The increasing utilization of millimeter-wave spectrum in 5G New Radio (NR) has stimulated the development of antennas supporting multiple FR2 frequency bands, particularly at 28 and 39 GHz. However, conventional electromagnetic (EM) simulation-based design optimization is computationally intensive when a large number of antenna geometric configurations are evaluated. This paper presents a data-driven methodology for predicting the reflection coefficient (S11) and optimizing a dual-band semicircular microstrip antenna. An opposed-semicircular radiator is investigated, and a dataset comprising 16,000 antenna configurations is generated using CST Studio Suite by varying the principal geometrical parameters over the 20–50 GHz frequency range. Six regression-based ML algorithms are comparatively assessed to establish a surrogate model for S11 prediction. Artificial Neural Network (ANN) exhibits the best predictive performance, with a Mean Squared Error (MSE) of 0.0948 dB and an R-squared value of 0.9957. Based on the predictions provided by ANN model, a dual-band millimeterwave antenna for 5G communications is designed with a planar size of 11 mm x 11 mm on a Rogers RT5880 substrate with a thickness of 0.79 mm. The final design exhibits dual resonances at 28 and 39 GHz, corresponding to the 5G NR FR2 bands around n257/n261 and n260, respectively. The results indicate that the proposed ANN-based surrogate model can effectively support rapid S11 prediction and geometry optimization for dual-band millimeter-wave antenna design.

ML,dual-band antenna,milimeter wave,5G