Digital Signal Processing Reference
In-Depth Information
Proof.
The mean square error can be expressed as
E
e
⊥
=
E
[
e
⊥
e
∗
⊥
x
)
∗
]=
E
[
e
⊥
x
∗
]
]=
E
[
e
⊥
(
x
−
−
(using Eq. (16.70))
|
2
=(1
−
α
)
E
|
x
(using Eq. (16.69))
so that
E
e
⊥
=(1
−
α
)
E
x
,
(16
.
74)
which proves Eq. (16.73). But in view of Eq. (16.68), we also have from
Eq. (16.69)
α
)
2
E
x
+
E
e
⊥
=(1
−
E
τ
,
(16
.
75)
where we have used the fact that
α
is real. Equating the right-hand sides
of Eqs. (16.74) and (16.75) we get Eq. (16.71). Dividing (16.74) by (16.71)
yields Eq. (16.72).
16.A.1 Bias-removed estimates
From Eq. (16.67) we see that the expected value of the estimate
x,
given the
transmitted symbol
x,
is given by
E
[
x
|
x
]=
αx
+
E
[
τ
|
x
]
.
(16
.
76)
Usually
τ
is a zero-mean random variable,
E
[
τ
]=0
,
(16
.
77)
and it is statistically independent of
x
. In particular, therefore,
E
[
τ
|
x
]=0
,
and
we see that
E
[
x
|
x
]=
αx
=
x.
(16
.
78)
Since the conditional expectation of
x
is different from
x
, we say that the estimate
is
biased
. An obvious way to remove the bias is to just “kick it out.” That is,
define the bias-removed estimate
x
br
=
x/α
. This has the form
x
br
=
x
+
τ
α
(16
.
79)
where “
br
” is a reminder for “bias-removed.” Clearly the mean squared error of
the bias-removed estimate is
E
τ
/α
2
,
which simplifies to
E
e
br
=
E
e
br
=
E
e
α
(16
.
80)
using Eq. (16.72). Since
α<
1 (unless the estimate is already unbiased), there
is an
increase
in the MSE due to bias removal.
Search WWH ::
Custom Search