An Improved Practical State-Space FDAF with Fast Recovery of Abrupt Echo-Path Changes

Abstract

A variable step-size echo cancellation algorithm in the frequency domain based on soft-decision and status information of the filter is established for enhancing tracking capability. The first-order Markov model is initially used to represent the time-variant acoustic path. In the meantime, the step size of the filter is replaced by the Kalman gain. Afterward, the magnitude squared coherence function of the reference signal and error signal are calculated as the filter status. The status of the filter is then mapped to a soft-decision value and is used to weigh the channel transmission parameter. Finally, the acoustic path is tracked by the weighted channel transmission parameter, which facilitates the reconvergence of the adaptive filter. The theoretical and experimental analysis is demonstrated on the validity of the algorithm. Compared to similar algorithms, the proposed algorithm has less misalignment but can also converge faster during abrupt echo-path changes. Compared to the state-space frequency domain adaptive filters (SS-FDAF) algorithm, the convergence time is 44% shorter. In addition, the amount of misalignment decreases by 8 dB. Moreover, the filter update time only increases by 15.38%.