Nonlinear system design for image processing and data communications

This thesis is concerned with the design of nonlinear structures for two purposes: to remove Gaussian and impulsive noise from 2-D signals and to remove echo and intersymbol interference in full duplex data communication systems.Three different filter structures are designed to remove Gaussian and impulsive noise from 2-D signals using the Fast Orthogonal Search algorithm. The two non-recursive filters are chosen to be isotropic so as to decrease the complexity of the problem.Nonlinear echo cancellers are designed for baseband full duplex data transmission over the subscriber loop telephone network. The Fast Orthogonal Search algorithm enabled the removal of the impairment with almost no error in an adaptive fashion. (Abstract shortened by UMI.)The use of the Fast Orthogonal Search algorithm introduces two important advantages. It minimizes the filter complexity for a given level of filter performance which in effect can be chosen to be optimal. This algorithm also guarantees a unique solution for a given filter structure.