT,ν=m∆f . (23)
−1 −1
M

N ∑
2

Xp [n, m]e−j2π( N − M ) . 2) Apply the window function Wrx [n, m] to Y [n, m] to


nk ml
xp [k, l] = (18)
n=0 m=− M
obtain the time-frequency function YW [n, m] and then
2
periodize the result to obtain the periodic (N, M ) signal
The inverse transform is Xp [n, m] = SFFT−1 (x[k, l]) for Yp [n, m]:
1 ∑
x[k, l]ej2π( N − M ) ,
nk ml
Xp [n, m] = (19) YW [n, m] = Wrx [n, m] Y [n, m],
MN ∞
l,k ∑ (24)
where l = 0, . . . , M − 1, k = 0, . . . , N − 1. If the support Yp [n, m] = YW [n − kN, m − lM ].
of X[n, m] is time-frequency limited to Z0 = {(n, m); 0 ≤ k,l=−∞

m ≤ M − 1, 0 ≤ n ≤ N − 1} then Xp [n, m] = X[n, m] 3) Apply the symplectic Fourier transform on the periodic
for (n, m) ∈ Z0 and the inverse transform (19) recovers the sequence Yp [n, m]:
original signal X[n, m]. It is shown in [9] that for X1 [n, m]
and X2 [n, m] periodic 2D sequences with period (M, N ), x̂[l, k] = y[l, k] = SFFT(Yp [n, m]). (25)
SFFT(X1 [n, m] ~ X2 [n, m]) = The last step can be interpreted as a projection of the time-
SFFT(X1 [n, m]) · SFFT(X2 [n, m]), (20) frequency modulation symbols onto the two-dimensional or-

Authorized licensed use limited to: UNIVERSITE DE GABES. Downloaded on May 16,2024 at 12:25:09 UTC from IEEE Xplore. Restrictions apply.
thogonal basis functions bk,l (n, m) as follows: the OTFS transform pre- and post-processing blocks at the
−1 N −1
transmitter and receiver respectively. We simulate the wireless

M ∑
x̂[l, k] = Yp (n, m)b∗k,l (n, m), fading channel according to the TDL-C channel model (delay
m=0 n=0 (26) spread of 300 ns, Rural Macro, and low correlation MIMO),
−j2π ( lm
L − K )
one of the standardized channel models in 3GPP. The details of
b∗k,l (n, m)
kn
=e .
the simulation parameters used are summarized in Figure 2. In
It is shown in [9] that the estimated sequence x̂[k, l] of order to focus on the intrinsic performance of each modulation,
information symbols obtained after demodulation is given by we simulate near-optimal receivers for both systems; a sphere
the two-dimensional periodic convolution of the input QAM decoder (reduced-complexity maximum likelihood) receiver
sequence x[n, m] and a sampled version of the windowed for OFDM and a turbo equalizer in the OTFS receiver. We
impulse response hw (·): assume ideal channel estimation (i.e. that H[n, m] is known)
M −1 N −1 ( ) at the receiver for both systems as the issues of channel
1 ∑ ∑ k−n l−m estimation are outside the scope of this paper. We simulate
x̂[k, l] = x[n, m]hw , , (27)
M N m=0 n=0 N T M ∆f high speed mobility scenarios, hence we assume only long
term average channel state information at the transmitter and
where focus on Block Error Rate (BLER) comparisons.
( )
k−n l−m
hw , = hw (ν ′ , τ ′ )|ν ′ = k−n ,τ ′ = l−m (28)
N T M ∆f NT M ∆f

for hw (ν ′ , τ ′ ) the circular convolution of the channel response


with a windowing function1 :
∫∫
hw (ν ′ , τ ′ ) = h(τ, ν) w(ν ′ − ν, τ ′ − τ )e−j2πντ dτ dν.
(29)
In (29) the windowing function w(τ, ν) is the symplectic dis-
crete Fourier transform (SDFT) of the time-frequency window
W [n, m], defined as

M −1 N
∑ −1 Fig. 2. Simulation Parameters for Performance Results
w(τ, ν) = W [n, m]e−j2π(νnT −τ m∆f ) , (30)
m=0 n=0 Figure 3 shows the BLER performance for a 1x1 system
for W [n, m] = Wtx [n, m] Wrx [n, m] the product of the trans- with mobile speeds of 120 km/h (444 Hz Doppler spread)
mit and receive window. Note that as the window W [n, m] with a modulation of 16 QAM rate and 64 QAM rate (solid
increases its support over time and frequency, hw (·, ·) more and dashed lines respectively). Notice that OTFS shows gains
closely approximates the channel impulse response h(·, ·). in the rage of 2-3 dB in this figure, confirming its robustness
From (27) we see that over a given frame, each demodulated to Doppler effects.
symbol x̂[l, k] for a given l and k experiences the same channel
gain hw (0, 0) on the transmitted symbol x[l, k]. Moreover,
cross-symbol interference nearly vanishes if
( )
k−n l−m
hw , ≈ 0 ∀n ̸= k, m ̸= l. (31)
N T M ∆f
Whether this condition is satisfied depends on the channel’s
delay and Doppler spread and the window design. An equalizer
at the receiver extracts the full channel diversity. Common
equalization architectures can be employed for this purpose;
turbo equalization is used in the simulation results below. More
details on OTFS equalization are discussed in [9].

IV. P ERFORMANCE R ESULTS


Fig. 3. BLER for 1x1 system, 16QAM/64QAM, Code rate R=1/2, 120kmph.
In this section we present some simulation results to illus-
trate the performance of an OTFS system as compared with
Figure 4 shows the BLER performance for a 1x1 system
an OFDM system. We simulate a coded modulation system
with mobile speeds of 30 km/h if only 4 resource blocks (48
and we choose the FEC code as well as all other PHY layer
subcarriers), called PRBs in LTE, are occupied by the user of
parameters to comply with the 4G LTE specification (ETSI TS
interest out of a total of 50 resource blocks (600 subcarriers).
36.211 and ETSI TS 36.212). For the OTFS system we add
This corresponds to a short packet length. Notice the increased
1 To be precise, the window w(τ, ν) is circularly convolved with the channel diversity gain of OTFS in this case resulting in gains of 4
impulse response h(τ, ν) times the complex exponential e−j2πντ . dB or more as SNR increases. This diversity gain is because

Authorized licensed use limited to: UNIVERSITE DE GABES. Downloaded on May 16,2024 at 12:25:09 UTC from IEEE Xplore. Restrictions apply.
the OTFS transform spreads each QAM symbol over all time fails while OTFS is able to multiplex two MIMO streams in
and frequency dimensions of the channel and then extracts the this environment.
resulting full diversity, while OFDM limits the transmission to
a narrow subchannel of 48 subcarriers.

Fig. 6. BLER for 2x2 system, 4QAM/16QAM/64QAM, Code rate R=1/2,


500kmph.
Fig. 4. BLER for short packet length (4 PRBs out of 50), 1x1 system,
16QAM/64QAM, Code rate R=1/2, 30kmph.
V. C ONCLUSION
Figure 5 depicts performance of a 2x2 system for a mobile We have developed OTFS, a novel two-dimensional mod-
speed of 120 km/h. For MIMO transmission, we adopt a ulation scheme for wireless communications with significant
constant modulus open loop precoding scheme as this is advantages in performance over existing modulation schemes.
more realistic for moderate and high speeds. Specifically, for OTFS operates in the delay-Doppler coordinate system and we
OFDM we use Large-Delay CDD per LTE Transmission Mode show that with this modulation scheme coupled with equaliza-
3, whereas OTFS uses an identity precoder. Notice that a tion, all modulated symbols experience the same channel gain
performance difference of up to 4 dB is observed in the MIMO extracting the full channel diversity. OTFS is shown to exhibit
case of Figure 5. significantly lower block error rates than OFDM over a wide
range of constellation sizes and channel Dopplers (for vehicle
speeds ranging from 30 km/h to 500 km/h). The robustness to
high-Doppler channels (500 km/h vehicle speeds) is especially
notable, as OFDM performance breaks down completely in
this use case.
R EFERENCES
[1] W.C Jakes, Jr., Microwave Mobile Communications, Wiley, New York,
1974.
[2] P. Bello, “Characterization of randomly time-variant linear channels,”
IEEE Trans. Commun., Vol. 11, No. 4, pp 360-393, Nov. 1963.
[3] A. Sayeed and B. Aazhang, “Joint multipath-Doppler diversity in mobile
wireless communications,” IEEE Trans. Commun., Vol. 47, No. 1, pp.
123–132, Jan. 1999.
[4] K. Liu, T. Kadous, and A.M. Sayeed, “Orthogonal time-frequency signal-
ing over doubly dispersive channels,” IEEE Trans. Inform. Theory, Vol.
50, No. 11, pp. 2583–2603, Nov. 2004.
Fig. 5. BLER for 2x2 system, 16QAM/64QAM, Code rate R=1/2, 120kmph.
[5] X. Ma and G. B. Giannakis, “Maximum-diversity transmissions Over
doubly selective wireless channels,” IEEE Trans. Inform. Theory, Vol.
Finally, in Figure 6 we compare the performance of OTFS 49, No. 7, pp. 1832–1840, July 2003.
in an extreme mobility case with a mobile speed of 500 Km/h. [6] W. Kozek and A. F. Molisch, “Nonorthogonal pulseshapes for multicarrier
communications in doubly dispersive channels,” IEEE J. Select. Areas
This is the so called high speed train use case considered in the Commun., Vol 16, No. 8, pp. 1579–1589, Aug. 1998.
context of 5G standardization and relates to offering service [7] S. Das and P. Schniter, “Max-SINR ISI/ICI-shaping multicarrier commu-
to passengers on a high speed train. The Doppler spread in nication over the doubly dispersive channel,” IEEE Trans. Signl. Proc.
Vol. 55, No.12, pp. 5782-5795, Dec. 2007.
this case is high (1820 Hz) relative to the subcarrier spacing [8] T. Strohmer and S. Beaver, “Optimal OFDM design for time-frequency
(15 KHz) resulting in intercarrier interference (ICI). Figure 4 dispersive channels,” IEEE Trans. Commun., Vol. 51, No. 7 pp. 1111-
shows the performance difference of the two systems assuming 1122, July 2003.
[9] R. Hadani, S. Rakib, M. Tsatsanis, A. Monk, A.J. Goldsmith, A.F.
(i) the receiver does not perform any explicit ICI cancellation Molisch, A.R. Calderbank, “Orthogonal Time Frequency Space Mod-
(ii) the receiver performs ICI cancellation which results in per- ulation,” In preparation for submission to the IEEE Trans. Wireless
fect removal of all ICI. Notice that the performance difference Commun..
[10] W. Mecklenbrauker, A Tutorial on Non-Parametric Bilinear Time-
increases for higher constellations and higher SNRs where Frequency Signal Representations, Time and Frequency Representation
ICIs detrimental effects are more pronounced. Further notice of Signals and Systems (Eds. G Longo and B. Picinbono), vol. 309, pp.
that for 64QAM and no ICI cancellation, OFDM completely 11-68, 1989.

Authorized licensed use limited to: UNIVERSITE DE GABES. Downloaded on May 16,2024 at 12:25:09 UTC from IEEE Xplore. Restrictions apply.

You might also like