Robust Design of Antenna Structures by Means of Domain-Confined Metamodels

Statistical analysis is of practical importance for high-frequency engineering when it comes to estimating the effects of manufacturing tolerances and other types of uncertainties on system properties. One of its bottlenecks, in the case of antenna structures, is a high computational cost of full-wave electromagnetic (EM) analysis normally utilized for reliable performance evaluation. This paper capitalizes on the recently introduced concept of surrogate modeling in constrained domains and delivers a simple yet computationally-efficient procedure for statistical analysis and yield optimization of multi-band antennas. The foundation of our methodology is an appropriate definition of the parameter space region selected for a construction of the fast surrogate model replacing EM analysis for the purpose of Monte Carlo simulations. The metamodel domain is spanned by the vectors corresponding to the most significant changes of antenna resonant frequencies. It is characterized by a small volume (as compared to the traditional interval-like domain), which translates to reduced expenses of training data acquisition, and, consequently, the overall cost of the yield optimization process. Our approach is validated using a triple-band dipole antenna with the yield-optimum design identified at the cost of only a few dozens of EM simulations of the antenna structure.