Nonlinear acoustic echo cancellation with fast converging memoryless preprocessor

CONVERGING MEMORYLESS PREPROCESSOR A. Stenger and W. Kellermann University of Erlangen-N urnberg, Tele ommuni ations Laboratory Cauerstr. 7, D-91058 Erlangen, Germany stenger nt.e-te hnik.uni-erlangen.de ABSTRACT Lowost audio omponents in hands-free telephone appli ations all for nonlinear adaptive e ho an ellation (AEC). It has been demonstrated that a as ade of a polynomial and an FIR lter an an el su h nonlinear e hoes [1℄. Another te hnique employs a hardlipping urve with LMS adapted saturation parameter [2℄. For both as aded systems we derive an LMStype adaptation using a ommon framework, and propose stepsize normalizations for both approa hes. To a hieve suÆ iently fast onvergen e for pra ti al use, an RLS-type adaptation for the polynomial is derived and experimentally veri ed. Both te hniques are ompared using real hardware and spee h signals, and show robust onvergen e behaviour and an e ho redu tion gain of up to 10 dB ompared to a linear AEC.