Analysis of nonlinear residual echo suppressors for telecommunications [voice communication applications]

This paper analyzes the performance of nonlinear processors (NLP) that are commonly used for attenuating residual echoes in adaptive echo cancellers for voice communications. The NLP (or residual echo suppressor) implemented as a center clipper produces undesired nonlinear distortions. In this paper, the center-clipping effects are analyzed using Fourier analysis, which decomposes the clipped signal as the summation of a set of rectangular waves and a pulse-amplitude modulated signal. The theoretical analysis shows that the clipping of a sinusoidal signal produces extra odd harmonics, which can be reduced by using the uncompressed center clippers. The analysis results can be extended to speech signals since they can be modeled as the summation of multiple sinusoidal components.