Adaptive Volterra filters for nonlinear acoustic echo cancellation

More than 20 years ago it was shown that adaptive Volterra filters can be used for telephone echo cancellation, but the inherent need for a large number of coefficients has inhibited their practical use. Recently, first attempts have been made to combat nonlinearities in the acoustic echo path of hands-free telephones, which are caused by low-cost audio components. As these attempts are based on a memoryless nonlinear model, they are constrained to saturation effects, which can not describe the nonlinear behavior of small loudspeakers accurately enough. In this contribution, we report real time measurements with an adaptive Volterra filter. At first, we verify that loudspeaker nonlinearities encountered in the echo path of hands-free telephones have to be modelled with memory. Based on these results, we propose an adaptive Volterra filter structure with a reduced number of coefficients. The new structure is experimentally compared to echo cancellers with a memoryless nonlinear model using a real acoustic echo cancelling setup with a small loudspeaker. With the new technique, an echo reduction improvement of 7 dB over conventional linear adaptive filters is achieved. Furthermore it is shown that the proposed adaptive Volterra filter structure outperforms the memoryless approaches.