Digital Signal Processing Reference
In-Depth Information
dJ
dw
¼ 0:96 ð0:032Þ¼0:992
Finally, substituting w
z
w
3
¼ 0:992 into the MSE function, we get the minimum J
min
as
J
min
z
40 20w þ 10w
2
w¼0:992
¼ 40 20 0:992 þ 10 0:992
2
¼ 30:0006
As we can see, after three iterations, the filter coefficient and minimumMSE values are very close to the theoretical
values obtained in Example 10.1.
w
3
¼ w
2
m
Application of the steepest descent algorithm still needs an estimation of the derivative of the MSE
function that could include statistical calculation of a block of data. To change the algorithm to do
sample-based processing, an LMS algorithm must be used. To develop the LMS algorithm in terms of
sample-based processing, we take the statistical expectation out of
J
and then take the derivative to
obtain an approximation of
dJ
dw
, that is,
2
2
J ¼ e
ðnÞ¼ðdðnÞwxðnÞÞ
(10.9)
dw
¼
2
ðdðnÞwxðnÞÞ
dðdðnÞwxðnÞÞ
dJ
¼
2
eðnÞxðnÞ
(10.10)
dw
Substituting
dJ
dw
into the steepest descent algorithm in Equation
(10.8)
, we achieve the LMS algorithm
for updating a single-weight case as
w
nþ
1
¼ w
n
þ
2
meðnÞxðnÞ
(10.11)
where
m
is the convergence parameter controlling speed of convergence. For example, let us choose
2
m ¼
0
:
01. In general, with an adaptive FIR filter of length
N
, we extend the single-tap LMS algo-
rithm without going through derivation, as shown in the following equations:
yðnÞ¼w
n
ð
0
ÞxðnÞþw
n
ð
1
Þxðn
1
Þþ
/
þ w
n
ðN
1
Þxðn N þ
1
Þ
(10.12)
for
i ¼
0
;
/
; N
1
w
nþ
1
ðiÞ¼w
n
ðiÞþ
2
meðnÞxðn iÞ
(10.13)
The convergence factor is chosen to be
1
NP
x
0
< m <
(10.14)
where
P
x
is the input signal power. In practice, if the ADC has 16-bit data, the maximum signal
amplitude should be
A ¼
2
15
. Then the maximum input power must be less than
P
x
<
2
15
2
¼
2
30
Search WWH ::
Custom Search