Associating Nonlinear Effects in Nlms Adaptation with Dynamic Weight Behavior

Nonlinear effects, or non-Wiener behavior, have been observed in several adaptive filtering applications, such as adaptive equalization of wideband communication signals contaminated by narrowband interference, adaptive narrowband noise cancellation, and adaptive linear prediction of narrowband processes. The nonlinear effects have been especially prominent when narrowband processes have been involved, and revealed by performance better than expected from the conventional Wiener filter for that scenario. Furthermore, in the adaptive noise-canceling scenario, almost deterministic semi-periodic dynamic behavior of the adaptive filter weights has been observed. In the adaptive equalization and prediction scenarios the weight behavior is also dynamic, albeit of a different nature. The dynamic weight behavior associated with nonlinear effects is shown to be the result of how the conventional NLMS adaptive filter tracks the equivalent model for the desired signal.