Performance of the variable-length stochastic gradient lattice filter in adaptive noise cancellation of nonstationary signals—A simulation case study

Abstract

The variable-length filter for adaptive noise cancellation of nonstationary signals is introduced. The optimum length of number of filter stages is shown to be chosen in terms of the algorithm noise. It is shown that such an optimal filter can always act positively even when the cross-correlation between the input channels is close to zero. The performance of the variable-length stochastic gradient (SG) lattice filter is theoretically studied for a given nonstationary (time variable) system simulated on a computer. The results of the simulation are then compared with theory. It is shown that, for nonstationary signals, the variable-length filter yields a better noise reduction than the conventional LMS (Widrow et al., 1976) and lattice filters.