Convergence of an adaptive echo cancellation system with an augmented predictor

The primary input to an adaptive filter often contains a component unrelated to the reference input or the parameters being estimated. Since this component interferes with error measurement, we introduce two new algorithms that remove it by linear prediction. These are the least mean square algorithm with augmented predictor and the sign algorithm with augmented predictor. Describing these methods in the context of echo cancellation, we present their theoretical learning curves and a simulation example that illustrates the improvement they can give.