770 IEEE ANTENNAS AND WIRELESS PROPAGATION LETTERS, VOL. 15, 2016
Time-Frequency Multi-Invariance ESPRIT
for DOA Estimation
Jincheng Lin, Xiaochuan Ma, Shefeng Yan, Senior Member, IEEE, and Chengpe
ng Hao, Member, IEEE
Abstract—In this letter, a new ESPRIT-type algorithm is
presented for estimating the direction of arrival (DOA). We
reconstruct the received signal to form a time-frequency -
data model with a multi-invariance (MI) property, and then a -
MI Estimation of Signal Parameters via Rotational Invariance
Technique (ESPRIT) algorithm for DOA estimation is proposed.
This algorithm constructs a spatial - distribution matrix in
terms of variable temporal and frequency points and combines
multiple subarrays rearranged from a uniform linear array (ULA)
to estimate far-field narrowband signals. Considering the model of
multiple narrowband chirp sources, simulation results illustrate
that
- MI-ESPRIT outperforms - ESPRIT and MI-ESPRIT
under different signal-to-noise ratio (SNR) and snapshot numbers.
Index Terms—Direction-of-arrival (DOA) estimation, Estima-
tion of Signal Parameters via Rotational Invariance Techniques
(ESPRIT), multi-invariance (MI), time-frequency
- .
I. INTRODUCTION
C
ONSIDERING a uniform linear array (ULA), Esti-
mation of Signal Parameters via Rotational Invariance
Techniques (ESPRIT) [1] shows a unique superiority for direc-
tion-of-arrival (DOA) estimation. ESPRIT needs no beamscan,
so it has a more convenient computation complexity than other
subspace-based methods [2], [3]. Notably, the multi-invariance
ESPRIT (MI-ESPRIT) [4]–[8] exploits the inherent multi-in-
variance structure of array signals and presents a high-accuracy
estimation performance. For many decades, time-frequency
- signal processing methods were only applied to analyze
the nonstationary signal of a single-sensor receiver. The
-
characteristics of array signals were fully analyzed in [9] and
[10]. The spatial time-frequency distribution (STFD) is very
effective for nonstationary or wide-frequency signals. In [11],
the
- ESPRIT method has been analyzed and shows a more
accurate estimation than conventional ESPRIT.
In recent years, diverse parameter estimation algorithms
based on ESPRIT have been proposed in [12]–[20]. Precisely,
[13]–[15] utilize the properties of multiple subarrays to improve
the performance of sources localization. Two-dimensional ES-
PRIT-type algorithms, focusing on the different signal sources
of particular forms, have been proposed in [16]–[18]. The joint
time-frequency characteristics, for the ESPRIT-based DOA
estimation, have been utilized in recent research [17]–[20].
Manuscript received July 05, 2015; accepted August 23, 2015. Date of publi-
cation August 27, 2015; date of current version March 17, 2016. This work was
supported by the National Natural Science Foundation of China under Grants
61222107 and 61471352.
The authors are with the Institute of Acoustics, Chinese Academy of Sciences,
Beijing 100190, China (e-mail: ljcmym@163.com; maxc@mail.ioa.ac.cn;
sfyan@ieee.org; haochengp@mail.ioa.ac.cn).
Digital Object Identifier 10.1109/LAWP.2015.2473664
However, in - ESPRIT, the multi-invariance property of a
ULA has not been utilized.
Here, we present a new method based on both STFD and
multi-invariance of a ULA for DOA estimation. A model of nar-
rowband chirp signals is considered in numerical simulations. It
will be shown that the proposed approach outperforms ESPRIT
and MI-ESPRIT in terms of root mean square error (RMSE).
II. N
OTATION
Let us first list some notational conventions that will be used
in this letter:
•
: the number of array elements.
•
: the number of sources.
•
: the number of sampling points of each temporal
window.
•
: the number of temporal windows for STFD.
•
: is the signal sequence
matrix of
sources, is the signal sequence of the
th source.
•
: is the array mani-
fold matrix for
array elements and sources.
•
: is the channel matrix for array elements and
sources.
•
: denotes the receiving sequences of array
element, and the length of each sequence is
.
III. S
IGNAL MODEL
Consider signals impinging on a ULA of sensors. The
channel matrix through which the signals propagate is expressed
as follows [7]:
.
.
.
.
.
.
.
.
.
.
.
.
(1)
The data received by the
th array element is
(2)
where
. is a matrix operator,
which forms a diagonal matrix using the
th row of the processed
matrix.
In this letter, multiple far-field narrowband signals are as-
sumed in the signal model, i.e., chirps whose bandwidth is
1536-1225 © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.