Comparison of LMS-based adaptive audio filters for identification
In the field of audio signal processing, logarithmic frequency resolution IIR filters, such as fixed-pole parallel filters and Kautz filters, are often used. These proven structures can efficiently approximate the frequency resolution of hearing, which is a highly desired property in audio applications. In recursive adaptive filtering however, the FIR structure with LMS algorithm is the most common. Since the linear frequency resolution of FIR filters is not well-suited for audio applications, in this paper we explore the possibility of combining the logarithmic frequency resolution IIR filters with the LMS algorithm. To this end the LMS algorithm is applied to fixed-pole parallel and Kautz filters, and the resulting structures are compared in terms of convergence properties.
Authors: Kristóf Horváth (Budapest University of Technology and Economics) and Balázs Bank (Budapest University of Technology and Economics)