Channel Estimation Of Sparse Multipath Based On
Compressed Sensing Using Golay Sequences
Zhiqiang Yao
The College of Information
Engineering
Xiangtan University
Xiangtan, China
Email: yaozhiqiang@xtu.edu.cn
Guanglong Li
The College of Information
Engineering
Xiangtan University
Xiangtan, China
Email: lglxtu@163.com
Shiguo Wang
The College of Information
Engineering
Xiangtan University
Xiangtan, China
Email: sgking@xtu.edu.cn
Yu Zheng
Department of Electrical and
Computer Engineering
University of Alberta
Edmonton, Canada
Email: yzheng2@ualberta.ca
Abstract—For the novel channel estimation method using
compressed sensing (CS), many strategies have been proposed for
its performance improvements. Nevertheless, random sequences
are usually used to construct sensing matrix, but it is difficult
and expensive to be implemented in hardware, and may cause
large Peak-to-Average Power Ratio (PAPR) in OFDM system.
Consequently, in this paper, the method that to obtain deter-
ministic sensing matrix by deterministic sequence named as
Golay Complementary Sequence is detailed investigated. Golay
sequence has the properties as perfect periodic auto-correlations,
simultaneously realizing timing acquisition and channel estima-
tion. Also PAPR can be limited within 3dB in OFDM system
by the method. Simulation results show that the deterministic
sensing matrix can offer reconstruction performance similar to
random matrix.
Index Terms—Channel Estimation; Compressed Sensing(CS);
Golay Complementary Sequences; Sparse Multipath
I. INTRODUCTION
Since compressive sensing was put forward by Donoho
[1], [2], Candes [3] and Chinese scientist Tao [4], [5], it
has been widely used in various areas. So far compressive
sensing has been successfully applied in channel estimation.
Compared to the conventional channel estimation methods
(LS) which do not explicitly account for the underlying
multipath sparsity, reliable estimation of the channel impulse
response in these settings can lead to significant reductions in
energy transmission and improvements in spectral efficiency.
As a matter of fact, the length of training sequence can be
shortened when compared with the linear estimation methods.
Compressed channel sensing was originally presented by Ba-
jwa [6], [7]. Existing results show that if entries of test vectors
are independent realizations of random variables with certain
distributions as zero-mean Gaussian with high probability,
the resulting observations sufficiently encode the information
in the unknown signal and recovery can be accomplished
by solving a tractable convex optimization problem. Based
on this, many researchers take up to investigate the sparse
multipath channel estimation. In literature [6], the scholars
apply random sequence to construct a sensing matrix [7]–
[13]. As an example, in [7], the training sequence is con-
structed by an independent identical distribution variable. Each
entry in Rademacher vector independently takes the value
+1/
√
𝑁 or −1/
√
𝑁 with probability 1/2 respectively. In
[10], the training sequence is used randomly to construct a
measurement matrix with toeplitz structure. Huge memory for
storage and high computational cost for signal reconstruction
is required.Besides, fully random matrices are often difficult or
expensive in hardware implementation. In the OFDM system,
if taking the random sequence as training sequence, the
corresponding PAPR is asymptotically 10 log
10
𝑁. But the
golay sequence can limit the PAPR within 3dB. Taking these
issues into account, based on the theory of previous articles,
in this paper, we will show that good autocorrelation can be
realized by using a deterministic golay sequence to construct
a sensing matrix.
The rest of this paper is organized as follows. The consid-
ered system model is presented in section II. Golay sequence
channel estimation is investigated in section III . Simulation
results are provided in Section IV. Finally concluding remarks
are given in Section V.
II. C
OMPRESSED CHANNEL SENSING MODEL
A. Compressed Sensing
Compressed Sensing (CS) is a new framework for simul-
taneous sampling and compression of signals, which utilizes
sparsity in representations to reduce the number of linear
measurements needed for signal encoding [1]–[3]. CS mainly
involves a problem of recovering a K- sparse signal 𝑥 ∈ 𝑅
𝑁
from a relatively small number of its measurements in the
form as follow:
𝑦 =Φ𝑥 ∈ 𝑅
𝑀
. (1)
Where Φ is the the sensing matrix and 𝑀<<𝑁.
As shown in [1]–[4], for a CS sensing matrix Φ,itsuffi-
ciently should satisfy the restricted isometry property (RIP),
in which sensing matrix is made act as a near isometry on all
K-sparse vectors. For example, an 𝑀 × 𝑁 sensing matrix Φ
satisfy the RIP with parameters (𝑘, 𝛿).For𝛿 ∈ (0, 1) , the RIP
of sensing matrix is with parameters
(1 − 𝛿) ∥𝑥∥
2
≤∥Φ𝑥∥
2
≤ (1 + 𝛿) ∥𝑥∥
2
, for all 𝑥 ∈ Ω.
(2)
Where Ω denotes the set of all length-𝑁 vectors with 𝐾
non-zero coefficients. As shown in [1] and [4], if Φ is a
976978-1-4799-8058-1/15/$31.00 ©2015 IEEE.