The application of the median LMS algorithm to ADPCM systems

The use of the median least-mean-square (MLMS) adaptive filter in adaptive differential pulse-code modulation (ADPCM) systems is examined. The MLMS is a modification of the LMS algorithm designed to protect the adaptive filter coefficients from the effects of impulsive noise. A modified MLMS-ADPCM system is proposed in which the MLMS replaces the LMS in the predictor updates. Objective SNR-based performance measures are used to compare MLMS with LMS in the context of a 32 kb/s system. It is shown that for ideal channels the performance of the MLMS-ADPCM is comparable to that of the conventional LMS-ADPCM. Using independent bit errors, it is shown that when the channel is degraded, MLMS outperforms LMS and appears to obviate the need for coefficient leakage in the predictor update