An adaptive non-Gaussian filtering using pattern recognition approach

An approach to adaptive non-Gaussian filtering based on the approximate maximum likelihood estimation, the so-called M-estimation, and a pattern recognition methodology has been considered in the paper. The proposed pattern recognition approach is based on the generation of a suitably chosen learning set, the appropriate selection of pattern vectors and the reduction of their dimension, as the k nearest neighbors classification procedure. A possibility of constructing an expert system for adaptive filtering is also discussed. The feasibility of the approach is demonstrated with simulations.