Digital Signal Processing Reference
In-Depth Information
The equation above can also be obtained using (7.14),
]
[ ]
=
() ()
]
[
]
[
()
[
()
()
=
()
(
)
w
0
xn
yn
xn
xn
-
1
w
0
w
1
()
(
)
w
1
xn
-
1
which reduces to
()
=
()()
+-
(
) ( )
yn
xnw
0
xn
1
w
1
which can also be obtained by summing the signals at the node of the two-weight
diagram shown in Figure 7.8.
As can be seen in Figure 7.8, the linear combiner with a single input is just an
FIR filter with adjustable coefficients. Although this is a very simple configuration,
it can handle many of the adaptive applications.
7.4 PERFORMANCE FUNCTION
In the preceding section we provided a structure for the filter whose characteristics
may be changed by adjusting the weights. However, we still need a way to judge
how well the filter is operating—a performance measure is needed. The perfor-
mance function will be based on the error, which is obtained from the block diagram
in Figure 7.1, with the time index incorporated:
()
=
()
-
()
en
dn
yn
(7.15)
The square of this function is
()
=
()
-
()()
+
()
2
2
2
en dn
2
dnyn yn
(7.16)
which is the instantaneous squared-error function. In terms of the weights, it
becomes
()
=
()
-
() ()
+
() ()
en dn
2
2
2
dn
XWWXXW
T
n
T
n
T
n
(7.17)
where the time index on the
W
has been dropped. Equation (7.17) represents a qua-
dratic surface in
W
, which means that the highest power of the weights is the squared
power. The strategy will be to adjust the weights so that the squared-error function
will be a minimum.
To understand the performance surface equation (7.17), consider the case of one
weight. The error surface then becomes
()
=
()
-
()()()
+
() ()
en dn
2
2
2
dnxnw
0
xnw
2
2
0
(7.18)
which is a second-order function in
w
(0). To find the minimum, set the derivative of
Search WWH ::
Custom Search