VOIP Acoustic Echo Cancellation Performance Improvement Using a Novel Time-Varying Step Size of NLMS Algorithm

الملخص

When multiple people participate in an online meeting, because the sound is in the space Factors such as medium transmission or reflection make the sound emitted by the speaker It is taken back by the microphone again, causing echo. if there is When a user's device generates an echo, the echo signal will Spread among participants, affecting all users ring. The echo delay time is difficult to predict, and because network transmission and Factors such as sound distortion cause the traditional echo cancellation mechanism to frequently fails in multi-person network chats. In VoIP phones, acoustic echo cancellers are generally used to cancel Echo generated during calls to improve voice quality.  adaptive filter and double-ended call detection filter. Adaptive filters eliminate echoes by adaptively modeling the echo path. sound, and its performance determines the effect of echo cancellation. In practical applications, adaptive. The most commonly used filter is the NLMS algorithm and its various improvements. algorithm, the reason is that the NLMS algorithm is simple and has low complexity. However, NLMS The algorithm has a fatal weakness, that is, when the input signal is like speech, etc. When there are signals with strong correlation, the convergence speed of the algorithm will be significantly reduced. And affect the quality of echo cancellation, effectively improving the convergence speed of the NLMS algorithm it is based on the instantaneous energy of the input signal the calculation step size will inevitably bring about random fluctuations in the steady-state offset error, thus Will affect the performance of the algorithm.

 This paper proposed a modified of NLMS method to the decorrelation significantly reduces the updating the decorrelation filter coefficients and through the step size determination machine proposed in this study control, which can effectively distinguish the difference between normal speech and echo. Effectively eliminates echo and exerts silence suppression (Silence Suppression) effect, blocking content that does not contain voice content packets. It is shown that our method can eliminate more than 90% of echoes When utilizing the suggested technique in place of the conventional NLMS algorithm, the ERLE boost is around 11 dB. Means that more robust variant of the algorithms and it exhibits a better balance between simplicity and performance than the conventional NLMS algorithm.