| 摘要 |
An echo cancellation unit includes a model store to store a current echo model from an adaptive filter when a real-time error occurs. The real-time error is detected by monitoring a convergence metric. In some embodiments, the convergence metric is echo return loss enhancement (ERLE). When a real-time error occurs, the current echo model is saved, and the adaptive filter is reset such that it will begin converging from the origin. As a new model emerges in the adaptive filter, it is compared to the saved model in the model store at several time lags. If a match is found, the emerging model is replaced with the saved model at the appropriate time lag. The result is faster convergence of the adaptive filter than if the adaptive filter were left to converge on its own.
|
