Académique Documents
Professionnel Documents
Culture Documents
. Introduction
The massive multi-input multi-output (M-MIMO)
technique deploys hundreds of transmit antennas in base
station to accommodate a large number of users over
spatial domain. In frequency division duplexed (FDD)
M-MIMO systems, channel reciprocity cannot be
exploited and hence the downlink pilots should be
transmitted, thereby leading to substantial pilot overhead
in downlink.
In this paper, we introduce a new downlink pilot
reduction method that exploits the common sparsity
structure of the channel impulse response (CIR) over a
local period of time. Our simulation results show that
we achieve significant reduction in pilot overhead while
maintaining good channel estimation performance.
,1 , , , , = [1 , , ] and
is the k th coordinate vector. Note that
represents the system matrix connecting the time
domain CIR vector and the received pilot vector .
Therefore, the channel estimation can be considered as
the recovery of the CIR vector, , from the linear
noisy observation model in (3). Note that due to the
randomly located pilots, the matrix in (3)
corresponds to the random sensing matrix constructed
by taking randomly selected rows of the DFT matrix
via .
. Main subject
1. LOW DENSITY PILOT ALLOCATION
The proposed pilot allocation is depicted by Fig. 1,
where a small number of pilots are randomly located in
frequency domain. As compared to the conventional
pilot allocation where pilots are uniformly allocated, the
proposed strategy avoids uniform allocation to ensure
good performance in channel estimation.
A vector of the frequency-domain received pilot
symbols, 1 for the nth pilot-containing OFDM
symbol can be expressed as
(1)
= ( ) +
(2)
= ( )
+
= (
) +
where is
the size
DFT
(3)
matrix, =
486
(4)
= + , ,
+
2016
We rearrange the channel vectors + , , such
that the ith entry for each of + , , form a
new vector , . We also rearrange the columns of the
(+ , , ) accordingly.
sensing
matrix
Denoting the sensing matrix corresponding to , as
, , (4) can be rewritten by
+
1 ,1
(5)
= ,1 , +
,
where a deterministic binary vector = [1 , , ] is
defined as
1
(6)
=
0
1 ||,
(1) 2
||2
(7)
487