Cost-efficient behavioral modeling of antennas by means of global sensitivity analysis and dimensionality reduction

Abstract

Abstract Computational tools, particularly electromagnetic (EM) solvers, are now commonplace in antenna design. While ensuring reliability, EM simulations are time-consuming, leading to high costs associated with EM-driven procedures like parametric optimization or statistical design. Various techniques have been developed to address this issue, with surrogate modeling methods garnering particular attention due to their potential advantages. One key benefit is the promise of unprecedented acceleration in handling design problems that require repetitive system evaluations. However, behavioral modeling of antennas is an intrinsic endeavor. Challenges include the curse of dimensionality and the high nonlinearity of antenna characteristics. Moreover, design utility necessitates that the models are defined across wide ranges of frequency, geometry dimensions, and material parameters, posing a significant bottleneck for existing modeling frameworks. This paper introduces an innovative approach to constructing design-ready behavioral surrogates for antenna structures. Our methodology involves a rapid global sensitivity analysis (GSA) algorithm developed to determine a set of parameter space directions that maximize antenna response variability. The latter are obtained from spectral analysis of the GSA-based sensitivity indicators, and employed to define a reduced-dimensionality domain of the metamodel. The dependability of the model constructed in such a domain is superior over conventional surrogates while being suitable for design purposes. These benefits have been conclusively showcased using several microstrip antennas and illustrated by a number of design scenarios involving antenna geometry optimization for a variety of performance specifications.