Simulation-driven design of planar filters using response surface approximations and space mapping

A robust approach for low-cost design optimization of planar microwave filters is presented. Our technique exploits response surface approximation (RSA) models of filter cells as well as space mapping (SM) as an optimization engine. The RSA models are constructed beforehand using sampled EM-simulation data of the cells of interest. The filter structure of a required order is subsequently assembled by cascading selected models. The process of adjusting the geometry parameters of the cells is realized by means of surrogate-based optimization with the combined RSA model as an underlying low-fidelity model, which is corrected using input SM and frequency scaling. The design optimization cost is very low and corresponds to a few EM simulations of the entire filter structure. The proposed technique is illustrated using several examples of bandpass filters constructed using three types of basic cells: stubs, transmission lines, and stepped impedance resonators.