Analysis of filter bank transfer functions

Parallel filter banks are used for realtime signal analysis, e.g. spectrum or speech analysis. The problem to be solved in filter bank realisation is to achieve as high selectivity of individual filters as possible and simultaneously meet the amplitude and phase characteristic specification of the bank. A design method for a bank of analogue filters is presented in the paper. An algorithm for the analysis of the amplitude, phase and group delay characteristics of the bank is given. The applicability of various types of transfer function (Bessel, Scanlan, Coung-Neirynck, Butterworth and Cheby¿chev linear phase approximation) is investigated. To study the other possibilities of realising a filter bank, the so-called modified Scanlan filter has been formed by applying an automated optimisation procedure. This filter retains the performance of the Scanlan filter, but the Q factor of the critical pole pair is reduced. The examples of uniform and constant percentage bandwidth filter banks are presented to illustrate the comparative characteristics of filter banks designed on the basis of these six approximations with 12th-order filters.