OTFS Interleaved OFDM With Block CP
OTFS Interleaved OFDM With Block CP
Vivek Rangamgari*, Shashank Tiwari*, Suvra Sekhar Das*, and Subhas Chandra Mondal†
*G.S Sanyal School of Telecommunication, Indian Institute of Technology, Kharagapur, India
† Wipro Limited, Bangalore, India
Abstract—Orthogonal time frequency space (OTFS) modula- namely (i) Linear receivers such as in [6] and (ii) Non-
tion is a recently proposed waveform for reliable communication linear receivers such as in [7]. Non-linear receivers have lower
in high-speed vehicular communication scenarios. It has better error probability but have higher computational complexity
resilience to inter-carrier interference (ICI) than orthogonal
frequency division multiplexing (OFDM). In this work, we than linear receivers. We limit our work to linear receiver
describe OTFS as block-OFDM with a cyclic prefix and time considering practical realizability.
interleaving. This interpretation helps one visualize OTFS in In order to deal with high ICI due to Doppler spread and
the light of OFDM as well as it also helps in analyzing the phase noise, 5G NR has adopted a variant of contemporary
gain obtained by OTFS over OFDM. Further, we compare the edition of OFDM, which is variable subcarrier bandwidth
performance of OTFS with its contender 5G new radio (NR)’s
OFDM configuration of variable subcarrier bandwidth (VSB- OFDM (VSB-OFDM) [8] [9] [10]. The VSB reconfigurability
OFDM) while considering practical forward error correction is expressed in terms of ‘numerology’. It is established that
codes and 3GPP high-speed channel model. This provides realistic VSB-OFDM has higher resilience to ICI [11]. Comprehensive
performance comparison, which is highly desired for technology comparison of VSB-OFDM with OTFS is important to select
realization. Considering practical channel estimation, we find that efficient waveform for the environments where Doppler and
OTFS outperforms VSB-OFDM with 5G NR parameter by about
5dB. We also present results on peak to average power ratio phase noise effects are strong.
(PAPR) due to specific pilot structure used in OTFS for channel The gain of OTFS over VSB-OFDM is mainly attributed
estimation. to diversity achieved by time-frequency spreading of signal
Index Terms—OFDM, OTFS, 5G NR, ASB, Doppler, ICI, ISI, [5]. On the other hand, practical systems use forward error
phase noise, time-varying channel, performance comparison correction (FEC) codes which improve error performance by
virtue of diversity introduced through redundancy and ad-
I. I NTRODUCTION vanced singal processing at receiver. Thus, the overall diversity
There is an increasing demand for providing good quality gain that coded OTFS has over coded VSB-OFDM needs
of service in high speed vehicular scenarios [1] such as in to be studied, which is one of the objectives of this work.
vehicle to vehicle communications (V2V), unmanned aerial In [5], VSB-OFDM is compared with CP-OTFS. Uncoded
vehicle communications, etc in 5G. Orthogonal frequency di- performance of fixed subcarrier bandwidth OFDM is compared
vision multiplexing (OFDM), which is a popular transmission with RCP-OTFS [7] and CP-OTFS [12], whereas performance
technology, is limited in providing reliable connection in high of coded RCP-OTFS with VSB-OFDM (5G NR) is yet to
speed vehicular scenarios. This is due to its high sensitivity receive attention, which is provided in this work.
to inter carrier interference (ICI), caused by Doppler spread The contributions in this paper can be summarized as,
and phase noise. Orthogonal time frequency space (OTFS)
[2] is shown to be superior to OFDM in such high mobility • We show that OTFS can be interpreted as a block OFDM
environments, which is an integral part of 5G’s operating with a single CP and time interleaving of the samples of
scenarios. all OFDM symbols. This interpretation simplifies the way
In OTFS, the user data (constellation symbols) is placed we understand OTFS signal generation thereby paving the
in the delay-Doppler (De-Do) domain as opposed to time- path for a reconfigurable transceiver architecture.
• Through this novel interpretation of OTFS, we establish
frequency grid in OFDM. The data is then spread across the
time-frequency grid using a unitary transform. This is followed that diversity gain in OTFS is due to time-interleaving.
by an OFDM [2] or block OFDM [3] modulator. Cyclic prefix We also seggregate the gains due to the use of advanced
(CP) [2] or block CP [3] is added to absorb the channel delay interference cancellation receiver and time-interleaving.
• In order to present realistic results for performance com-
spread. When OFDM modulation is used with CP, it is called
as CP-OTFS whereas, OTFS with block OFDM modulation parison, we evaluate the performance of both transmis-
and block CP is termed as reduced CP OTFS (RCP-OTFS) [4]. sion technologies with
In this paper, we consider the RCP-OTFS, which is spectrally (i) channel profile specified in [13] by 3GPP.
more efficient than CP-OTFS. (ii) practical channel estimation algorithms, and
The time-frequency spread OTFS signal, which suffers from (iii) LDPC codes
inter symbol interference (ISI) and ICI in linear time varying • The pilot structure used for channel estimation in OTFS
(LTV) channel [5], is processed using advanced interference is different from VSB-OFDM. We hypothesize that such
cancellation receiver. Such receivers can be of two types, pilot structure should affect the peak to average power
Authorized licensed use limited to: BMS College of Engineering. Downloaded on August 06,2024 at 08:58:20 UTC from IEEE Xplore. Restrictions apply.
ratio (PAPR) of OTFS, hence present the related results. where, g(t) is transmitter pulse of duration T . In this paper,
We follow the notations described below throughout the we consider g(t) to be a rectangular pulse as in [3],i.e,
paper. We use Z, z, z as Matrix, Vector and scalar respec-
1 0≤t≤T
tively. ()T and ()† denote transpose and hermitian operations. g(t) = (4)
WL and IN represents L order normalized Inverse Discrete 0 otherwise
Fourier Transform(IDFT) matrix and N order Identity matrix. To denote the above system in its equivalent matrix-vector
Kronecker product operator is given by ⊗. The operator form, we obtain the discrete version of OTFS system by
diag{x} creates a diagonal matrix with the elements of sampling s(t) at the sampling interval of M T
and s =
vector x. Circulant matrix is represented by circ{x} whose [s(0) s(1) · · · s(M N − 1)] is formed from the samples of s(t).
first column is x. Notations −and − are floor and ceil If the De-Do symbols x(k, l) are arranged in M × N matrix
operators respectively. Column-wise vectorization
√ of matrix as,
(X) is represented by vec{X} and j = −1. N[a b] denotes ⎡
x(0, 0) x(1, 0) ··· x(N − 1, 0) ⎤
set of natural numbers between a and b.
⎢ x(0, 1) x(1, 1) ··· x(N − 1, 1) ⎥
X=⎢
⎣ .. .. .. .. ⎥
⎦
II. RCP-OTFS . . . .
x(0, M − 1) x(1, M − 1) ··· x(N − 1, M − 1)
A. OTFS Transmission (5)
We consider RCP-OTFS system operating with time- Then, the time domain signal can also be written as matrix-
frequency resource of total Tf seconds duration and B Hz. vector multiplication,
The Bandwidth is divided into M number of sub-carriers s = Ax (6)
having Δf sub-carrier bandwidth and we transmit N number where, x = vec(X) and A = WN ⊗ IM denotes the OTFS
of symbols having T symbol duration, thus B = M Δf transform ISFFT matrix. A cyclic prefix (CP) of length L ≥ lτ
and Tf = N T . Furthurmore, OTFS is critically sampled, i.e is appended at the start of the s.
T Δf = 1
The source bitstream is encoded using LDPC codes and B. OTFS as interleaved Block OFDM
then passed through symbol mapper. The QAM modulated The time-frequency signal obtained in (2) can be expresesed
data and pilot symbols are arranged over De-Do lattice Λ = in matrix form Z as,
{( NkT , MΔf
l
)}, k ∈ N[0 N − 1], l ∈ N[0 M − 1] as shown
in Fig.2a. De-Do signal can be given as, Z = WM † XWN (7)
⎧ The OTFS time domain signal s is obtained by applying IDFT
⎪
⎪ xp , k = Kp & l = Lp
⎪
⎨0 w.r.t frequency domain and written as,
Kp − 2kν ≤ k ≤ Kp + 2kν &
x(k, l) = (1) s = vec{WM Z} (8)
⎪
⎪ L p − lτ ≤ l ≤ L p + lτ
⎪
⎩
d(k, l), otherwise which can be simplified as,
Authorized licensed use limited to: BMS College of Engineering. Downloaded on August 06,2024 at 08:58:20 UTC from IEEE Xplore. Restrictions apply.
and Nbof dm = M OFDM symbols. Therefore, the subcarrier D. Receiver
bandwidth for the equivalent Block OFDM sytem changes to After removal of CP at the receiver, the received signal can
Δfbof dm = M N Δf and OFDM symbol duration changes to be written as [6],
N
Tbof dm = M T . In general M > N , thus Δfbof dm > Δf r = Hs + n (16)
which implies that OTFS will have increased capability to
combat Doppler as opposed to an OFDM system having Δf where, n is white Gaussian noise vector of length M N with
sub-carrier bandwidth. Fig. 1 shows the time domain signal elemental variance σn2 and H is a M N × M N channel matrix
generation for different waveforms and depicts the RCP-OTFS and can be given as,
as block OFDM with time interleaving.
P
H= hp Πlp Δkp (17)
p=1
The delay and Doppler values for pth path is given as where p ∈ N[1 P̂ ] and P̂ is the number of taps detected.
τp = MΔf
lp k
and νp = NpT where lp ∈ N[0 M − 1] and The estimate of matrix H is then given as,
kp ∈ N[0 N − 1] are delay and Doppler bin number on De-Do
P̂
lattice Λ for pth path. In this work, we assume that N and Ĥ = ĥp Πl̂p Δk̂p
M are sufficiently large so that there is no effect of fractional p=1
delay and Doppler on the performance [7].
b) Equalization: The De-Do data is estimated by equal-
We also define time varying frequency response of channel as,
izing the received time domain signal using MMSE [6] as,
τmax νmax
htf (f, t) = h(τ, ν)ej2π(νt−f τ ) dνdτ (13) x̂ = Hmmse r (21)
0 −νmax 2
where Hmmse = (ĤA)† [(ĤA)(ĤA)† + I]−1 .
σn
which simplifies as, σd2
To decode the equalized data using LDPC decoder, the log-
P likelihood ratios (LLRs) are passed to the decoder which are
htf (f, t) = hp ej2π(νp t−f τp ) (14) calculated from the equalized symbols as,
p=1
||x̂(η) − s||2 ||x̂(η) − s||2
LLR(bjη |x̂(η)) ≈ (min0 ) − (min )
Its discrete version h́(m, n), the channel coefficient at mth sSj σ 2 (η, η) sSj1 σ 2 (η, η)
subcarrier of nth time symbol, used later in III-B, is defined (22)
as, where x̂(η) is the η th element of x̂ mapped from the bits
b0η b1η · · · bK−1
η , K is the number of bits per symbol and
h́(m, n) = htf (f, t) (15)
f =mΔf,t=nT
σ 2 (η, η) is the element of σ 2 = σn2 (Hmmse H†mmse ). Sj0
Authorized licensed use limited to: BMS College of Engineering. Downloaded on August 06,2024 at 08:58:20 UTC from IEEE Xplore. Restrictions apply.
shown in Fig.2b. Then the PRBs are arranged to fill the time-
frequency grid of B Hz and Tf sec. The number of PRBs in
a given frame is,
B Tf
NP RB =
12.2μ .Δf 14.2−μ .T
The source bits are encoded using LDPC code and then passed
through the symbol mapper. The modulated QAM symbols
are arranged in the NP RB number of PRBs to form the time-
frequency signal x́(m, n), m ∈ N[0 2−μ M − 1] and n ∈
N[0 2μ N − 1] in a frame. Then, the nth time domain OFDM
symbol can be given as,
(a) RCP-OTFS 2−µ
M−1
µ
Δf (t−n2−µ T )
śn (t) = x́(m, n)ej2πm2 (23)
m=0
and Sj1 denotes the set of constellation symbols where the bit
B. Receiver
bjη = 0 and bjη = 1 respectively for j = 0, 1, · · · , K − 1.
These LLRs are then fed into the LDPC decoder to decode We assume that the CP duration is greater than maximum
data. Let L denotes a matrix where L(η, j) = LLR(bjη |x̂(η)) excess delay of channel and OFDM system’s symbol duration
for η = 1, 2, · · · , M N and j = 0, 1, · · · , K − 1. L is reshaped and subcarrier bandwidth are less than coherence time and
to Lcl × Ncw matrix where Lcl and Ncw denote the LDPC coherence bandwidth of the channel respectively. Then, after
codeword length and number of codewords respectively. Each removal of CP and applying DFT on the residual signal, the
column of L subsequently regenerates message word mι for received time-frequency symbol ý(m, n) can be expressed as,
ι = 1, 2 · · · , Ncw using the Min-Sum algorithm [15] employed ý(m, n) = h́(m, n)x́(m, n) + v́(m, n) (25)
by the LDPC decoder and is collected as the recovered data.
where, m ∈ N[0 2−μ M − 1] is the subcarrier index and
III. VSB OFDM OVERVIEW n ∈ N[0 2μ N − 1] is the OFDM symbol index. h́(m, n) ∈ C
defined in (15) is the channel coefficient and v́(m, n) ∈ C is
We consider a VSB-OFDM system with total frame duration white guassian noise with variance σV2 at mth subcarrier of
Tf sec. and bandwidth B Hz. We have a base sub-carrier band- nth OFDM symbol.
width of Δf Hz and variability in the sub-carrier bandwidth For channel estimation, the RS in the PRB are used to obtain
is introduced by the parameter μ. For a given μ, the subcarrier the estimates at the pilot location using MMSE estimation as,
bandwidth is 2μ Δf Hz. We have total 2−μ M number of sub-
carriers and 2μ N number of OFDM symbols having 2−μ T ˆ x́(mRS , nRS )† ý(mRS , nRS )
h́(mRS , nRS ) = (26)
symbol duration , thus B = M Δf and Tf = N T . x́(mRS , nRS ) 2 + σV2
where (mRS , nRS ) = {(0, 0), (6, 0), (3, 4), (9, 4), (0, 7), (6, 7),
A. Transmitter
(3, 11), (9, 11)} are the locations of RS in a PRB.
We use the Physical Resource Block (PRB) frame structure The obtained channel estimates at the RS locations in
of 5G NR for VSB-OFDM system. Each PRB consists of 12 PRBs are interpolated to get the channel estimates at the data
subcarriers × 14 time slots with 8 reference signals (RS) as locations using DFT interpolation along frequency axis and
Authorized licensed use limited to: BMS College of Engineering. Downloaded on August 06,2024 at 08:58:20 UTC from IEEE Xplore. Restrictions apply.
linear interpolation along time axis as described in [16]. OFDM system, we consider same frame size as RCP-OTFS.
The estimate of data symbols are obtained at the receiver as, Time-frequency slots for block OFDM can be evaluated as
ý(m, n) already described in Sec.II-B, thus Mbof dm = 128, Nbof dm =
ˆ
x́(m, n) = (27) 512, and Δfbof dm = 60 KHz. It may be noted that we use the
ˆ
h́(m, n) same LMMSE based interference cancellation equalizer for
The estimated symbols are used to generate the channel LLR RCP-OTFS and block OFDM so that any difference in BLER
values of bits corresponding to the symbol by substituting performance can be attributed to interleaving only. The SNR
T B
σ 2 = diag{vec{σV2 ef f }} in (22) where σV2 ef f (m, n) = losses due to CP, which are given as 10log10( cp M ) = 0.3
2 Tcp B
σV dB for VSB-OFDM and 10log10 ( MN ) = 0.0023 dB for
ˆ . Then, the LLRs are used to recover the data as
h́(m,n)2 OTFS. These SNR losses are also adjusted in the results.
described in II-D. Moreover, same transmit power is ensured for all the transmis-
IV. P ILOT P OWER IN OTFS AND VSB-OFDM sion schemes. Doppler is generated following Jakes spectrum.
The CP is chosen long enough to accommodate the maximum
In this section we describe the pilot structure to be used in excess delay of the channel.
OTFS as in [14]. We also describe the pilot structure used
for evaluating VSB-OFDM. In the performance evaluation,
we intend to keep same total transmit power for OTFS and 10 0
Parameter Value
Carrier Frequency(fc ) 6 GHz
Bandwidth(B) 7.68 MHz
10 -1
Frame Time(Tf ) 10 ms
Subcarrier Bandwidth(Δf ) 15 KHz
RCP-OTFS parameters M=512, N=128, Kp =80, Lp =16
BLER
Authorized licensed use limited to: BMS College of Engineering. Downloaded on August 06,2024 at 08:58:20 UTC from IEEE Xplore. Restrictions apply.
ΔP = 34 dB and ΔP = 28 dB in Fig.3. At ΔP = 34 OTFS is less sensitive to estimation error. Furthermore, the
dB, which arises when we consider same PT for VSB-OFDM percentage resource used for pilots and CP overhead in OTFS
and RCP-OTFS. It can be observed that PAPR of RCP-OTFS is less than OFDM, and hence the spectral efficiency (SE)
is nearly 3.5 dB higher than that of VSB-OFDM (μ = 0). of OTFS is higher than OFDM. Finally, it can be said that
We therefore reduce ΔP so that PAPR may be reduced. We due to OTFS’s high SE, better resilience to ICI because of
have found that for given configuration, if ΔP ≤ 28 dB, interference canceling receiver and interleaving, which is not
there is no further reduction in PAPR of RCP-OTFS signal usually available in an OFDM system with single tap equalizer,
s than that of OFDM, and it is very close to that of VSB- OTFS may be recommended as transmission technology for
OFDM with μ = 0. Therefore, we have kept the ΔP = 28 future generations of wireless communication systems.
dB in the performance evaluation presented here. Further, the
reduction in ΔP implies that power of data symbols in RCP- R EFERENCES
OTFS is higher than that of VSB-OFDM by 0.154dB/symbol. [1] L. Liang, H. Peng, G. Y. Li, and X. Shen, “Vehicular Communications:
Therefore, there is an additional gain in SNR for OTFS. A Physical Layer Perspective,” IEEE Transactions on Vehicular Tech-
Figure 4 presents the performance of different waveforms nology, vol. 66, no. 12, pp. 10 647–10 659, Dec. 2017.
[2] R. Hadani, S. Rakib, M. Tsatsanis, A. Monk, A. J. Goldsmith, A. F.
for 16 QAM modulation at the vehicular speed of 500 kmph. Molisch, and R. Calderbank, “Orthogonal Time Frequency Space Mod-
It can be observed that performance of VSB-OFDM becomes ulation,” in 2017 IEEE Wireless Communications and Networking Con-
better with increasing value of μ as expected and supported by ference (WCNC), Mar. 2017, pp. 1–6.
[3] P. Raviteja, Y. Hong, E. Viterbo, and E. Biglieri, “Practical Pulse-
literature. This is because the effect of Doppler spread reduces Shaping Waveforms for Reduced-Cyclic-Prefix OTFS,” IEEE Transac-
with increased sub-carrier bandwidth as post processing SNR tions on Vehicular Technology, vol. 68, no. 1, pp. 957–961, Jan. 2019.
(Γ) for OFDM under Doppler, ∝ ( νΔf )2 [17]. It can also [4] P. Raviteja, E. Viterbo, and Y. Hong, “OTFS Performance on Static
max Multipath Channels,” IEEE Wireless Communications Letters, pp. 1–1,
be observed that OTFS outperforms VSB-OFDM even with 2019.
μ = 3. [5] R. Hadani and A. Monk, “OTFS: A New Generation of
At BLER of 10−1 , OTFS has a SNR gain of 3.5 dB Modulation Addressing the Challenges of 5g,” arXiv:1802.02623
[cs, math], Feb. 2018, arXiv: 1802.02623. [Online]. Available:
over VSB-OFDM (μ = 3). This gain further increases to http://arxiv.org/abs/1802.02623
nearly 5 dB at BLER of 10−2 . This result can be attributed [6] S. Tiwari, S. S. Das, and V. Rangamgari, “Low complexity LMMSE
to two reasons (i) OTFS has an interleaving gain and, (ii) receiver for OTFS,” IEEE Communications Letters, pp. 1–1, 2019.
[7] P. Raviteja, K. T. Phan, Y. Hong, and E. Viterbo, “Interference Can-
OTFS uses ICI cancellation LMMSE receiver whereas VSB- cellation and Iterative Detection for Orthogonal Time Frequency Space
OFDM uses a simple single-tap receiver. Block OFDM has Modulation,” IEEE Transactions on Wireless Communications, vol. 17,
a SNR gain of around 2.5 dB and 3 dB at BLER 10−1 and no. 10, pp. 6501–6515, Oct. 2018.
[8] S. S. Das and I. Rahman, “Method for transmitting data in a wireless
10−2 respectively. This gain is due to the use of interference network, indian patent, granted, number 260999,” Patent 260 999, 2005.
cancelling receiver in block OFDM. Further, at the BLER of [9] S. S. Das, E. De Carvalho, and R. Prasad, “Variable Sub-Carrier Band-
10−1 and 10−2 , OTFS has gain of 1 dB and 2 dB respectively widths in OFDM Systems,” in 2007 IEEE 65th Vehicular Technology
Conference - VTC2007-Spring, Apr. 2007, pp. 1866–1870, iSSN: 1550-
over block OFDM. As both block OFDM and RCP-OTFS 2252.
use FEC and LMMSE equalizer while the difference in the [10] S. Das, M. Rahman, and F. Fitzek, “Multi rate orthogonal frequency
trasmitted signal is the time-interleaving, henceforth it can division multiplexing,” in IEEE International Conference on Communi-
cations, 2005. ICC 2005. 2005, vol. 4, May 2005, pp. 2588–2592 Vol.
be said that this gain is primarily due to the interleaving. 4, iSSN: 1550-3607, 1938-1883.
We have observed similar results for other vehicular speeds [11] S. S. Das, E. D. Carvalho, and R. Prasad, “Variable sub-carrier band-
as well as for other modulation schemes and chose to not width in OFDM framework,” Electronics Letters, vol. 43, no. 1, pp.
46–47, Jan. 2007.
include them here for brevity. Thus, it can be concluded [12] F. Wiffen, L. Sayer, M. Z. Bocus, A. Doufexi, and A. Nix, “Comparison
that OTFS performs better than other contender waveforms in of OTFS and OFDM in Ray Launched sub-6 GHz and mmWave Line-
vehicular scenarios by virtue of its interleaving and LMMSE of-Sight Mobility Channels,” in 2018 IEEE 29th Annual International
Symposium on Personal, Indoor and Mobile Radio Communications
interference cancellation. (PIMRC), Sep. 2018, pp. 73–79, iSSN: 2166-9589, 2166-9570.
[13] 3GPP, Study on Channel Model for Frequencies from 0.5 to 100 GHz
VI. C ONCLUSIONS (Release 15), TR 38.901, V 15.0, 3GPP, 2018.
In this work, we have detailed the signal generation of OTFS [14] P. Raviteja, K. T. Phan, and Y. Hong, “Embedded Pilot-Aided Channel
Estimation for OTFS in DelayDoppler Channels,” IEEE Transactions on
and explained how it could be viewed in an alternative form as Vehicular Technology, vol. 68, no. 5, pp. 4906–4917, May 2019.
block OFDM with time-interleaving. From performance com- [15] Jianguang Zhao, F. Zarkeshvari, and A. H. Banihashemi, “On imple-
parison results, we conclude that block OFDM outperforms mentation of min-sum algorithm and its modifications for decoding low-
density parity-check (ldpc) codes,” IEEE Transactions on Communica-
VSB-OFDM by 3 dB due to the use of interference cancelling tions, vol. 53, no. 4, pp. 549–554, April 2005.
receiver. Inspite of using the same receiver as block OFDM, [16] M. J. Fernandez-Getino Garcia, J. M. Paez-Borrallo, and S. Zazo,
OTFS outperforms block OFDM. The only difference being “Dft-based channel estimation in 2d-pilot-symbol-aided ofdm wireless
systems,” in IEEE VTS 53rd Vehicular Technology Conference, Spring
the time-interleaving of the transmitted signal which provides 2001. Proceedings (Cat. No.01CH37202), vol. 2, May 2001, pp. 810–
an additional gain of 2 dB. Overall, OTFS has 5 dB gain over 814 vol.2.
VSB-OFDM for 16 QAM in LTV channel with FEC. [17] S. S. Das, E. D. Carvalho, and R. Prasad, “Performance Analysis of
OFDM Systems with Adaptive Sub Carrier Bandwidth,” IEEE Transac-
We have also found that the pilot power of OTFS can tions on Wireless Communications, vol. 7, no. 4, pp. 1117–1122, Apr.
be made lower than OFDM, since the channel estimation in 2008.
Authorized licensed use limited to: BMS College of Engineering. Downloaded on August 06,2024 at 08:58:20 UTC from IEEE Xplore. Restrictions apply.