Efficiency of evolutionary search for analog filter synthesis

Design automation presents a trade off between: using expert knowledge to restrict the possible solutions examined; or spending time searching through many, possibly ineffective, solutions. Incorrect assumptions or misapplication of constraints can miss otherwise superior results. Automated search of possible solutions has been shown effective, limited by the available computational resources. This paper presents an evolutionary search method for finding optimized circuit topologies and component values. An example analog filter problem is exhaustively enumerated, to test the efficiency of different stochastic search methods in finding globally optimal solutions. Evolutionary methods are shown to be efficient for this problem. Impacts of varying parameters and techniques of evolutionary search are compared for a more complex asymmetric bandpass filter problem. Genetic Algorithm and Evolutionary Strategy methods are found to have similar performance. Hybrid evolutionary methods using Differential Evolution for component value optimization are found to be more efficient with limited: component count; or computational resources. The hybrid method used is shown to scale to more complex problems without changing parameters. Proposed comparison metrics, normalized for experimental variables, show the efficiency of this work improves upon published benchmarks. This is achieved without restricting the synthesized topologies to known structures, producing novel results more effective than prior works. •Circuit topology presents a challenging optimization problem.•Expert knowledge improves efficiency but can limit effectiveness of search methods.•Existing search methods have lacked comparisons of efficiency.•State of the art is advanced by critical selection of genetic algorithm parameters.