Open AccessVariable‐fidelity response feature surrogates for accelerated statistical analysis and yield estimation of compact microwave components
Author(s) -
Koziel Slawomir,
Bekasiewicz Adrian
Publication year - 2019
Publication title -
iet microwaves, antennas and propagation
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.555
H-Index - 69
eISSN - 1751-8733
pISSN - 1751-8725
DOI - 10.1049/iet-map.2019.0065
Subject(s) - fidelity , feature (linguistics) , microwave , variable (mathematics) , monte carlo method , algorithm , high fidelity , reliability (semiconductor) , computer science , process (computing) , electronic engineering , reliability engineering , mathematics , engineering , statistics , power (physics) , physics , mathematical analysis , telecommunications , philosophy , linguistics , quantum mechanics , electrical engineering , operating system
Accounting for manufacturing tolerances is an essential part of a reliable microwave design process. Yet, quantification of geometry and/or material parameter uncertainties is challenging at the level of full‐wave electromagnetic (EM) simulation models. This is due to inherently high cost of EM analysis and massive simulations necessary to conduct the statistical analysis. Here, a low‐cost and accurate yield estimation procedure for compact microwave couplers is proposed. Authors’ technique involves variable‐fidelity electromagnetic (EM) simulation models, as well as fast surrogates constructed using a response feature approach. In order to improve the computational efficiency of the analysis, the primary surrogate is obtained from the characteristic points of the low‐fidelity model and, subsequently, corrected using a single evaluation of the high‐fidelity model. Combination of both methods results in an extremely low cost of yield estimation being just a few high‐fidelity EM analyses. For the sake of demonstration, a compact hybrid rat‐race coupler operating at 1 GHz is considered. Yield estimation is carried out under several scenarios concerning various probability distributions of the geometry parameter deviations. Reliability of the approach is verified by comparing the results with direct Monte–Carlo analysis and single‐fidelity feature‐based yield estimation.