Advanced Optimization Techniques for Modern Filter Design – From Newton to Space Mapping

Abstract

In a 1967 review, Temes and Calahan advocated iterative optimization techniques for filter design. They demonstrated gradient and Newton-based methods, the Fletcher-Powell method and methods suitable for Chebyshev approximation. Benchmarks in later years include sequential quadratic programming approaches to constrained optimization, and Hald and Madsen’s powerful algorithms for minimax and L1 optimization. Engineering optimization techniques soon needed to evolve to handle complex circuits, including EM-validated designs. In 1994, Bandler et al. proposed space mapping, and in 1995 the Newton-based aggressive space mapping algorithm, to optimize circuits with EM accuracy in a handful of EM simulations. So-called “surrogate-based” optimization continues to flourish, with novel work in tuning space mapping. This workshop presentation uses successes in filter and multiplexer design to illustrate the current state of the art in such advanced optimization methodologies.