Adaptive filtering incorporating a local mean estimation substructure

Adaptive filters have been used successfully in many applications of signal processing. However, their performance in dealing with signals of nonzero mean, especially with sharp changes (edges), is problematic, which limits the extension of the use of adaptive filters in some important application areas. To overcome this problem, the authors introduce in this paper a scheme for adaptive filtering obtained by incorporating a local mean estimation substructure (denoted as AF-LME scheme). It is shown by theoretical analysis that, by handling the signal mean and the zero mean component (the residual signal) separately, the performance of adaptive algorithms (e.g. the LMS or the RLS) can be improved. Analysis is also given to show the weakness of an adaptive filter in dealing with the edges of the signal mean. This weakness can be overcome by incorporating the technique of low-pass filtering with an edge preserving property as the local mean substructure. A method for the implementation of this substructure is proposed. This implementation was satisfactorily used in computer simulations and representative examples simulations are presented