IEEE COMMUNICATIONS LETTERS, VOL. 23, NO.

12, DECEMBER 2019 2205

Low complexity LMMSE Receiver for OTFS

Shashank Tiwari , Suvra Sekhar Das, Member, IEEE, and Vivek Rangamgari

Abstract— Orthogonal time frequency space modulation is a posses computational burden for OTFS as the time-frequency
two dimensional (2D) delay-Doppler domain waveform. It uses grid size in OTFS is very large. Linear minimum mean
inverse symplectic Fourier transform (ISFFT) to spread the signal square error (LMMSE) receiver which is well known for its
in time-frequency domain. To extract diversity gain from 2D
spreaded signal, advanced receivers are required. In this work, interference cancellation capabilities [6], [9], is extended to a
we investigate a low complexity linear minimum mean square low complexity form in this work.
error receiver which exploits sparsity and quasi-banded structure Direct implementation of LMMSE receiver require com-
of matrices involved in the demodulation process which results plexity in the order of O(M 3 N 3 ), where M and N are
in a log-linear order of complexity without any performance total number frequency and time slots respectively. When the
degradation of BER.
values of M and N are in order of 100’s, the complexity of
Index Terms— Orthogonal time frequency space modulation LMMSE receiver becomes extraordinarily large. To the best
(OTFS), orthogonal frequency division multiplexing (OFDM), of the authors’ knowledge, not much attention has been paid
time-varying channel, minimum mean square error (MMSE)
receiver, LU factorization, low complexity. towards low complexity design of LMMSE receiver for OTFS
in literature. We present a low complexity LMMSE receiver
I. I NTRODUCTION which has a complexity in the order of O(M N log2 (N ))
without any degradation in BER performance.
F IFTH generation new radio (5G-NR) [1] uses multi-
numerology Orthogonal frequency division multiplexing
(OFDM) system to cater to different requirements of 5G
We use the following notations throughout the letter. We let
x, X and x represent vectors, matrices and scalars respec-
such as support for higher vehicular speed scenario and high tively. The superscripts (−)T and (−)† indicate transpose and
phase noise. Although sub-carrier bandwidth in 5G-NR can conjugate transpose operations, respectively. Notations 0, IN
be increased to combat Doppler spread, the provision of and WL represent zero matrix, identity matrix with order
proportional decrement of CP length to retain OFDM symbol N and L-order normalized inverse discrete Fourier transform
efficiency induces interference when both delay and Doppler (IDFT) matrix respectively. Kronecker product operator is
spreads are significant. Orthogonal time frequency space mod- given by ⊗. The operator diag{x} creates a diagonal matrix
ulation (OTFS) has been recently proposed in [2] to effi- with the elements of vector x. Circulant matrix is represented
ciently transfer data in such channel conditions. In OTFS, data by circ{x} whose first column is x. Notations E{−} and −
symbols are spread across available time-frequency resources are expectation and ceil operators respectively. Column-wise
which can be exploited to extract diversity gain. Different vectorization of matrix (X) is represented by vec{X}. Natural
receivers have been proposed in the literature [3]–[8], which numbers are denoted by N √ . Complex conjugate value of x is
achieve such diversity gain. given by x̄ whereas j = −1.
When exposed to a time variant channel (TVC), OTFS
suffers from inter-symbol and inter-carrier interference [5]. II. S YSTEM M ODEL
Hence a simple matched filter receiver as in [3] is unable We consider an OTFS system with M number of sub-
to suppress the interference sufficiently. There can be two carriers having Δf sub-carrier bandwidth and N number of
types of receivers, namely (i) linear receivers (LRx) and symbols having T symbol duration. Total bandwidth B =
(ii) non-linear receivers (NLRx). NLRx (such as in [4]–[7]) M Δf and total duration Tf = N T . Moreover OTFS system
have near maximum likelihood (ML) performance but have is critically sampled i.e. T Δf = 1.
iterative structure and high complexity. On the other hand
LRx are simple in the structure but have relatively poorer A. Transmitter
performance than non-linear receivers. As linear processing
QAM modulated data symbols, d(k, l) ∈ C, k ∈ N
requires inversion and multiplication of matrices, LRx still
[0 N −1], l ∈ N[0 M −1], are arranged over Doppler-delay lat-
Manuscript received September 5, 2019; accepted September 29, 2019. tice Λ = {( NkT , MΔfl ¯  , l )] =
)}. We assume that E[d(k, l)d(k
2  
Date of publication October 4, 2019; date of current version December 10, σd δ(k − k , l − l ), where δ is Dirac delta function. Doppler-
2019. The work was supported by the AWF project with contract number
IIT/SRIC/GS/AWF/2019-20/069 funded by WIPRO LIMITED, Bengaluru, delay domain data d(k, l) is mapped to time-frequency domain
India. The associate editor coordinating the review of this letter and data X(n, m) on lattice Λ⊥ = {(nT, mΔf )}, n ∈ N[0 N −1]
approving it for publication was N. I. Miridakis. (Corresponding author: and m ∈ N[0 M − 1] by using inverse symplectic fast Fourier
Shashank Tiwari.)
The authors are with the G. S. Sanyal School of Telecommunications, transform (ISFFT). X(n, m) can be given as [2],
Indian Institute of Technology Kharagpur, Kharagpur 721302, India (e-mail: −1 M−1
[email protected]; [email protected]; rkvivek97@ 1 
N 
d(k, l)ej2π[ N − M ] .
nk ml
gmail.com). X(n, m) = √ (1)
Digital Object Identifier 10.1109/LCOMM.2019.2945564 N M k=0 m=0
1558-2558 © 2019 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.

Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on December 05,2022 at 07:04:15 UTC from IEEE Xplore. Restrictions apply.
 2206 IEEE COMMUNICATIONS LETTERS, VOL. 23, NO. 12, DECEMBER 2019

Next, X(n, m) is converted to a time domain signal s(t) We take a 3GPP vehicular channel EVA [11] with vehicular
through a Heisenberg transform as, speed of 500 Kmph. Delay-Doppler channel lengths can be
−1 M−1 computed as, α = 20 M N = 65536 and β = 16

N 
s(t) = X(n, m)g(t − nT )ej2πmΔf (t−nT ) , (2) M N = 65536.
n=0 m=0

where, g(t) is transmitter pulse of duration T . It has C. Receiver

been shown in [10] that non rectangular pulse induces After removal of CP at the receiver, received signal can be
non-orthogonality which degrades BER performance. Thus, written as [10],
in
 this work, we assume a rectangular pulse i.e. g(t) =
1 if 0 ≤ t ≤ T r = Hs + n, (7)
.
0 otherwise where, n is white Gaussian noise vector of length M N with
To obtain discrete time representation of OTFS transmis- elemental variance σn2 and H is a M N × M N channel matrix
T
sion, s(t) is sampled at the sampling interval of M [10]. given as,
We collect samples of s(t) in s = [s(0) s(1) · · · s(M N − 1)]

P
and QAM symbols d(k, l) are arranged in M × N matrix as, H= hp Πlp Δkp , (8)
⎡ ⎤
d(0, 0) d(1, 0) · · · d(N − 1, 0) p=1
⎢ d(0, 1) d(1, 1) · · · d(N − 1, 1) ⎥
⎢ ⎥ with Π = circ{[0 1 0 · · · 0]T
MN ×1 } is a circulant delay matrix
D=⎢ .. .. . .. ⎥. 1 M N −1
⎣ . . . . . ⎦ and Δ = diag{[1 ej2π M N · · · ej2π M N ]T } is a diagonal
d(M − 1, 0) d(M − 1, 1) · · · d(N − 1, M − 1) Doppler matrix. We further process r through a LMMSE
(3) equalizer which results in estimated data vector,
2
d̂ = (HA)† [(HA)(HA)† + I]−1 r.
ρn
Using above formulations, s can be given as [10], ρd2
(9)
s = vec{DWN }. (4)
III. L OW C OMPLEXITY LMMSE R ECEIVER FOR OTFS
Alternatively, if d = vec{D}, transmitted signal can also be
When g(t) is rectangular, A in (5) is unitary. Thus (9) is
written as matrix-vector multiplication,
simplified to,
s = Ad, (5) Heq


where A = WN ⊗ IM is OTFS modulation matrix. Finally, † † †
d̂ = A H [HH +
2
ρn
I]−1 r. (10)
ρd2
a cyclic prefix (CP) of length α ≥ α − 1 is appended at the 

starting of the s, where α is channel delay length described rce =Heq r
in Sec. II-B. Thus LMMSE equalization can be performed as a two stage
equalizer. In the first stage, LMMSE channel equalization is
B. Channel performed to obtain rce = Heq r. Second stage is a OTFS
We consider a time varying channel with P paths having matched filter receiver to obtain d̃ = A† rce . We will show
hp complex attenuation, τp delay and νp Doppler value for in Sec. III-C that the implementation of d̃ = A† rce is simple
pth path where p ∈ N[1 P ]. Delay-Doppler channel spreading which requires MN 2 log2 (N ) complex multiplications (CMs).
function can be given as, But direct implementation of rce = Heq r requires inversion
ρ2
of Ψ = HH† + ρn2 I and multiplication of H† which need

P
d
h(τ, ν) = hp δ(τ − τp )δ(ν − νp ). (6) O(M 3 N 3 ) CMs. Thus, we need to reduce the complexity of
p=1 rce = Heq r. To do so, we investigate the structure of matrices
involved in channel equalization in Sec. III-A.
The delay and Doppler values for pth path is given as τp =
lp kp
MΔf and νp = N T , where lp ∈ N[0 M − 1] and kp ∈ N 2
A. Structure of Ψ = [HH† +
ρn
[0 N − 1] are dealy and Doppler bin number on Doppler- ρd2
I]
delay lattice Λ for pth path. We assume that N and M are
Using (8), HH† can be expressed as,
sufficiently large so that there is no effect of fractional delay
and Doppler on the performance. We also assume the perfect 
P 
P

knowledge of (hp , lp , kp ), p ∈ N[0 P − 1], at the receiver HH† = hp Δkp Πlp h̄s Δ−ks Π−ls . (11)
p=1 s=1
as in [3]–[8]. Let τmax and νmax be the maximum delay
and Doppler spread. Channel delay length α = τmax M  Since Π is a circulant matrix, it can be verified that Πlp =
and channel Doppler length, β = νmax N T . Typically, WΔ−lp W† . Therefore,
α M N as well as β M N which dictates the system
matrices to be sparse (as will be seen in Sec. III-A). For 
P 
P 
P
HH† = |hp |2 I + hp h̄s Πlp −ls Δkp −ks (12)
example, if we consider an OTFS system with Δf = 15 KHz, p=1 p=1 s=1
carrier frequency, fc = 4 GHz, N = 128 and M = 512. p=s p=s

Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on December 05,2022 at 07:04:15 UTC from IEEE Xplore. Restrictions apply.
 TIWARI et al.: LOW COMPLEXITY LMMSE RECEIVER FOR OTFS 2207

Algorithm 1 Computation of Y = Γ−1 X

1: Given : a lower triangular matrix ΓQ×Q and XQ×θ
−1
2: Output : YQ×θ = ΓQ×Q XQ×θ
3: for s = 0 : θ do
X(0,s)
4: Y(0, s) =
Γ(0,0)
5: for k = 1 : θ do
1

Structure of Ψ = HH† +
2
σn 6: Y(k, s) = Γ(k,k) X(k, s)− k−1i=1 Γ(k, k−i)Y(k−i, s)
Fig. 1. 2 I
σd
matrix and its LU factorization.
7: end for
8: for k = θ + 1 : Q do
Using (12), Ψ becomes,  −1
9: Y(k, s) = 1 X(k, s)− P i=1 Γ(k, k−i)Y(k−i, s)
Γ(k,k)

P
2 
P 
P
10: end for
(|hp |2 + hp h̄s Πlp −ls Δkp −ks .
ρn
Ψ= ρd2
)I + 11: end for
p=1 p=1 s=1
p=s p=s
(13) Algorithm 2 Computation of r(1) = L−1 r
Following (13), it can be concluded that the maximum shift 1: Given : a quasi banded lower triangular matrix LMN ×MN
of diagonal elements in Δ can be ±(α − 1). Additionally due and rMN ×1
(1) −1
to the cyclic nature of shift, Ψ is quasi-banded with bandwidth 2: Output : rMN ×1 = LMN ×MN rMN ×1
of 2α − 1 as depicted in Fig. 1. As discussed in Sec. II-B, (1)
3: r (0) = r(0)
α M N , Ψ is also sparse for typical wireless channel. 4: for k = 1 : α − 1 do
k−1
Since, we need to implement Ψ−1 in order to realize LMMSE 5: r(1) (k) = r(k) − i=1 L(k, k − i)r
(1)
(k − i)
receiver, we propose a low complexity LU decomposition of 6: end for
Ψ in Sec. III-B. 7: for k = α : Q do
α−1
8: r(1) (k) = r(k) − i=1 L(k, k − i)r
(1)
(k − i)
B. Low Complexity LU Factorization of Ψ 9: end for
10: for k = Q + 1 : M N − 1 do
To implement the low complexity LU factorization of Ψ, MN −1
we propose following partition of Ψ (by considering, θ = α−1 11: r(1) (k) = r(k) − i=1 L(k, k − i)r(1) (k − i)
and Q = M N − θ). 12: end for
     
TQ×Q BQ×θ L 0 U E
= Q×Q Q×θ × Q×Q Q×θ
Sθ×Q Cθ×θ Vθ×Q Fθ×θ 0θ×Q Gθ×θ Note on the Non-Singularity of L and U: For LMMSE

 
 

Ψ L U processing, L and U need to be inverted (as will be discussed
(14) in Sec. III-C). We next discuss the non-singularity of L and U.
As HH† is a hermitian matrix, its a positive semi-definite
Using the partition in (14), following equalities hold ρ2
matrix. Since ρn2 > 0 for finite SNR ranges, Ψ is a positive
T=LU (15) d
definite matrix; therefore, Ψ is invertible. As diagonal values
−1
E=L B (16) of L are unity, L is non-singular. Further, non-singularity of
V = S U −1 (17) U is a consequence of non-singularity of Ψ [13].
FG = C − VE (18)
C. Computation of r
Next we will discuss the solution of (15-18) to compute LU
factorization of Ψ. Since T is a banded matrix its LU decom- After LU decomposition of Ψ, rce is simplified to,
position can be computed using low complexity algorithm r(2)


presented in [12]. L−1 B can be computed using forward
rce = H† U−1 L

−1
r . (19)
substitution algorithm for lower triangular banded matrix as
explained in Algorithm 1. We can compute (17) in following r(1)

two steps. As U † is a lower triangular banded matrix, in the As L is a quasi-banded lower triangular matrix, r(1) = L−1 r
first step, we compute V† = (U † )−1 S† using Algorithm 1. can be computed using low complexity forward substitution as
Finally, V can be computed simply by taking hermitian of V† . explained in Algorithm 2. r(2) = U−1 r(1) can be computed
As θ M N , even a direct computation of (18) requires using Algorithm 3.
O(θ2 M N ) computations. As F is a lower triangular matrix Using the definition of H, rce = H† r(2) can be given as,
and G is a upper triangular matrix, F and G can be computed
using LU decomposition of (18). Pivotal Gaussian elimination 
P
rce = h̄p Δ−kp Π −lp (2)

r  (20)
algorithm [13] can be used to compute LU decomposition of
p=1 circular shift
(18) without much increase in complexity. It should be noted
that diagonal values of L and F are unity. Thus, diagonal To compute rce , r(2) is first circularly shifted by delay −lp
values of L are also unity. and then multiplied by h̄p diag{Δ−kp } by using point-to-point

Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on December 05,2022 at 07:04:15 UTC from IEEE Xplore. Restrictions apply.
 2208 IEEE COMMUNICATIONS LETTERS, VOL. 23, NO. 12, DECEMBER 2019

TABLE I
Algorithm 3 Computation of r(2) = U−1 r(1)
C OMPUTATIONAL C OMPLEXITY OF D IFFERENT
1: Given : a quasi banded upper triangular matrix UMN ×MN O PERATIONS IN O UR P ROPOSED R ECEIVER
(1)
and rMN ×1
(2) −1 (1)
2: Output : rMN ×1 = UMN ×MN rMN ×1
(1)
r (MN −1)
3: r(2) (M N − 1) = U(MN −1,MN −1)
4: for k = M N − 2 : M N −2α do
1 MN −k−1
5: r(2) (k) = U(k,k) r(1)(k)− i=1 U(k, k+i)r(2) (k+i)
6: end for
7: for k = α : Q do 
1
8: r(2) (k) = U(k,k) r(1) (k) − α i=1 U(k, k + i)r
(2)
(k + i)
MN −1 (2)
− r=MN −α U(k, r)r (r)
9: end for

TABLE II
C OMPUTATIONAL C OMPLEXITY OF D IFFERENT R ECEIVERS

Fig. 2. Our proposed low complexity OTFS-MMSE receiver.

multiplication for each path p. All vectors obtained in above

step are finally summed to obtain rce .
Instead of directly computing d̂ as A† rce , we first reshape
rce to a M × N size R matrix as,
⎡ ⎤
rce (0) rce (M ) · · · rce (M N − N )
⎢ rce (1) rce (M + 1) · · · rce (M N − N + 1)⎥
⎢ ⎥
R=⎢ .. .. .. .. ⎥.
⎣ . . . . ⎦
rce (M − 1) rce (2M − 1) · · · rce (M N − 1)
(21)
Then we perform
†
d̂ = vec{RWN }, (22)
which can be implemented using M number of N -point FFT
operations. Fig. 2 describes the signal processing steps of our
proposed low complexity LMMSE receiver.

D. LMMSE Receiver for OFDM Over TVC

Low complexity receiver discussed for OTFS can easily be Fig. 3. Computation complexity comparison of different receivers.
extended to OFDM by setting A = IN ⊗ WM . To do so,
rce = Heq r is performed by computing (14-20) as discussed
†
in Sec. III-B and III-C. Further, d̂ = (IN ⊗ WM )rce can be
FFT algorithm using 2c log2 (c) CMs [14]. The complexity of
computed using N number of M -point FFTs.
the proposed receiver can be computed using the structure
provided in Sec. III-B. Computation of c × c matrix-matrix
IV. R ESULT
multiplication, matrix inversion and LU decomposition require
A. Computational Complexity c3 2c3 2c3
2 , 3 and 3 CMs respectively. Total CMs required to
In this section, we present the computational complexity of compute different operations in our receiver are presented
our proposed LMMSE receiver. We calculate the complexity in Table I. CMs required for different receivers is presented
in terms of total number of complex multiplications (CMs). in Table II. It is evident that the our proposed receiver has
c-point FFT and IFFT can be implemented using radix-2 complexity of O(M N [log2 (N ) + α2 + P 2 β]).

Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on December 05,2022 at 07:04:15 UTC from IEEE Xplore. Restrictions apply.
 TIWARI et al.: LOW COMPLEXITY LMMSE RECEIVER FOR OTFS 2209

TABLE III Doppler is generated using Jake’s formula, νp = νmax cos(θp ),

S IMULATION PARAMETERS where θp is uniformly distributed over [−π π]. The CP is
chosen long enough to accommodate the wireless channel
delay spread. Figure 4 compares BER performance of our
proposed receiver with the direct ones using (9). It can be
observed that the proposed receiver does not suffer from
any performance degradation when compared with the direct
ones. It can also be observed that OTFS-LMMSE receiver can
extract diversity gain, for instance at the BER of 5 × 10−4 ,
OTFS-LMMSE receiver achieves an SNR gain of 13 dB over
OFDM-MMSE receiver.

V. C ONCLUSION
In this letter, we have proposed a low complexity LMMSE
receiver for OTFS waveform. The proposed technique exploit
sparsity and quasi banded structure of matrices involved
in LMMSE processing without incurring any performance
penalty. We have shown that our proposed receiver can achieve
upto 107 times complexity reduction over direct implementa-
tion. Such substantial reduction with linear receiver is expected
to provide an impetus for practical realization of future wire-
less OTFS based systems.

R EFERENCES
[1] NR Physical Channels and Modulation (Release 15), document TS
38.211, V 15.2, 3GPP, 2018.
[2] R. Hadani et al., “Orthogonal time frequency space modulation,” in Proc.
IEEE Wireless Commun. Netw. Conf. (WCNC), Mar. 2017, pp. 1–6.
[3] A. Farhang, A. R. Reyhani, L. E. Doyle, and B. Farhang-Boroujeny,
Fig. 4. BER comparison of our proposed receiver with the direct one using “Low complexity modem structure for OFDM-based orthogonal time
(9) for 4 QAM modulation. frequency space modulation,” IEEE Wireless Commun. Lett., vol. 7,
no. 3, pp. 344–347, Jun. 2018.
[4] K. R. Murali and A. Chockalingam, “On OTFS modulation for high-
Doppler fading channels,” in Proc. Inf. Theory Appl. Workshop (ITA),
To evaluate the complexity reduction achieved by our pro- Feb. 2018, pp. 1–10.
posed receiver, we consider an OTFS system with Δf = [5] P. Raviteja, K. T. Phan, Y. Hong, and E. Viterbo, “Interference
15 KHz, fc = 4 GHz and vehicular speed of 500 kmph. cancellation and iterative detection for orthogonal time frequency
space modulation,” IEEE Trans. Wireless Commun., vol. 17, no. 10,
We consider two 3GPP vehicular channel models [11] namely pp. 6501–6515, Oct. 2018.
(i) extended vehicular A (EVA) with P = 9 and τmax = [6] R. Hadani and A. Monk, “OTFS: A new generation of modula-
2.51 μ sec , and (ii) extended vehicular B (EVB) with P = 6 tion addressing the challenges of 5G,” Feb. 2018, arXiv:1802.02623.
[Online]. Available: http://arxiv.org/abs/1802.02623
and τmax = 20 μ sec. Two block durations are assumed [7] R. Hadani, S. S. Rakib, A. Ekpenyong, C. Ambrose, and S. Kons,
namely (i) small block with N = 16 and Tf = 1.1 msec. “Receiver-side processing of orthogonal time frequency space modulated
and (ii) large block with N = 128 and Tf = 8.85 msec. signals,” U.S. Patent 2019 0 081 836 A1, Mar. 14, 2019.
[8] L. Li, Y. Liang, P. Fan, and Y. Guan, “Low complexity detection
The complexity presented in Table II is plotted in Figure 3 algorithms for OTFS under rapidly time-varying channel,” in Proc. IEEE
for M ∈ [2 4096]. It is evident from the figure that for EVA 89th Veh. Technol. Conf. (VTC-Spring), Apr. 2019, pp. 1–5.
channel our proposed receivers require up-to 107 and 105 times [9] Y. Jiang, M. K. Varanasi, and J. Li, “Performance analysis of ZF and
MMSE equalizers for MIMO systems: An in-depth study of the high
lower CMs than direct ones using (9) for large and small block SNR regime,” IEEE Trans. Inf. Theory, vol. 57, no. 4, pp. 2008–2026,
respectively. Whereas for EVB channel, our proposed receiver Apr. 2011.
need 2.5×105 and 3000 times lesser CMs over the direct ones [10] P. Raviteja, Y. Hong, E. Viterbo, and E. Biglieri, “Practical pulse-shaping
waveforms for reduced-cyclic-prefix OTFS,” IEEE Trans. Veh. Technol.,
using (9) for large and small block respectively. This reduction vol. 68, no. 1, pp. 957–961, Jan. 2019.
in complexity gain for EVB channel as compared with EVA [11] Guidelines for Evaluation of Radio Interface Technologies for IMT-
channel is due to increase in α. We can conclude that our Advanced, document Rep. ITU 2135-1, 2009.
[12] D. W. Walker, T. Aldcroft, A. Cisneros, G. C. Fox, and W. Furmanski,
proposed receivers achieve a significant complexity reduction “LU decomposition of banded matrices and the solution of linear
over direct implementation of (9). systems on hypercubes,” in Proc. 3rd Conf. Hypercube Concurrent
Comput. Appl. (CP), vol. 2. New York, NY, USA: ACM, 1988,
pp. 1635–1655. doi: 10.1145/63047.63124.
B. BER Evaluation [13] G. H. Golub and C. F. Van Loan, Matrix Computations, vol. 3.
Baltimore, MD, USA: The Johns Hopkins Univ. Press, 2012.
Here we present BER performance of the proposed receiver [14] R. E. Blahut, Fast Algorithms for Signal Processing. New York, NY,
in EVA channel. Simulation parameters are given in Table III. USA: Cambridge Univ. Press, 2010.

Authorized licensed use limited to: Indian Institute of Technology (ISM) Dhanbad. Downloaded on December 05,2022 at 07:04:15 UTC from IEEE Xplore. Restrictions apply.