Network structures for nonlinear digital filters

Mapping neural networks based on a piecewise-linear (PWL) function approximation scheme are useful in signal processing, i.e. nonlinear filtering. However, the traditional canonical PWL model has a drawback that limits the usefulness of these networks. To overcome this limitation, three more general PWL models with their network implementation structures are introduced in this paper. As the first application of the models in signal processing, the modelling, the unification, and the generalization of the useful nonlinear filter family, the order statistic filters are considered.<
>