Nonlinear signal processing filters: A unification approach

The paper addresses the problem of classification of nonlinear filters for digital signal processing and of unification of approaches to their analysis, design and implementation. It is assumed that the filters work in a moving window and, in each position of the window, produce an output signal sample as an estimate obtained from input signal samples within the window. The keys of the classification and unification are notions of estimation and neighborhood building operations. On the base of thorough analysis of a large variety of known filters, a set of estimation and multi stage (in general) neighborhood building operations is introduced that can serve as building blocks for the filter design. In conclusion, application of the approach to the filter design and implementations is briefly discussed.