HARQ Performance Over Massive MIMO Channels Using Chase Combining and CZA Pilot Decontamination
HARQ Performance Over Massive MIMO Channels Using Chase Combining and CZA Pilot Decontamination
HARQ Performance Over Massive MIMO Channels Using Chase Combining and CZA Pilot Decontamination
1
2
3
HARQ performance over massive MIMO channels
4
5
6
using chase combining and CZA pilot
7
8
decontamination
9
10
Maroua Boudaya Ines Kammoun Mohamed Siala
11
LETI laboratory LETI laboratory MEDIATRON laboratory
12
ENIS, University of Sfax ENIS, University of Sfax SUP’COM, University of Carthage
13
Sfax, Tunisia Sfax, Tunisia Tunisia, Tunisia
14
[email protected] [email protected] [email protected]
15
16
17
18
19
Abstract—Pilot contamination is known to seriously limit the proved in [3] that CZA reaches the single-user performance
20
performance of massive MIMO systems due to degradation when the numbers of symbols per transmitted frame and of
of channel estimation. Channel Zooming Algorithm (CZA) has
21
receive antennas are large and that CZA performance degrades
been proposed as an efficient iterative algorithm to ensure the
22
mitigation of the pilot contamination effect. This algorithm is when the transmitted packets are of lengths [4].
23
able to zoom through iterations on the signal of the strongest Nevertheless, intelligent terminals such as smart phones and
24
user by taking advantage of the difference in space signatures tablets produce an increasing number of data packets of
25
and transmitted data sequences. CZA has been proved to be lengths. A solution to enhance transmission reliability in
26
efficient for pilot contamination in the massive MIMO systems this case is the use of Automatic Repeat request (ARQ)
in the case of large number of antennas and long data packets.
27
In this paper, we propose a novel version of CZA that utilizes and hybrid ARQ (HARQ) retransmission protocols. The two
28
the benefits of retransmitted packets combining. We consider HARQ combining schemes are incremental redundancy (IR),
29
a truncated Hybrid Automatic Repeat reQuest (HARQ), where where the retransmissions consist of new parity bits from the
30
the number of retransmissions is limited using Chase combining channel encoder, and Chase combining, where the retransmis-
31
(CC). We evaluate the Packet Error Rate (PER) performance of sions are identical copies of the original transmission. These
32
the proposed approach for different order modulations as BPSK,
QPSK, 16-QAM and 64-QAM. Simulation results show that very retransmission schemes have been considered for MIMO and
33
good performances can be achieved after very few transmission massive MIMO systems in some research works [7-10].
34
attempts with our HARQ scheme using CC and CZA, despite In [7], a linear multi-user detection algorithm with HARQ with
35
possible different contamination during each retransmission. incremental redundacy (HARQ-IR) is proposed for MIMO-
36
Index Terms—Massive MIMO, pilot contamination, Channel systems. Its aim is to improve the throughput and the bit error
37
Zooming Algorithm, pilot decontamination, automatic repeat
rate (BER) by considering MIMO with a simple retransmission
38
request, Chase combining
method. In [8], Hybrid schemes with CC and IR have been
39
considered in the uplink of a multicell multiuser SIMO system
40
I. I NTRODUCTION
with large-scale antenna arrays at the base station (BS) for
41
Massive multiple-input multiple-output (MIMO) systems improving the spectral efficiency. In [8], the author showed
42
are equipped with a large number of antennas at the base that the throughput and the outage performance could be
43
station (BS), which can significantly improve the spectrum impressively improved when the number of antennas increases
44
efficiency, system capacity and coverage of the system [1-2]. in massive MIMO. In [9], authors reveal that HARQ with
45
It is regarded as the most promising technology in the fifth- the adaptation of codeword length or transmit power decrease
46
generation (5G) wireless communication system. However, the power consumption. In [10], the authors consider the
47
the pilot contamination caused by sharing the pilot sequences three incremental schemes ARQ, HARQ-CC and HARQ-IR
48
among users is considered to be a bottleneck of these systems for MIMO systems. They propose energy-efficient adaptive
49
[3-4]. To address this problem, several approaches have been power allocation to minimize the outage probability under a
50
proposed to mitigate the effect of pilot contamination in constraint on average energy consumption per data packet.
51
massive MIMO systems [3-6]. Their study has been evaluated considering small numbers of
52
We are interested in this paper on the semi-blind algorithm, transmit and receive antennas.
53
proposed in [3], which is called ”Channel Zooming Algo- In this paper, we consider the uplink of massive MIMO
54
55
rithm”, that ensures the mitigation of pilot contamination. The systems and propose a novel HARQ scheme using Chase
56
idea of this algorithm is to make a zoom, through iterations, on combining and CZA pilot decontamination. Our idea is that
57
the signal of the strongest user by taking advantage of differ- after a first transmission, the contaminated packet transmitted
60
ence in received powers from contaminated users. It has been by the strongest user is recovered with CZA. If this fails,
61
62
63
64
65
1 on June 19,2021 at 08:16:27 UTC from IEEE Xplore. Restrictions apply.
Authorized licensed use limited to: Univ of Calif Santa Barbara. Downloaded
the receiver combines the retransmitted packets using Chase distributed according to CN√(0, N0 ), where N0 is the channel
combining CC. Since contamination can occur during each noise variance. Let xk = E k (xk0 , xk1 , ..., xk,N −1 ) be the
retransmission, we propose a modified version of CZA that transmitted sequence from user k, where its components xkn
uses the benefits of retransmissions. We evaluate the perfor- are normalized symbols.
mance of the proposed scheme in term of packet error rate for
different symbol modulations, namely BPSk, QPSK, 16-QAM
and 64-QAM. III. P ROPOSED HARQ USING CC AND CZA
The rest of the paper is organized as follows. In section II, In this section, we introduce our approach for retransmission
we present the system model. In section III, we introduce the and combining of unsuccessfully detected packets and pilot
novel channel zooming algorithm that utilizes the benefits of decontamination with CZA.
retransmission with HARQ scheme with CC. In section IV,
we provide some simulation results. Finally, in section V, we A. Description of the mechanism for retransmission
summarize the main results of this paper and propose some
We assume that during the first transmission of a given
perspectives.
packet by a user of interest, indexed as user 1, a pilot
Notations contamination occurs between user 1 and U users located
The superscripts (.)H , (.)∗ and (.)T denote the Hermitian in neighboring cells, indexed as users k, k = 1, . . . , U . The
transpose operator, the conjugate operator and the transpose received signal at the base station from these users, is given
operator, respectively. Moreover, ha, bi denotes the scalar by
2
product of two vectors a and b and k.k denotes the Euclidean U p
square norm of a vector. Each matrix A (respectively vector
p X
y1n = E1 h11 x1n + E2 h1k xkn + b1n , (2)
a) in the system model will be split into Ad (resp. ad ) for data k=2
part and Ap (resp. ap ) for the part emanating from pilots.
for n = 1, . . . , N − 1.
II. S YSTEM MODEL The CZA is applied to recover the packet of user 1. We recall
We consider a multi-cell multi-user massive MIMO system that the CZA idea is to zoom, through the iterations, on the
in uplink transmission. In each cell, we assume that the BS is signal of the strongest user of interest [3]. We notice that the
equipped with M antennas and serves K single antenna users. contaminating users, which are located in neighboring cells to
The transmitted signal from each user is arranged into bursts the cell of interest, have necessarily less received powers at
of N modulated symbols each, comprising Np pilot symbols the base station of the cell of interest than the user of interest
and Nd data symbols, with common transmitted energy EkT x which is directly connected to the base station of interest and
and positions pn = nTs , n = 0, 1, . . . , N − 1, where Ts is located in the cell of interest.
is the symbol period. Without loss of generality, we assume When the CZA algorithm fails in pilot decontamination, the
for all users that the beginning of each burst is dedicated to BS requests a retransmission of the same replica at the next
pilot transmission. Each link between a user and the BS is slot and when the packet of user 1 is retransmitted, it can be
assumed to borrow a Rayleigh flat fading channel. Thus, its contaminated or not. The worst case is that when it will be
coefficients are constant during the transmission of at least contaminated during its transmission. Then we will propose a
one burst. Let hkj be the propagation channel attenuation solution for this problem.
0
between user k and receive antenna j, j = 0, 1, . . . , M −1. The Let U be the set containing the indices of contaminating users
channel coefficients hkj are statistically Independent, Identi- transmitting during the next slot.
cally Distributed (i.i.d.) complex Gaussian random variables The received signal at the base station at this slot, is given by
distributed according to CN (0, 1). We assume that at each U p 0
Nd −1 ∗ ∗ (20)
1 X E1 hz1 , h21 i h11 + β2 E1 hh23 , h21 i h11
x∗ bmn ≈ 0, (14) + 2
Nd n=0 kn (E1 kh11 k + N0 )
∗
E2 hh23 , h21 i h12
+ 2 .
for k = 1, 2 and 3 and m = 1, 2. (E1 kh11 k + N0 )
Yet again, the next four terms in equation (13) can be
neglected. Moreover, since the vectors h11 , h12 , h21 , h23 By developing (11) for m = 2 to estimate the channel h21
and zm are independent of bmn and 2Nd is large, we can seen by user 1 during the retransmission, we get
neglect the next four terms in v. The last term in v can be (1) (1)
also neglected since b1 and b2 are independant. Therefore, ĥ2 = h21 + ρ2 h23 + w2 , (21)
all the terms in the equation (13) could be neglected, a claim
where
that will also be confirmed through simulation results. Since
2
the number of data symbols Nd is large and according to the β2 E3 kh23 k + N0
central limit theorem, we have the following approximation ρ2 = . (22)
2 2
E1 kh21 k + kh11 k + N0
for the fourth term in (12)
Nd −1
At the l-th iteration of the CZA, the channel estimate is
1 X (0)H (0) obtained as follows
b1n (ĥ1 b1n )∗ ≈ N0 ĥ1 . (15)
Nd n=0 Nd −1
1 X
λ(l) (l)
m ĥm = ymn rn(l−1) ∗ (23)
2 Nd n=0
Moreover, we can use the approximations
xdk
≈ Ek Nd ,
for k = 1, . . . , 3, since Nd is large. We note that, for phase (l)
modulations, these approximations become equalities. and the decision variable rn on the n-th symbol of the
Then, we get strongest user is improved as follows
P2 (l)
1 m=1 rmn
rn(l)
(1) (1) 2 ∗
λ1 ĥ1 = E1 kh11 k + hz1 , h11 i + β1 hh11 , h12 i h11
∗ = , (24)
2 2
1 2 ∗ ∗
where
+ E2 β1 kh12 k + hz1 , h12 i + hh11 , h12 i h12
2 (l)
1 rmn = ĥ(l) H
m ymn . (25)
2 ∗ ∗
+ E1 kh21 k + hz2 , h21 i + β2 hh21 , h23 i h11
2 (l−1)
N0 After that, by replacing yn and rn by their expressions in
+ (h11 + β1 h12 + z1 ) (23), we show that
2
(16)
By choosing the multiplicative factor λ(1) as (l)
ĥ1 = h11 + ρl1 h12 + w1
(l)
(26)
2 2
E1 kh11 k + E1 kh21 k + N0 and
λ(1) = , (17)
2 (l)
ĥ2 = h21 + ρl2 h23 + w2 ,
(l)
(27)
C. Decision metric
For uncoded symbols, the decision on signal x11 is given
simply by
!
1 ĥH 11 d 1 ĥH 21 d
x̂11 = Γ Y + Y , (36)
2 ||ĥ11 ||2 1 2 ||ĥ21 ||2 2
where Γ (.) is the decision function of a threshold de-
d
tector depending on the used modulation and Ym =
(y0 , y1 , . . . , yN −1 ) is the M ∗ N received matrix.
For the coded system, the soft decisions ĥH d
m1 Ym , where
m = 1, 2 can be used by an error-correction Viterbi decoding
10-2 10-1
PER
PER
10-3
10-4 10-2
Fig. 1. PER as a function of SNR for BPSK modulation, θ = 1/2 and Fig. 3. PER as a function of SNR for 16-QAM modulation, θ = 1/3 and
M = 500. M = 500.
100 100
10-1 10-1
PER
PER
10-2 10-2
l=1, CZA-HARQ-CC
10-3 10-3
l=2,CZA-HARQ-CC
l=5, CZA-HARQ-CC M=500, HARQ-CZA
l=1, CZA M=500, CZA
l=2,CZA M=300, HARQ-CZA
l=5, CZA M=300, CZA
10-4 10-4
-19 -18 -17 -16 -15 -14 -13 -12 -11 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2
SNR [dB] SNR [dB]
Fig. 2. PER as a function of SNR for QPSK modulation, θ = 1/3 N = 1024 Fig. 4. PER as a function of SNR for 64-QAM modulation, θ = 1/6 and
and M = 500. N = 1024.