Department of Computer Science and Electrical Engineering

Signal Prewhitening Schemes for Detection-Guided LMS Estimates

Vanessa Edward

Abstract:
The use of handsfree communication in cars, computer applications and video conferencing has created a demand for high quality acoustic echo cancellation. In these applications the acoustic channel has typically a long impulse response, and hence the tap length of adaptive FIR filters are particularly long for acoustic echo cancellation. In order to reduce the complexity and also to improve the convergence rate, the method employed in this thesis suggests the detection of the active FIR filter taps. I consider an active tap detection algorithm proposed for white signals.

Typically, however, the signals (voice) in echo cancellation are highly correlated. To make the tap detection algorithm suitable I explore different signal prewhitening schemes. The study shows a definite asymptotic performance increase for a coloured input modelled as an AR process when using the active tap detection and an AR prewhitening scheme. Other simulations and conclusions are based upon the results for the standard LMS algorithm, and both AR and affine projection prewhitening schemes, with and without the zero tap detection algorithm.

Complete Thesis:
thesis.pdf

Conference Paper:
conference_paper.pdf

Additional material:
code images

About the Author

Email: s341346@student.uq.edu.au

Page last updated: 9 October, 1999.