Excitation-dependent stepsize control of adaptive volterra filters for acoustic echo cancellation

This paper proposes a new updating technique for adaptive Volterra kernels employed as nonlinear models for the identification of unknown feedback paths in acoustic nonlinear echo cancellation tasks. Considering that nonlinear distortions are mainly introduced for high input levels, the effective step size is shaped according to the instantaneous envelope of the excitation signal. The loudspeaker-enclosure-microphone setup is modeled using measured linear and quadratic Volterra kernels. The efficiency of the proposed method is compared to the one of the second-order Volterra filter that uses only the Normalized Least-Mean-Square algorithm for kernel adaptation in terms of Echo Return Loss Enhancement. Simulations for different input signals show a superior behavior of the new approach in terms of convergence speed, considering the same steady-state error.