Data-driven (or approximation) surrogate models have been gaining popularity in many areas of engineering and science, including high-frequency electronics. They are attractive as a way of alleviating the difficulties pertinent to high computational cost of evaluating full-wave electromagnetic (EM) simulation models of microwave, antenna, and integrated photonic components and devices. Carrying out design tasks that involve massive EM simulations, including optimization or uncertainty quantification, might be impractical or simply prohibitive. Shifting the computational burden onto the faster representations can mitigate the problem. However, construction of surrogates for high-frequency components is challenging due their highly nonlinear characteristics and wide ranges of operating conditions the models should be able to cover in order to be practically useful. A recently proposed nested kriging framework offers a remedy to these issues by focusing the modeling process on a small region of the parameter space, which contains designs that are of high quality with respect to the performance figures of interest. A result is considerable reduction of the number of training data samples necessary to build up the surrogate. This paper proposes an enhanced design of experiments scheme for nested modeling which further improves predictive power of the surrogate as demonstrated using a dual-band dipole antenna and a miniaturized impedance matching transformer. The accuracy improvement is up to 20 percent for the models constructed within wide ranges of geometry parameters and operating conditions of the respective structures.
Authors
Additional information
- DOI
- Digital Object Identifier link open in new tab 10.1108/ec-02-2019-0054
- Category
- Publikacja w czasopiśmie
- Type
- artykuł w czasopiśmie wyróżnionym w JCR
- Language
- angielski
- Publication year
- 2019
