Proceedings of ICCT2013
DOA Estimation Based on Sparse Representation
via Folding OMP
Xiaohuan Wu, Jun Yan, Ying Ji, Wei-ping Zhu
Institute of Signal Processing and Transmission, Nanjing University of Posts and Telecommunications,
Nanjing 210003, China
{1010010518, yanj, 1012010528, zwp}@njupt.edu.cn
Abstract:
In this paper, a new direction-of-arrival (DOA)
estimation method is proposed based on the array
cross-correlation vector (ACCV) model which can
decrease the computational complexity of multiple
measurement vectors (MMV) model. Firstly, the ACCV
model is refined to accommodate the correlated signal
scenario. Then by properly incorporating the folding
scheme in the compressive sensing (CS) framework, a
new algorithm termed folding orthogona matching
pursuit (FOMP) is proposed in the reconstruction of CS
framework. The method has a lower computational
complexity and higher performance compared with other
existing DOA algorithms. Numerical results are
presented to verify the efficiency of the proposed
method.
Keywords: Direction-of-arrival (DOA) estimation;
sparse representation; multiple measurement vectors;
Orthogonal Matching Pursuit
1 Introduction
Direction-of-arrival (DOA) estimation of far-field
narrowband signals has been of interest in the past few
decades [1] and played a fundamental role in many areas
in engineering, involving radar based antenna arrays and
microphone arrays [2]. Many high-resolution algorithms
have been proposed, including MUSIC [3], ESPRIT [4]
and their variations [5] [6]. Unfortunately, those
methods may fail to provide an accurate estimation of
the direction of arrival of correlated sources because
these techniques rely on the eigenvectors of the noise
subspace of the covariance matrix which is theoretically
singular in the presence of correlated arrivals. Some
decorrelated techniques for reducing the effective size of
the array, have been put forward and employed to detect
correlated signals successfully.
The technique of sparse representation provides a new
perspective for DOA estimation. The DOAs of signals
are usually very sparse relative to the entire spatial
domain which can be divided into a large number of
spaces. In this way, the DOA estimation problem can be
solved by finding the sparsest representation of the data
which can be solved by many existing algorithms. Those
algorithms are divided into two classes: those with a
single measurement vector (SMV) and those with
multiple measurement vectors (MMV). The sparse
representation with SMV model is simple but with a low
accuracy. The MMV model with multiple snapshots is
investigated in [7] and is then proved to be of a high
accuracy. But the MMV model survives more
computational cost, and thus the DOA estimation is
unreachable for realtime scenario. Based on focal
underdetermined system solver (FOCUSS), proposed by
Gorodnitsky et al. in [8], M-FOCUSS [7] algorithm has
been introduced to solve the MMV problem. By
combining the singular value decomposition (SVD) step
of the subspace algorithms with a sparse recovery
method based on ℓ1-norm minimization, Malioutov et al.
proposed the method of L1-SVD [9]. Hyder and Mahata
introduced their ℓ0-norm-based joint sparse
approximation technique into DOA estimation and
presented a method called joint ℓ0 approximation DOA
(JLZA-DOA) in [10] to resolve the MMV problem.
One of the latest algorithms is the L1-ACCV proposed
in [11], which transforms the MMV problem into an
SMV model through an array cross-correlation vector
(ACCV), and then solves the problem by the
second-order cone (SOC) programming framework, thus
decreasing the computational complexity. However, this
method fails to localize the correlated sources.
In this paper, the ACCV model is refined to cope with
correlated signal scenario. Then a new algorithm named
folding orthogonal matching pursuit (FOMP) is
proposed by folding the measurement matrix to reduce
the computational cost. With the combination of the
ACCV model and the FOMP algorithm, we present a
method named CV-FOMP to obtain the DOA estimation
with a high accuracy and a low computational
complexity.
In this paper, the notations
[]E
,
[]
H
]
,
,
,
†
denote expectation operator, conjugate transpose,
round up operator, round down operator, pseudo-inverse,
respectively.
2 Signal Model
Suppose that K far-field stationary and narrowband
signals impinge on an uniform linear array (ULA) of M
omnidirectional elements from direction angles
12
[, , , ]
K
θ
. The incident signals
()ts
are
corrupted by additive white Gaussian noise
()tn
. Let
12
() [ (), (), , ()]
T
M
txtxt xtx
be the signals received
by the array. Then the output of the array can be written
as
____________________________________
978-1-4799-0077-0/13/$31.00 ©2013 IEEE