On computationally-efficient reference design acquisition for reduced-cost constrained modeling and re-design of compact microwave passives

Abstract

Full-wave electromagnetic (EM) analysis has been playing a major role in the design of microwave components for the last few decades. In particular, EM tools allow for accurate evaluation of electrical performance of miniaturized structures where strong cross-coupling effects cannot be adequately quantified using equivalent network models. However, EM-based design procedures (parametric optimization, statistical analysis) generate considerable computational expenses. These can be mitigated using fast surrogate models, yet their construction is hindered by the curse of dimensionality but also the utility requirements: a practically useful model needs to cover sufficiently broad ranges of geometry/material parameters as well as operating conditions. The recently proposed constrained modeling methods—both forward and inverse—work around the above issues by setting up the surrogate only in the relevant regions of the parameter space, i.e., containing designs that are of high quality with respect to the assumed performance measures. The model domain is established using pre-optimized sets of reference points. The high cost of generating such designs may significantly diminish the computational savings achieved by operating in confined domains. This article discusses a technique for fast reference design acquisition, involving inverse gradients, and expedited local refinement aided by the response feature technology. The presented approach is validated using a branch-line coupler and miniaturized rat-race coupler. It is also demonstrated to considerably reduce the cost of constructing performance-driven surrogates as well as setting up efficient procedures for fast geometry scaling of microwave components.