Received April 30, 2021, accepted May 8, 2021, date of publication May 12, 2021, date of current version

May 24, 2021.

Digital Object Identifier 10.1109/ACCESS.2021.3079429

Downlink Synchronization for OTFS-Based

Cellular Systems in High Doppler Environments
MOHAMMED SAQUIB KHAN 1 , (Student Member, IEEE), YEONG JUN KIM 2,

QASIM SULTAN 1 , JINGON JOUNG 1 , (Senior Member, IEEE),

AND YONG SOO CHO 1 , (Senior Member, IEEE)
1 Department of Electrical and Electronics Engineering, Chung-Ang University, Seoul 06974, South Korea
2 LG Electronics, Seoul 06772, South Korea
Corresponding author: Yong Soo Cho ([email protected])
This work was supported by the National Research Foundation of Korea (NRF) Grant through the Korean Government [Ministry of
Science and ICT (MSIT)] under Grant 2021R1A4A2001316 and Grant 2018R1A2B2002621.

ABSTRACT This paper proposes a downlink synchronization and cell identity (CID) estimation technique
for cellular systems based on orthogonal time-frequency space (OTFS). In the proposed technique, each
base station (BS) transmits one detection preamble (DP) and two OTFS symbols (pilot and secondary-CID
(SCID) signal (SS)). The DP based on a linear frequency-modulated (LFM) waveform, called proposed-LFM
(pLFM), carries the primary-CID (PCID) for symbol timing (ST) synchronization, whereas the SS based
on the Zadoff-Chu (ZC) sequence mapped in the delay-Doppler domain carries the SCID. The pLFM is
obtained by discretizing the LFM waveform, increasing the frequency sweeping parameter beyond the limit
of the operational bandwidth, and incorporating the PCID into the LFM waveform. The autocorrelation
and cross-correlation functions of the pLFM are analyzed to examine its correlation properties under the
influence of a high Doppler shift. To reduce the undesirable correlation properties (side peak and time
ambiguity), a receiver processing (RP) scheme is developed for the pLFM. It is found that the pLFM with RP
is suitable for the DP design because it can provide an accurate ST in high-mobility scenarios. In addition,
the influence of OTFS modulation on the ZC sequence is derived and a low-complexity detection algorithm
based on message passing is applied to detect SCID at the user equipment. The simulation results demonstrate
that the proposed downlink synchronization and CID estimation techniques are suitable for OTFS-based
cellular systems in high-Doppler environments.

INDEX TERMS Downlink synchronization, Doppler, high-mobility, OTFS, pLFM.

I. INTRODUCTION Moreover, the CFO is typically caused by the oscillator

Future wireless communication systems are expected to mismatch between the Tx and Rx or Doppler spread (dou-
accommodate numerous diverse requirements and usage sce- bly selective/dispersive) of time-varying channels owing to
narios, including high-mobility scenarios with a maximum the swift movement of Tx/Rx [4]. Although conventional
velocity of 500 km/h (high-speed trains) [1], [2]. Currently, transceivers can compensate for the Doppler shift in an ideal
the dominant modulation technique of the fourth-generation environment, the OFDM performance significantly deterio-
and fifth-generation New Radio (5G NR) is orthogonal rates owing to the multiple reflections with corresponding
frequency division multiplexing (OFDM), which provides CFOs in multipath environments. As the velocity increases,
broadband services to a large population [3]. However, this effect worsens, thus restricting the use of OFDM in
OFDM requires accurate frequency synchronization between high-mobility scenarios.
the transmitter (Tx) and receiver (Rx). In high-mobility In order to eliminate or mitigate the impact of ICI, some
scenarios, it may encounter significant intercarrier inter- modifications to conventional OFDM transceivers have been
ference (ICI) and cause carrier frequency offset (CFO). proposed. For instance, linear [5] and non-linear [6] equal-
izations at Rx, pulse shaping [7] and polynomial cancelation
The associate editor coordinating the review of this manuscript and coding [8] at Tx, and a multiple access scheme based on
approving it for publication was Yiming Huo . channel-independent block spreading [9] using both Tx and

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
VOLUME 9, 2021 73575
 M. S. Khan et al.: Downlink Synchronization for OTFS-Based Cellular Systems

Rx processing have been proposed. In 5G NR [10], a solu- the time-domain and then the residual ISI was minimized in
tion based on variable sub-carrier spacing, called numerol- the delay-Doppler domain. Thus, OTFS is a promising mod-
ogy, is adopted to overcome the high Doppler effect in the ulation scheme for next-generation communication systems
above-6 GHz bands. Furthermore, Doppler-insensitive or tol- in high-mobility scenarios.
erant waveforms/sequences for cell selection [11], random To use OTFS in a cellular system, initial downlink synchro-
access [12], and positioning [13] for OFDM-based systems nization and cell searching are necessary. However, it is not
have been proposed. Alternatively, some modulation tech- easy to perform symbol timing (ST) and cell identity (CID)
niques, such as the frequency oversampling technique [14] detection with the received preamble in a high-Doppler envi-
and vector OFDM [15], have been proposed. ronment. A high Doppler shift produces a large ambiguity in
Recently, a novel two-dimensional (2D) modulation tech- the timing estimation and CID detection. The delay-Doppler
nique driven by a delay-Doppler channel, called orthogo- channel estimation, MIMO, multiple access, detection, and
nal time-frequency space (OTFS), has been proposed [16]. equalization techniques of OTFS described in the abovemen-
Instead of time-frequency processing as in OFDM, chan- tioned studies considered an ideal synchronization condition
nel representation and information symbol multiplexing in or synchronization with the conventional OFDM-based tech-
OTFS are performed in the delay-Doppler domain, which is nique to detect OTFS information symbols. In 5G NR [10],
a 2D Fourier dual of the time-frequency channel. The key the user equipment (UE) is used to detect the primary
advantage of delay-Doppler domain processing in OTFS is synchronization signal (PSS) and secondary synchronization
that it represents the definite geometry of a wireless channel signal (SSS) transmitted from the base station (BS) in a syn-
and transforms the well-known time-varying time-frequency chronization signal block for ST and CID detection. However,
channel into a 2D time-invariant delay-Doppler channel. the conventional OFDM-based NR system is sensitive to high
Owing to the advantages of OTFS over OFDM, several Doppler and fails to provide accurate timing synchronization
studies on channel estimation, detection, and multiple-input in high-mobility scenarios. By increasing the subcarrier spac-
multiple-output (MIMO) for OTFS have been proposed. The ing, the Doppler sensitivity can be reduced but at the cost of
authors in [16], [17] showed that the packet error rates, increased bandwidth (BW). In addition, additional blocks for
peak-to-average power ratio, and MIMO of OTFS are better OFDM-based synchronization and cell searching are required
in high-mobility scenarios than in OFDM-based systems. for the initial downlink synchronization and cell searching for
A multiple access mechanism for an OTFS-based communi- OTFS systems. To the best of our knowledge, the ST and CID
cation system by assigning non-overlapping resource blocks estimation for OTFS-based cellular systems have not yet been
was proposed in [18], making the system multi-user interfer- explored.
ence free. In [19], a random access preamble (RAP) design An energy-based method proposed in [19] for random
technique was proposed to achieve uplink timing synchro- access with OTFS can be considered for ST and CID esti-
nization and estimate the timing advance for OTFS-based mation. In this method, RAPs are allocated to nonover-
cellular systems. In [20], a nonorthogonal multiple access lapping contiguous rectangular delay-Doppler resource
(NOMA)-based communication system incorporating OTFS element (DDRE) groups in the delay-Doppler domain. For
modulation for a heterogeneous mobility scenario was pro- each RAP, the energy is transmitted to only one DDRE in
posed. Here, the users with different mobility patterns are its group. Each DDRE group spans the entire delay-domain.
grouped for implementing NOMA, and a suitable perfor- If this method is used in the downlink, the required frame size
mance is achieved in both uplink and downlink by minimiz- for cell search will be very large because of the requirement
ing the intersymbol interference (ISI). of generating a large number of CIDs (1008 in NR [10])
In [21]–[27], estimation techniques for the delay-Doppler in the cellular systems. Moreover, unlike the UE, the BS
channel were investigated because the delay-Doppler channel transmits signals (synchronization signal, pilot, and user data)
estimation is necessary at the Rx to perform OTFS demodula- continuously to serve multiple UEs in a cell. The UEs at
tion. For single-input single-output OTFS systems, the most the cell boundary may receive signals from adjacent BSs.
straightforward method that can be used to estimate the Thus, the energy received from the BSs at a UE is normally
delay-Doppler channel from the received signal is a thresh- high, which makes the energy detection approach in downlink
old method [21] or training pilots, such as pseudo-random synchronization inefficient.
noise sequences [22]. In [22], [23], channel estimation was This paper proposes a downlink synchronization and CID
performed in the time-frequency domain, and it was found estimation technique for OTFS-based cellular systems in a
that this estimation incurs higher implementation complex- high-Doppler environment. Here, each BS transmits a detec-
ity than the delay-Doppler channel estimation performed in tion preamble (DP) based on a linear frequency-modulated
[24], [25]. For massive MIMO-OTFS systems, the authors (LFM) waveform and two OTFS symbols, which comprise
in [26] proposed an uplink-aided channel estimation algo- two delay-Doppler resource blocks (DDRBs): a pilot DDRB
rithm. In [27], after proving that the 3D channel is sparse, and a secondary-CID (SCID) signal (SS) DDRB based on
the sparsity was exploited to estimate the channel as a sparse Zadoff-Chu (ZC) sequence. The DP carries the primary-CID
recovery problem. In addition, a two-stage equalization algo- (PCID), which is also used for ST estimation, whereas the SS
rithm was proposed in [28], which first equalized the ISI in carries the SCID, like PSS and SSS in NR. By combining

73576 VOLUME 9, 2021

M. S. Khan et al.: Downlink Synchronization for OTFS-Based Cellular Systems

the PCID and SCID obtained from DP and SS, respec- 3 such that X [n, m] element of X denotes the nth symbol
tively, the UE derives the CID. The DP must be designed transmitted on mth subcarrier for 0 6 n < N , 0 6 m < M .
such that the ST and PCID are correctly detected in a Thus, the complex baseband OFDM signal is given as
high-Doppler environment without any prior information N −1 M −1
1 XX
(noncoherent detection), such as the PSS in NR. In the pro- s (t) = X [n, m] ej2πm1f (t−nT ) (1)
posed technique, the LFM waveform is used for the DP NM
n=0 m=0
design because the DP should be detected in a high-Doppler
Practically, (1) is implemented in the discrete time-domain
environment. The LFM waveforms are widely used for target
by inverse discrete Fourier transform [29]. Then, s(t) in
detection in high-Doppler environments such as RADAR and
(1) passes over a wireless multipath channel comprising P
SONAR systems because the LFM waveform is Doppler-
discrete propagation paths. In most cases where the UE
insensitive. However, the location of its correlation peak is
and/or scatters are immobile or have low velocity, the OFDM
shifted depending on the magnitude of the Doppler shift.
system is considered linear time-invariant (LTI). When the
Thus, the DP designed with the conventional LFM (cLFM)
transmitted signal is received by passing through multiple
waveform can be successfully detected in a high-Doppler
scatters, it is temporally smeared out. However, if the channel
environment but cannot produce the correct timing. In this
impulse response (CIR) is time-invariant, a one-dimensional
paper, a proposed-LFM (pLFM)-based DP design technique
delay-domain CIR, h(τ ), is sufficient to characterize the
and receiver processing technique is presented to provide
time-dispersive channel (or frequency-selective channel).
accurate ST estimation and PCID detection in a high-Doppler
In systems where the UE is highly mobile and oper-
environment. The autocorrelation and cross-correlation func-
ates at a higher carrier frequency (such as millimeter-
tions of the pLFM-based DP are analyzed to examine its
wave), the assumption of LTI CIRs may no longer be
correlation properties under the influence of a high Doppler
valid. Thus, several previous studies have adopted a linear
shift. Subsequently, the effect of OTFS modulation on the ZC
time-variant (LTV) channel model [30] for high-mobility
sequence is derived for the SS design. In addition, a detection
scenarios. Because of the presence of the Doppler effect
algorithm based on message passing (MP) is introduced to
in high-mobility scenarios, the LTV channels induce a fre-
coherently detect SSs at the UE in the presence of multicell
quency shift, which causes the signal spectrum to be smeared
interference and noise. Finally, the performance of the pro-
(i.e., frequency-dispersive). In fact, as the Doppler effect
posed downlink synchronization technique is examined with
and dispersive multipath propagation coexist in the LTV
the detection probability in low- and high-mobility cellular
channels, they are usually doubly selective. Therefore, in a
environments and compared with the results obtained using
high-mobility environment, the received signal at the UE can
the OFDM system.
be expressed as
The remainder of this paper is organized as follows.
In Section II, the system model and concept of the proposed Z∞
downlink synchronization technique for OTFS-based cellular r (t) = h (t, τ ) s (t − τ ) dτ + w (t) , (2)
systems are described. In Section III-A, the pLFM-based DP −∞
design technique and receiver processing technique are pro-
where h (t, τ ) is the LTV time-delay CIR:
posed to provide accurate ST estimation and PCID detection
in a high-Doppler environment. In Section III-B, the effect P
X
of OTFS modulation on the ZC sequence is derived for the h (t, τ ) = βi e−j2πvi t δ (τ − τi (t)), (3)
SS design. In addition, an MP-based detection algorithm is i=1
described to coherently detect SSs at the UE. In Section IV, P, βi , vi , τi , δ (·), and w (t) denote the number of resolv-
the performance of the proposed technique is evaluated using able paths, complex gain, Doppler (relative to the carrier
a simple model of an OTFS-based cellular system and com- frequency), delay of ith path, impulse function, and additive
pared with the conventional OFDM system. The conclusions white Gaussian noise (AWGN), respectively. In an OFDM
are presented in Section V. based system, the LTV channel can be represented in the
time-delay domain, h (t, τ ), or equivalently time-frequency
II. PRELIMINARIES domain, H (t, f ) [30]. The time-frequency channel, H (t, f ),
In an OFDM system, the synchronization signal is can be obtained by taking the Fourier transform (FT) of the
mapped in the time-frequency resource grid where time-delay domain channel, h (t, τ ), along τ as follows:
the time-domain is partitioned into N symbols and Z∞
the frequency-domain is partitioned into M subcarriers.
H (t, f ) = h (t, τ ) e−j2πf τ dτ (4)
The entire time-frequency plane is subdivided into NM
resource elements, where each time-frequency resource −∞
element (TFRE) is T sec and 1f Hz in width along Although the time-delay and time-frequency domain repre-
the time-domain and frequency-domain, respectively, sentations of LTV channels have been extensively used under
i.e., 3 = {(nT , m1f ) , 0 6 n < N , 0 6 m < M }. Let X ∈ high-mobility scenarios, they have limited support and are
CN ×M denote a packet of complex symbols mapped on characterized by a maximum delay spread (∝ 1/coherence

VOLUME 9, 2021 73577

 M. S. Khan et al.: Downlink Synchronization for OTFS-Based Cellular Systems

time) and maximum Doppler spread (∝ 1/coherence BW).

Owing to the high mobility, the channel in the time-delay or
time-frequency domain changes rapidly, which makes chan-
nel estimation difficult and incurs huge overhead. There-
fore, LTV channels are physically meaningful and intuitively
characterized in terms of time delays and Doppler shifts.
By expressing the FT of the time-delay channel, h (t, τ ),
along t, the delay-Doppler channel can be obtained as
Z∞
h (τ, v) = h (t, τ ) e−j2πvt dt (5)
−∞

The delay-Doppler domain representation of the channel cap-

tures the actual geometry of the channel by characterizing
the reflectivity of scatterers associated with all paths having
an independent delay value (with respect to the scatterer’s
relative distance) and Doppler value (with respect to the
scatterer’s relative velocity). FIGURE 1. Example of a cellular system with two BSs.
Motivated by the properties of the delay-Doppler domain
channel representation in high-mobility scenarios, the authors
in [16] developed the OTFS modulation scheme. OTFS mod-
ulation efficiently transforms the time-varying fading channel
into a time-independent nonfading channel, as follows:
Z∞ Z∞
FIGURE 2. Downlink frame structure.
h (τ, v) = H (t, f ) e−j2π(vt−f τ ) dtdf (6)
−∞ −∞

In OTFS, the symbols are mapped on a discretized N × M BSs, the UE needs to perform synchronization and CID esti-
resource block in the delay-Doppler domain, called DDRB. mation to detect the serving cell. An accurate ST estimation
The symbols are T sec and 1f Hz in width along is also required for the channel estimation. In the proposed
the Doppler-domain and delay-domain, respectively. Each technique, each BS transmits a downlink synchronization
DDRE in the DDRB is 1/NT Hz and 1/M 1f sec in width frame comprising a DP, a pilot, and an SS, as shown in Fig. 2.
along thenDoppler-domain
 and delay-domain, orespectively, The DP should be found at the Rx by using a noncoherent
i.e., 0 = NT k
, M 1f
l
, 0 6 k < N , 0 6 l < M . The infor- detection scheme because the channel state information and
mation symbols in the delay-Doppler domain, x [k, l], are timing information are not available at the start of synchro-
then transformed in the familiar time-frequency domain, nization. The DP is used to determine the ST and PCID
X [n, m], using a transform called the 2D inverse symplectic at the Rx by maximizing the cross-correlation between the
finite FT (ISFFT): transmitted and received signals. Using the ST information,
a pilot is detected in the delay-Doppler domain. Using the
N −1 M −1
channel information obtained by the pilot, the SS (SCID)
 
1 XX −j2π ml nk
M−N
X [n, m] = x [k, l] e (7) is detected coherently in the delay-Doppler domain. The
MN
k=0 l=0 CID of the BS is then obtained by combining the PCID
Thus, the OTFS resultant waveform in time-domain is the and SCID.
fluctuating pulse train, shifted in time equal to
 the delay, The selection of synchronization signals is crucial because
lp M 1f , and in frequency equal to Doppler kp NT , where

they are the first signals that the UE needs to identify without
p is the index of DDRE. In other words, the time-domain any prior information before accessing the network. In LTE,
waveform of OTFS inherits the combinational and benefi- the ZC sequence is used as the PSS and is always mapped
cial properties of well-known multiple access techniques. to 72 subcarriers at the center of the carrier BW. In 5G
The time-domain OTFS waveform is locally localized in NR, the m-sequence is used as a PSS and can be mapped
the time-domain like time-division multiple access, globally to 127 active subcarriers at multiple locations on the entire
localized in the Doppler-domain like OFDM, and spreads like carrier BW [10]. These sequences possess good autocorre-
code-division multiple access in the delay-Doppler domain. lation properties for accurate ST estimation [31]. However,
Let us consider a cellular system in which a UE performs they may be unsuitable for synchronization in high-mobility
cell search by receiving the downlink frame transmitted from environments because of their sensitivity to the high Doppler
multiple BSs located at their predefined locations, as shown effect. For the ZC sequence, the influence of Doppler shift is
in Fig. 1. With the downlink signals received from adjacent studied and the solutions are provided in [31], [32]. However,

73578 VOLUME 9, 2021

M. S. Khan et al.: Downlink Synchronization for OTFS-Based Cellular Systems

the solutions in [31], [32] are not applicable to high-frequency

bands, and correspondingly, higher Doppler shifts. Hence,
these sequences may cause high ambiguity in ST estimation
in a high-Doppler environment and degrade the performance
of the initial synchronization. Therefore, this paper proposes
a pLFM waveform to provide accurate ST estimation and
PCID detection in a high-Doppler environment. After com-
pleting the ST synchronization with the DP, the pilot trans-
mitted in the delay-Doppler domain (OTFS modulation) is
used to estimate the delay-Doppler channel. Finally, the SS
in the delay-Doppler domain is used to detect the SCID with
the delay-Doppler channel estimated by the pilot. Because
the ZC sequence possesses good correlation properties and
can be used to generate a large number of SCIDs, it is
selected as an SS where the SCID is mapped to its root
index.
When the signal transmitted from the cth BS is sc (t) and
the number of adjacent BSs is C, the signal received at the
UE in the time-domain is given by FIGURE 3. (a) Pulse, (b) conventional-LFM (cLFM), (c) proposed-LFM
(pLFM), and (d) ambiguity functions of pulse, cLFM (b = B = 15.36 MHz),
C−1
XZ Z and pLFM (receiver processing, α = 5 and NpLFM = 1024).
r (t) = hc (τ, v) sc (t − τ ) ej2πv(t−τ ) dτ dv + w (t)
c=0
(8) waveform with a duration of T follows a triangular function
Here, the delay-Doppler channel is defined as [22], [33], [34]: along the delay axis (Fig. 3d). At zero delay, the AF of the
pulse waveform follows a sinc function along the Doppler
P
axis and becomes zero at a Doppler mismatch of ±1 τ Hz.

X
hc (τ, v) = βic δ τ − τic δ v − vci



(9) Therefore, the peak of the pulse waveform’s AF remains
i=1 at the correct timing when the Doppler mismatch is small
where 0 6 τic 6 τmax and −vmax 6 vci 6 vmax . Here, < 1 τ . However, in the case of a nonnegligible Doppler

vmax and τmax are the maximum Doppler and delay spreads, shift, the pulse waveform cannot be detected.
taps for the ith path
respectively. The delay and Doppler shift
 The cLFM waveform is extensively used for surveillance
are τi = li M 1f and vi = ki + κi NT , respectively.
c c c c c and tracking applications in which the target must be detected
Here, 0 6 lic 6 lmax , −kmax 6 kic 6 kmax , lmax , and in a high-mobility environment. A time-domain cLFM wave-
kmax denote the delay index of the ith path, Doppler index form with a frequency sweeping parameter b and symbol
of the ith path, largest index of delay tap, and largest index duration T is given as [35]
of Doppler tap, respectively. In addition, κic ∈ [−0.5, 0.5] b 2
denotes the fractional Doppler frequency associated with the xLFM (t) = ejπ T t , 0 6 t < T , 0 < b 6 B (11)
ith path. where B denotes the operational BW. Unlike the pulse wave-
form, the parameters of the cLFM waveform (b and T ) can be
III. PROPOSED DOWNLINK SYNCHRONIZATION used independently to regulate the delay resolution and pulse
TECHNIQUE FOR OTFS energy. The AF of the cLFM waveform is a triangular ridge
A. DETECTION PREAMBLE (DP) DESIGN skewed in the delay-Doppler plane (Fig. 3d). It can provide
To analyze and characterize the properties and behavior of significant peaks at the output of the matched filter for a broad
the waveforms paired with the matched filter, the ambiguity range of Doppler shifts, which follow a sinc function along
function (AF) is extensively used, which can be expressed as the Doppler axis at zero delay. However, its peak is shifted in
Z∞ time proportional to the Doppler shift. The timing accuracy
χ (τ, v) = s (t − τ ) s∗ (t) ej2πvt dt (10) obtained by the cLFM waveform is constrained by b and T ;
it improves as b increases and/or T decreases. Generally,
−∞
the value of b in cLFM is set to B because it is known to
The AF is useful for examining the resolution, side-lobe exhibit the best performance of the LFM waveform. The AF
behavior, and ambiguities in both Doppler and delay of the cLFM waveform (skewed triangular ridge) is beneficial
(time shift) domains of a given waveform. Fig. 3 shows for surveillance applications (only for detection). However,
the time-domain waveforms (pulse in Fig. 3a and cLFM for ST estimation, AF of a waveform that provides high
in Fig. 3b); their corresponding AF contours are presented peaks at the correct timing regardless of any Doppler shift
in Fig. 3d. When the Doppler shift is zero, the AF of the pulse value is beneficial. Thus, an ideal AF for an accurate ST

VOLUME 9, 2021 73579

 M. S. Khan et al.: Downlink Synchronization for OTFS-Based Cellular Systems

estimation in high-Doppler environments would be ‘‘a line’’ where −NpLFM < mpLFM < NpLFM . Note that the correlation
centered on the horizontal (Doppler shift) axis, as shown operation is performed over the full period of pLFM (NpLFM ).
in Fig. 3d. This subsection (III-A) aims to identify a wave- In (14), when mpLFM = 0 and v = 0, RupLFM 1 u1
(0, 0) =
form that has an AF (pLFM in Fig. 3d) close to the ideal AF, 0
NpLFM . If u1 6= u1 , u1 = 1, cross-correlation function can
so that the DP is detected with accurate ST in high-Doppler be derived as follows:
environments.
u1 u01
mpLFM , v


The pLFM is obtained by discretizing the cLFM wave- RpLFM
form, increasing the frequency sweeping parameter beyond ωm2pLFM ∞ j2πnpLFM    
the limit of B and incorporating PCID in the cLFM waveform. jπ NpLFM
X
NpLFM ω npLFM −mpLFM +vT
,e e
By selecting the frequency sweeping parameter value as an npLFM =−∞
integer multiple of B, the pLFM can be defined as  2
vT −ωmpLFM
ωm2pLFM −jπ 2ωNpLFM
u1 ωn2
pLFM jπ e
NpLFM
u1
jπ NpLFM =e
xpLFM npLFM = e (12)
q 
−j2ω NpLFM

 and ω = αbTs , and α ∈ N. ( )

where u1 ∈ {1, −1} is the PCID, ∞
X NpLFM n2pLFM npLFM vT −ωmpLFM
−jπ
Here, t = npLFM Ts and Ts = T NpLFM . Ts and NpLFM denote e 2ω ej2π 2ω

the sampling time and number of samples in T , respectively. npLFM =−∞

As ω exceeds one, the signal in (12) becomes a repetitive
  2 
jπ vT −ωmpLFM
ωm2
pattern because the pLFM is discretized in the time-domain. pLFM −

NpLFM 2ω

For instance, when ω = 5 in (12), the cLFM waveform with e

=
b = B and symbol duration T /5 repeats five times over a
q 
−j2ω NpLFM
given duration T , as shown in Fig. 3c. The DP is constructed !
π vT − ωmpLFM


NpLFM
using a pLFM signal at the BS.
ϑ3 , e−jπ 2ω
At the Rx (UE), the ST and PCID are obtained noncoher- 2ω
ently by correlating the received signal with the locally gen- (15)
erated reference DPs. The correlation between the received
∞ n2pLFM
signal with v and the reference DP is given by where ϑ3 (z1 , z2 ) ≡
P
ej2z1 npLFM is the Jacobi
z2
npLFM =−∞
u1 u01
= π vT − ωmpLFM 2ω and


mpLFM , v


RpLFM theta-3 function with z1
∞
u1
xpLFM npLFM

z2 ≡ ejπ (−NpLFM 2ω) [36].
Fig. 4 shows the autocorrelation and cross-correlation
X
j2πvT npLFM
, h u01
i∗ NpLFM
npLFM =−∞ xpLFM npLFM − mpLFM e functions of the pLFM waveform when NpLFM = 1024,
ωu01 m2
T = 66.67µs, b = B = 15.36 MHz, and v = 0. In this figure,
−jπ pLFM
NpLFM ω = 1 corresponds to the case of cLFM. As ω increases,
=e
the autocorrelation function (Fig. 4a) in the region of origin
ωu1 npLFM − ωu01 npLFM
 
n
∞ jπ NpLFM  (mpLFM = 0) becomes sharper (although only slightly),
+2ωu01 mpLFM + 2vT

X pLFM
e (13) which implies a slight increase in the timing estimation
npLFM =−∞ accuracy. Fig. 4b shows that increasing ω does not degrade
the cross-correlation function property. The mean value of
where ∗ denotes the complex conjugate. If u1 = u01 , the auto- cross-correlation function (for ω ∈ [1, 10]) is 0.066 ≈
p
correlation function of pLFM can be derived as 2 NpLFM . Moreover, the analytical results for both auto-
correlation function in (14) and cross-correlation function
RupLFM
1 u1
mpLFM , v


in (15) agree well with the simulation results presented
u1 ωm2
pLFM
NpLFM −1 j2πnpLFM   in Figs. 4a and 4b, respectively.
−jπ u1 ωmpLFM +vT
As ω increases, the time–BW product of pLFM increases,
X
NpLFM NpLFM
,e e
npLFM =0 which in turn decreases its Doppler sensitivity. However,
jπvT (1+mpLFM NpLFM ) jπu1 ωmpLFM
 an increase in ω also increases the side peaks in the autocor-
, e  e
π (u1 ωmpLFM +vT )(NpLFM −mpLFM )
 relation function (Fig. 4a), which may produce time ambi-

 sin NpLFM guity. To reduce the effect of side peaks and obtain a higher
, mpLFM > 0



 
π (u 1 ωm pLFM +vT )
 time–BW product, a receiver processing (RP) scheme is
sin


NpLFM developed for pLFM, as shown in Fig. 5. Here, the received
π (u1 ωmpLFM +vT )(NpLFM +mpLFM ) into α fragments each
 
 sin DP of NpLFM samples is segmented
 NpLFM j . k
, mpLFM < 0 α
of NpLFM = NpLFM α samples. Then, each fragment


π (u1 ωmpLFM +vT )

  

 sin NpLFM is correlated with a locally generated DP with ω (NpLFM α

(14) samples) to reduce the Doppler effect by α. When the RP

73580 VOLUME 9, 2021

M. S. Khan et al.: Downlink Synchronization for OTFS-Based Cellular Systems

FIGURE 4. Comparison of simulation and analytic results of correlation property of pLFM with v = 0.

scheme is applied to pLFM, (14) becomes

 
α
RupLFM
1 u1
,RP mpLFM , v
 . 
jπ αv T 1+mαpLFM NpLFM
α ω α
= e ejπu1 α mpLFM
π u1 ωmα α α
   !
pLFM +vT NpLFM −mpLFM

 sin α
αNpLFM
, mαpLFM > 0



π u1 ωmα
  !
pLFM +vT



 sin α
αNpLFM

π u1 ωmα α +mα
  !
pLFM +vT NpLFM pLFM
sin


 α
αNpLFM
, mαpLFM < 0


π u1 ωmα

  !
pLFM +vT


 sin α
αNpLFM

(16)

where −NpLFMα < mαpLFM < NpLFMα . In (16), when mαpLFM =

0 and v = 0, the correlation output of each correlator gives
a peak value of RupLFM
1 u1 α
,RP (0, 0) = NpLFM . Using only one FIGURE 5. Receiver processing for STS and PCID detection.
α
correlator, the performance of pLFM is restricted to NpLFM ,
which is α times less than that of cLFM (when no Doppler the time shift, m̃αpLFM , corresponding to the maximum value
effect exists). Because pLFM is robust to the Doppler effect,
RupLFM
1 u1
,RP (0, v) can be expressed as
each correlator’s output is coherently combined to achieve a
performance gain equal to NpLFM (same as the performance u1 ωTv
of cLFM in the absence of Doppler effect). This process is m̃αpLFM = − ,
 α 
2
repeated for each value of u1 and the PCID is estimated. α
When v 6 = 0, (16) can be viewed as a time-shifted RupLFM
1 u1
,RP (0, v) = sinc m̃pLFM
version of a sinc function. With the proposed RP, (17)

VOLUME 9, 2021 73581

 M. S. Khan et al.: Downlink Synchronization for OTFS-Based Cellular Systems

FIGURE 6. Comparative simulation and analytic results of pLFM with varying v .

Eq. (17) indicates that when B and T are fixed, the time sample. Similarly, for the pLFM without RP (with the same
shift increases and decreases proportionally with v and α, v = ±8 kHz), when ω ∈ {2, 5}, the maximum magnitude
respectively. Moreover, RupLFM
1 u1 α
,RP (0, v) decreases as m̃pLFM
is {0.41, 0.63} and the time shift is {512, 205} samples.
increases. The ST estimation should be sufficiently accurate However, for the pLFM with RP, the maximum magnitude
because it is used to detect the start of OTFS symbols. Other- is {0.68, 0.65, 0.64} and the time shift of one sample occurs
wise, the symbol timing offset (i.e., time shift) will produce a at v ∈ ± {16, 38, 76} kHz when ω ∈ {2, 5, 10}. That is,
cyclic shift in delay-domain, thereby introducing interference for the pLFM with RP, the time shift does not occur until v
in the received OTFS symbols in the delay-Doppler domain. reaches {16, 38, 76} Hz when ω ∈ {2, 5, 10}. For cLFM,
Fig. 6 illustrates the comparative simulation and analytic a time shift occurs when v = ±8 kHz. Thus, the time
results of cLFM and pLFM (with and without RP) in the exis- shift decreases as ω in pLFM with RP increases. Because
tence of a Doppler shift; that is, v 6 = 0. In the simulation, for a an accurate ST estimation in a high-Doppler environment is
fair comparison, the same values of B and T are used in both crucial in downlink synchronization, pLFM with RP is used
cLFM and pLFM. Figs. 6a and 6b show the maximum magni- for DP design.
tude and time shift in the autocorrelation function of pLFM, When the C BSs transmit their DPs, the signal received at
respectively, with varying v, NpLFM = 1024, T = 66.67µs, the UE can be expressed as
b = B = 15.36 MHz, and ω ∈ {1, 2, 5}. Fig. 6a and the first 0 0
 0
 j2πvc0 t−τ c0 
two figures in Fig. 6b indicate that the analytical results in rpLFM (t) = βic scpLFM t − τic e i i
+ w̃ (t)
(14) and (16) agree well with the simulation results. The third C−1 P
βic scpLFM t − τic ej2πvi (t−τi ) +w (t)
X X c c
w̃ (t) =


figure in Fig. 6b indicates a detailed version of the time shift
of cLFM and pLFM with RP. These figures indicate that when c=0,c6 =c0 i=1
v = 0, the maximum magnitude is one for both cLFM and (18)
pLFM (for any value of ω) and no time shift occurs. However,
c0 u01
the maximum magnitude fluctuates following a sinc function where spLFM (t) = (t) denotes the pLFM signal trans-
xpLFM
proportional to the Doppler shift, and a corresponding time mitted from the desired BS c0 and w̃ (t) denotes the inter-
shift occurs. For cLFM (i.e., ω = 1), when v = ±8 kHz, ference signals from other BSs and noise. ST and PCID are
the maximum magnitude is 0.68 and the time shift is one acquired at the UE when the DP is detected at the output of

73582 VOLUME 9, 2021

M. S. Khan et al.: Downlink Synchronization for OTFS-Based Cellular Systems

The relation between delay-Doppler ZC and time-

frequency ZC is given as
N −1 M −1  
u2 1 X X u2 j2π nk ml
N −M
XSS [n, m] = xSS [k, l] e
MN
k=0 l=0
M −1 
1 X − j2π (mN +n)l − jπu2 −û2 n2 +û2 n

= e MN e MN 2
M
l=0
N −1 
1 X u2 
xSS k − û2 n M , l
 
N
k=0
M −1  
(û2 n(û2 n−1)) 1 − j2π
jπu2 mN +n
X u l
FIGURE 7. Frame structure of OTFS for synchronization (o: zeros and =e MN X̂SS2 [0, l] e M N

: elements of ZC sequence). M
 l=0 
u2 ∗   u2 mN + n
= xSS 0, û2 n x̂SS ,0 (23)
N
the pLFM correlator, as follows:
Here û2 is the multiplicative inverse of u2 such that
h i 2
0
u01 , t̃ c = arg max R̃cpLFM (t) >γ (19) u2 û2 mod MN = 1. Hence, the ISFFT of a ZC sequence
c,t∈[0,TW ] becomes another ZC sequence. Now, the time-frequency ZC,
u2
XSS [n, m] in (23), is converted to time-domain to show the
0
where t̃ c , TW , and γ denote the estimated ST of the desired relation between delay-Doppler ZC and time-domain ZC as
BS c0 , the length of the correlator window, and a predefined
threshold value, respectively. The output of the pLFM corre- suSS2 (t)

M −1 N −1
lator for the cth BS can be expressed as follows:
[n, m] ej2πm1f (t−nT )

u2
X X
0 6 t < NT

 XSS
=
ZTW h i∗ 
 m=0 n=0
0 Otherwise
R̃cpLFM (t) = rpLFM (t + τ ) scpLFM (τ ) dτ (20)
M −1 
0 X jπu2
m2 M 2 N 2 +mMN −2û2 mMN 1ft


= e− MN
B. SECONDARY-CID SIGNAL (SS) DESIGN m=0
After transmitting the DP, each BS transmits two OTFS sym- N −1  
X u2 ∗   u2 mN + n
bols comprising two OTFS DDRBs, as shown in Fig. 7. In the xSS 0, û2 n − u2 mMN x̂SS ,0
N
first DDRB, a pilot signal is mapped at one DDRE in 0 as: n=0
M −1 
e− MN (mMN −û2 1ft )(mMN −û2 1ft+1)−
jπu2
û22 1f 2 t 2 −û2 1ft
( X 

1 if k = kpc , l = lpc =
c
xp [k, l] = (21) m=0
0 Otherwise
N −1 
u2
X
û2 n − u2 mMN , m
 
  XSS
where kpc , lpc is a DDRE allocated to the cth BS. For the UE n=0

= e MN (û2 1ft (û2 1ft−1)) xSS

located at the cell edge, the delay of certain paths from the jπu2
u2
[0, 0]
serving BS to the UE can be as high as τmax . Thus, the signal
M −1
received through these paths at the UE may possess an offset
e− MN (mMN −û2 1ft )(mMN −û2 1ft+1)
X jπu2

in the delay-Doppler domain and cause intercell interference

m=0
unless the pilot of each BS does not overlap in the DDRB. u2 ∗   u2 u
To ensure that the pilot can be received at the UE with- = xSS 0, û2 1ft xSS [0, 0] sSS2 (0) (24)
out interference, each BS’s pilot has some DDREs reserved Eqs. (23) and (24) indicate that the duality property of the
around it such that kic+1 − kic > 2kmax , lic+1 − lic > lmax . ZC sequence is conserved between the delay-Doppler and
Following the pilot, the SS generated using a ZC sequence time-frequency domains and between the time-frequency
with root index u2 and sequence length MN is mapped in the domain and time-domain, respectively. Further, (24) indicates
next DDRB, such that that time-domain ZC can be obtained from delay-Doppler
u2 u
jπu2 ZC by scaling it by constants xSS [0, 0] and sSS2 (0). In addi-
u2 (l+kM )(l+kM +1)
xSS [k, l] = e− MN , 06k <N, 0 6 l <M tion, the elements of the ZC sequence in the proposed
(22) SS are quantized before being mapped to DDRB (such as
u2  u
Re xSS [k, l] > 0 = 1, Re xSS2 [k, l] < 0 = −1,
u2 u
The SCID is mapped to the root index u2 of the ZC sequence. Im xSS [k, l] > 0 = j, and Im xSS2 [k, l] < 0 = −j) to

VOLUME 9, 2021 73583

 M. S. Khan et al.: Downlink Synchronization for OTFS-Based Cellular Systems

Using (8) and (9), the SS received at the UE can be

expressed as
C−1 P
βic scSS t − τic ej2πvi (t−τi ) + w (t) (25)
XX c c
rSS (t) ,

c=0 i=1

The signal rSS (t) in (25) sampled at fs = 1 Ts is given by



τc
 

C−1 P
τic j2π υic Ts ς 0−Tis
XX  
rSS ς =
0
βi sSS ς −
c c 0
+w ς 0


e
Ts
c=0 i=1
(26)
where 0 6 ς 0 < MN . The n0 th OTFS symbol comprising M
samples can be collected from (26) as follows:
rSS n0 M + `


c
c  j2πvc T n0 M +`− τi
 
C−1 P
τ

XX i s Ts
= βic scSS n0 M + ` − i e
Ts
c=0 i=1
+ w n0 M + `


(27)
where 0 6 n0 < N and 0 6 ` < M . Now, (27) is transformed
into the time-frequency domain using discrete FT as
YSS n0 , m0
 

M −1
1 X m0 `
rSS n0 M + ` e−j2π M


=
M
`=0
M −1 C−1 P
τic
 
1 X XX
= βi sSS n M + ` −
c c 0
M Ts
`=0 c=0 i=1
τc
 
j2πvci Ts n0 M − Tis m0 `
e e−j2π M
M −1
1 X m0 `
w n0 M + ` e−j2π M


FIGURE 8. Correlation property of the ZC sequence.
+
M
`=0
τc
 
u2 C−1 P τic
take advantage of the MP algorithm. Thus, [k, l] takes
xSS XX j2πvci Ts n0 M − Tis m0
βic e e−j2π M n ,m
c
 0 0
= Ts XSS
the values from A ∈ {1 + j, 1 − j, −1 + j, −1 − j}. In con-
c=0 i=1
trast to OFDM, where the Doppler spread in high-mobility
+ W n0 , m0
 
environments induces ICI and degrades the performance, (28)
OTFS demonstrates no performance degradation because the Using SFFT, (28) is transformed back to the delay-Doppler
efficient delay-Doppler channel fairly shifts the symbols from domain as
one DDRE to another without any loss of information [16],
ySS k 0 , l 0
 
[22], [33].
N −1 M −1 
m0 l 0 n0 k 0

Fig. 8 shows the correlation property of the ZC sequence as X X  j2π M − N
YSS n0 , m0 e

=
a function of lag m̄. Figs. 8a and 8b show the autocorrelation
n0 =0 m0 =0
function (u2 = 11 and u02 = 11) and cross-correlation func- 
τc

N −1 M −1 C−1 P τic
tion (u2 = 11 and u02 = 19), respectively, for both unquan- X X XX j2πvci Ts n0 M − Tis m0
= βic e e−j2π M Ts
tized and quantized ZC sequences. As shown in Fig. 8a,
n0 =0 m0 =0 c=0 i=1
the autocorrelation function of the ZC sequence preserves its
 0 0  j2π mM0 l 0 − nN0 k 0
 
characteristics and exhibits good autocorrelation when quan- XSS n , m e
c
tized to A. As shown in Fig. 8b, the normalized mean value N −1 M −1
 0 0  j2π mM0 l 0 − nN0 k 0
 
of cross-correlation
.√ function for unquantized ZC sequences is
X X
+ W n ,m e
0.0313 (i.e., 1 MN ). The normalized mean value for quan- n0 =0 m0 =0
tized ZC sequences is 0.0347, which is slightly higher than C−1
XX P
βic xSS
c
k 0 − kic , l 0 − lic + ŵ k 0 , l 0


  
that for quantized ZC sequences. Thus, the figures clearly = N M
show that the correlation properties of the ZC sequence hold c=0 i=1
even after the elements are quantized to A. (29)

73584 VOLUME 9, 2021

M. S. Khan et al.: Downlink Synchronization for OTFS-Based Cellular Systems

where (·)M and (·)N represent modulo M and N operations, where  [A (g)]| A (g) ∈ A is the probability mass func-
respectively. tion (PMF) of symbols in A and σ is the noise variance. The
As seen in (29), the desired SS needs to be detected in the message passed from xSS (e) to ySS (d) is the PMF
presence of intercell interference and noise because multiple q+1 q q−1
SSs from adjacent BSs are received at the UE. Inspired by the ed = ξ ed [A (g)] + (1 − ξ ) ed [A (g)]
Y
MP algorithm in [34], a low-complexity algorithm is applied Pr ySS (η) xSS (e) = A (g) , HSS


∝ (35)
for SS detection. Let the OTFS symbol vector and channel η∈ψe ,η6 =d
matrix of the cth BS be denoted by xcSS ∈ CMN ×1 and HcSS ∈
CMN ×MN , respectively. Here, it is assumed that the channel is where
correctly estimated using one of the algorithms in [21]–[27]. |ySS (η)−µqηe −HSS (η,e)A(g)|2
−
Then, (29) can be written as Pr ySS (η) xSS (e) = A (g) , HSS (σηe )
q 2


∝e
ySS = HSS xSS + w (30) (36)
h i The convergence rate can be enhanced using the damp-
where ySS ∈ CMN ×1 , HSS = H0SS , H1SS , · · · , HC−1
SS , ing factor ξ ∈ (0, 1] [37]. The messages keep
T
passing from ySS (d) to xSS (e), and vice versa, until


T
T   T
and xSS = x0SS , x1SS , · · · , xC−1
SS . Then, (30) q+1 q
arg max ed [A (g)] − ed [A (g)] < 4. Here, 4 is a small
can be modeled as a sparsely connected factor graph with e,d,A(j)
MN variable and observation nodes corresponding to xSS number or q = qmax . Now, the decisions on the transmitted
and ySS , respectively. Let the d th row of HSS , eth column SSs are expressed as
of HSS , variable node, and observation node be denoted 1
x̃SS (e) = arg max Pr ySS xSS (e) = A (g) , HSS


by ψd , ψe , xSS (e), and ySS (d), respectively. In this factor
A(g)∈A |A|
graph, each xSS (e) is linked to the set of observation nodes ≈ arg max e [A (g)]
ySS (η) , η ∈
 ψe and each ySS (d) is linked to the set of vari- A(g)∈A
able nodes xSS (η) , η ∈ ψd . For optimal detection of the Y
Pr ySS (η) xSS (e) = A (g) , HSS


transmitted SSs xSS , the joint maximum a posteriori (MAP) ≈ arg max
A(g)∈A η∈ψ
detection rule can be considered as follows: e
(37)
x̃SS = arg max Pr xSS ySS , HSS


(31)
xSS ∈AMN Then, the detected SSs in (37) are correlated with the ref-
erence SSs at the UE. The correlator’s output for BS c is
However, (31) incurs a large computational complexity and is expressed as:
difficult to be solved even for very small values of M and N .
MN −1
Hence, the symbols-in-succession MAP detection rule is con- X ∗
RcSS (m̄) = x̃SS (e) xSS (e + m̄)
 c
(38)
sidered for 0 6 e < MN to detect the SS. With the initial-
e=0
ization of the iteration, i.e., q = 0, the message passed from
ySS (d) to xSS (e) is a Gaussian probability density function where xSSc
(e) is the 1D form of (22).
which can be computed as Finally, the desired SCID, u02 , is detected from the received
SS by rearranging the elements of (38) into a 2D form with
ySS (d) = xSS (e) HSS (d, e) + Ide (32) 0 6 k 0 < N and 0 6 l 0 < M , such that RcSS k 0 , l 0 ∈ CN ×M ,
as follows:
xSS (η) HSS (d, η) + w (d) represents the
P
where Ide = " #
η∈ψd  0 0 2
interference-plus-noise term approximated as a Gaussian ran- u2 = arg max arg max RSS k , l
0 c
>γ (39)
u2 k 0 ,l 0
dom variable with mean and variance:
|A| IV. SIMULATION
q q
X X
µde = E [Ide ] = ηd [A (g)] A (g) HSS (d, η) In this section, the performance of the proposed downlink
η∈ψd ,η6=e g=1 synchronization technique for an OTFS-based cellular system
(33) in high-Doppler environments is evaluated and compared
q
2
σde = Var (Ide ) with that of an OFDM-based cellular system. A simple cellu-
  lar model shown in Fig. 1 is used for the computer simulation,
|A|
q 2 which assumes that the UE moves from BS0 to BS1. The
ηd
[A (g)] |A (g)| HSS (d, η) 
2
P
carrier frequency (fc ) and 1f are set to 28 GHz and 15 kHz,

X  g=1 
=  2 respectively. For the channel, the 3GPP tapped delay line-C
 P |A| 
η∈ψd ,η6=e  − q
ηd [A (g)] A (g) HSS (d, η) (TDL-C) model is used [38].
For the cth BS, the Doppler

g=1 shift (vci ) is set to vmax cos θic for the ith path, where θic ∈
+σ 2
(34) [0, 2π) is an independent and uniformly distributed random

VOLUME 9, 2021 73585

 M. S. Khan et al.: Downlink Synchronization for OTFS-Based Cellular Systems

TABLE 1. OTFS Parameter set.

variable [34]. The simulation parameters are summarized

in Table 1.
Fig. 9 shows the detection probability of DP designed
with cLFM and pLFM (with and without RP). The two
BSs (BS0, BS1) transmit DPs with NpLFM = 1024 and
PCID u1 ∈ {1, −1}. Here, it is assumed that the UE moves
at ν ∈ {15, 75, 150} mph, which corresponds to vmax ∈
{1.4, 7, 14} kHz at fc = 28 GHz. Moreover, the UE is placed
at the cell boundary of serving BS (BS0), implying that the
signal-to-intercell interference ratio is 0 dB. In these figures,
the detection probability of the DP transmitted from BS0 is
shown. When both the ST and PCID detected at the UE
are correct, the detection is considered ‘‘successful’’. Fig. 9a
shows that, when the UE velocity is low, the detection prob-
abilities are the same, i.e., 100% at an SNR of −12 dB,
regardless of the waveform (cLFM and pLFM). However,
the detection probability of cLFM degrades by 3 dB when
ν = 75 mph (vmax = 7 kHz) and falls to zero when ν = 150
mph (vmax = 14 kHz), as shown in Figs. 9b and 9c. The
detection probability of pLFM without RP degrades by 3 dB
when vmax = 7 kHz and ω = 2, degrades significantly when
vmax = 7 kHz and ω = 10, and converges to zero when
ν = 150 mph (vmax = 14 kHz). The detection probability of
the pLFM with RP does not degrade for any ω when vmax = 7
kHz, degrades by 3 dB when vmax = 14 kHz and ω = 2,
and does not degrade when vmax = 14 kHz and ω = 10.
These results can be expected from Fig. 6b because, for the
cases of cLFM and pLFM without RP, a time shift occurs in
the presence of Doppler shift. In these cases, the detection
probability decreases in proportion to the UE velocity. How-
ever, the performance degradation is significantly reduced in
the case of pLFM with RP because an accurate ST can be
FIGURE 9. Detection probability of DP under high-mobility two-cell
obtained in high-Doppler environments, as shown in Fig. 6b. multipath environment.
Note that the detection probability obtained by the pLFM
with RP (ω = 10) in a high-mobility environment (ν = 150
mph) is similar to that obtained in a low-mobility environ- ‘‘successful’’ when the SCID detected at the UE is cor-
ment (ν = 15 mph). Therefore, the effect of Doppler shift on rect. Fig. 10 shows that, for all three Doppler frequencies,
ST and PCID detection can be significantly reduced using the the detection probability reaches 100% at an SNR of −21 dB
proposed DP (pLFM with RP). using the proposed technique (OTFS). In addition, in the
Fig. 10 shows the detection probability of the SS designed figure, the detection probability is compared with the results
with two OTFS symbols (pilot DDRB and SS DDRB). obtained by the conventional technique (OFDM). In conven-
Here, the delay-Doppler channel is assumed to be correctly tional techniques, the ZC sequence is mapped to an OFDM
estimated using the pilots discussed in Section III-B, and symbol. For a fair comparison, the parameters such as ZC
the SSs are generated at BS0 and BS1 with u2 = 11 sequence length, 1f , and fc are maintained constant. The
and u2 = 19, respectively. Here, the detection is declared figure (OFDM) shows that when the Doppler frequency is

73586 VOLUME 9, 2021

M. S. Khan et al.: Downlink Synchronization for OTFS-Based Cellular Systems

BS can be correctly detected using the proposed technique

(OTFS). However, the detection probability becomes zero
over all SIR regions when OFDM is used in high-Doppler
environments (vmax = 14 kHz).

V. CONCLUSION
This paper proposes a downlink synchronization and CID
estimation technique for OTFS-based cellular systems in
high-Doppler environments. To provide accurate ST esti-
mation and a large number of CIDs in high-Doppler envi-
ronments, a frame structure for downlink synchronization
is proposed, which comprises a DP, a pilot, and an SS.
The DP is designed using the pLFM because it can provide
accurate ST and PCID in high-Doppler environments with-
out any prior information (channel, ST). The autocorrelation
FIGURE 10. Detection Probability of SS using the OFDM and proposed
OTFS techniques. and cross-correlation functions of pLFM are examined to
investigate their correlation properties when subjected to a
high Doppler shift. To reduce the undesirable properties of
the autocorrelation function (side peak and time ambiguity),
an RP scheme is developed for pLFM. After completing
the ST synchronization with DP, the pilot transmitted in the
delay-Doppler domain is used to estimate the delay-Doppler
channel. The SS is designed with a ZC sequence and detected
coherently in the delay-Doppler domain. After deriving the
duality property of the ZC sequence, it is shown that the
properties of the ZC sequence are conserved when trans-
formed from the delay-Doppler domain to the time-frequency
domain and time-frequency domain to time-domain. Finally,
a low-complexity MP-based detection algorithm is applied to
detect the SCID at the UE. The simulation results confirm that
the pLFM with RP is suitable for DP design because it can
provide an accurate ST in the presence of a high Doppler shift.
FIGURE 11. Detection Probability of SS in a high-mobility two-cell The cross-correlation level of pLFM is small, which implies
environment. that the intercell interference is not significant. The ZC-based
SS is suitable for OTFS-based cellular systems because
low (i.e., vmax = 1.4 kHz), a 100% detection probability it can provide a large number of CIDs in high-Doppler
can be obtained at an SNR of −21 dB. However, as the environments and produce a small intercell interference
Doppler frequency increases, the time-domain ZC sequence (cross-correlation function). Moreover, the detection proba-
will be distorted and time-shifted, resulting a decrease in bility decreases significantly in high-mobility scenarios when
detection probability. This time shift causes a cyclic shift in OFDM is used. The proposed downlink synchronization tech-
the frequency-domain and results in zero correlation between nique can be successfully used for OTFS-based cellular sys-
the desired and cyclic-shifted ZC sequences. Therefore, when tems because the detection probabilities of DP and SS do not
vmax = 7 kHz, the detection probability degrades by 6 dB degrade in high-Doppler environments.
as compared to vmax = 1.4 kHz and falls to zero at all
SNR regions when vmax = 14 kHz. These results indicate REFERENCES
that the SS designed with the proposed technique (OTFS) is [1] Technical Specification Group Radio Access Network; 5G; Study on
robust in high-Doppler environments, whereas the detection Scenarios and Requirements for Next Generation Access Technologies
(Release 16), document 3GPP 38.913, Rev. 16.0.0, Jul. 2020.
probability of the OFDMcase is reduced to zero in all SNR [2] M. Muhammad and G. A. Safdar, ‘‘Survey on existing authentication
regions when vmax > 1f 2. issues for cellular-assisted V2X communication,’’ Veh. Commun., vol. 12,
Fig. 11 shows the detection probability of SS when the UE pp. 50–65, Apr. 2018.
moves from BS0 to BS1 at ν = 150 mph. It can be seen [3] LTE, Evolved Universal Terrestrial Radio Access (E-UTRA); Base Station
(BS) Radio Transmission and Reception, Standard 38.104, Rev. 8.6.0,
from this figure that as the signal-to-interference ratio (SIR) Jul. 2009.
increases, the cell detection probability of the target BS (i.e., [4] Y. S. Cho, J. Kim, W. Y. Yang, and C. G. Kang, MIMO-OFDM Wireless
BS1) increases. Here, SIR represents the ratio of the signal Communications With MATLAB. Hoboken, NJ, USA: Wiley, Aug. 2010.
[5] S. Das and P. Schniter, ‘‘Max-SINR ISI/ICI-shaping multicarrier commu-
power from the target BS to that from the serving BS (i.e., nication over the doubly dispersive channel,’’ IEEE Trans. Signal Process.,
BS0). From this figure, it can be concluded that the target vol. 55, no. 12, pp. 5782–5795, Dec. 2007.

VOLUME 9, 2021 73587

 M. S. Khan et al.: Downlink Synchronization for OTFS-Based Cellular Systems

[6] K. Fang, L. Rugini, and G. Leus, ‘‘Low-complexity block turbo equaliza- [31] F. Khan, LTE for 4G Mobile Broadband: Air Interface Technologies and
tion for OFDM systems in time-varying channels,’’ IEEE Trans. Signal Performance. Cambridge, U.K.: Cambridge Univ. Press, 2009.
Process., vol. 56, no. 11, pp. 5555–5795, Aug. 2008. [32] X. Liu and D. Qiao, ‘‘Location-fair beamforming for high speed rail-
[7] K. Liu, T. Kadous, and A. M. Sayeed, ‘‘Orthogonal time-frequency sig- way communication systems,’’ IEEE Access, vol. 6, pp. 28632–28642,
naling over doubly dispersive channels,’’ IEEE Trans. Inf. Theory, vol. 50, May 2018.
no. 11, pp. 2583–2603, Nov. 2004. [33] P. Raviteja, K. T. Phan, Y. Hong, and E. Viterbo, ‘‘Interference cancellation
[8] Y. Zhao and S.-G. Haggman, ‘‘Intercarrier interference self-cancellation and iterative detection for orthogonal time frequency space modulation,’’
scheme for OFDM mobile communication systems,’’ IEEE Trans. Com- IEEE Trans. Wireless Commun., vol. 17, no. 10, pp. 6501–6515, Oct. 2018.
mun., vol. 49, no. 7, pp. 1185–1191, Jul. 2001. [34] P. Raviteja, K. T. Phan, Q. Jin, Y. Hong, and E. Viterbo, ‘‘Low-complexity
[9] G. Leus, S. Zhou, and G. B. Giannakis, ‘‘Orthogonal multiple access over iterative detection for orthogonal time frequency space modulation,’’ in
time- and frequency-selective channels,’’ IEEE Trans. Inf. Theory, vol. 49, Proc. IEEE Wireless Commun. Netw. Conf. (WCNC), Barcelona, Spain,
no. 8, pp. 1942–1950, Aug. 2003. Apr. 2018, pp. 1–6.
[10] Technical Specification Group Radio Access Network; NR; Phys. Channels [35] B. R. Mahafza, Radar Systems Analysis and Design Using MATLAB. Boca
Modulation (Release 15), Standard 38.211, Rev. 15.2.0, Jun. 2018. Raton, FL, USA: CRC Press, 2013.
[11] M. Asim, R. Pec, T. H. Im, and Y. S. Cho, ‘‘Cell search tech- [36] Z. G. Liu, ‘‘Residue theorem and theta function identities,’’ Ramanujan J.,
niques for underwater acoustic cellular systems,’’ IEEE Access, vol. 7, vol. 5, pp. 129–151, Jun. 2001.
pp. 106019–106033, Aug. 2019. [37] M. Pretti, ‘‘A message-passing algorithm with damping,’’ J. Stat. Mech.:
[12] M. S. Khan, R. Pec, C. H. Park, and Y. S. Cho, ‘‘Random access preamble Theory Exp., vol. 2005, no. 11, Nov. 2005, Art. no. P11008.
design for high-velocity user in millimeter-wave cellular networks,’’ IEEE [38] Technical Specification Group Radio Access Network; Study Channel
Access, vol. 6, pp. 66047–66054, Nov. 2018. Model for Frequencies From 0.5 to 100 GHz (Release 16), docu-
[13] M. S. Khan, C. H. Park, J. Joung, and Y. S. Cho, ‘‘Doppler-tolerant ment 3GPP 38.901, Rev. 16.1.0, Dec. 2019.
sequence design for positioning high-speed vehicles in millimeter-wave
cellular systems,’’ Veh. Commun., vol. 30, Aug. 2021, Art. no. 100358.
[14] Z. Wang, S. Zhou, G. B. Giannakis, C. R. Berger, and J. Huang,
‘‘Frequency-domain oversampling for zero-padded OFDM in underwater
acoustic communications,’’ IEEE J. Ocean. Eng., vol. 37, no. 1, pp. 14–24,
Jan. 2012.
[15] X.-G. Xia, ‘‘Precoded and vector OFDM robust to channel spectral nulls
and with reduced cyclic prefix length in single transmit antenna systems,’’
IEEE Trans. Commun., vol. 49, no. 8, pp. 1363–1374, Aug. 2001.
[16] R. Hadani, S. Rakib, M. Tsatsanis, and A. Monk, ‘‘Orthogonal time fre-
quency space modulation,’’ in Proc. IEEE Wireless Commun. Netw. Conf.
(WCNC), San Francisco, CA, USA, Mar. 2017, pp. 1–6.
[17] G. D. Surabhi, R. M. Augustine, and A. Chockalingam, ‘‘Peak-to-average MOHAMMED SAQUIB KHAN (Student Mem-
power ratio of OTFS modulation,’’ IEEE Commun. Lett., vol. 23, no. 6, ber, IEEE) was born in Mumbai, Maharashtra,
pp. 999–1002, Jun. 2019. India, in 1994. He received the B.E. degree in elec-
[18] V. Khammammetti and S. K. Mohammed, ‘‘OTFS-based multiple-access tronics and telecommunication engineering from
in high Doppler and delay spread wireless channels,’’ IEEE Wireless Mumbai University, Mumbai, in 2015, and the
Commun. Lett., vol. 8, no. 2, pp. 528–531, Apr. 2019. Ph.D. degree in electrical and electronics engi-
[19] A. K. Sinha, S. K. Mohammed, P. Raviteja, Y. Hong, and E. Viterbo, neering from Chung-Ang University, Seoul, South
‘‘OTFS based random access preamble transmission for high mobility Korea, in 2020.
scenarios,’’ IEEE Trans. Veh. Technol., vol. 69, no. 12, pp. 15078–15094, During his Ph.D., he worked as a Research
Dec. 2020.
Assistant with the Mobile Communications Labo-
[20] Z. Ding, R. Schober, P. Fan, and H. Vincent Poor, ‘‘OTFS-NOMA: An
ratory, Chung-Ang University, Seoul, where he is currently engaged in Post-
efficient approach for exploiting heterogenous user mobility profiles,’’
IEEE Trans. Commun., vol. 67, no. 11, pp. 7950–7965, Nov. 2019. doctoral Research. His research interests include wireless communication
[21] P. Raviteja, K. T. Phan, and Y. Hong, ‘‘Embedded pilot-aided channel esti- systems, digital signal processing, and machine learning.
mation for OTFS in delay–Doppler channels,’’ IEEE Trans. Veh. Technol., Dr. Khan is a member of the Korea Institute of Communications and
vol. 68, no. 5, pp. 4906–4917, May 2019. Information Sciences. He received the Chung-Ang University Young Scien-
[22] K. R. Murali and A. Chockalingam, ‘‘On OTFS modulation for high- tist Scholarship (CAYSS) for his M.S. and Ph.D. degrees and the Samsung
Doppler fading channels,’’ in Proc. Inf. Theory Appl. Workshop (ITA), San Humantech Paper Encouragement Award, in 2020.
Diego, CA, USA, Feb. 2018, pp. 1–10.
[23] A. Fish, S. Gurevich, R. Hadani, A. M. Sayeed, and O. Schwartz, ‘‘Delay-
Doppler channel estimation in almost linear complexity,’’ IEEE Trans. Inf.
Theory, vol. 59, no. 11, pp. 7632–7644, Nov. 2013.
[24] M. Kollengode Ramachandran and A. Chockalingam, ‘‘MIMO-OTFS in
high-Doppler fading channels: Signal detection and channel estimation,’’ YEONG JUN KIM was born in South Korea,
in Proc. IEEE Global Commun. Conf. (GLOBECOM), Abu Dhabi, UAE, in 1981. He received the B.S., M.S., and Ph.D.
Dec. 2018, pp. 206–212. degrees in electrical and electronics engineering
[25] R. Hadani and S. S. Rakib, ‘‘OTFS methods of data channel characteriza- from Chung-Ang University, Seoul, South Korea,
tion and uses thereof,’’ U.S. Patent 9 444 514 B2, Sep. 13, 2016. in 2005, 2007, and 2011, respectively.
[26] Y. Liu, S. Zhang, F. Gao, J. Ma, and X. Wang, ‘‘Uplink-aided high mobility From 2005 to 2011, he was a Research Assis-
downlink channel estimation over massive MIMO-OTFS system,’’ IEEE J. tant with the Mobile Communications Labora-
Sel. Areas Commun., vol. 38, no. 9, pp. 1994–2009, Sep. 2020.
tory, Chung-Ang University. From 2011 to 2015,
[27] W. Shen, L. Dai, J. An, P. Fan, and R. W. Heath, ‘‘Channel estimation for
he worked at LIG NEX1, South Korea, as a Senior
orthogonal time frequency space (OTFS) massive MIMO,’’ IEEE Trans.
Signal Process., vol. 67, no. 16, pp. 4204–4217, Aug. 2019.
Research Engineer. He was involved in developing
[28] C. Jin, Z. Bie, X. Lin, W. Xu, and H. Gao, ‘‘A simple two-stage equalizer modems for multiband and multirole radio. Since 2015, he has been working
for OTFS with rectangular windows,’’ IEEE Commun. Lett., vol. 25, no. 4, as the Deputy Principal Research Engineer with LG Electronics Company,
pp. 1158–1162, Apr. 2021, doi: 10.1109/LCOMM.2020.3043841. Seoul. He has been involved in standardization of 5G NR and development
[29] R. Nee and R. Prasad, OFDM for Wireless Multimedia Communications. of MODEM chipset for LTE-A and 5G. He is the author of more than
Norwood, MA, USA: Artech House, 2000. 50 conference and articles. His research interests include the area of mobile
[30] F. Hlawatsch and G. Matz, Wireless Communications Over Rapidly Time- communication and digital signal processing, especially for MIMO-OFDM
Varying Channels. New York, NY, USA: Academic, 2011. and 5G.

73588 VOLUME 9, 2021

M. S. Khan et al.: Downlink Synchronization for OTFS-Based Cellular Systems

QASIM SULTAN was born in Pakistan, in 1994. YONG SOO CHO (Senior Member, IEEE) was
He received the B.S. degree in electrical and born in South Korea. He received the B.S. degree
telecommunications engineering from COMSATS in electronics engineering from Chung-Ang Uni-
University, Islamabad, Pakistan, in 2017, and the versity, Seoul, South Korea, in 1984, the M.S.
M.S. degree in electrical and electronics engi- degree in electronics engineering from Yonsei
neering from Chung-Ang University, Seoul, South University, Seoul, in 1987, and the Ph.D. degree
Korea, in 2019, where he is currently pursu- in electrical and computer engineering from The
ing the Ph.D. degree in electrical and electronics University of Texas, Austin, TX, USA, in 1991.
engineering. In 1984, he was a Research Engineer with
Since 2017, he has been working as a Research Goldstar Electrical Company, Osan, South Korea.
Assistant with the Mobile Communications Laboratory, Chung-Ang Univer- In 2001, he was a Visiting Research Fellow with the Electronics and
sity. His research interests include areas of wireless communication systems Telecommunications Research Institute. Since 1992, he has been a Professor
and digital signal processing. He was a recipient of Chung-Ang University with the School of Electrical and Electronics Engineering, Chung-Ang Uni-
Young Scientist Scholarship (CAYSS). versity, Seoul. He is the author of 12 books, more than 400 conference and
articles, and more than 120 patents. His research interests include the area of
mobile communication and digital signal processing, especially for MIMO
OFDM and 5G.
Dr. Cho served as the President of the Korean Institute of Communications
JINGON JOUNG (Senior Member, IEEE) and Information Sciences, in 2016. He was a recipient of Dr. Irwin Jacobs
received the B.S. degree in radio communication Award, in 2013.
engineering from Yonsei University, Seoul, South
Korea, in 2001, and the M.S. and Ph.D. degrees in
electrical engineering and computer science from
KAIST, Daejeon, South Korea, in 2003 and 2007,
respectively.
He was a Postdoctoral Fellow with KAIST,
South Korea and UCLA, CA, USA, in 2007 and
2008, respectively. From 2009 to 2015, he was
a Scientist with the Institute for Infocomm Research (I2 R), Agency for
Science, Technology and Research (A*STAR) Singapore. In 2016, he joined
Chung-Ang University (CAU), Seoul, as a faculty member. He is currently
an Associate Professor with the School of Electrical and Electronics Engi-
neering, CAU, where he is also the Principal Investigator with the Intelligent
Wireless Systems Laboratory. His research interests include communication
signal processing, numerical analysis, algorithms, and machine learning.
Dr. Joung was a recipient of the First Prize of the Intel-ITRC Student Paper
Contest, in 2006. He was recognized as the Exemplary Reviewers of the IEEE
COMMUNICATIONS LETTERS, in 2012 and the IEEE WIRELESS COMMUNICATIONS
LETTERS, in 2012, 2013, 2014, and 2019. He served as the Guest Editor for
the IEEE ACCESS, in 2016. He served on the Editorial Board of the APSIPA
Transactions on Signal and Information Processing, from 2014 to 2019, and
served as a Guest Editor for the Electronics (MDPI), in 2019. He is currently
serving as an Associate Editor for the IEEE TRANSACTIONS ON VEHICULAR
TECHNOLOGY and Sensors (MDPI).

VOLUME 9, 2021 73589