Nonlinear Acoustic Echo Cancellation Based on Piecewise Linear Approximation with Amplitude Threshold Decomposition

Abstract:
We propose a new nonlinear acoustic echo cancellation scheme for the cascade echo path model of memoryless nonlinearity and linear filter. The scheme has two features. First, a class of piecewise linear (PWL) approximation based on amplitude threshold decomposition is applied to the memoryless nonlinearity. Unlike other PWL classes conventionally used for this problem, the threshold decomposition-based PWL class enables more intuitive implementation with a compact parameter set. The second is a new joint adaptive algorithm, which is normalized by taking into account the difference in the numbers of the nonlinear and linear model parameters. The simulation results show that the proposed scheme more stably cancels the nonlinear echo than the conventional cascade scheme, even when the numbers of parameters are changed. Index Terms — Nonlinear acoustic echo cancellation, memoryless nonlinearity, piecewise linear approximation, adaptive filtering, normalized LMS.