Digital Signal Processing Reference
In-Depth Information
the matrix inversion lemma formulated in Lemma 4.1. If we set the equivalences
A
x T
λ
A n 1 , D
1, B
x
(
n
)
, and C
(
n
)
, by applying them to ( 4.20 ) we
arrive at:
1 λ 2 A 1
A 1
n
x T
(
)
(
)
1 x
n
n
A 1
n
= λ 1 A 1
n
n
1
.
(5.50)
A 1
n
+ λ 1 x T
1
(
n
)
1 x
(
n
)
Replacing ( 5.50 )in( 5.49 ) we obtain:
1 b n 1 λ 1 A 1
x T
A 1
n
1 x
(
n
)
(
n
)
1 b n 1
n
A 1
n
+ λ 1 A 1
n
w
(
n
) =
1 x
(
n
)
d
(
n
)
A 1
n
+ λ 1 x T
1
(
n
)
1 x
(
n
)
λ 2 A 1
A 1
x T
n 1 x
(
n
)
(
n
)
n 1 x
(
n
)
d
(
n
)
.
(5.51)
A 1
n
+ λ 1 x T
(
)
(
)
1
n
1 x
n
A 1
n
Firstly, we notice that w
1 b n 1 . Also notice that the last two terms in
the previous equation can be combined into k
(
n
1
) =
(
n
)
d
(
n
)
, where:
λ 1 A 1
n
1 x
(
n
)
k
(
n
) =
) .
(5.52)
A 1
n
1
+ λ 1 x T
(
n
)
1 x
(
n
Using these observations, Eq. ( 5.51 ) can be written as:
d
x T
w
(
n
) =
w
(
n
1
) +
k
(
n
)
(
n
)
(
n
)
w
(
n
1
)
.
(5.53)
x T
Defining
ζ(
n
)
d
(
n
)
(
n
)
w
(
n
1
)
as the a priori estimation error [ 7 ], the LS
solution to ( 5.43 ) takes the form:
w
(
n
) =
w
(
n
1
) +
k
(
n
)ζ(
n
).
(5.54)
The operation of the algorithm can be summarized as follows:
0. Compute A 1
1 = δ 1 I L .
Set n
=
0. Choose w
(
1
) =
0 ,0
1 and
δ>
At time instant n , after receiving d
(
n
)
and x
(
n
)
, calculate:
λ 1 A 1
n
1 x
(
n
)
k
(
n
) =
.
A 1
n
1
+ λ 1 x T
(
n
)
1 x
(
n
)
A 1
= λ 1 A 1
n
1 λ 1 k
A 1
n
x T
(
)
(
)
n
n
n
1
x T
ζ(
n
) =
d
(
n
)
(
n
)
w
(
n
1
)
w
(
n
) =
w
(
n
1
) +
k
(
n
)ζ(
n
)
 
Search WWH ::




Custom Search