Multilayer perceptron structures applied to adaptive echo canceller for ISDN subscriber loops

Echo cancellation is the main method used to realize two wire full-duplex transmission on ISDN subscriber loops. In the data echo canceller there exists nonlinear echo distortion which comes mainly from the A/D or D/A convertor. A conventional FIR filter is not sufficient to cancel this nonlinear distortion. In this paper a nonlinear echo canceller using a multilayer neural network is described. The coefficients of the multilayer perceptron are adaptively adjusted by a back-propagation algorithm. The computer simulation shows that by using a neural echo canceller nonlinear echo distortion can be effectively cancelled. The performance of the neural echo canceller is better than that of the conventional FIR echo canceller.<>