RLS-adapted polynomial for nonlinear acoustic echo cancelling

Low-cost audio components in hands-free telephone applications call for nonlinear adaptive echo cancellation. It has been demonstrated that a cascade of a polynomial and an FIR filter can cancel echoes due to nonlinearities of typical low-cost applications [1]. Overcoming the slow convergence of an NLMS-based adaptation [5], we derive an RLS-type adaptation for the polynomial preprocessor. Furthermore, we present a robust control of the RLS algorithm. Experiments with double-talk scenarios and real audio hardware show more than 8 dB overall echo reduction gain and fast and robust convergence of the polynomial. In addition, an implementation with reduced computational complexity and similar robustness is presented.