Accurate Design-Oriented Modeling of Compact Microwave Couplers in Constrained Domains

Fast replacement models are indispensable in the design of high-frequency structures. Unfortunately, construction of data-driven (approximation) surrogates is hindered by curse of dimensionality and wide range of geometry/material parameters that need to be covered in order to yield practically useful models. Both factors lead to a prohibitively large number of training data points necessary to build reliable models. Here, a technique for defining a surrogate model domain spanned by triangulated reference designs optimized w.r.t. selected figures of interest is described. Our approach allows for dramatic reduction of the (volume-wise) size of the design space that need to be samples without formally reducing the geometry parameter ranges. As a result, reliable surrogate can be constructed using a practically acceptable number of data samples. The proposed approach is demonstrated using a compact rat-race coupler modeled for operating frequencies from 1 GHz to 2 GHz and power split of -4dB to 0dB. Benchmarking and application examples for coupler design optimization are also discussed.