An indirect model selection algorithm for nonlinear active noise control

Model structure selection is a crucial task in applications where nonlinear black-box models are used, in order to reduce the model size and the associated computational effort. One such application is Active Noise Control (ANC), where nonlinear effects arise due \emph{e.g.} to saturation and distortion of microphones and loudspeakers. Both parameter estimation and model selection are complex in the general nonlinear case if standard algorithms of the Least Mean Squares (LMS) type are used, due to the inherent difficulties in the gradient calculation when the secondary path is nonlinear. A model selection method is here proposed that employs a gradient-free parameter estimation algorithm to tackle the secondary path issue. A virtualization scheme is used to estimate the performance of the model subject to various different structural modifications, in order to select the most appropriate one to apply to the actual control filter. Some simulation examples are discussed to show the effectiveness of the algorithm.