Accurate Modeling of Antenna Structures by Means of Domain Confinement and Gradient-Enhanced Kriging

Abstract

High cost of electromagnetic (EM)-simulation-based antenna design fosters the incorporation of surrogate modeling techniques as the most promising tools for expediting the procedures such as parametric optimization or uncertainty quantification. Data-driven metamodels are by far the most popular ones due to their versatility and easy handling even by inexperienced users. Notwithstanding, construction of reliable surrogates for contemporary antenna structures is, in a large part, hindered by the curse of dimensionality and nonlinear relationships between the geometry parameters and antenna characteristics. A possible workaround are performance-driven modeling methods. The idea it to restrict the construction of the surrogate to small parts of the parameter space, containing designs that are of high quality with respect to the relevant figures of interest. This leads to a significant reduction of the required number of training samples, the improvement of the model predictive power, and a possibility of representing antenna outputs over broad ranges of operating conditions. The model domain is established using the sets of so-called reference designs that need to be pre-optimized beforehand. The CPU cost of this acquisition adds to the overall expenses of surrogate model setup and may undermine the computational benefits of the performance-driven modeling paradigm. This paper proposes an alternative approach, where gradient-enhanced kriging is employed to reduce the number of required reference points, thus leading to lowering the cost of the model setup without compromising its accuracy. A quasi-Yagi antenna example is provided for demonstration purposes.