Novel and Efficient Download Test for Two Path Echo Canceller

The two-path echo cancellation technique is a popular method for handling the double-talk problem in acoustic and line echo cancellation applications. The method uses two filters. A so-called background adaptive filter adapts its coefficients to predict echo all or most of the time regardless of signal activity on the near-end. A second foreground filter, that also predicts the echo, receives it coefficients from the background filter, but only when the background is performing better than the foreground. Only the foreground residual echo is sent to the far-end, so any background divergence due to double-talk is not observed by the user. The key to good two-path performance is in the definition of the background-to-foreground coefficient download tests. These typically contain a suite of various measures that attempt to ascertain the convergence state of the two filters. In this paper we present a novel, simple statistic that directly and accurately estimates a filter's convergence state. With the aid of this new statistic we show significant improvement in the overall performance of the two path echo canceller.