Application of echo--Cancelling techniques to audioconference

The least mean squares (LMS) and short time spectral analysis (SSA) identification algorithms are considered with respect to their applicability to echo-cancelling in audio-conference systems. The SSA method turns out to be more suitable, expecially from the point of view of the computation complexity, which is strictly associated with the real-time implementability. The performances of this algorithm are analyzed by reproducing the operating conditions that may be actually expected in audioconference rooms. It´s particularly shown that a reasonably fast identification time can be obtained with respect to audioconference requirements, while ensuring to the echo-return path a loss enhancement (ERLE) of up to 20 dB. The effects of echo on stability are also dealt with in detail by calculating the frequency dependence of the incremental loss. It is shown that the stability condition is fulfilled almost at all frequencies, excepting some isolated points corresponding to sharp dips in the frequency response. Anyhow, the loss of the dips itself should likely prevent the system from howling.