A Double-talk Robust and Fast Converging Echo Canceller

In the application of network communication, echo cancellers with faster convergence and more robustness to double-talk are highly expected. The robustness can be realized by an echo canceller with two echo path models to avoid false adaptation problem under double-talk and algorithms based on PNLMS (proportionate normalized least mean square), which were reported recently, are ones to achieve more rapid convergence. In this paper a new echo canceller with two echo path models is presented, which will choose appropriate algorithm between PNLMS and NLMS to update the background filter coefficients according to the difference of two echo path models. This echo canceller can take advantage of both PNLMS and NLMS so that it converges as fast as PNLMS and shows better performance after fast initial convergence. In our experiments and applications, better performance during double-talk is obtained with proposed echo canceller with two echo path models compared with conventional one. The real-time realization of the new echo canceller is implemented on a fixed-point DSP from ADI