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

作     者：Koziel, Slawomir Pietrenko-Dabrowska, Anna 

作者机构：Reykjavik Univ Engn Optimizat & Modeling Ctr IS-101 Reykjavik Iceland Gdansk Univ Technol Fac Elect Telecommun & Informat PL-80233 Gdansk Poland 

出 版 物：《SCIENTIFIC REPORTS》 (Sci. Rep.)

年 卷 期：2025年第15卷第1期

页      面：1-21页

核心收录：

基　　金：Icelandic Centre for Research Dassault Systemes, France Icelandic Research Fund [2022/47/B/ST7/00072] National Science Centre of Poland 

主　　题：Antenna design Behavioral modeling Global sensitivity analysis Spectral analysis Dimensionality reduction EM-driven design 

摘      要：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.