Reliable Modeling of Antenna Input Characteristics by Means of Domain Confinement and Variable-Fidelity EM Simulations

摘要

Electromagnetic (EM) simulation is the single most important tool in the design of modern antenna structures. Its versatility and reliability comes at a price of considerable evaluation cost, which may be impractical or even prohibitive when multiple simulations are to be executed, e.g., pertaining to parameter tuning or uncertainty quantification. Fast replacement models (or surrogates) may mitigate this problem but modeling of antenna characteristics is a challenging endeavor due to their nonlinearity and typically high-dimensionality of the parameter space. Conventional data-driven modeling methods fail under such circumstances. In this paper, a novel technique is presented which combines the nested kriging framework and variable-fidelity computational models. The former is employed to confine the surrogate model domain to a region containing designs that are of high-quality from the point of view of the considered performance figures. This dramatically reduces the number of required training data samples. Incorporation of low- and high-fidelity simulations is realized using co-kriging. The proposed approach is illustrated using a wideband monopole antenna modeled for wide ranges of substrate permittivity (2.0 ≤ εr ≤ 5.0) and height (0.5 mm ≤ h ≤ 1.5 mm). Benchmarking indicates that constrained variable-fidelity model outperforms both the conventional surrogates but also the constrained surrogate constructed using high-fidelity EM data only.