RIS-aided OTFS Modulation in High-Doppler Channels

Gandhodi Harshavardhan, Vighnesh S Bhat, and A. Chockalingam

Department of ECE, Indian Institute of Science, Bangalore 560012

Abstract—Two research topics, namely, reconfigurable intelli- multicarrier modulation schemes such as orthogonal frequency
gent surfaces (RIS) and orthogonal time frequency space (OTFS) division multiplexing (OFDM) in high-mobility scenarios [7]-
modulation, have drawn significant interest in recent wireless [9]. In high-mobility scenarios, the channel witnesses rapid
research. While RIS technology can aid communication by favor-
ably controlling the electromagnetic wave propagation towards variations in time. This introduces inter-carrier interference
the receiver, OTFS modulation can offer robust performance (ICI) in OFDM, which causes severe performance degradation.
in high-Doppler channels. Both RIS and OTFS are envisioned Whereas, OTFS is robust to rapid time variations of the
to be promising technologies for 6G and beyond. This paper channel as OTFS signaling is carried out in the delay-Doppler
investigates a RIS-aided OTFS system. A novel and useful (DD) domain in contrast to OFDM signaling which is carried
contribution in this paper is the detailed derivation of the end-
to-end delay-Doppler (DD) domain input-output relation of a out in the time-frequency (TF) domain.
RIS-aided OTFS system, which can serve as a basic system Owing to their inherent strengths and performance benefits,
model that can aid the investigation of efficient transceiver tech- both RIS and OTFS are envisioned to be promising technolo-
niques/algorithms, performance evaluation, and implementation. gies for 6G and beyond. For the same reasons, there has been
We consider maximization of the Frobenius norm of the effective a lot of interest in RIS and OTFS research. An increasing body
end-to-end DD channel matrix of the RIS-aided OTFS system to
choose the phase vector at the RIS. Our simulation results show of research has been appearing in the literature on these two
that RIS-aided OTFS performs significantly better compared to topics (e.g., [10]- [17] and several other papers in the litera-
RIS-aided OFDM. ture), where they are treated independently highlighting their
Index Terms—Reconfigurable intelligent surface, OTFS mod- individual merits. Investigation of RIS-aided OTFS systems
ulation, delay-Doppler domain, RIS-aided OTFS, input-output is of interest. An observation of interest in this regard is that
relation, RIS-aided OFDM. RIS-aided OFDM systems have been investigated in the recent
literature in [12]- [14], where RIS has been shown to improve
I. I NTRODUCTION OFDM performance. It is of interest to investigate RIS-aided
OTFS in high-mobility scenarios and assess its performance in
The key performance requirements in next generation wire-
comparison with that of RIS-aided OFDM. This paper focuses
less communication systems include increased energy and
on this aspect, which has not been reported before. Our new
spectral efficiency, high reliability, low latency, and high
and novel contributions can be summarized as follows.
mobility. Two promising research directions have emerged
• We derive the end-to-end DD domain input-output rela-
recently to meet these diverse requirements. They are recon-
figurable intelligent surfaces (RIS) [1]- [4] and orthogonal tion of a RIS-aided OTFS system. This input-output rela-
time frequency space (OTFS) modulation [5]- [9]. RIS tech- tion can serve as a basic system model that can aid the in-
nology is a paradigm that provides energy efficient wireless vestigation of efficient transceiver techniques/algorithms,
communications and OTFS technology is a paradigm that performance evaluation, and implementation.
• We propose the maximization of the Frobenius norm of
provides reliable high-mobility support. In a RIS assisted
communication system, smart reflecting surfaces consisting of the effective end-to-end DD channel matrix of the RIS-
an array of passive tunable reflecting elements are placed in aided OTFS system to choose the RIS phase vector.
• Our results show that 1) RIS improves the performance
the propagation environment to aid communication between
the transmitter and the receiver. This is achieved by favorably of OTFS, 2) the performance improvement gets enhanced
controlling the electromagnetic wave propagation towards the for increased number of reflecting elements, and 3) RIS-
receiver by tuning the reflection characteristics of the elements aided OTFS performs significantly better compared to
in the surface. By tuning the phases of the reflecting elements, RIS-aided OFDM in high-mobility channels.
the incident signal from the transmitter is reflected in such a The rest of the paper is organized as follows. The RIS-
way that the directivity of the reflected signal is maximized in aided OTFS system model is introduced in Sec. II. The end-to-
the direction of the receiver, leading to significant SNR gains. end DD domain input-output relation of the RIS-aided OTFS
The passive nature of the reflecting elements renders the RIS is derived in Sec. III. Simulation results and discussions are
approach to be an energy efficient approach [10]. presented in Sec. IV. Conclusions are presented in Sec. V.
OTFS modulation is a recently introduced modulation II. RIS- AIDED OTFS SYSTEM MODEL
scheme [5] which significantly outperforms contemporary
Consider a RIS-aided OTFS system consisting of an OTFS
This work was supported by the J. C. Bose National Fellowship, Depart- transmitter, an OTFS receiver, and a RIS as shown in Fig.
ment of Science and Technology, Government of India. 1. A single antenna is assumed both at the transmitter and
the receiver. The RIS consists of multiple reflecting elements {x[k, l]} {X[n, m]} Heisenberg x(t)
whose phases can be tuned so that the reflected rays are ISFFT
transform
steered towards the receiver. When the number of elements {hr (τ, ν)}Lr=1
are large, the reflection power towards the receiver can be
maximized at the expense of a large overhead in the reflection
phase design. The reflection phase design is simplified by RIS
grouping adjacent elements with highly correlated channel
coefficients into sub-surfaces that share a common reflection
coefficient [12]. Let K denote the number of reflecting el- {y r [k, l]} {Y r [n, m]} {g r (τ, ν)}Lr=1
Wigner
Wigner
ements in the RIS that are grouped into L sub-surfaces, so SFFT transform {y r (t)}
that each sub-surface contains Ns = K/L adjacent reflecting
elements. Let ϕr denote the reflection coefficient of the rth
sub-surface, r = 1, · · · , L. A controller connected to the Fig. 1: RIS-aided OTFS system.
RIS tunes the sub-surfaces with these reflection coefficients. respectively. The received signal at the rth sub-surface of RIS,
At the OTFS transmitter, M N information symbols from a denoted by z r (t), can be written as
modulation alphabet A (e.g., QAM/PSK), denoted by x[k, l], Z Z
k = 0, · · · , N − 1, l = 0, · · · , M − 1, are multiplexed in z r (t) = hr (τ, ν)x(t − τ )ej2πν(t−τ ) dτ dν. (4)
the DD grid with N Doppler bins and M delay bins. The ν τ
DD grid is defined as Γ = ( NkT , M ∆f l
) , where M ∆f The DD channel response of the RIS-to-receiver link (i.e., 2nd
is the system bandwidth, N T is one OTFS frame duration, link) associated with the rth sub-surface is written as
and ∆f = 1/T . The information symbols in the DD domain
P2
are converted into symbols in the time-frequency (TF) grid, X
g r (τ, ν) = gqr δ(τ − τqr,2 )δ(ν − νqr,2 ), (5)
denoted by X[n, m], n = 0, · · · , N − 1, m = 0, · · · , M − 1,
q=1
using inverse symplectic finite Fourier transform (ISSFT).
The TF domain symbols are converted to time domain signal where P2 is the number of DD channel paths in the 2nd link,
x(t) using Heisenberg transform, and transmitted through the gqr , τqr,2 , and νqr,2 are the fade coefficient, delay, and Doppler
channel. The RIS elements reflect the incident signal towards of the qth path associated with rth sub-surface in the 2nd link,
αr,1 β r,1 αr,2
the receiver. The received time-domain signal is converted to respectively. Define τpr,1 ≜ Mp∆f , νpr,1 ≜ NpT , τqr,2 ≜ Mq∆f ,
TF domain (using Wigner transform), which is then converted β r,2
and νqr,2 ≜ NqT , where αpr,1 , βpr,1 , αqr,2 , and βqr,2 are assumed
to DD domain (using SFFT) where the information symbols
to be integers.
are detected.
Let y r (t) denote the received time domain signal at the
The TF domain symbols X[n.m] at the output of the ISFFT
receiver reflected from the rth sub-surface of the RIS. The
at the transmitter is given by
received signal at the receiver associated with the rth sub-
N −1 M −1
surface is given by
1
x[k, l]ej2π( N − M ) ,
X X nk ml
X[n, m] = √ (1)
Z Z
M N k=0 l=0 y r (t) = ϕr g r (τ2 , ν2 )z r (t − τ2 )ej2πν2 (t−τ2 ) dτ2 dν2
Zν2 Zτ2 Z Z
and the time domain signal x(t) at the output of the transmitter = ϕr g r (τ2 , ν2 ) hr (τ1 , ν1 )x(t − τ1 − τ2 )
after Heisenberg transform is given by ν 2 τ2 ν1 τ1
i
ej2πν1 (t−τ1 −τ2 ) dτ1 dν1 ej2πν2 (t−τ2 ) dτ2 dν2 , (6)
N
X −1 M
X −1
x(t) = X[n, m]gtx (t − nT )ej2πm∆f (t−nT ) , (2) where ϕr = γr ejθr is the reflection coefficient, γr ∈ [0, 1]
n=0 m=0 is the reflection amplitude, and θr ∈ [−π, π] is the reflection
phase, respectively, of the rth sub-surface.
where gtx (t) is the transmit pulse. Let hr (τ, ν) denote the
DD domain channel response of the transmitter-to-RIS link III. D ERIVATION OF DD INPUT- OUTPUT RELATION
(i.e., 1st link) associated with the rth element of the RIS,
r = 1, · · · , L, which can be written as In this section, we derive the end-to-end input-output rela-
tion in the DD domain for the system considered in Sec. II.
P1
X We first consider the time domain-to-TF domain conversion
hr (τ, ν) = hrp δ(τ − τpr,1 )δ(ν − νpr,1 ), (3) then the TF domain-to-DD domain conversion at the receiver.
p=1 Time domain-to-TF domain: The time domain received
signal y r (t) is converted to TF domain signal Y r [n, m] using
where P1 is the number of DD channel paths in the 1st link, Wigner transform as
hrp , τpr,1 , and νpr,1 are the fade coefficient, delay, and Doppler
of the pth path associated with rth sub-surface in the 1st link, Y r [n, m] = Y r (t, f )|t=nT,f =m∆f , (7)
where Y r (t, f ) is the cross-ambiguity function between the where yr , x ∈ CM N ×1 , the (k + N l)th entry of x, xk+N l =
received signal y r (t) and the receive pulse grx (t), given by x[k, l], k = 0, · · · , N − 1, l = 0, · · · , M − 1 and x[k, l] ∈
r
Z
′
A, where A is the modulation alphabet. Similarly, yk+N l =
r ∗
Y (t, f ) = grx (t′ − t)y r (t′ )e−j2πf (t −t) dt′ . (8) y [k, l], k = 0, · · · , N − 1, l = 0, · · · , M − 1, and Hr ∈
r
t′
CM N ×M N is the effective cascaded channel matrix for the
The input-output relation in TF domain, ignoring noise, can rth sub-surface. Combining the reflected signals from all the
be written as [9] L sub-surfaces at the receiver, we obtain the overall end-to-end
N
X −1 M
X −1 DD domain input-output relation for RIS-aided OTFS as
Y r [n, m] = ϕr r
Hn,m [n′ , m′ ]X[n′ , m′ ], (9) L
X
n′ =0 m′ =0 y= ϕr Hr x + v, (17)
r
where Hn,m [n′ , m′ ]is given by r=1
Z Z Z Z where y is the combined received vector from all the sub-
r
Hn,m [n′ , m′ ] = g r (τ2 , ν2 ) hr (τ1 , ν1 ) surfaces and v ∈ CM N ×1 is the additive noise vector at the
ν 2 τ2 ν 1 τ1
  receiver.
Agrx ,gtx (n − n′ )T − (τ1 + τ2 ), (m − m′ )∆f − ν1 − ν2
 A. Reflection phase design
′ ′
ej2πm ∆f ((n−n )T −(τ1 +τ2 )) ej2πν1 (nT −(τ1 +τ2 )) dτ1 dν1 We consider a reflection phase vector design at the RIS
that maximizes the Frobenius norm of the effective end-
ej2πν2 (nT −τ2 ) dτ2 dν2 , (10) to-end DD channel matrix of the RIS-aided OTFS system.
Let γr = 1 and θr ∈ [−π, π], r = 1, · · · , L. We are
where Agrx ,gtx (t, f ) is the cross ambiguity function between
interested in the reflection phase vector Θ = [θ1 θ2 · · · θL ]
the transmit and receive pulses, defined as
that maximizes thePFrobenius norm of the effective channel
L
Z
∗ ′ matrix, given by ∥ r=1 ejθr Hr ∥2 . We use a random Monte
Agrx ,gtx (t, f ) = grx (t′ − t)gtx (t′ )e−j2πf (t −t) dt′ . (11)
t′ Carlo sampling approach to obtain an approximate solution
The derivation of (10) is given in Appendix A. It is assumed to this maximization problem. Let θri denote the ith realiza-
that both the transmit pulse and the receive pulse are ideal tion of randomly generated phase uniformly distributed in
whose cross-ambiguity function is given by [9] [−π, π], so that the ith realization of the phase vector is
( Θi = [θ1i θ2i · · · θL
i
]. A large number of such phase vector
1, n = 0, m = 0 realizations are generated and that vector which gives the
Agrx ,gtx (t, f ) = (12) ∗
0, otherwise, maximum Frobenius norm is chosen, i.e., choose Θi where
L i
i∗ = arg maxi ∥ r=1 ejθr Hr ∥2 .
 P
for t ∈ (nT − τmax , nT + τmax ) and f ∈ (m∆f −
νmax , m∆f +νmax ), where τmax and νmax are the maximum IV. R ESULTS AND DISCUSSIONS
delay and the maximum Doppler, respectively. It is clear In this section, we present the simulated bit error rate (BER)
r
from (12) that Hn,m [n′ , m′ ] is non-zero only when n′ = n, performance of the considered RIS-aided OTFS system. Table
′
m = m for the ideal pulses satisfying bi-orthogonal property. I shows the system parameters used in the simulations. A
Therefore, (9) reduces to carrier frequency (fc ) of 4 GHz, a subcarrier spacing (∆f )
of 3.75 kHz, and BPSK modulation are used. The maximum
Y r [n, m] = ϕr Hn,m
r
[n, m]X[n, m]. (13)
Doppler considered is 1.875 KHz, which corresponds to a
TF domain-to-DD domain: The TF signal Y r [n, m] is maximum speed of 506.25 km/h for the considered carrier
transformed to the DD domain using SFFT as frequency of 4 GHz. The channel fade coefficients hri s and
N −1 M −1 gir s are assumed to be i.i.d and distributed as CN (0, 1/P ),
1 X X nk ml
where P denotes the number of DD channel paths in the
y r [k, l] = √ Y r [n, m]e−j2π( N − M ) . (14)
M N n=0 m=0 corresponding link. In all the simulations, the same DD profile
is considered for all the sub-surfaces, and 30 random phase
The end-to-end DD domain input-output relation associated
vector realizations are taken in the Frobenius norm based
with the reflected signal from the rth sub-surface is derived
choice of reflection phases.
as (please see Appendix B for the derivation)
Performance of OTFS without and with RIS: Figure 2 shows
P2 P1
X r,2 r,2 X r,1
(τpr,1 +τqr,2 )
the BER performance of OTFS without RIS and RIS-aided
y r [k, l] = ϕr gqr e−j2πνq τq
hrp e−j2πνp OTFS and for L = 5, 15, M = N = 2, P1 = P2 = 2,
q=1 p=1 and maximum likelihood (ML) detection. From Fig. 2, it is
x[[k − (βpr,1 + βqr,2 )]N , [l − (αpr,1 + αqr,2 )]M ], (15) observed that RIS-aided OTFS for L = 15 achieves the best
performance followed by RIS-aided OTFS for L = 5. OTFS
where [.]N is the modulo N operation. Now, the input-output
without RIS has the least performance. For example, at a BER
relation in (15) can be written in a vectorized form as
of 10−4 , RIS-aided OTFS has SNR gain of about 20 dB and
yr = ϕr Hr x, (16) 25 dB for L = 5 and L = 15, respectively, compared to OTFS
 Parameter Value 100
Frame size (M, N ) (2, 2)
DD (τi , νi ) profile for 2 paths (0, 0), ( M1∆f , N1T )
(0, 0), (0, N1T ),
DD (τi , νi ) profile for 4 paths
( M1∆f , 0), ( M1∆f , N1T )
Maximum speed 506.25 km/h 10-2
Modulation BPSK

TABLE I: Simulation parameters.

100
10-4

10-2 5 10 15 20 25 30 35 40 45 50

Fig. 3: Effect of number of sub-surfaces L on the BER

performance of RIS-aided OTFS.
10-4
102

100
-6
10
-40 -30 -20 -10 0 10 20 30
10-2
Fig. 2: BER performance of OTFS without RIS and RIS-aided
OTFS with L = 5, 15. 10-4

without RIS, demonstrating the beneficial role of RIS. Also,

a diversity slope of one is observed in all the cases. 10-6
Effect of number of sub-surfaces on performance: Figure 3
shows the performance of RIS-aided OTFS as a function of 10-8
-40 -30 -20 -10 0 10 20
number of sub-surfaces L for M = N = 2 P1 = P2 = 2
at SNR values of -10 dB, -5 dB, and 0 dB. The performance
Fig. 4: BER performance of RIS-aided OTFS without and with
without RIS at these SNR values are also plotted for compari-
precoding at the transmitter.
son. As expected, the use of RIS results in significant bit error
performance gains, and the performance gain improves with the diversity order of RIS-aided OTFS without precoding
increase in the number of sub-surfaces. For example, at 0 dB is one, the precoding operation at the transmitter achieves
SNR, the BER with RIS improves by two orders compared to a significantly better BER performance with an increased
without RIS when L = 5 and the improvement becomes three diversity order.
orders when L = 50. Performance comparison between RIS-aided OTFS and RIS-
Performance of RIS-aided OTFS with precoding at the aided OFDM: In Fig. 5, we present a performance comparison
transmitter: Figure 4 shows the effect of precoding at the between RIS-aided OTFS and RIS-aided OFDM for practi-
transmitter on the BER performance of RIS-aided OTFS for cally large values of M and N considering the power-delay
M = N = 2, L = 5, P1 = 2 and P2 = 4. The precoding profile defined in extended vehicular A (EVA) model [18].
1 M N −1
matrix used is Φ = diag{1, ej M N , · · · , ej M N } [15]. In Accordingly, the parameters considered are M = 64, N = 12,
addition to simulated BER plots, corresponding upper bounds P1 = P2 = 9, and L = 5. The Doppler shift corresponding
on the bit error probability Pb are also plotted. The upper to the ith path is generated using νi = νmax cos θi , where
bound on Pb is obtained from pairwise error probability (PEP) θi ∈ [−π, π] is uniformly distributed and νmax = 1.34 kHz is
using union bound as the maximum Doppler shift. Since ML detection is not feasible
for large values of M and N , we use minimum mean square
Z Z
1 X X error (MMSE) detection. From Fig. 5, we observe that the
Pb ≤ d(xi , xj )P (xi → xj (18)
), performance of RIS-aided OTFS is significantly better than
ZM N log2 |A| i=1
j=1,j̸=i
that of RIS-aided OFDM. For example, at a BER of 10−3 ,
where A denotes the modulation alphabet, Z = |AM N |, RIS-aided OTFS performs better by about 9 dB compared to
d(xi , xj ) is the Hamming distance between the bit mappings RIS-aided OFDM. Figure 6 shows the effect of the number
of OTFS signal vectors xi and xj , and P (xi → xj ) is of sub-surfaces on the BER performance of RIS-aided OTFS
the PEP between xi and xj . It can be seen that the upper and RIS-aided OFDM at SNR of 5 dB. It can be seen that the
bounds are tight at high SNRs. It is also observed that while performance for both RIS-aided OTFS and RIS-aided OFDM
 Dopplers and practical pulses, which can be taken up as future
100 work. Also, efficient techniques for RIS phase optimization in
RIS-aided OTFS can be explored as future work.
10-1
A PPENDIX A
D ERIVATION OF (10)
10-2 r
In this appendix, we derive the expression for Hn,m [n′ , m′ ]
r r
in (10). Substituting y (t) from (6) in (8), Y (t, f ) can be
10-3 written as
Z Z Z
r ∗ ′
Y (t, f ) = ϕr grx (t − t) g r (τ2 , ν2 )
10-4 t′ ν2 τ2
Z Z
′
hr (τ1 , ν1 )x(t′ − τ1 − τ2 )ej2πν1 (t −τ1 −τ2 ) dτ1 dν1
10-5 ν1 τ1
-40 -30 -20 -10 0 10 20 30 ′ ′
ej2πν2 (t −τ2 ) dτ2 dν2 e−j2πf (t −t) dt′ . (19)
Fig. 5: BER performance comparison between RIS-aided Substituting x(t) from (2) in (19), we get
OTFS and RIS-aided OFDM. Z Z Z Z Z
10 -2 Y r (t, f ) = ϕr ∗
grx (t′ − t) g r (τ2 , ν2 )
t′ ν2 τ2 ν1 τ1
 NX
−1 M
X −1
r
h (τ1 , ν1 ) X[n′ , m′ ]gtx (t′ − τ1 − τ2 − n′ T )
10-3
n′ =0 m′ =0
 
′ ′ ′ ′
ej2πm ∆f (t −τ1 −τ2 −n T ) ej2πν1 (t −τ1 −τ2 ) dτ1 dν1
10-4 
′ ′
ej2πν2 (t −τ2 ) dτ2 dν2 e−j2πf (t −t) dt′ . (20)
10-5 Rearranging the terms in (20), we get
N
X −1 M
X −1 Z Z
r ′ ′
10 -6 Y (t, f ) = ϕr X[n , m ] g r (τ2 , ν2 )
5 10 15 20 25 30 35 40 n′ =0 m′ =0 ν2 τ2
Z Z Z
r ∗
Fig. 6: Effect of number of sub-surfaces on the BER perfor- h (τ1 , ν1 ) grx (t′ − t)gtx (t′ − τ1 − τ2 − n′ T )
ν1 τ1 t′
mance of RIS-aided OTFS and RIS-aided OFDM. j2πm′ ∆f (t′ −τ1 −τ2 −n′ T ) −j2πf (t′ −t) j2πν1 (t′ −τ1 −τ2 )
e  e  e
improve as L is increased. However, the performance of RIS- ′
ej2πν2 (t −τ2 ) dt′ dτ1 dν1 dτ2 dν2 . (21)
aided OTFS is significantly better compared to RIS-aided
OFDM. Also, the performance gap between RIS-aided OTFS Evaluating Y r (t, f ) in (21) at t = nT and f = m∆f and
and RIS-aided OFDM becomes large for large values of L. comparing with (9), Hn,m r
[n′ , m′ ] can be written as
For example, there is about one order BER performance gap Z Z Z Z
in favor of RIS-aided OTFS at L = 5, and the performance r
Hn,m [n′ , m′ ] = g r (τ2 , ν2 ) hr (τ1 , ν1 )
gap increases to three orders for L = 20. ν 2 τ2 ν1 τ1
Z
V. C ONCLUSIONS ∗
grx (t′ − nT )gtx (t′ − τ1 − τ2 − n′ T )
We investigated the performance benefits of using RIS in t′
j2πm′ ∆f (t′ −τ1 −τ2 −n′ T ) −j2πm∆f (t′ −nT )
OTFS modulated systems in high-Doppler channels. In this e e  
effort, we derived the end-to-end DD domain input-output e j2πν1 (t′ −τ1 −τ2 ) j2πν2 (t′ −τ2 )
e ′
dt dτ1 dν1 dτ2 dν2 . (22)
relation in RIS-aided OTFS systems. This derivation is a new
contribution. The derived input-output relation can aid further Now, changing the variable of integration t′ in the inner
investigations in RIS-aided OTFS relating to transceiver tech- integral of (22), i.e., t′ − τ1 − τ2 − n′ T = t′′ , we obtain
niques/algorithms, performance evaluation, and implementa- Z Z Z Z
r ′ ′ r
tion. The reflection phases at the RIS were obtained by max- Hn,m [n , m ] = g (τ2 , ν2 ) hr (τ1 , ν1 )
imizing the Frobenius norm of the effective end-to-end DD ν2 τ2 ν1 τ1
Z
channel matrix. Our simulation results showed that RIS-aided grx∗
(t′′ − (n − n′ )T + τ1 + τ2 )gtx (t′′ )
OTFS achieves significantly better performance compared to t′′
′ ′′ ′′
RIS-aided OFDM in high-Doppler channels. In this work, −(n−n′ )T +τ1 +τ2 )
ej2πm ∆f t e−j2πm∆f (t
integer delay-Dopplers and ideal biorthogonal pulses were
 
j2πν1 (t′′ +n′ T ) j2πν2 (t′′ +τ1 +n′ T ) ′′
considered. This can be readily extended for fractional delay- e e dt dτ1 dν1 dτ2 dν2 . (23)
Rearranging the terms in (23) and using algebraic manipula- The hrw [k, l] in (30) can be viewed as the sampled version of
tions, we obtain l
hrw (ν, τ ) at ν = NkT , τ = M ∆f , where hrw (ν, τ ) is given by
Z Z Z Z
r N −1 M −1  Z Z
[n′ , m′ ] = g r (τ2 , ν2 ) hr (τ1 , ν1 )
Z Z
Hn,m X X
ν 2 τ2 ν1 τ1 hrw (ν, τ ) = g r (τ2 , ν2 ) hr (τ1 , ν1 )
Z n=0 m=0 ν2 τ2 ν1 τ1
∗
grx (t′′ − (n − n′ )T + τ1 + τ2 )gtx (t′′ )

−j2πm∆f (τ1 +τ2 ) j2πν1 (nT −(τ1 +τ2 ))
t′′
 e e dτ1 dν1
−j2π((m−m′ )∆f −ν1 −ν2 )(t′′ −(n−n′ )T +τ1 +τ2 ) ′′
e dt

 ej2πν2 (nT −τ2 ) dτ2 dν2 e−j2πnT ν ej2πm∆f τ . (31)
′ ′
ej2πm ∆f ((n−n )T −(τ1 +τ2 )) ej2πν1 (nT −(τ1 +τ2 )) dτ1 dν1 Rearranging the terms in (31), we get
 NX
−1 M −1
ej2πν2 (nT −τ2 ) dτ2 dν2 . (24)
Z Z Z Z X
r r r
hw (ν, τ ) = g (τ2 , ν2 ) h (τ1 , ν1 )
The inner integral of (24) w.r.t t′′ is the cross-ambiguity ν 2 τ2 ν1 τ1 n=0 m=0
function between the transmit pulse and the receive pulse −j2π(ν−(ν1 +ν2 ))nT j2π(τ −(τ1 +τ2 ))m∆f
e e 
defined in (11). Therefore, (24) can be rewritten as −j2πν1 (τ1 +τ2 ) −j2πν2 τ2
Z Z Z Z e e dτ1 dν1 dτ2 dν2 , (32)
r
Hn,m [n′ , m′ ] = g r (τ2 , ν2 ) hr (τ1 , ν1 )
ν2 τ2 ν1 τ1
which can be written as
Z
′ ′
Z Z Z
Agrx ,gtx ((n − n )T − (τ1 + τ2 ), (m − m )∆f − ν1 − ν2 ) r
hw (ν, τ ) = r
g (τ2 , ν2 ) hr (τ1 , ν1 )

j2πm′ ∆f ((n−n′ )T −(τ1 +τ2 )) j2πν1 (nT −(τ1 +τ2 )) ν2 τ2 ν1 τ1
e e dτ1 dν1
w(ν − (ν1 + ν2 ), τ − (τ1 + τ2 ))e−j2πν1 (τ1 +τ2 )
j2πν2 (nT −τ2 )

e dτ2 dν2 . (25) −j2πν2 τ2
e dτ1 dν1 dτ2 dν2 , (33)
which is same as (10).
where
A PPENDIX B N
X −1 M
X −1
D ERIVATION OF (15) w(ν, τ ) = 1.e−j2π(νnT −τ m∆f ) (34)
r n=0 m=0
Substituting Y [n, m]) from (13) in (14), the received signal
y r [k, l] in the DD domain can be written as is the window function and hrw (ν, τ ) is the circular convolution
of the channel response with the SFFT of a rectangular
N −1 M −1
ϕr X X r nk ml windowing function in TF domain. Substituting for hr (τ1 , ν1 )
y r [k, l] = √ Hn,m [n, m]X[n, m]e−j2π( N − M ) .
M N n=0 m=0 and g r (τ2 , ν2 ) from (3) and (5), we obtain
(26) Z Z X P2
Substituting for X[n, m] from (1), we obtain hrw (ν, τ ) = gqr δ(τ2 − τqr,2 )δ(ν2 − νqr,2 )
ν2 τ2 q=1
N −1 M −1  NX −1 M −1
r ϕr X X r X
y [k, l] = Hn,m [n, m] x[k ′ , l′ ] P1
Z Z X
M N n=0 m=0 ′ ′
k =0 l =0
hrp δ(τ1 − τpr,1 )δ(ν1 − νpr,1 )
 ν1 τ1 p=1
′ ml′
j2π( nk − ) −j2π( nk
− ml
)
e N M e N M , (27) w(ν − (ν1 + ν2 ), τ − (τ1 + τ2 ))

Rearranging the terms in (27), we get e−j2πν1 (τ1 +τ2 ) e−j2πν2 τ2 dτ1 dν1 dτ2 dν2 . (35)
N −1 M −1  NX −1 M −1
ϕr X X X Using the sifting property of the impulse function, (35) can
y r [k, l] = x[k ′ , l′ ] r
Hn,m [n, m]
MN ′ be written as
k =0 l′ =0 n=0 m=0
 P2 P1
′ l−l′ r,2 r,2 X r,1 r,1 r,2
X
−j2πnT ( k−k ) j2πm∆f ( M ) hrw (ν, τ ) = gqr e−j2πνq τq hrp e−j2πνp (τp +τq )
e N T e ∆f , (28)
q=1 p=1
which can be further simplified as w(ν − (νpr,1 + νqr,2 ), τ − (τpr,1 + τqr,2 )). (36)
N −1 M −1 Substituting for w(ν, τ ) in (36), we get
ϕr X X
y r [k, l] = x[k ′ , l′ ]hrw [k − k ′ , l − l′ ], (29)
MN ′ ′
P2 P1
k =0 l =0 r,2 r,2 r,1
(τpr,1 +τqr,2 )
X X
hrw (ν, τ ) = gqr e−j2πνq τq
hrp e−j2πνp
where q=1 p=1
N −1 M −1 N −1 M −1
r,1
k l
+νqr,2 ))nT −(τ −(τpr,1 +τqr,2 ))m∆f )
X X X X
hrw [k, l] = r
Hn,m [n, m]e−j2πn N ej2πm M , (30) e−j2π((ν−(νp . (37)
n=0 m=0 n=0 m=0
Separating the summations, we get Using (44) in (29), we obtain
P2 P1 N −1 M −1 P2
X
X X r,2 r,2
r,2 r,2 r,1 r,1 r,2
y r [k, l] = ϕr x[k ′ , l′ ] gqr e−j2πνq τq
X X
hrw (ν, τ ) = gqr e−j2πνq τq hrp e−j2πνp (τp +τq )
q=1 p=1 k′ =0 l′ =0 q=1
N −1 M −1 P1
r,1
(τpr,1 +τqr,2 )
X
r,1
+νqr,2 ))nT r,1
+τqr,2 ))m∆f hrp e−j2πνp δ[[k − k ′ − (βpr,1 + βqr,2 )]N ]
X X
e−j2π(ν−(νp ej2π(τ −(τp , (38)
n=0 m=0 p=1

′
Now, define δ[[l − l − (αpr,1 + αqr,2 )]M ] . (45)
N −1
X r,1
+νqr,2 ))nT Using the sifting property of the impulse function in (45), we
G(ν, νpr,1 + νqr,2 ) ≜ e−j2π(ν−(νp , (39)
obtain (15).
n=0
R EFERENCES
M −1
j2π(τ −(τpr,1 +τqr,2 ))m∆f
X
[1] S. Hu, F. Rusek, and O. Edfors, “Beyond massive MIMO: the potential
F(τ, τpr,1 + τqr,2 ) ≜ e . (40) of data transmission with large intelligent surfaces,” IEEE Trans. Signal
m=0 Process., vol. 66, no. 10, pp. 2746–2758, Mar. 2018.
[2] C. Liaskos, A. Tsioliaridou, A. Pitsillides, S. Ioannidis, and I. Akyildiz,
Therefore, (38) can be written as “Using any surface to realize a new paradigm for wireless communica-
tions,” Commun. ACM, vol. 61, no. 11, pp. 30-33, Nov. 2018.
P2 P1
r,2 r,2 r,1 [3] E. Basar, M. D. Renzo, J. de Rosny, M. Debbah, M.-S. Alouini, and
(τpr,1 +τqr,2 )
X X
hrw (ν, τ ) = gqr e−j2πνq τq hrp e−j2πνp R. Zhang, “Wireless communications through reconfigurable intelligent
q=1 p=1 surfaces,” IEEE Access, vol. 7, pp. 116753-116773, Aug. 2019.
[4] Q. Wu and R. Zhang, “Intelligent reflecting surface enhanced wireless
G(ν, νpr,1 + νq )F(τ, τpr,1
r,2
+ τqr,2 ). (41) network via joint active and passive beamforming,” IEEE Trans. Wireless
Commun., vol. 18, no. 11, pp. 5394-5409, Nov. 2019.
From (29) it is clear that we need hrw [k − k ′ , l − l′ ] in order to [5] R. Hadani et al., “Orthogonal time frequency space modulation,” Proc.
IEEE WCNC’2017, pp. 1-6, Mar. 2017.
find the input-output relation in the DD domain, and therefore [6] R. Hadani and A. Monk, “OTFS: a new generation of mod-
′
l−l′
hrw (ν, τ ) is evaluated at ν = k−k
N T and τ = M ∆f . Considering ulation addressing the challenges of 5G,” [Online] Available:
′ https://arxiv.org/abs/1802.02623, Feb. 2018.
(39) and evaluating at ν = k−k N T and substituting for νp and
r,1
[7] R. Hadani et al., “Orthogonal time frequency space modulation,” [On-
r,2
νq , we get line] Available: https://arxiv.org/abs/1808.00519v1, Aug. 2018.
[8] K. R. Murali and A. Chockalingam, “On OTFS modulation for high-
N −1 Doppler fading channels,” Proc. ITA, pp. 1-10, Feb. 2018.
k − k ′ r,1
 
′ r,1
2π
+βqr,2 ))n
X
G , νp + νqr,2 = e−j N (k−k −(βp [9] P. Raviteja, K. T. Phan, Y. Hong, and E. Viterbo, “Interference can-
NT n=0
cellation and iterative detection for orthogonal time frequency space
modulation,” IEEE Trans. Wireless Commun., vol. 17, no. 10, pp. 6501-
′ r,1
+βqr,2 ))
e−j2π(k−k −(βp −1 6515, Aug. 2018.
= 2π ′ r,1 r,2 [10] C. Huang, A. Zappone, G. C. Alexandropoulos, M. Debbah, and
e−j N (k−k −(βp +βq )) −1 C. Yuen, “Reconfigurable intelligent surfaces for energy efficiency in
wireless communication,” IEEE Trans. Wireless Commun., vol. 18, no.
(
′
N, [k − k − (βpr,1 + βqr,2 )]N = 0 8, pp. 4157-4170, Aug. 2019.
= (42)
0, otherwise. [11] S. Atapattu, T. A. Tsiftsis, R. Fan, P. Dharmawansa, G. Wang, and J.
Evans, “Reconfigurable intelligent surface assisted two–way communi-
cations: performance analysis and optimization,” IEEE Trans. Commun.,
l−l′
In a similar way, evaluating (40) at τ = M ∆f and substituting vol. 68, no. 10, pp. 6552-6567, Oct. 2020.
for τpr,1 and τqr,2 , we get [12] B. Zheng and R. Zhang, “Intelligent reflecting surface-enhanced OFDM:
channel estimation and reflection optimization,” IEEE Wireless Commun.
Lett., vol. 9, no. 4, pp. 518-522, Apr. 2020.
M −1
l − l′ r,1
  X 2π ′ r,1
+αr,2
[13] Y. Yang, B. Zheng, S. Zhang, and R. Zhang, “Intelligent reflecting
F , τ + τqr,2 = ej M (l−l −(αp q ))m
surface meets OFDM: protocol design and rate maximization,” IEEE
M ∆f p m=0 Trans. Commun., vol. 68, no. 7, pp. 4522-4535, Jul. 2020.
j2π(l−l′ −(αr,1 r,2 [14] B. Zheng, C. You, and R. Zhang, “Fast channel estimation for IRS-
e p +αq )) −1 assisted OFDM,” IEEE Wireless Commun. Lett., vol. 9, no. 3, pp. 580-
= 2π ′ r,1 r,2 584, Mar. 2021.
(ej M (l−l −(αp +αq )) −1 [15] G. D. Surabhi, R. M. Augustine, and A. Chockalingam, “On the diversity
′
M, [l − l − (αpr,1 + αqr,2 )]M = 0 of uncoded OTFS modulation in doubly-dispersive channels,” IEEE
= (43) Trans. Wireless Commun., vol. 18, no. 6, pp. 3049-3063, Jun. 2019.
0, otherwise. [16] P. Raviteja, K. T. Phan, and Y. Hong, “Embedded pilot-aided channel
estimation for OTFS in Delay–Doppler channels,” IEEE Trans. Veh.
Using (42) and (43) in (41), we obtain Tech., vol. 68, no. 5, pp. 4906-4917, May 2019.
[17] S. Tiwari, S. S. Das, and V. Rangamgari, “Low complexity LMMSE re-
P2 P1 ceiver for OTFS,” IEEE Commun. Lett., vol. 23, no. 12, pp. 2205–2209,
X r,2 r,2 X
Dec. 2019.
hrw (k − k ′ , l − l′ ) = gqr e−j2πνq τq
hrp [18] Evolved Universal Terrestrial Radio Access (E-UTRA); Base Station
q=1 p=1 (BS) Radio Transmission and Reception, Version 14.3.0, Release 14,
−j2πνpr,1 (τpr,1 +τqr,2 ) ′ document TS 36.104, 3GPP, Apr. 2017.
e N δ[[k − k − (βpr,1 + βqr,2 )]N ]
M δ[[l − l′ − (αpr,1 + αqr,2 )]M ]. (44)