Our analysis of the problems with Spitz's method suggests that the prediction
filter at a given frequency should be computed from the data component
of that frequency.
However, prediction filters computed from high-frequency components of data may
suffer from contamination by spatially aliased energy. This section
first shows that spatially aliased energy creates ambiguities
regarding the zeros of the *z*-transform of the prediction filters and
describes how to use a neural net to remove these ambiguities.

- The zeros of prediction filters
- Dealiasing prediction filters with a neural net
- Summary of the algorithm

11/18/1997