Digital Signal Processing Reference
In-Depth Information
dJ
dw
dJ
dw
dJ
dw
dJ
dw
Case
0
and
0
Case
0
and
0
J
J
J
n
J
n
1
J
n
J
n
1
J
min
J
min
w
*
w
n
1
w
n
w
w
n
w
n
1
w
w
*
FIGURE 10.6
Illustration of the steepest descent algorithm.
coefficient
w
nþ
1
will be decreased to approach the optimal value
w
. When
dJ
dw
¼
0, the best coef-
ficient
w
nþ
1
is reached.
EXAMPLE 10.2
Consider the following quadratic MSE function for the Wiener filter:
J ¼ 40 20w þ 10w
2
Use the steepest decent method with an initial guess of w
0
¼ 0 and m ¼ 0:04 to find the optimal solution for w
and determine J
min
by iterating three times.
Solution:
Taking a derivative of the MSE function, we have
dJ
dw
¼20 þ 10 2w
n
When n ¼ 0, we calculate
dw
¼ 0:04 ð20 þ 10 2w
0
Þ
w
0
¼0
¼0:8
Applying the steepest decent algorithm, it follows that
dJ
m
dJ
dw
¼ 0 ð0:8Þ¼0:8
w
1
¼ w
0
m
Similarly for n ¼ 1, we get
dw
¼ 0:04 ð20 þ 10 2w
1
Þ
w
1
¼0:8
¼0:16
dJ
m
dJ
dw
¼ 0:8 ð0:16Þ¼0:96
w
2
¼ w
1
m
and for n ¼ 2, it follows that
dw
¼ 0:04 ð20 þ 10 2w
2
Þ
w
2
¼0:96
¼0:032
dJ
m
Search WWH ::
Custom Search