Design of finite word length FIR digital filter using a parallel genetic algorithm

The application of a parallel genetic algorithm implemented with a Hypercube computer to design optimal finite-wordlength finite-impulse-response digital filters is described. Two examples of this application to the design of a Chebyshev low-pass digital filter are given. One example, describing a filter with a length of 40 coefficients and 10-b wordlength, is used for comparison to a previously reported identical design problem which was solved with a general-purpose integer programming algorithm. This comparison reveals that the genetic algorithm produces a better Chebyshev approximation to the desired frequency response. The second example describes the results of the application of the genetic algorithm to the design of a high-order filter developed for the processing of seismic signals. The characteristics of this second filter with 140 coefficients and a 16-b wordlength are compared to those obtained by a rounded-off method.<>