This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LWC.2018.2890643, IEEE Wireless
Communications Letters
1

OTFS Performance on Static Multipath Channels

P. Raviteja, Emanuele Viterbo, and Yi Hong

Abstract—Orthogonal time frequency space (OTFS) modula- performance target and detection complexity. Finally, we apply
tion is known to achieve excellent error performance in delay– a message passing (MP) algorithm to OTFS/A-OFDM and
Doppler channels. However, its performance over static multipath show that it outperforms OFDM as well as A-OFDM with
channels has never been fully investigated. In this letter, we show
that, in static multipath channels, the system structure of OTFS is ZF and MMSE detectors in [1].
equivalent to the asymmetric orthogonal frequency division mul- Notations: We denote scalar, vector, and matrix by a, a,
tiplexing (A-OFDM), a scheme proposed in [1], bridging between and A, respectively. We let a(i) and A(i, j) represent the i th
cyclic prefix single carrier (CPSC) and traditional OFDM. We element of a and (i, j)th element of A. We denote the set
derive a condition on the parameters of OTFS to guarantee that of M × N dimensional matrices with each entry from the
all the transmitted symbols experience uniform channel gains,
as in CPSC. Finally, we apply a low-complexity message passing complex plane by C M×N . Let A = circ[A0, · · · , A N −1 ] and
detection to OTFS/A-OFDM and show a significant performance diag[A0, · · · , A N −1 ] ∈ C M N ×M N represent the block circulant
improvement over ZF and MMSE detection originally proposed matrix with first column block as {A0, · · · , A N −1 } ∈ C M×M ,
for A-OFDM. and the block diagonal matrix with {A0, · · · , A N −1 } as diago-
Index Terms—OTFS, OFDM, static multipath channels, mes- nal blocks, respectively. We let superscript H denote Hermitian
sage passing transposition, the ⊗ operation denote Kronecker product, and
the vec(·) operation denote the column vectorization of an
I. I NTRODUCTION M × N (complex) matrix into an M N × 1 (complex) column
vector. Finally, we let Fn = { √1n e2π jkl/n }k,l=0
n−1 and FH be the
n
Recently proposed orthogonal time frequency space (OTFS)
modulation offers great potential to handle multiple high n-point DFT and the IDFT matrices, and the term I M be a
Doppler shifts in time-varying multipath wireless channels M-dimensional identity matrix.
[3]–[5]. The key idea of OTFS is to transmit information sym-
bols in the delay–Doppler domain to better resolve the delay II. OTFS IN S TATIC M ULTIPATH C HANNELS
and Doppler multiple paths of a time-varying wireless channel. We consider an OTFS system with single antenna trans-
The information symbols placed in delay-Doppler domain mitter and receiver over static multipath channels, i.e., the
can be transformed to the standard time-frequency domain, channel consists of P zero-Doppler multipaths with the i th
used in traditional modulation schemes such as orthogonal path delay denoted by τi , for i = 1, 2, · · · , P. We assume that
frequency division multiplexing (OFDM). Such transformation a total of Nc = M N symbols are transmitted in an OTFS
is realized via a set of two-dimensional (2D) bi-orthogonal frame of duration Nc Ts , where Ts is the sampling interval.
basis functions shaped by ideal pulse-shaping waveforms. As Let τmax = (L − 1)Ts denote the maximum delay of an L-tap
a result, all information symbols experience a constant flat channel. The static multipath channel is represented by the L
fading equivalent channel [3]. tap coefficients [h0, h1, · · · , h L−1 ], where only P elements are
Since the first paper on OTFS in [3], a number of system non-zero.
improvements were proposed in [5]–[18]. All these works Let x = vec(X) ∈ C Nc ×1 denote one OTFS frame contain-
have studied OTFS over time-variant (high Doppler) wireless ing Nc transmitted information symbols, each with average
channels, yet the complete analysis of OTFS over time- energy Es , where the matrix X ∈ C M×N represents the two-
invariant (static) multipath channels has never been explored. dimensional information symbols transmitted in the delay-
In this letter, we study OTFS over static multipath channels Doppler plane. The transmitted time domain signal in OTFS
and reveal that the system structure of OTFS is equivalent can be obtained by first applying the (2D) ISFFT on X
to the asymmetric orthogonal frequency division multiplexing followed by Heisenberg transform [3]. Assuming rectangular
(A-OFDM), a scheme proposed in [1] that generalizes OFDM transmit waveform, the output of the Heisenberg transform can
and cyclic prefix single carrier (CPSC) by exploiting a layered be written as [8]
FFT structure. Next, we derive a necessary and sufficient
S = FHM (F M XFHN ) = XFHN (1)
condition on the number of subcarriers in OTFS to guarantee
that all the transmitted symbols experience uniform channel The transmitted time domain signal can be generated by
gains, as in CPSC (a special case of OTFS/A-OFDM). We also column-wise vectorization of S:
show that OTFS offers a tradeoff between spectral efficiency
s = vec(S) = (FHN ⊗ I M )x (2)
and maximum peak-to-average power ratio (PAPR), for a given
We assume a CP of length (L − 1) is added to s before trans-
The authors are with the Department of Electrical and Computer Sys-
tems Engineering„ Monash University, Clayton, VIC 3800, Australia. Email: mission. The received signal in time domain, after discarding
{raviteja.patchava, emanuele.viterbo, yi.hong,}@monash.edu. This work was the CP, can be written as
supported by the Australian Research Council through the Discovery Project
under Grant DP160100528. r = Hs + w, (3)

2162-2337 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LWC.2018.2890643, IEEE Wireless
Communications Letters
2

0 0 0 0 0 0
M 1 1 1 M 1
IFFT FFT
1 1
(N − 1)M (N − 1)M
S/P CP P/S Channel S/P CP P/S Detector
M −1 M −1
2M − 1 IFFT 2M − 1 FFT
Nc − 2 Nc + L − 3 Nc + L − 3 Nc − 2
M Nc − 1 Nc + L − 2 Nc + L − 2 M Nc − 1
NM − 1 NM − 1

Fig. 1. OTFS/A-OFDM for static multipath channels

where H = circ[h0, h1, · · · , h L−1, 0, · · · , 0] ∈ C Nc ×Nc is the Therefore, from (1), (4), (6), and (7), we can conclude that,
circulant matrix, and w ∈ C Nc ×1 is the i.i.d. Gaussian noise under static multipath channels, OTFS and A-OFDM systems
vector with the i th entry, wi ∼ CN(0, σ 2 ). share the same transmitter and receiver structure (see Fig. 1).
At the receiver, the received signal r is devectorized into an Note that OTFS/A-OFDM uses M copies of an N-point
M × N matrix R, followed by a Wigner transform as well as IFFT and FFT at transmitter and receiver, respectively. Com-
a SFFT, yielding paring to a conventional OFDM with Nc = M N sub-
carriers, the complexity of OTFS/A-OFDM reduces from
Y = FHM (F M R)F N = RF N (4) M N log2 (M N) to M N log2 N complex multiplications and the
maximum PAPR reduces from M N to N.
Finally, the input–output relation of OTFS in the information
domain can be obtained by column-wise vectorization of (4): III. D ETECTION OF OTFS/A-OFDM
y = vec(Y) = (F N ⊗ I M )r In this section, we first review traditional ZF and MMSE
detections, originally proposed for A-OFDM in [1], [2] and
= (F N ⊗ I M )H(FHN ⊗ I M )x + w
also applicable for OTFS in static multipath channels. Further,
e
= Heff x + w
e (5) we derive a necessary and sufficient condition on the number
of subcarriers in OTFS to guarantee that all the transmitted
where Heff = (F N ⊗ I M )H(FHN ⊗ I M ) is the effective channel
symbols experience uniform channel gains, as in CPSC (a
e = (F N ⊗ I M )w
matrix. Since (F N ⊗ I M ) is a unitary matrix, w
special case of OTFS/A-OFDM). We then apply the low-
preserves the same statistical properties of w.
complexity message passing (MP) detection algorithm for
Let us consider the following two special cases for OTFS
OTFS (see [10]) with improved error performance over ZF
over static multipath channels.
and MMSE detections.
1) If M = 1 then (4) reduces to y = F N HFHN x + w e, i.e., a
conventional N-subcarrier OFDM system, when a CP is A. ZF detection
added to x.
It was identified in [1, Theorem 1] for A-OFDM that
2) If N = 1 then (4) reduces to y = Hx + w e, i.e., a
the effective channel matrix has a block diagonal structure,
conventional CPSC system.
Heff = diag[H̆0, H̆1, · · · , H̆ N −1 ] with H̆0, · · · , H̆ N −1 ∈ C M×M .
This shows that OTFS can be seen as a generalization of both Further, each H̆n , for n = 0, · · · , N − 1, can be diagonalized as
OFDM and CPSC systems. H̆n = FHM Dn F M . Therefore, from (5), received symbols can
be simplified as
A. Relation between OTFS and A-OFDM yn = H̆n xn + w
en (8)
Now we are ready to reveal the relation between OTFS and ŷn = F M yn = Dn F M xn + F M w
en (9)
A-OFDM systems in static multipath channels. for n = 0, 1, · · · , N −1. Here, yn, xn, and w
e n , are the subvectors
Specifically, at the transmitter of an A-OFDM system [1, formed by taking nM to (n+1)M −1 elements from y, x, and w e,
Fig. 1], the input data of length Nc is arranged into an M × N respectively. Hence, the estimated symbols after ZF detection
matrix and a N-point IFFT is then applied to each row. The can be written as
transmitted outputs after IFFT are read out column-wise and
can be written as x̂n = FHM D−1
n ŷn (10)

S̆ = XFHN , (6) B. MMSE detection

From (9), the estimated symbols after MMSE detection can
which yields the transmitted time domain signal, s̆ = (FHN ⊗
be written as
I M )x.  −1
σ2

At the receiver of A-OFDM, the Nc received signals are
x̂n = F M Dn Dn Dn +
H H H
I M ŷn (11)
converted to a M × N matrix and a N-point FFT is applied Es
to each row. Similar to the transmitter, the receiver output of Note that the complexity of ZF and MMSE detectors is of
A-OFDM can be written as the order of O(M log2 M). However, ZF and MMSE linear
detectors do not fully exploit the available system diversity.
Y̆ = RF N
Finally, these detection methods do not take advantage of the
y̆ = (F N ⊗ I M )r̆ (7) sparsity of H̆n .

2162-2337 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LWC.2018.2890643, IEEE Wireless
Communications Letters
3

C. Message passing detection 100

N c = 1024, M = 128, N = 8, L = 72, P=1
For OTFS in static multipath channels, we first estab- 16-QAM
P=2
P=4
10-1
lish the relation between Heff and H using the following P=8
AWGN
lemma, which is based upon the observation that H = 10-2
circ[H0, H1, · · · , H N −1 ], is a block circulant matrix, where

BER
Hn, 0 ≤ n ≤ N − 1, are M × M submatrix. 10-3

Lemma 1: H̆n (i, j), 0 ≤ i, j ≤ N − 1, is equal to the nth

10-4
element in the FFT of u(i, j) , [H0 (i, j), · · · , H N −1 (i, j)].
Proof: Since H is a block circulant matrix of N blocks of 10-5
15 20 25 30 35
size M × M, it can be block-diagonalized using (F N ⊗ I M ) and SNR in dB

(FHN ⊗ I M ) [8], [19], and the result follows from (5).  Fig. 2. BER of OTFS for different P with Nc = 1024, M = 128, N = 8, L =
Next, using Lemma 1, we prove the following theorem on 72, and 16-QAM
the minimum value of M in OTFS to guarantee that all the Detector: Since the OTFS input–output relation for static
transmitted symbols experience uniform channel gains, as in multipath channels is sparse (13), we propose to use the
CPSC. MP algorithm presented in our earlier works [9], [10]. The
Theorem 1: The input–output relation in an OTFS system complexity of MP algorithm for each block is O(niter M PQ),
of N M transmitted symbols is equivalent to N parallel CPSCs where niter is the number of iterations in MP and Q is the
of length M with the identical time-domain channel, except modulation alphabet size. In general, even the value of L is
for an additional phase shift, if and only if M ≥ L. large, but the value of P can be small, for example, in LTE
Proof: vehicular (EVA) channel model, L = 72 and P = 9.
(If) – For M ≥ L, the entries of the vectors u(i, j) , 0 ≤ i, j ≤ Remark 1: OTFS has the same performance and detection
M − 1, become complexity as N consecutive blocks of CPSCs of length M, but
has higher spectral efficiency, since OTFS only requires one
 [h , 0, · · · , 0] if 0 ≤ (i − j) ≤ (L − 1)
 i−j CP, whereas CPSC requires N CPs. On the other hand, OTFS


u(i, j) = [0, h L+(i−j), · · · , 0] if − (L − 1) ≤ (i − j) < 0

has a higher PAPR = N than CPSC (PAPR = 1). Therefore,
 [0, 0, · · · , 0]

otherwise OTFS offers a tradeoff between spectral efficiency and PAPR.


Taking the FFT’s of the above we have D. Channel estimation
We now propose an embedded pilot channel estimation
 [hi−j , hi−j , · · · , hi−j ] if 0 ≤ (i − j) ≤ (L − 1)
method to estimate the P non-zero channel coefficients for



 0 1 N −1
−j2π N
, e−j2π N , · · · , e−j2π N ] OTFS with M ≥ L. In this method, we allocate first M

= h L+(i−j) · [e


v(i, j)

 if − (L − 1) ≤ (i − j) < 0 symbols of x as a header and the remaining M(N −1) symbols
for data. In the header, we transmit a known pilot symbol x p


 [0, 0, · · · , 0]

otherwise
 followed by M − 1 zeros. Therefore, from (8) and (13), y0
(12)
reduces to
According to Lemma 1 and (12), we obtain
y0 (m) = hm x p + w
e0 (m) for 0 ≤ m ≤ M − 1 (14)
 h0 0 · · · h1 e−j2π Nn 
 h1 h0 · · · h2 e−j2π Nn 
  and hm can be estimated using the threshold method proposed
H̆n =  . . .. in [16], [17]. Note that the pilot power |x p | 2 can be M
.. ...
(13)
 
 ..

 . 
 times higher than the data signal power without increasing
 0 · · · h1 h0  the average transmitted power.
 
Hence, from (8) and (13), we can conclude that OTFS input- Also note that OTFS enables simple correction of any carrier
output relation is equivalent to N parallel CPSCs of length M frequency offset (CFO). This is due to the fact that the CFO
over identical channels, except for an additional phase shift effect is equivalent to applying a single Doppler shift to all
n
e−j2π N . Moreover, there are only P non-zero entries in each the paths in the OTFS channel [8], [9]. This can be easily
row and column of H̆n . detected and corrected in the channel estimation using a pilot
(Only if) – For M < L, we can easily see that u(i, j) has signal and thus enables to compensate for much larger CFOs
at least two non-zero entries for some 0 ≤ i, j ≤ M − 1. For than OFDM.
example, for M = 2, the value of H̆n becomes
n 
IV. N UMERICAL RESULTS AND DISCUSSIONS
v (n) v(1,0) (n)e−j2π N

H̆n = (0,0) In this section, we compare BER of OTFS with OFDM and
v(1,0) (n) v(0,0) (n)
CPSC for different P and M. In all simulations, we consider
where, v(0,0) and v(1,0) are the FFT’s of u(0,0) = [h0, h2, · · · , 0] Nc = 1024 and 16-QAM modulation alphabet (Q = 16).
and u(1,0) = [h1, h3, · · · , 0], respectively. In order to obtain BER, we consider 105 different channel
Therefore, due to FFT operation, the entries of H̆n differ in realizations in Monte-Carlo simulations.
both amplitude and phase for each n, and lower gain channels Fig. 2 illustrates the BER performance of OTFS for different
effect the overall system performance (similar to OFDM).  P = 1, 2, 4, and 8 with M = 128, N = 8, L = 72. Note
that we consider M > L in the figure so that all transmitted

2162-2337 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LWC.2018.2890643, IEEE Wireless
Communications Letters
4

100 M = 1, OFDM
to A-OFDM, a scheme proposed in [1]. Further, we have
N c = 1024, L = 72, 16-QAM M=2
M=4 derived a necessary and sufficient condition (M ≥ L) in
M = 128, 256, 1024; MP
10-1 CPSC, MMSE OTFS to guarantee that all the transmitted symbols experi-
M = 128; A-OFDM, ZF
M = 128; A-OFDM, MMSE
AWGN
ence uniform channel gains, as in CPSC. We apply a low-
complexity MP detection algorithm to OTFS and show that
BER

10-2
OTFS with MP detection performs similarly to CPSC, but
better than OTFS/A-OFDM with ZF and MMSE detections.
10-3
We also show that the performance of OTFS using channel
estimation with embedded pilots approaches the performance
10-4
15 20 25 30 35 40 with ideal channel state information at the receiver.
SNR in dB

Fig. 3. BER of OTFS for different M with Nc = 1024, L = 72, and 16-QAM
R EFERENCES
10-1
[1] J. Zhang, A. D. S. Jayalath, and Y. Chen, “Asymmetric OFDM systems
based on layered FFT structure,” IEEE Signal Proces. Lett., vol. 14, no.
11, pp. 812-815, Nov. 2007.
10-2 [2] L. Luo, J. Zhang and Z. Shi, “BER analysis for asymmetric OFDM
systems,” in Proc. 2008 IEEE Global Telecommunications Conference
BER

N c = 1024, M = 128, L = 72,

16-QAM
(GLOBECOM), New Orleans, LO, USA, 2008.
[3] R. Hadani, S. Rakib, M. Tsatsanis, A. Monk, A. J. Goldsmith, A.
10-3 SNRp = 30 dB
F. Molisch, and R. Calderbank, “Orthogonal time frequency space
SNRp = 35 dB
modulation,” in Proc. IEEE Wireless Communications and Networking
SNRp = 40 dB
Ideal
Conference (WCNC), San Francisco, CA, USA, March 2017.
10-4 [4] R. Hadani, S. Rakib, S. Kons, M. Tsatsanis, A. Monk, C. Ibars, J.
15 16 17 18 19 20 21 22 23 24
SNR in dB Delfeld, Y. Hebron, A. J. Goldsmith, A.F. Molisch, and R. Calder-
bank, “Orthogonal time frequency space modulation,” Available online:
Fig. 4. BER of OTFS for different SNRp with Nc = 1024, M = 128, L = 72, https://arxiv.org/pdf/1808.00519.pdf.
and 16-QAM [5] R. Hadani and A. Monk, “OTFS: A new generation of
symbols experience equal channel gains. We assume that the modulation addressing the challenges of 5G,” OTFS Physics
P paths are uniformly distributed in L, for example, when White Paper, Cohere Technologies, 7 Feb. 2018, available online:
https://arxiv.org/ftp/arxiv/papers/1802/1802.02623.pdf.
P = 4, we assume only h0, h23, h46 , and h69 have non-zero [6] A. Farhang, A. RezazadehReyhani, L. E. Doyle, and B. Farhang-
coefficients. Moreover, if P = 1 then it reduces to a flat fading Boroujeny, “Low complexity modem structure for OFDM-based orthog-
channel. The channel coefficients of the P paths are generated onal time frequency space modulation,” IEEE Wireless Commun. Lett.,
vol. 7, no. 3, pp. 344-347, June 2018.
using i.i.d. complex Gaussian distribution, CN(0, 1/P). Here [7] A. RezazadehReyhani, A. Farhang, M. Ji, R. R. Chen, and B. Farhang-
we adopt MP detection algorithm and assume perfect channel Boroujeny, “Analysis of discrete-time MIMO OFDM-based orthogonal
state information (CSI) is available at the receiver. We observe time frequency space modulation,” in Proc. 2018 IEEE International
Conference on Communications (ICC), Kansas City, MO, USA, 2018.
from Fig. 2 that as P increases, the BER slope improves. This [8] P. Raviteja, Y. Hong, E. Viterbo, and E. Biglieri, “Practical pulse-
diversity advantage is due to the fact that each information shaping waveforms for reduced-cyclic-prefix OTFS,” accepted in IEEE
symbol experience the channel gains from P paths. Transactions on Vehicular Technology, Engine, North America, 2018.
[9] P. Raviteja, et al., “Interference cancellation and iterative detection for
In Fig. 3, we present the performance of OTFS for different orthogonal time frequency space modulation,” IEEE Trans. Wireless
M with L = 72 and P = 9. We consider LTE EVA channel Commun., vol. 17, no. 10, pp. 6501-6515, Oct. 2018.
model for generating channel tap coefficients (hl ) and assume [10] P. Raviteja, et al., “Low-complexity iterative detection for orthogonal
time frequency space modulation,” in Proc. IEEE Wireless Communica-
perfect CSI is available at the receiver.. We observe that, for tions and Networking Conference (WCNC), Barcelona, April 2018.
M = 128, 256, (M > L), the performance of OTFS using [11] L. Li, et al., “A simple two-stage equalizer with simplified orthogonal
MP detection improves with M and achieves the performance time frequency space modulation over rapidly time-varying channels,”
available online: https://arxiv.org/abs/1709.02505.
similar to CPSC of M = 1024, which agrees to Theorem 1. [12] T. Zemen, M. Hofer, and D. Loeschenbrand, “Low-complexity equal-
Moreover, OTFS using MP detection outperforms OTFS/A- ization for orthogonal time and frequency signaling (OTFS),” available
OFDM using MMSE detection by approximately 5 dB, and online: https://arxiv.org/pdf/1710.09916.pdf.
[13] T. Zemen, M. Hofer, D. Loeschenbrand, and C. Pacher, “Iterative
OFDM by 15 dB 1 . This is due to the fact that MP detection detection for orthogonal precoding in doubly selective channels,” available
is approximate to maximum likelihood detection and better online: https://arxiv.org/pdf/1710.09912.pdf.
exploits the full channel diversity, when compared to MMSE. [14] K. R. Murali and A. Chockalingam, “On OTFS modulation for high-
Doppler fading channels,” in Proc. ITA’2018, San Diego, Feb. 2018.
Fig. 4 compares the BER of OTFS for different pilot SNRs, [15] M. K. Ramachandran and A. Chockalingam, “MIMO-OTFS in high-
SNRp = |x p | 2 /σ 2 , with M = 128, N = 8 and L = 72. We Doppler fading channels: signal detection and channel estimation,” avail-
adopt a threshold of 3σ. We observe that BER performance able online: https://arxiv.org/abs/1805.02209.
[16] P. Raviteja, K. T. Phan, Y. Hong, and E. Viterbo, “Embedded delay-
improves as SNRp increases and approaches the performance Doppler channel estimation for orthogonal time frequency space modu-
of the perfect CSI (ideal) case for SNRp = 40 dB. lation,” in Proc. IEEE VTC2018-fall, Chicago, USA, August 2018.
[17] P. Raviteja, K. T. Phan, and Y. Hong, “Embedded pilot-aided channel es-
V. C ONCLUSION timation for OTFS in delay-Doppler channels,” submitted for publication
in IEEE Trans. Veh. Technol.
In this letter, we have studied an M × N OTFS in static [18] A. Nimr, M. Chafii, M. Matthe, and G. Fettweis, “Extended
multipath channels and showed that its structure is equivalent GFDM framework: OTFS and GFDM comparison,” available online:
https://arxiv.org/pdf/1808.01161.pdf.
1 We observe very similar behavior of OTFS for other modulation schemes [19] P. J. Davis, Circulant Matrices. New York: John Wiley & Sons, 1970.
(BPSK, 4-QAM), which are not reported due to lack of space.

2162-2337 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.