A Low Complexity Channel Estimation Algorithm For Massive MIMO System
A Low Complexity Channel Estimation Algorithm For Massive MIMO System
A Low Complexity Channel Estimation Algorithm For Massive MIMO System
1. Introduction
In the foreseeable future, the requirements of high rate wireless applications and the high
density of wireless devices are expected to lead to enormous new challenges in terms of the
efficient exploitation of the achievable spectral resources. Among the prospective airinterface techniques, Massive MIMO (known as Large-scale MIMO) in the base station can
boost the cell throughput, improve the link quality and decrease the transmission power [1, 2].
Massive MIMO is characterized that Base station can distinguish the objective user not only
the position in X/Y planar but also the height in Z axis. To achieve the alignment of three
dimensions, it will configure two-dimension rectangle array which has more antenna than the
traditional one-dimension array [3]. Massive MIMO can precisely concentrate the
transmitting power on the height of the target user so as to improve the link quality and
reduce the interference to other users. On the hand Massive MIMO can support spatial
multiplexing of multiple users with different heights, so can boost the system capacity.
The most common method to acquire MIMO channel knowledge in the receiver is channel
estimation aided by pilots or preambles [4, 5]. The transmitter delivers pilots from every
antenna or different beams, and the receiver estimates channel information of the
corresponding antenna or beam using LS (Least Squares), LMMSE (Linear Minimum Mean
Square Error), 2D-MMSE (Two-dimension Minimum Mean Square Error) or 2D (twodimension) Wiener filtering algorithms. With antenna and distinguishable beam increasing,
the overhead of pilots or preambles will linearly increase and greatly decrease the system
throughput. For Massive MIMO with more antenna and more precise beams, new channel
estimation algorithms should be studied on how to reduce pilot overhead and improve the
quality of channel estimation.
The following literatures have studied on how to utilize the channel correlation to reduce
pilot overhead and improve the quality of channel estimation. Literature [6] proposed 2DWiener channel estimation algorithm which achieved channel estimation interpolation using
time-frequency channel correlation. It can significantly reduce pilot overhead and improve the
channel estimation quality comparing LS, LMMSE algorithms. The three dimensional Wiener
filter (referred to as 3D-Wiener) channel estimation algorithm was considered in the literature
[7]. It achieved channel estimation interpolation using the space-time-frequency correlation of
MIMO channel and proved that 3D-Wiener is superior to 2D-Wiener of approximately 5dB
with same pilot overhead. Literature [8] proposed a three-dimensional MMSE channel
estimation algorithm using the channel space-time-frequency correlation. But those literatures
were for the conventional one-dimensional antenna array and havent defined the space
correlation of arbitrary two elements for the two-dimension antenna array. As 2D-Wiener
dont use spatial correlation of MIMO channel, pilot will deliver from every antenna to
estimate channel information and its pilot overhead is linearly increased with the number of
antenna. 3D-Wiener and 3D-MMSE dont use two-dimension spatial correlation of MIMO
channel, pilot will deliver from every raw of the antenna array to estimate channel
information and its pilot overhead is increased with the number of rows. For Massive MIMO,
the space-time-frequency correlation of the two-dimension antenna array and new channel
estimation algorithm still need more studied.
On the study of the space-time-frequency correlation with the two-dimensional uniform
rectangular array, literature [9] presented and simulated the solving of the spatial correlation
of uniform rectangular array, but hasnt considered the integrated solving scheme of spacetime-frequency correlation function. Literature [10] proposed the space-time-frequency
correlation solution with arbitrary antenna arrays, but its solution is very complicate and hard
to be applied in the actual channel estimation algorithm. Literature [11] proposed a Massive
MIMO channel model for IST-Wiener channel simulation but didnt analyze the channel
correlation.
In this paper, we extend one-dimension space filtering of original 3D-Wiener algorithms to
two-dimension space filtering and propose a multi-dimension Wiener filter channel
estimation algorithm. We achieves the channel estimation interpolation with multi-dimension
filtering and gets the optimal Wiener filtering weights using the channel correlation functions
of the time, frequency, raw and column of the antenna array. The proposed algorithm doesnt
need to transmit the pilot from every antenna element contrast to 2D-Wiener or even every
raw contrast to 3D-Wiener and can be significantly reduced the pilot or preamble overhead.
At last based on theoretical analysis and simulation results, its shown that the proposed
algorithm has the better quality of channel estimation than 2D-Wiener and 3D-Wiener
algorithm with the same pilot overhead. What is more, the proposed algorithm acquires the
optimal throughput with considering both the channel estimation quality and pilot overhead.
82
axis and d y to y-axis. The receiver configures uniform linear array (ULA), as shown in Figure
2, with M antenna elements, which the spacing of every two adjacent units is d x to x-axis.
Assume that time-frequency resources of MIMO+ OFDM system are divided into a
plurality of same resource blocks and every resource block includes K subcarriers, L OFDM
symbols and applies the same pilot transmission scheme. In this paper a resource block will
be the object to describe the multi-dimension Wiener filter channel estimation algorithms. For
th
th
the t OFDM symbol and the subcarrier, the received signal of a physical resource block
is can be expressed as:
Y t , H t , X t , nt,
(1)
Es
1 and the
dy
1,11,21,3
2,1
3,1
1 N col
dx
nrowncol
NrowNcol
X
dx
2
m
M
X
83
In this paper we focus on the estimation of the channel impulse response (CIR) without
considering the impact of large-scale fading. It is commonly assumed that the distance from
the antenna array to scatters a much larger than the antenna spacing. Therefore, propagation
waveforms in the scattering environment are plane waves. There are I multi-paths from the
transmitter to the receiver which every path is distributed to Rayleigh distribution. For the
t th OFDM symbol and the th subcarrier, the channel matrix H t, can be expressed as:
I
H t, a R i ,i i t , aT i ,i
i 1
(2)
Where denotes the transpose of the matrix or vector. I is the total number of multiT
path, and i t , is CIR of the i th path. aT i , i is the steering vector of URA in the
transmitter, i is the angel between the x-axis and the direction-of departure (DOD) of the
i th path from the transmitter and commonly known as the azimuth angle i is the angel
between the z-axis and the DOD of the i th path and commonly known as the elevation angle.
a R i , i is the steering vector of the ULA in the receiver. aT i , i is defined as the
following formula (3):
row
,..., e j N row 1 N col 1
e
the
column
steering
(3)
vector
and
a Ncol 1, e j , , e j ( Ncol 1)
a R i , i is the steering vector of the receiving antenna array and the M 1 dimension
a R i , i 1, e j , , e j ( M 1)
(4)
84
Null Signal
Pilots
Antenna n0+ds
Space
Dt
Time
Antenna n0
Frequency
Df
0th sub-carrier and the n0th transmitting antennas. CIR can be estimated by LS algorithm to
estimate from the receiving signal in this time-frequency resource and is expressed as:
Y t0 , 0 , m0 , n0
H t0 , 0 , m0 , n0
X t0 , 0 , m0 , n0
(5)
The channel information of other antenna and other time-frequency resources are obtained
by the interpolation using Wiener filtering. Wiener interpolation filter achieve the minimum
H t , , m, n w opt H t 0 , 0 , m, n0
1
Rdp R pp
H t 0 , 0 , m, n0
(6)
The channel correlation matrix Rdp and R pp are the four-dimensional function, so the
proposed algorithm called multi-dimension Wiener filter channel estimation algorithm. R pp is
the LK M M auto-correlation matrix of the estimated CIR at the pilot position. For the
t 0th OFDM symbols, the 0th sub-carrier, the m0th receiving antenna and the n0th transmitting
antenna, the autocorrelation matrix R pp is calculated as follows:
Rpp t0 , 0 , m, n0 E H t1 , 1 , m, n1 H t0 , 0 , m, n0
(7)
Wherein denotes conjugate. Rdp is the cross correlation matrix between output values
of CIR in data position and the estimated CIR in pilot positions and is calculated as follows:
Rdp t , , m, n E H t , , m, nH * t0 , 0 , m, n0
(8)
85
system bandwidth. Where t is denoted the time correlation function between the
channel of symbol t and symbol t t ; is denoted the frequency correlation
function between the channel of sub-carrier and sub-carrier
.Therefore, t
and are irrelevant with the spatial correlation function. We utilize formula (5) to
estimate CIR of every receiving antenna, so it is only necessary to consider the spatial
correlation of the transmitter. Through the above analysis, formula (8) is the independent
product of spatial correlation function, the time correlation function and the frequency
correlation function [6-8]:
Rdp RT n1, n2 t
Where
(9)
th
th
between the n transmit antennas and the n1 transmit antennas.
(10)
Where f max is the maximum Doppler shift, Tsym is OFDM symbol length, and J 0
represents the zero order Bessel function. The frequency correlation function and is calculated
as follows [5, 6]:
L
Pi e j 2
i 1
(11)
th
n1row
row
th
1th
and ncol
column of the transmitting antenna array, the spatial correlation function of the n
transmitting antenna and the n1th transmitting antenna can be calculated according to the
following formula:
RT n1 , n2
an
row , ncol
, an
1
1
row , ncol
0 0
0 -
0 -
, p , dd
0 0
0 -
86
0 -
j y sin sin
e jx cos sin dd
(12)
Where p , is the joint probability density function of the azimuth angle and the
elevation angle. Assume that the azimuth angle is independent distribution with the elevation
angle, so p , p p . is the angel spread of the azimuth angle and 0 is the
average azimuth angle, is the angel spread of the elevation angle and 0 is the average
elevation angle. If the azimuth angle
4 0 - 0 -
0 0 cos sin sin
dd
RT n1 , n2
(13)
J 2 n (2 z sin )d
J n2 z
(14)
(15)
J 02 ( )
2
2
2
J n sin 0 - sin 0 -
2
n 1 2n
J 2 1 cos 0 - - cos0
n 2
n 1 2n
2
(16)
We calculate the cross correlation matrix Rdp using formula (10), (11), (16), and estimate
channel information of every transmitting antenna in the resource block based on formula (6).
87
Values
Parameters
Values
2.4GHz
Bandwidth
10MHz
Micro
Sampling
65.1 ns
Urban
duration
Transmitter: 4 4 URA,
Receiver: 4 ULA. Antenna spacing: / 2
IST-Wiener Model, B1 NLOS (16 paths)
Scenario
Antenna configurations
Channel Model
Physical Resource Blocks
e2
K L N M
1
e2 t , , m, n
MNLK 1 t 1 n 1 m 1
t , , m, n Rdp t , , m, n
E H t , , m, n w opt
2
t , , m, n wopt t , , m, n
Rdp
w opt
R pp t , , m, n w opt
(17)
According the solution of formula (6), the latter two can be canceled each other out. Then
formula (16) can be simplified as the following formula (18):
e2
t , , m, nRdp t , , m, n
E H t , , m, n w opt
2
(18)
-1
10
MSE
10
-3
10
-4
10
10
15
SNR/dB
20
25
88
120
100
Throughput/Mbps
80
60
40
2D-Winner
3D-Winner
Multi-dimension Winner with pilot transmission scheme I
Multi-dimension Winner with pilot transmission scheme II
20
10
11
12
SNR/dB
13
14
15
16
89
5. Conclusions
In this paper we analyze Massive MIMO channel model and the space-time-frequency
correlation of URA and put forward a simple and closed solution of the space-time-frequency
correlation function between arbitrary two pairs of transmitting antenna and receiving
antenna. Further the multi-dimension Wiener filtering algorithm obtains the optimum filter
weights using the above-mentioned correlation function and achieves the channel estimation
interpolation with multi-dimensional filtering in the time, frequency, rows and columns
domain. This algorithm can reduce by half or three-quarters of pilots overhead while
obtaining a good channel estimation performance. Based on simulation results, it is proved
that the proposed algorithm can acquire higher system throughput comparing with 3D-Wiener
and 2D-Wiener channel estimation algorithm for Massive MIMO system with a uniform
rectangular array. This algorithm can be applied in Massive MIMO system with good
research and practical value.
Acknowledgements
This paper is sponsored by National 863 Program (2014AA01A705), National Natural
Science Foundation of China (61102047, 61201194) and the science research project of
Shaanxi Provincial Department of Education (11JK1016).
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
90
[17] X. Gao, F. Tufvesson, O. Edfors and F. Rusek, Measured propagation characteristics for very-large MIMO
at 2.6 GHz, European Conference on Antennas and Propagation (EUCAP), (2013), pp. 433-437.
[18] H. Q. Ngo, E. G. Larsson and T. L. Marzetta, Energy and spectral efficiency of very large multiuser MIMO
Authors
Jiang Jing received M. Sc. from the Xi Dian University in 2005 and a
Ph.D. in Information and Communication Engineering from North
Western Polytechnic University, China in 2009. She had been a
researcher and a project manager from 2006 to 2012 at ZTE Corporation
in China, and currently as an associate professor of School of
Communication and Information Engineering in Xi'an University of
Posts & Telecommunications, China. Her research interests include
wireless communication theory and MIMO system design. She has been
a Member of the IEEE, 3GPP.
91
92