Abstract:
In teleconferencing, the introduction of the acoustic echo canceller by using an adaptive transversal filter is the most effective technique of controlling the acoustic echoes. However, the requirement of a high order filter for modeling the long acoustic echo impulse responses results in difficulties in convergence and hardware implementation. This problem becomes even worse in multiple microphone systems. A Partial Adaptive Process (PAP) is developed and studied in this thesis. This method utilizes the exponential decay characteristics of the echo impulse response; therefore, the significant portion of the echo impulse response is modelled by a smaller size filter. Analyses and simulations have been made to estimate the performance of this finite length filter in the steady-state. Results show that the echo canceller will have smaller residual error for speech signal than white noise. Acoustic theory is used to effectively demonstrate how the minimum mean squared error is affected by the dynamically changing impulse response and the effect of using a finite filter order. Results indicate that a large filter will slow down the convergence rate and increase the system error, thus preventing further system performance in acoustic echo cancellation. The small order filter using the PAP method has been shown to be effective. Other filter structures and algorithms in complement to the PAP are introduced. Their improvement in terms of performance and computational savings are discussed. Finally, a new adaptive notch filter is developed and is applied to control the howling phenomenon found in the speakerphone systems.
