Digital Signal Processing Reference
In-Depth Information
to this application, and then we widened the scope to the analysis of more
general nonlinear structures, as Volterra filter and artificial neural networks.
As far as artificial neural networks are concerned, we focused on two
main structures, MLP and RBF, and their corresponding algorithms: back-
propagation and
k
-means.
An important issue of this chapter was the analysis of the equalization
problem as a classification task. This allowed us to employ neural networks
as a method of solution. And, it was also possible to accomplish the deriva-
tion of the Bayesian equalizer. As discussed in Chapter 1, this equalizer
is the fundamental reference of optimal performance, but it cannot easily
be derived in the framework of linear filtering theory, as the suboptimal
solutions based on MSE and similar criteria. On the other hand, nonlin-
ear approaches as, for instance, the RBF can provide practical methods for
implementing the Bayesian filter.
