On low-fidelity models for variable-fidelity simulation-driven design optimization of compact wideband antennas

The paper addresses simulation-driven design optimization of compact antennas involving variable-fidelity electromagnetic (EM) simulation models. Comprehensive investigations are carried out concerning selection of the coarse model discretization density. The effects of the low-fidelity model setup on the reliability and computational complexity of the optimization process are determined using a benchmark set of three ultra-wideband antennas designed for the best matching and for the minimum size. The optimization algorithm of choice is trust-region gradient search with low-fidelity model correction realized through frequency scaling and multiplicative output space mapping. The results indicate that appropriate low-fidelity model setup can be assessed using a correlation analysis based on the selected characteristic points of the antenna responses and sparsely allocated test designs, and the results are consistent across the considered benchmark set.