Colloquium - Details

Master-Vortrag: Dekorrelation für die Echokompensation mittels Kalman-Filter

Tobias Hübschen
19. Oktober 2016
14:00 Uhr
Hörsaal 4G IKS

Although Acoustic Echo Control (AEC) has been a research topic for several decades, the amount of ongoing research has not ceased since the system requirements keep changing especially due to the observed trend towards multi-channel acoustic setups. A more recent research result is the formulation of the Kalman filter in the frequency domain as an adaptive algorithm for the estimation of the required acoustic impulse responses. Due to the close relationship between the Kalman filter and the Least Mean Square (LMS) algorithm, the Kalman filter yields a performance equivalent to the performance of the LMS algorithm with optimal stepsize control. Nevertheless, it was shown that the identification performance of the Kalman filter still depends on the correlation properties of the excitation signals. This, consequently, motivates the introduction of decorrelation techniques to Kalman filter based AEC.

This thesis considers the decorrelation by Linear Prediction (LP) filtering technique introduced for Normalized Least Mean Square (NLMS) based AEC. It is shown that the technique may be adapted for an operation in the frequency domain. This fact is used to derive a modified observation matrix to formulate the Kalman filter with LP filtering in closed form. The performance of this algorithm for both mono and stereo applications is simulatively evaluated. Furthermore, a criterion for the determination of the coefficients representing the decorrelation filter is introduced which, in contrast to the LP criterion, explicitly addresses the cross-correlation of the excitation signals to improve the conditioning of the stereo estimation problem.

back