Globalized Parametric Optimization of Microwave Passive Components Using Simplex-Based Surrogates

Optimization-based parameter adjustment involving full-wave electromagnetic (EM) simulation models is a crucial stage of present-day microwave design process. In fact, rigorous optimization is the only reliable mean permitting to simultaneously handle multiple geometry/material parameters, objectives, and constraints. Unfortunately, EM-driven design is a computationally intensive endeavor. While local tuning is usually manageable, direct global search is most often prohibitively costly. At the same time, global optimization is often required: design of compact microwave passives, multimodal problems, and circuit redesign across a broad range of operating conditions are a few examples. A popular mitigation method is to employ a surrogate modeling approach. Still, constructing reliable models is impeded by the curse of dimensionality. This article presents a novel algorithm for globalized design of microwave devices, which employs simplex-based regression models built at the level of operating parameters and performance figures of the circuit of interest. The computational efficiency of our technique stems from topological simplicity of the surrogate and its updating rules, as well as nearly linear dependence between the circuit dimensions and its operational parameters. Furthermore, the updating rules of the regression model guarantee convergence of the optimization process. The global search capability of our technique is validated through repetitive optimization of three microstrip circuits. A perfect success rate is demonstrated along with a low (average) computational cost of just about 80 EM analyses of the circuit at hand. Numerical results are supported by experimental validation of selected designs.