Automating IIR filter design by genetic algorithm

The design of digital IIR filters is a multistage process, involving the optimisation of coefficient values, coefficient wordlengths, structure and section ordering. These are traditionally regarded as separate operations, and, as such, can in general only produce filters which are optimal in certain aspects, but not optimal overall. By exploiting the multiple criterion optimisation abilities of the genetic algorithm, we show that it is possible to perform several of these steps simultaneously. This allows the designer to specify the relative importance of each area of the design, for example, the frequency response or roundoff noise effects, thereby permitting the design of filters from a few initial specifications without requiring detailed knowledge of the individual design steps.