An empirical weight update approach for nonlinear active noise control

Nonlinear effects, such as saturation and distortion of microphones and loudspeakers, are often experienced in complex Active Noise Control (ANC) problem settings. A nonlinear control filter must be employed to adequately address these issues, together with an ad hoc adaptive algorithm to tune its parameters. The nonlinear ANC (NANC) problem is particularly involved if the secondary path is nonlinear, since in that case the controller weight update mechanism of Least Mean Squares (LMS)-type algorithms must account for the secondary path gradient with computationally heavy recursions. A simpler and computationally less demanding approach is here proposed that avoids the use of the error gradient and relies on direct cost function evaluations. The proposed method can also tackle the model structure selection problem, which is crucial to reduce the on-line computational effort. Some simulations are reported to show the effectiveness of the algorithm.