Design and optimization of broadband circular polarized miniaturized UWB antenna using machine learning

This paper presents a broadband circularly polarized (CP) miniaturized ultra-wideband coplanar waveguide-fed antenna. The antenna comprises an asymmetrically fed hexagonal-shaped patch radiator, which is rotated to achieve wide impedance bandwidth. In addition, two inverted L-shaped stubs and an inverted L-shaped slot are embedded with the radiator to achieve CP. It has been observed that the antenna measured -10 dB impedance bandwidth of 112.12% (2.9–10.3 GHz) and 3-dB axial ratio (AR) bandwidth of 33.33% (8–11.2 GHz) with a peak gain of 2.0 dBi. The antenna performance is further optimized in terms of reflection coefficient ($\left|S_{11}\right|$) and AR using various machine learning (ML) models. The outcomes of these models are found to be closely aligned with the simulated results generated from the CST Microwave Studio. Among the ML models, the Ensemble Bagged Tree (EBT) for $\left|S_{11}\right|$ and Trilayered Neural Network (TNN) for AR achieved the lowest error of 0.5402, 0.659 and the highest R-squared score of 0.9935, 0.9956, respectively showcasing their superior generalization performance.