A Multiband Structure based on Hammerstein Model for Nonlinear Audio System Identification

In recent years, the increasing computational power has enabled the large-scale use of techniques for modeling real world systems and their non-linear characteristics. Hammerstein model is a well known structure for non-linear identification of memoryless systems. The proposed work aims at showing two innovative approaches based on multiband structure adaptive filters for Hammerstein adaptive identification process. In particular two multiband structures have been developed using the discrete cosine transform and the wavelet transform for the filterbank construction. The proposed approaches have been compared with a fullband normalized least mean square (NLMS) Hammerstein model, taking into consideration synthetic and real world audio devices.