Digital Signal Processing Reference
In-Depth Information
10
-1
10
-2
10
-3
MLP
10
-4
Proposed NN
10
-5
0
10
20 30
× 5,000 Iterations
40
50
FIgure 5.10
Fault detection: learning curves.
0.7
g
1
(
x
)
0.6
NN
1
(
x
)
g
1
(
x
)
0.5
NN
1
(
x
)
0.4
Before change
0.3
0.2
After change
0.1
0
0
0.5
1
1.5
2
x
FIgure 5.11
g
1
(
x
) and
NN
1
(
x
) (normalized) before and after the change.
shown in the figure) were correctly identified as well. We have studied other MIMO sys-
tems with higher dimensions (3 × 3 and 4 × 4 systems) in which we simulated changes in
more than one nonlinearity; the proposed NN model was always capable of character-
izing the changes.
On the other hand, the black-box MLP structure could not characterize the change. By
looking at the MLP learning curve (Figure 5.10), one could only predict that “something
Search WWH ::
Custom Search