A Direction-of-Arrival Estimation Method via Joint
Diagonalization
Xian-feng Xu
1,*
, Ying-ying Li
1
, Ding-yi Wang
1
, Lai-jun Liu
2
1. School of Electronic & Control Engineering, Chang’an University, Xi’an, China
2. School of Highway, Chang’an University, Xi’an, China
*corresponding author
Email address: xuxianfeng1982@163.com, yingyingli@163.com, dingyiwang@163.com, ljliu@chd.edu.cn
Abstract—A new Direction-of-Arrival (DOA) estimation
method based on a Joint Diagonalization tool for Fast Blind
Source Separation (FBSS-DOA) is proposed in this paper. A
group of correlation matrices possessing diagonal structure is
generated. A cost function of joint diagonalization for blind
source separation is introduced. For solving this cost function, a
fast multiplied iterative algorithm in complex-valued domain is
utilized. The demixing matrix is then estimated and the
estimation of DOA can furthermore be realized. Compared with
familiar algorithms, the algorithm has more generality and better
estimation performance. Simulations results illustrate its
efficiency.
Keywords—DOA estimation; blind source separation (BSS);
joint diagonalization (JD); FBSS-DOA
I. INTRODUCTION
The Direction-of-Arrival (DOA) is a critical parameter of
source signals. The estimated DOA utilizing data received by
an array of sensors plays an important part in the array system
such as radar, communication, sonar, seismic detection, and so
on [1]-[2]. The technique only utilizing the receiving signals by
an array of sensors and the statistical property of the source
signals to estimate the channel parameters and to retrieve the
source signals, without any other prior knowledge on them, is
named as Blind Source Separation (BSS), also named as Blind
Signal Separation (BSS) [3]. Joint Diagonalization (JD) is one
of the most efficient tools for BSS [3]-[8]. In the JD algorithms,
it is desired to seek a matrix which is the estimation of the
mixing matrix, often named a diagonalizer, to diagonalize
simultaneously a set of square target matrices. This estimated
mixing matrix comprises the information of DOA. Thus JD
could also be considered as a tool to realize the estimation of
DOA [9]-[10].
Pourrostam applied the second-order blind identification
(SOBI) algorithm proposed for BSS to realize the estimation of
DOA, named as SOBI-DOA. This algorithm assumes that at
least one of the target matrices is positive definite in order to
construct the whitening matrix. However, the target matrices
are inevitable corrupted by some estimation error due to the
finite observation snapshots and received noise in practice.
This pre-whitening step is thus never exact. What is worse, this
preliminary error cannot be corrected by the following
operations. This limitation reduces the performance of SOBI-
DOA. The Triply Iterative Algorithm (TIA) [5] which has
avoided prewhitening operation was utilized for the estimation
of DOA in [10], named TIA-DOA. This algorithm has greatly
improved the estimation performance. However, the TIA-DOA
strictly restrains that the number of sensors should be more
than that of sources. Furthermore, the existed non-symmetricity
between the estimated left mixing matrix and the estimated
right mixing matrix has introduced the estimated errors to
DOA [6]. In order to conquer this, a fast algorithm for
complex-valued domain BSS is adopted in array signal
processing to realize the estimated DOA, named FBSS-DOA.
In FBSS-DOA, a multiplicative update was adopted to
minimize the Frobenius-norm formulation [3], [4], [8]-[9] of
the joint diagonalization problem. At each of multiplicative
iterations, a strictly diagonally-dominant updated matrix is
obtained. This scheme ensures the invertibility of the
diagonalizer. Then dig the information of DOA from the
estimated manifold matrix. Compared with SOBI-DOA [9],
this proposed FBSS-DOA algorithm discarded the whitening
operation and the damage to performance by whitening was
avoided. Compared with TIA-DOA [10] which required that
the number of sensors should be strictly lager than that of
sources, FBSS-DOA only demanded that the number of
sensors not less than that of sources. Therefore, the proposed
FBSS-DOA algorithm exhibited more general utilization areas
and more accurate estimated DOA performance.
II. SIGNAL MODEL
Consider a uniform linear array composed by
sensors
on which
narrow band noncoherent far field signals are
impinging with different DOAs. The signals arriving at the
sensor is
2
( 1) sin( )
1
( ) ( ) ( )
v
N
j u d
u v u
v
x t s t e n t
(1)
here
(
) is the interval between two adjacent sensor
elements.
is the wavelength and
is the DOA from the
source to the
sensor .
The received signals of the sensors at the snapshot
could be
expressed as
1
( ) [ ( ), , ( )] ( ) ( )
T
M
t x t x t t t x As n
. (2)
Here the array manifold matrix
12
( ), ( ), , ( )
N
A a a a
is column full rank.
This work was supported in part by the National Natural Science
Foundation of China (Grant No. 61201407, No. 61473047), in part by China
Postdoctoral Science Foundation (Grant No. 2013M542309), and in part by
the Special Fund for Basic Scientific Research of Central Colleges, Chang'an
University (Grant No. 0009-2014G1321038).