Koziel Slawomir, Pietrenko-Dabrowska Anna, Leifsson Leifur
Engineering Optimization & Modeling Center, Reykjavik University, 101, Reykjavik, Iceland.
Faculty of Electronics, Telecommunications and Informatics, Gdansk University of Technology, 80-233, Gdansk, Poland.
Sci Rep. 2024 Aug 22;14(1):19465. doi: 10.1038/s41598-024-70246-4.
Behavioral models have garnered significant interest in the realm of high-frequency electronics. Their primary function is to substitute costly computational tools, notably electromagnetic (EM) analysis, for repetitive evaluations of the structure under consideration. These evaluations are often necessary for tasks like parameter tuning, statistical analysis, or multi-criterial design. However, constructing reliable surrogate models faces several challenges, including the nonlinearity of circuit characteristics and the vast size of the parameter space, encompassing both dimensionality and design variable ranges. Additionally, ensuring the validity of the model across broad geometry/material parameter and frequency ranges is crucial for its utility in design. The purpose of this paper is to introduce an innovative approach to cost-effective and dependable behavioral modeling of microwave passives. Central to our method is a fast global sensitivity analysis (FGSA) procedure, which is devised to identify correlations between design parameters and quantify their impacts on circuit characteristics. The most significant directions identified through FGSA are utilized to establish a reduced-dimensionality domain. Within this domain, the model may be constructed using a limited amount of data samples while capturing a significant portion of the circuit response variability, rendering it suitable for design purposes. The outstanding predictive capability of the proposed model, its superiority over traditional techniques, and its readiness for design applications are demonstrated through the analysis of three microstrip circuits of diverse characteristics.
行为模型在高频电子领域引起了广泛关注。它们的主要功能是替代昂贵的计算工具，特别是电磁（EM）分析，用于对所考虑结构的重复评估。这些评估对于诸如参数调整、统计分析或多准则设计等任务通常是必要的。然而，构建可靠的替代模型面临着几个挑战，包括电路特性的非线性以及参数空间的巨大规模，这涵盖了维度和设计变量范围。此外，确保模型在广泛的几何形状/材料参数和频率范围内的有效性对于其在设计中的实用性至关重要。本文的目的是介绍一种创新方法，用于微波无源器件的经济高效且可靠的行为建模。我们方法的核心是一种快速全局灵敏度分析（FGSA）程序，该程序旨在识别设计参数之间的相关性并量化它们对电路特性的影响。通过FGSA确定的最显著方向用于建立一个降维域。在这个域内，可以使用有限数量的数据样本构建模型，同时捕获电路响应变化的很大一部分，使其适用于设计目的。通过对三个具有不同特性的微带电路的分析，证明了所提出模型出色的预测能力、相对于传统技术的优越性以及其在设计应用中的适用性。
