Digital Signal Processing Reference
In-Depth Information
is a weight function that depends on the density generator
g
(
) of the underlying cir-
cular CES distribution. For the CN distribution (i.e., when
g
(
d
)
¼
exp (
d
)), we
have that
w
ml
;
1, which yields the SCM
C
as the MLE of
S
. The MLE for
T
k
,
n
dis-
tribution (cf. Example 2.2), labeled MLT(
n
), is obtained with
2
k
þ
n
nþ
2
x
:
w
ml
(
x
)
¼
(2
:
24)
Note that MLT(1) is the highly robust estimator corresponding to MLE of
S
for the
complex circular Cauchy distribution, and that MLT(
n
)
!C
(as
T
k
,
n
!F
k
)as
n !1
, thus the robustness of MLT(
n
) estimators decrease with increasing values
of
n
.
We generalize (2.22), by defining the
M
-estimator of scatter, denoted by
C
w
, as the
choice of
C
[
PDH(
k
) that solves the estimating equation
n
X
n
1
w
(
z
i
C
1
z
i
)
z
i
z
i
C ¼
(2
:
25)
i¼
1
where
w
is any real-valued weight function on [0,
1
). Hence
M
-estimators constitute a
wide class of scatter matrix estimators that include the MLE's for circular CES distri-
butions as important special cases.
M
-estimators can be calculated by a simple iterative
algorithm described later in this section.
The theoretical (population) counterpart, the
M
-functional of scatter, denoted by
C
w
(
z
), is defined analogously as the solution of an implicit equation
C
w
(
z
)
¼ E
[
w
(
z
H
C
w
(
z
)
z
)
zz
H
]
:
(2
:
26)
Observe that (2.26) reduces to (2.25) when
F
is the empirical distribution
F
n
, that is,
the solution
C
w
of (2.25) is the natural plug-in estimator
C
w
(
F
n
). It is easy to show that
the
M
-functional of scatter is equivariant under invertible linear transformation of the
data in the sense required from a scatter matrix. Due to equivariance,
C
w
(
F
S
)
¼ s
w
S
,
that is, the
M
-functional is proportional to the parameter
S
at
F
S
, where the positive
real-valued scalar factor
s
w
¼ s
w
(
d
) may be found by solving
E
[
w
(
d=s
w
)
d=s
w
]
¼ k
(2
:
27)
where
d
has density (2.12). Often
s
w
need to be solved numerically from (2.27) but in
some cases an analytic expression can be derived. Since parameter
S
is proportional to
underlying covariance matrix
C
(
F
S
), we conclude that the
M
-functional of scatter is
also proportional to the covariance matrix provided it exists (
i.e.
,
g
[
G
1
). In many
applications in sensor array processing, covariance matrix is required only up to a
constant scalar (see e.g. Section 2.7), and hence
M
-functionals can be used to
define a robust class of array processors.
Search WWH ::
Custom Search