Nonlinear acoustic echo cancellation using a Hammerstein model

In hands-free telephone or video conference application, there exists an acoustic feedback coupling between the loudspeaker and microphone, which creates the acoustic echo. Linear acoustic echo cancellers (AECs) are commonly used to remove this echo. However, they are unable to effectively cancel nonlinear distortions. This paper employs a Hammerstein model to describe the acoustic channel of a nonlinear system concatenated with a linear faded echo path. A feed-forward neural network is used to model the static nonlinearity and a Finite Impulse Response (FIR) structure is used to model the linear dynamic system. The formed nonlinear model is applied to real data collected in an anechoic chamber and it performs slightly better than linear models. Although the improvement is small, the results show some interesting insights on the characteristic of a loudspeaker’s nonlinearities and their effect on the performance of an AEC.