Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Lagrange Vandermonde Division Multiplexing

Kamel Tourki, Rostom Zakaria, and Mérouane Debbah


Mathematical and Algorithmic Sciences Lab, Huawei Technologies France SASU, France
Email: firstname.lastname@huawei.com

Abstract—Next generation networks will support diverse use This explains the complex receiver that has been adopted
cases that require higher flexibility in the resource allocation. To where the inversion of the matrices Vm and Γm will cost
this end, we propose a new waveform referred to as Lagrange O(P 3 ) of operations [5], [6]. Furthermore, the signature roots,
Vandermonde division multiplexing (LVDM) that generalizes
the zero padding orthogonal frequency division multiplexing needed to build Gm and Vm , have not been optimized since
(ZP-OFDM) while guaranteeing the perfect recovery of the the optimization of K signature roots should be performed
transmitted signal. The LVDM transceiver design relies on the over CK where the complexity drastically increases with K.
signature roots that have been judiciously selected to 1) provide a Consequently, the signature roots have been spread over a unit
flexible resource allocation and 2) keep the complexity low where −1
circle, thus Vm reduces to a discrete Fourier transform (DFT)
one-tap equalization is adopted. Carrying out the simulations
in both frequency selective and 3GPP channels, our proposed matrix [1].
waveform outperforms the ZP-OFDM where performance results
B. Contributions
are shown in terms of bit error ratio (BER). The LVDM achieves
a signal-to-noise ratio gain of 5 dB over the ZP-OFDM when The 3rd Generation Partnership Project (3GPP) decided to
applying the optimized power allocation over subcarriers. use OFDM-based waveforms [7] while relevant technologies,
Index Terms—Lagrange polynomials, Vandermonde matrix, which have been considered too immature for 5G, should be
multicarrier, B5G.
reconsidered. In this paper, we propose:
I. I NTRODUCTION • A new multicarrier waveform, referred to as Lagrange

As fifth generation (5G) research is maturing towards a Vandermonde division multiplexing (LVDM), that gener-
global standard, the research community starts focusing on alizes the zero padded (ZP)-OFDM while guaranteeing
the development of beyond 5G (B5G) solutions. Hence, new the perfect recovery (PR) of the transmitted signal. By
waveform and duplexing technologies beyond orthogonal fre- defining the signature roots, the modulation and demod-
quency division multiplexing (OFDM) will be investigated in ulation matrices are determined through the Lagrange
the coming years while identifying the key drivers, challenges polynomials’ coefficients and the corresponding Vander-
and essential research questions related to B5G. Furthermore, monde matrix, respectively.
• The signature roots are chosen to be distributed over the
flexibility enablers for transceiver architecture will be needed j2πk
to boost the data rate. circle of radius a, ρk = a e K , which reduces the
problem to a single variable optimization.
A. Related Works • We propose a balanced implementation for the
Lagrange-Vandermonde scheme has been proposed in lit- transceiver, where one tap equalization is performed
erature for code-division multiple access (CDMA) systems which drastically reduces the transceiver complexity.
in 1999 and 2000 [1]–[4]. The most generalized frame- • The signature roots provide further implementation flex-
work for quasi-synchronous blind CDMA has been proposed ibility which allow for uniform and optimized power
and referred to as a mutually-orthogonal usercode-receiver allocation over subcarriers.
(AMOUR) system. Therein, every user transmits K symbols • We also provide signature roots refinement that further
through frequency selective channels while using K spreading boosts the system performance.
codes of length Q = M (K + L) + L, where M is the • We validate our results and show that LVDM outperforms
number of users and L is the channel delay spread. These ZP-OFDM in frequency and doubly selective channels.
spreading codes have been derived from Lagrange polynomi- The remainder of this paper is organized as follows. ZP-
als. However, at the receiver side, the mth user uses a de- OFDM is revisited in Section II while useful mathematics and
spreading Vandermonde matrix Gm of size P × Q, where notations introduce in Section III the LVDM waveform that
P = K + L and the pth row presents the coefficients of the will be detailed in Section IV. In Section V, LVDM is analyzed
pth receive filter. The detection is made through the inverse of and simulation results are depicted in Section VI. Finally, the
−1
a Vandermonde (P × P ) matrix, Vm , then a zero forcing results and future works are summarized in Section V.
(ZF) given by the pseudo-inverse of the Toeplitz channel
matrix Γm of size (K × P ). It is worth mentioning that II. R EVISITING ZP-OFDM
the aforementioned schemes have been proposed for single In this section, we provide a brief overview of the ZP-
carrier systems where perfect recovery (PR) condition has not OFDM systems. The baseband discrete-time block equivalent
been respected and single tap equalization cannot be used. model of a standard ZP-OFDM system is depicted in Fig. 1,

978-1-7281-5089-5/20/$31.00 ©2020 IEEE

Authorized licensed use limited to: Carleton University. Downloaded on August 01,2020 at 13:30:44 UTC from IEEE Xplore. Restrictions apply.
Fig. 1: ZP-OFDM Transceiver

Fig. 2: LVDM Transceiver

where the nth K × 1 information block s(n) is first precoded III. U SEFUL M ATHEMATICS AND N OTATIONS
by the inverse discrete Fourier transform (IDFT)
√ K ×K matrix
From a set of K distinct nonzero complex points {ρk }K−1
k=0 ,
FHK with (k, q)th entry exp(j2πkq/K)/ K, to yield the so-
that we will refer to as signature roots, we build
called time domain block vector s̃(n) = FH K s(n), where (.)
H

denotes conjugate transposition. Then, L trailing zeros are • A K × P Vandermonde matrix, given by
padded at each precoded block to yield the P × 1 transmitted ⎡ ⎤
1 ρ−1 · · · ρ1−P
vector s̃zp (n) = Fzp s(n), where Fzp = [IK ; 0]FH K of size
0 0
⎢1 ρ−1 · · · ρ1−P ⎥
P ×K where P = K +L. The received block symbol is given ⎢ 1 1 ⎥ 1−p
VK×P = ⎢ . .. .. ⎥ , vk,p = ρk−1
by ⎣ .. . . ⎦
yzp (n) = H Fzp s(n) + HIBI Fzp s(n − 1) + η(n), (1) 1 ρ−1
K−1 · · · ρ1−P
K−1
(4)
where H is the P × P lower triangular Toeplitz matrix with
first column [h0 · · · hL 0 · · · 0]T with (.)T denoting transpo- • The Lagrange basis polynomials (K polynomials), given
sition, HIBI is the P × P upper triangular Toeplitz matrix by
with first row [0 · · · 0 hL · · · h1], which captures the inter-block K−1 K−1
interference (IBI); and η(n) denotes the P × 1 additive white 1 − ρn z −1
Fk (z) = κk = ri,k z −i
Gaussian noise (AWGN) vector. The ZP block advantage lies n=0
1 − ρn ρ−1
k i=0
n=k
in the all zero L × K matrix 0 which eliminates the IBI, since 
−1
HIBI Fzp = 0. Thus, letting H = [H0 Hzp ] denote a partition = 1 z ··· z 1−K rk (5)
of the P × P convolution matrix H between its first K and
last L columns, the equation (1) reduces to where κk is a tuning factor that normalizes the corre-
sponding filter (Fk ) energy.
yzp (n) = H0 FH
K s(n) + η(n), (2) • Using (5), a Lagrange matrix is formed and given by
The frequency domain (FD) received signal is given by 
R = r0 r1 · · · rK−1
⎡ ⎤
x(n) = F̃yzp (n) = diag ([H0 , · · · , HK−1 ]) s(n) + F̃η(n), r0,0 r0,1 ··· r0,K−1
(3) ⎢ r1,0 r1,1 ··· r1,K−1 ⎥
⎢ ⎥
where we append FK with its first L columns such that =⎢ . . . .. ⎥ (6)
2πk/K
⎣ .. .. .. . ⎦
F̃ K FK (:, 1 : L)] and Hk = C(e
L= [F−j2πkl/K ) =
rK−1,0 rK−1,1 ···
rK−1,K−1
l=0 hl e is the frequency channel response at the
kth subcarrier. However, it is obvious that the kth transmitted  
It is worth mentioning that if ρk = exp j 2πk K , VK×K
symbol in s(n) cannot be recovered when it is hit by a channel reduces to the DFT matrix. Furthermore, using the equations
zero (Hk = 0). In this case, the PR condition could not be (4-6), we can formulate the following identities
satisfied.
The natural question is whether it is possible to 1) design Fk (ρl ) = κk δ(k − l), ∀k, l ∈ {1, · · · , K}, (7)
a precoder/modulation scheme with perfect IBI cancellation VK×K R = diag ([κ0 , · · · , κK−1 ])  Ω, (8)
using a linear receiver with one-tap equalization (low com-
plex transceiver implementation) while 2) satisfying the PR and with a full-banded Toeplitz matrix, H0 (defined above in
condition. We start by providing some useful mathematical Section II) satisfies the following identity
basics and notations that will help us introducing the LVDM
−(P −1) −(K−1)
scheme. [1 ρ−1
k · · · ρk ]H0 = C(ρk )[1 ρ−1
k · · · ρk ], (9)

Authorized licensed use limited to: Carleton University. Downloaded on August 01,2020 at 13:30:44 UTC from IEEE Xplore. Restrictions apply.
where C(z) is given by Furthermore, using (15) and (16), the one-tap equalization
L gives
C(z) = hl z −l , (10) ŝ = D−1 ỹ = s + u, (20)
l=0
and P = K + L. where u = D−1 E η. Therefore, the equalized data symbols
are corrupted by the noise vector u whose statistics depend
IV. S YSTEM M ODEL AND P ROBLEM F ORMULATION on |ρk |, and the mean squared error (MSE) is given by
As depicted in Fig. 2, we propose a multicarrier transceiver 
where the transmitter (Tx) modulates the transmit symbol MSE = K −1 E Trace(uuH ) . (21)
vector using a Lagrange matrix (R), while the receiver (Rx) We aim to select the signature points {ρk }K−1 k=0 that
demodulates the received symbol vector using a Vandermonde minimize the MSE in (21) (or maximize (19)) while
matrix (E = VK×P ). Using (6), the modulated signal is satisfying (13). However, it is worth noting that the possible
padded with L zeros and transmitted over a frequency selective combinations of (κk ), k = 0, ..., K − 1 depend on {ρk }K−1
k=0 ,
channel. Similarly to (3), the FD received signal is given by while for a given set of {ρk }K−1
k=0 , we can find an infinity
ỹ(n) = E(n)H0 (n)R(n)s(n) + η(n). (11) of {κk }K−1
k=0 sets that satisfy (13). Our proposal is then
introduced in the following lemma.
It is worth mentioning that the matrices E(n), H0 (n) and
R(n) are assumed to be constant during the transmission block Lemma 1: If {ρk }K−1
k=0 are uniformly distributed on a circle
s(n), and could change from a block to another. Subsequently, of radius a > 0, such that ρk = a e2πk/K , the transmit filters
we will omit the index n for these matrices. Furthermore, using Fk , given by (5), can be rewritten as
Eq. (9), the following identity is true for all distinct nonzero
K−1 K−1
complex points {ρk }Kk=1 such that κk 2πkq κk
Fk (z) = aq ej K z −q = ρqk z −q , (22)
E H0 R = diag ([C(ρ0 ), · · · , C(ρK−1 )])  C. (12) K q=0
K q=0

Consequently, the PR is satisfied once the set {ρk }K−1


k=0 does
and will have same energy Ek = E0 given by
not contain the channel zeros. For this reason we refer to K−1
{ρk }K−1 1
k=0 as the signature roots of C(z) [2]. Moreover, in E0 = a2K . (23)
(5), κk should be chosen to satisfy the overall transmit power K2
k=0
constraint, given by
Lemma 2: The LVDM scheme depends on the parameter
Trace(RH R) = K. (13) a = |ρk | that should be optimized as
1
Let Ek be the energy of the transmit filter κk Fk , the aopt = arg min MSE (24)
a
constraint (13) can be rewritten as
K−1
V. P ERFORMANCE A NALYSIS
2
Ek |κk | = K, (14) In this section, we carry out the performance analysis of the
k=0 scheme under consideration. Following Lemma 1 by choosing
K−1 ρk = a e2πk/K , the Lagrange matrix in (6) reduces to
where Ek = i=0 |ri,k |2 /|κk |2 . Furthermore, the demodu-
lated signal is given by R = G Ω, (25)
ỹ(n) = D s + E η, (15) where G is a Vandermonde matrix given by
where ⎡ ⎤T
1 ρ0 · · · ρK−1
0
D = diag ([κ0 C(ρ0 ), · · · , κK C(ρK )]) . (16)
1 ⎢
⎢1 ρ1 · · · ρK−1
1


It follows that the noise variance and signal-to-noise ratio G= ⎢ .. .. .. ⎥ (26)
K ⎣. . . ⎦
(SNR) per subcarrier are given by
1 ρK−1 ··· ρK−1
K−1
K+L−1
σk2 = ση2 |ρk |−2i , (17) and the precoding matrix Ω (see Fig. 2) by (8)
i=0 Remark 1: It is worth noting that when a = 1, (23) gives
and 1 1 − a2K
|κk C(ρk )|2 σs2 E0 = , (27)
Λk = , (18) K 2 1 − a2
σk2 while a = 1 reduces the LVDM scheme to ZP-OFDM. In the
respectively, and we define the equivalent SNR as sequel, we will limit our analysis derivation to the cases when
K−1 −1 a = 1.
SNReq = K Λ−1 (19) Furthermore, we will perform (Step 1) the optimization
k
k=0
of the signature roots using (24) either by considering a

Authorized licensed use limited to: Carleton University. Downloaded on August 01,2020 at 13:30:44 UTC from IEEE Xplore. Restrictions apply.
Fig. 3: LVDM scheme with two consecutive steps: (Step 1) aopt optimization followed by (Step 2) signature roots refinement.

uniform power allocation (UPA) over subcarriers or an op- It turns out that (33) is a convex optimization problem and
timized power allocation (OPA), then (Step 2) proceed with the optimal solution is given by
the refinement of the resulting signature roots given by ρk =  
K−1
aopt ej2πk/K using the gradient descent algorithm (GDA). E0 i=0 |C(ρi )|−1
xk = , (34)
K|C(ρk )|
A. Uniform Power Allocation over Subcarriers
where, the optimal κk can be given by
Considering a UPA over subcarriers, we have κk = κ ∀k,  −1
K−1
where 2 −1
 |κk | = K E0 |C(ρk )| |C(ρi )| . (35)
1 1 − a2 i=0
κ= √ =K , (28)
E0 1 − a2K Plugging (35) into (32), and using the expression of E0 as
and the precoding matrix reduces to Ω = κIK . The MSE given by (27), the MSE is minimized and given by
expression in (21) for UPA case is given by  2
ση2 (1 − a−2P )(1 − a2K ) K−1 −1
M SEmin = 4 2 |C(ρk )| .
ση2 (1 − a2K )(1 − a−2P ) K−1 −2 K σs (1 − a−2 )(1 − a2 )
MSE = |C(ρk )| , (29) k=0
σs2 K 3 (1 − a2 )(1 − a−2 ) (36)
k=0
Finally, aopt is determined using (24), while Algorithm 1
and the optimized signature roots radius can be derived using summarizes the LVDM block-building using either UPA or
(24). OPA allocation.
Lemma 3: For UPA allocation over subcarriers, Λk in (18)
is given by Algorithm 1: LVDM block-building in Step 1
K 2 |C(ρk )|2 (2 − a−2 − a2 ) Inputs: K, L, Channel: hl , l = 0, ..., L.
Λk = σs2 2 , (30) Outputs: R Ω, and E
ση (1 − a−2(K+L) )(1 − a2K ) K−1
1: Use {ρk = a ej2πk/K }k=0 and {hl }L l=0 to compute (21)
where once plugged in (19), we get 2: Compute aopt such that (24)
K−1
1 3: (Rx) Use {ρk = aopt ej2πk/K }k=0 to compute E
SNReq = . (31) 4: Feedback aopt to Tx
MSE
K−1
B. Optimized Power Allocation over Subcarriers 5: (Tx) Use {ρk = aopt ej2πk/K }k=0 to compute R

As stated above, for a given set of {ρk }K−1 k=0 , we can find
an infinity of (κk ) sets that satisfy (13). Subsequently, one can It is worth mentioning that the explicit expressions given by
optimize {ρk }K−1
k=0 that minimizes a given cost function under (21) will be provided in Section V following the considered
the overall power constraint. We take the MSE in (21) as the case; UPA or OPA.
cost function, that can be rewritten as Remark 2: Likewise, the minimized MSE of the ZP-OFDM
K−1 scheme can be found by setting a = 1, and given by
ση2 (1 − a−2P )  −2   2
MSE = κk | |C(ρk )−2 . (32)
σs2 (1 − a−2 )K ση2 (K + L) K−1 j2πk
−1
k=0 M SEmin |a=1 = |C(e K )| . (37)
−1
K 3 σs2
k=0
We set xk = |κk |−1 |C(ρk )| , and we formulate the
optimization problem as follow C. Signature Roots Refinement
K−1 The first step is determined using the optimized radius aopt .
minimize x2k The second step uses the GDA algorithm, to update the values
x
k=0
of ρ−1
k as follow
(33)
K−1
E0 ∂M SE
subject to = K. ρ−1 −1
k ← ρk − α (38)
|C(ρk )| x2k
2 ∂ρ−1
k
k=0

Authorized licensed use limited to: Carleton University. Downloaded on August 01,2020 at 13:30:44 UTC from IEEE Xplore. Restrictions apply.
where α > 0 is the step factor of the algorithm and the -2 3 0.18

MSE expression in (21) can be rewritten after some algebraic -1.5


2.5
0.16

0.14

manipulations as -1

k
0.12

Imaginary part of ρ
2

SNReq = 1/MSE
-0.5

ση2
0.1
0 1.5

−2 H
MSE = Trace(D E E ),
0.08
(39) 0.5

K 1 0.06

  1
0.04

where D−2 = diag κ−2 0 |C(ρ0 )|


−2
, · · · , κ−2
K |C(ρK )]|
−2 0.5

. 1.5
0.02

2 0
Due to space limitation, we omit the derivation steps of the -2 -1.5 -1 -0.5 0
Real part of ρ
0.5
k
1 1.5 2 0.5 0.6 0.7 0.8 0.9 1
Value of a
1.1 1.2 1.3 1.4 1.5

GDA which is summarized in Algorithm 2.


Fig. 4: C(z) = 1 + z −4 , K = 16, and L = 4. (left) Λk as
function of ρk and (right) SNReq as a function of a.
Algorithm 2: GDA refinement using UPA or OPA
Inputs: K, L, Channel: hl , l = 0, ..., L, aopt
Outputs: ρk , k = 0, ..., K − 1.
-2 1

0.9
-1.5 6

0.8
-1
5
0.7

M SEmin ← M SE(aopt ) in (29) or (36)

k
Imaginary part of ρ
1:

SNReq =1/MSE
-0.5
4 0.6

j2πk
ρk ← aopt e K .
0 0.5
2: 0.5
3
0.4

3: M SE1 ← M SEmin . 1
2 0.3

0.2

4: Repeat 1.5
1
0.1

5: Compute ∂M SE/∂ρ−1k .
2
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
0
0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5

βk−1 ← ρ−1
Real part of ρ Values of a
k

6: k .
7: ρ−1 −1
k ← ρk − α∂M SE/∂ρk .
−1 (a) (b)
8: Compute M SE2 using (29) or (36) . Fig. 5: C(z) = 1 − z + z , K = 16, and L = 4. (left) Λk −1 −4
9: while M SE2 > M SE1 do as function of ρk and (right) SNReq as a function of a.
10: α ← α/10.
11: ρ−1 −1
k ← βk − α∂M SE/∂ρk .
−1

12: Compute M SE2 using (29) or (36) . and we compute the corresponding MSE given by (36) and
13: end while then derive aopt as depicted in the figure 6 (see Algorithm 1).
14: M SE1 ← M SE2 .
15: Until |M SE1 − M SE2 | <
300
1

0.8
250
0.6

VI. S IMULATION R ESULTS


Normalized MSE (dB)

0.4 200
Imaginary part

0.2
Optimized radius of ρ k

In this section, we validate the LVDM performance through 0

-0.2
Unity circle
Channel zeros
150
a opt = 1.058
MSE min = 0.3374

comparison to the ZP-OFDM with simulation results. We con- -0.4


100

sider single antenna transceivers (see Fig. 3) communicating


-0.6
50
-0.8

through frequency selective channels. -1


-1 -0.5 0 0.5 1
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Real part Radius of ρk (a=|ρk |)

A. Impact of the choice of a


Fig. 6: (left) Λk as function of ρk and (right) SNReq as a
We start by providing the significance of the optimization function of a, where K = 16, and L = 4.
block carried out in Step 1 and provide illustrative examples
in Figs. (4) and (5) where we present the SNR per subcarrier By applying the GDA (Algorithm 2) to {ρk =
Λk as function of ρk (left) and the equivalent SNR, SNReq , as aopt ej2πk/K }K−1
k=0 , the MSE drastically decreases while the
function of a (right). In Fig. (4), for a channel C(z) = 1+z −4 signature roots converge to new locations as shown by 7.
where the channel zeros are placed over the unit circle (a = 1),
the corresponding SNReq when a = 1 is at its minimum while 5

it achieves it maximum at a = 1.1. Therefore, for this chan-


1

0.8 4.5

nel realization, the LVDM will achieve better instantaneous 0.6


4
Normalized MSE (dB)

0.4

performance than ZP-OFDM (a = 1).


Imaginary part

3.5
0.2 Optimized radius of ρ
k
Unit circle

However, for another channel realization, where C(z) =


0 Channel zeros 3
Refined values of rho k
-0.2
2.5

1 − z −1 + z −4 , Fig. 5 shows that SNReq is maximized at -0.4

-0.6 2

a = 1, thus, LVDM reduces to ZP-OFDM. -0.8

-1
1.5

1
-1 -0.5 0 0.5 1 0 10 20 30 40 50 60 70 80

B. Impact of the Signature Roots Refinement Real part Iterations

Here, we will provide examples where Step 1 uses the OPA Fig. 7: (left) Λk as function of ρk and (right) SNReq as a
allocation over subcarrier followed by the GDA in Step 2. We function of a, where K = 16, and L = 4.
take a snapshot of a selective frequency channel with L = 4,

Authorized licensed use limited to: Carleton University. Downloaded on August 01,2020 at 13:30:44 UTC from IEEE Xplore. Restrictions apply.
In Fig. 9, we carry out simulations in 3GPP EVB chan-
100 nels [8] where the carrier frequency is set to fc = 700 MHz
and the velocity v = 60 Kmph and L = 32. Here, it is
LVDM (OPA)
LVDM (UPA)

10-1
ZP-OFDM (OPA) worth mentioning that EVB channels exhibit a maximum delay
ZP-OFDM (UPA)
spread of 5 microseconds. Furthermore, both LVDM and ZP-
OFDM use UPA and OPA allocation over subcarriers where
10-2 K = 128 with a subcarrier spacing Δf = 15 KHz that leads
BER

to a Doppler spread of 39 Hz. It is worth noting that LVDM is


more robust to the channel time selectivity even though a more
10-3
advanced receiver scheme should be adopted to overcome the
performance saturation.
10-4
VII. C ONCLUSIONS AND F UTURE W ORKS
In this work, we propose a new waveform, referred to as
10-5 LVDM, that generalizes yet outperforms the ZP-OFDM and
0 5 10 15 20 25 30 35 40
SNR (dB) guarantees the perfect recovery of the transmitted signals.
Relying on the signature roots optimization, we propose a
Fig. 9: LVDM and ZP-OFDM performance results comparison
complete framework on the design of the LVDM while keeping
in 3GPP EVB channels, where K = 128 and L = 32.
a low complexity transceiver implementation for frequency
selective channels. Furthermore, the numerical results revealed
that LVDM keeps outperforming the ZP-OFDM even within
C. BER Performance
doubly selective 3GPP EVB channels. However, more ad-
Here, we consider LVDM and ZP-OFDM systems trans- vanced receiver should be designed to dead with doubly
mitting QPSK symbols using K = 32 subcarriers through a selective channels in high mobility regime.
frequency selective channel with L = 4. The channel taps As a future research direction, we will propose advanced
follow a power delay profile (PDP) of factor ξ = 0.2. Both receivers that exploit the LVDM design flexibility to overcome
schemes have been carried out using UPA and OPA allocation the inter-carrier interference induced by the time selectivity in
over subcarriers. Fig. 8 shows that LVDM achieves an SNR high mobility scenario.
gain of 5 dB for BER at 10−5 compared to ZP-OFDM when
applying OPA. Furthermore, an additional SNR gain of 2 dB R EFERENCES
is achieved when LVDM applies the GDA algorithm. [1] G. B. Giannakis, Z. Wang, A. Scaglione, and S. Barbarossa, “Amour-
generalized multicarrier transceivers for blind CDMA regardless of mul-
tipath,” IEEE Transactions on Communications, vol. 48, no. 12, pp. 2064–
2076, Dec 2000.
100
[2] A. Scaglione, G. B. Giannakis, and S. Barbarossa, “La-
grange/Vandermonde MUI eliminating user codes for quasi-synchronous
CDMA in unknown multipath,” IEEE Transactions on Signal Processing,
vol. 48, no. 7, pp. 2057–2073, July 2000.
10-2 [3] G. B. Giannakis, Zhengdao Wang, A. Scaglione, and S. Barbarossa,
“Mutually orthogonal transceivers for blind uplink cdma irrespective of
multipath channel nulls,” in 1999 IEEE International Conference on
Acoustics, Speech, and Signal Processing. Proceedings. ICASSP99 (Cat.
BER

10-4 No.99CH36258), vol. 5, March 1999, pp. 2741–2744 vol.5.


[4] A. Scaglione and G. B. Giannakis, “Design of user codes in QS-
CDMA systems for MUI elimination in unknown multipath,” IEEE
LVDP (OPA)
Communications Letters, vol. 3, no. 2, pp. 25–27, February 1999.
LVDM (UPA) [5] D. Calvetti and L. Reichel, “Fast inversion of Vandermonde-like matrices
10-6
ZP-OFDM (OPA) involving orthogonal polynomials,” BIT Numerical Mathematics, vol. 33,
ZP-OFDM (UPA) no. 3, pp. 473–484, September 1993.
LVDM (GDA + OPA)
[6] M. R. Skrzipek, “Inversion of Vandermonde-like matrices,” BIT Numer-
ical Mathematics, vol. 44, no. 2, pp. 291–306, May 2004.
10-8 [7] ETSI TS 138 211, “5G ; NR; Physical channels and modulation,” 3GPP
0 5 10 15 20 25 30 35 40
SNR (dB)
TS 38.211 version 15.2.0 Release 15, July 2018.
[8] “Guidelines for evaluations of radio transmission technologies for IMT-
Fig. 8: LVDM and ZP-OFDM performance results comparison 2000,” ITU-R M.1225, 1997.
in frequency selective channel, where K = 32 and L = 4.

Authorized licensed use limited to: Carleton University. Downloaded on August 01,2020 at 13:30:44 UTC from IEEE Xplore. Restrictions apply.

You might also like