performance degradation. In order to improve the
robustness against above problems, a new type of beam-
former based on combination of INC matrix reconstruction
and SV is proposed [9–12].In[9], the INC matrix recon-
struction was performed based on the Capon spectrum by
integrating over an angular sector where the interference
SVs locate. However, in [10], by exploiting the sparsity of
the directions of the source signals, the INC matrix was
reconstructed by solving a compressive sensing (CS) pro-
blem. Although the performance of above beamformers in
[9,10] can be effective against signal direction error, the
precise array manifold is required to provide. Practically,
the array calibration errors always exist due to antenna
location error, mutual coupling, and the channel error. In
this case, the reconstruction operation only considers the
pointing errors to ignore other potential errors. In [11], the
high-dimensional annulus uncertainty set of interference
SV was used to reconstruct the INC matrix. Moreover, the
SV is estimated by solving a quadratically constrained
quadratic programming (QCQP) problem. In [12], the
reconstruction operation had taken array manifold error
into account, however, it led to a large amount of calcu-
lation without considering the sparsity of interference
signal. More importantly, these beamformers based on
reconstruction paid little attention to the effectiveness of
reconstruction in different input SNRs.
In this paper, the idea of improve factor (IF) is intro-
duced to evaluate the effectiveness of reconstruction. From
the point of view of IF, a novel beamformer is proposed to
improve robustness against SV mismatch and the covar-
iance matrix contaminated by desired signal component.
The major work is presented in the following three
aspects. The first aspect is to introduce IF. By exploiting the
relationship between the IF and the input SNR, we find
that IF is small in low SNR, and in contrast, a large IF can be
obtained in high SNR. The second one is to estimate the
desired SV in advance, then the SNR is estimated. Because
the SV has been estimated, the relatively precise signal
power can be obtained. Instead of using convex optimi-
zation toolbox, the desired SV is estimated based on the
spherical uncertainty. The noise power can be approxi-
mately calculated by the minimum eigenvalue of the
sample covariance matrix. So far, the SV and SNR estima-
tion are completed. The third one is to reconstruct the INC
matrix based on the Capon spatial spectrum integration by
making use of the sparsity of interferences direction.
Meanwhile, the array calibration errors are taken into
account.
The rest of this paper is organized as follows. The sys-
tem model and some necessary background are demon-
strated in Section 2.InSection 3, a new beamformer based
on effectiveness of reconstruction is proposed. Then, the
desired SV is estimated based on a spherical uncertainty.
Next, the INC matrix is reconstructed by making use of
sparsity of interferences distribution. In Section 4, simu-
lation results are presented. Finally, some conclusions are
drawn in Section 5.
2. System model and problem background
By considering a uniform linear array consisting of M
isotropic antenna elements spaced by half wavelength, the
instantaneous output of a narrowband beamformer at
time k is expressed by [3]
yðkÞ¼w
H
xðkÞð1Þ
where w A C
M1
denotes the complex weight vector. The
symbol ðÞ
T
and ðÞ
H
denote the transpose and Hermitian
transpose, respectively. The array observation vector xðkÞ
can be denoted as [3]
xðkÞ¼sðkÞaþiðkÞþnðkÞ¼sðkÞþiðkÞþnðkÞð2Þ
where sðkÞ, iðkÞ, and nðkÞ denote the statistically indepen-
dent components of the desired signal, interference, and
noise, respectively. The a is the SV corresponding to the
direction of desired signal, s(k) is the complex signal
waveform. The optimal weight vector w
opt
can be derived
by maximizing the output signal-to-interference-plus-
noise ratio (SINR) [8]
SINR ¼
σ
2
s
∣w
H
a∣
2
w
H
R
i þ n
w
ð3Þ
where σ
s
2
represents the desired signal power, and
R
i þ n
A C
MM
is the INC matrix. As can be derived that the
maximization of (3) is equivalent to the Capon beamfor-
mer
min
w
w
H
R
i þ n
w s:t: w
H
a ¼ 1 ð4Þ
From (4), the well-known solution can be expressed as
follows:
w
opt
¼
R
1
i þ n
a
a
H
R
1
i þ n
a
ð5Þ
Correspondingly, the optimal SINR is given by [8]
SINR
opt
¼ σ
2
s
a
H
R
1
i þ n
a ð6Þ
From (6), the output SINR is not only related to the
desired signal steering vector a, but also the INC matrix
R
i þ n
. Unfortunately, neither of them is easy to obtain in
practice. The former is usually replaced by the nominal
(presumed) one
a, whereas, in practical circumstances,
some mismatches between the nominal and actual SV can
easily occur due to array imperfections such as look
direction, signal pointing errors, array calibration errors. If
the INC matrix R
i þ n
is available, adaptive beamformer can
be performed well against these mismatches. However,
the R
i þ n
requires an infinite number of pure snapshots
data without the desired signal. It is usually replaced by
sample covariance matrix
^
R
x
¼ð1=KÞ
P
K
k ¼ 1
xðkÞx
H
ðkÞ,
where K is the number of snapshots. In this case, it may
result in significant performance deterioration, especially
in the high input SNR [6]. As stated above, it is necessary to
estimate the desired signal SV and make the component of
the SOI free from the sample covariance matrix,
simultaneously.
Fr om the stan dpoint of cova rianc e matrix, recently, a
robust adaptive beamformer (RAB) technique based on the
Y. Zhang et al. / Signal Processing 120 (2016) 572 – 579 573