This thesis presents some solutions to non-linear acoustic echo cancellation using loudspeaker pre-processing. We first investigate an experimental and theoretical analysis of linear acoustic echo cancellation behaviour in non-linear environment. The analysis shows that linear echo cancellation performance is degraded by the presence of non-linearities. This leads to the requirement of non-linear solutions. In the first step, measurements taken from a real mobile phone confirm that the loudspeaker is the major source of non-linearities. The same measurements are then used to characterise and model a loudspeaker. In the second step, the loudspeaker model is used in two different structures for loudspeaker pre-processing to improve linear acoustic echo cancellation performance. Two new algorithms are then proposed. In the first structure the pre-processor is placed before the linear acoustic echo canceller. It aims to emulate loudspeaker behaviour so that non-linearities are taken into account by the linear echo canceller. An improved version of this structure is then proposed. It combines clipping compensation in the pre-processor and decorrelation filtering in the linear acoustic echo canceller. The objective is to improve the convergence and the tracking capability of the linear acoustic echo canceller. In the second structure the pre-processor is placed before the loudspeaker. In this case the pre-processor aims to linearise the loudspeaker output which results in more efficient linear acoustic echo cancellation.
Finally, we propose a comparative analysis of the different non-linear acoustic echo cancellers.
