Acoustic echo cancellation using fast RLS adaptive filters with reduced complexity

Abstract

Absb-uct -We consider the recently proposed Fast Newton Transversal Filter 0, a new family of Fast Recursive Least Squares (FRLS) transversal fiiters with reduced computational complexity. This new algorithm is a good candidate for acoustic echo cancellation: it is only slightly more complex than the LMS whereas it yields greatly enhanced performance. Additional features needed for stable operation on speech signals taken from FRLS practice arediscussed. We show that Fast Newton TransversalFilters with lengths appropriate for acoustic echo cancellation in hands free applications can be implemented on one standard floating point DSP. Experiments on artificial and real speech signals show that the FNTFperformance is very close to the standard FRLS'ones.