Digital Filter Design via Recurrent Cartesian Genetic Programming
Edward Ly, Julián Villegas · 2023
We introduce a method for the automatic program induction of Single-Input/Single-Output (SISO) Infinite Impulse Response (IIR) filters for Digital Signal Processing (DSP) applications. Recurrent Cartesian Genetic Programming (RCGP) evolves a population of DSP programs, represented as directed cyclic/acyclic graphs, to generate a filter whose magnitude response approximates that of a target filter. The Log-Spectral Distance (LSD) is used as a fitness measure to minimize the differences between the magnitude responses of these filters, and the filter with the smallest distance is output. We evaluated our method by generating a number of filters, and found that the accuracy of the generated filters depends on both the order of the target filter and the parameter values set for the RCGP algorithm.