Genetic Algorithms for Digital

Recursive digital lters are potentially less computationally expensive than their non-recursive counterparts. However, algorithms for adjusting the coeecients of recursive lters may produce biased or sub-optimal estimates of the optimal coeecent values. In addition, recursive lters may become unstable if the adaptive algorithm updates a feedback coeecient so that one of the poles remains outside the unit circle for any length of time. This paper details an adaptive algorithm for optimizing the coeecients of recursive digital lters based on the genetic algorithm. Stability considerations are addressed by implementing the population of adaptive lters as lattice structures which allows the entire feasible, stable coeecient space to be searched whilst ensuring that crossover and mutation do not produce invalid (unstable) lters. Results are presented showing the application of this technique to the tasks of system identiication and adaptive data equalization.