A HYBRID METHOD FOR BLIND SIGNAL DE-NOISING VIA INDEPENDENT COMPONENT ANALYSIS

Alessandra Budillon, Francesco Palmieri and Rosario Varriale
e-mail: falebudil,frapalmig@unina.it,rosvarri@tin.it

This paper presents a signal de-noising method based on the extraction of two independent components. Two FIR lters are adapted blindly and forced to produce outputs which are as independent as possible through time. Various cost func- tions are considered ranging from linear correlation to non linear correlation and joint entropy. The typical presence of a large number of local minima in the various cost functions has suggested a hybrid approach that trades computational complexity with solution accuracy. Two experiments on speech corrupted by independent noise and the separation of two AR processes are presented.