Acoustic echo cancellation for wideband VoIP

Acoustic echo cancellation is investigated in a wideband voice-over-IP (VoIP) framework. Simulations using fullband and subband adaptive filtering algorithms show that subband adaptive filters are an effective solution. Low residual echo levels observed in oversampled subband echo cancellers are attributed to the adaptive filters exploiting the subband signal correlation to lower the output error. In simulated changing acoustic environments the minimum echo return loss enhancement (ERLE) of the subband Normalised LMS (NLMS) is over 3 dB higher than fullband; the result is verified using experimental data. In the presence of narrowband near-end disturbances with an active doubletalk detector some bands in a subband system can continue to adapt resulting in deeper convergence and higher ERLE during doubletalk. Nonlinear distortion created by wideband VoIP vocoders can degrade the performance of linear echo cancellers by 5--10 dB. The fast tracking IP Affine Projection Algorithm (IP-APA) is shown to perform the best in the distorted channel, the differences between subband and fullband structures are not significant. Techniques from Fast Affine Projection algorithms are used to reduce the complexity of the IP Affine Projection Algorithm, making it more applicable for wideband acoustic echo cancellation.