Multidim Syst Sign Process
The rest of this paper is organized as follows. The background of adaptive beamforming
and several previous robust adaptive beamforming techniques are overviewed in Sect. 2.In
Sect. 3, the newly proposed robust adaptive beamforming algorithm is established. Section 4
shows and discusses our simulation results with different types of signal model mismatch.
Finally, the concluding remarks are contained in Sect. 5.
2 Signal model and preliminaries
Assume that a linear antenna array with N omni-directional antenna elements receives signals
from multiple narrowband sources. The array observation snapshot vector x
(
k
)
is given by
x
(
k
)
= s
(
k
)
a
0
+ i
(
k
)
+ n
(
k
)
(1)
= s
(
k
)
+ i
(
k
)
+ n
(
k
)
(2)
where k is the time index, s
(
k
)
, i
(
k
)
and n
(
k
)
denote the desired signal, interference
signal and noise component, respectively. s
(
k
)
and a
0
are the desired signal waveform and
corresponding steering vector, respectively. In the sequel, we assume that the desired signal,
interference, and noise waveforms are quasi-stationary zero-mean random processes and that
the desired signal waveform is uncorrelated with the interference and noise waveforms. Then
the beamformer output at the k th time index can be expressed as
y
(
k
)
= w
H
x
(
k
)
(3)
where w is the complex N ×1 beamforming weight vector and
(
·
)
H
is a Hermitian transpose
operator. The beamforming problem is formulated as finding the weight vector w which
maximizes the output signal-to-interference-plus-noise ratio (SINR) defined as
SINR =
σ
2
0
w
H
a
0
2
w
H
R
i+n
w
(4)
where σ
2
0
= E
|
s
(
k
)
|
2
is the power of the desired signal, R
i+n
= E
(
i
(
k
)
+ n
(
k
))
(i
(
k
)
+ n(k))
H
is the interference-plus-noise covariance matrix, and E
{
·
}
denotes the statistical
expectation. Alternatively, the maximization of problem (4) is mathematically equivalent to
solve the following optimization problem
min
w
w
H
R
i+n
w
s.t. w
H
a
0
= 1
(5)
Since the received snapshots contain the desired signal component and the training data
size is finite, R
i+n
is usually unavailable in practical applications. Instead, the sample covari-
ance matrix
ˆ
R =
1
K
K
k=1
x
(
k
)
x
H
(
k
)
(6)
should be used, where K is the number of snapshots collected. By replacing R
i+n
with
ˆ
R,
the solution of (5) is well known as the minimum variance distortionless response (MVDR)
beamformer or the Capon beamformer, which is given by
w
MVDR
=
ˆ
R
−1
a
0
a
H
0
ˆ
R
−1
a
0
(7)
123