Machine learning-assisted prediction of scattering parameters for a circularly polarized dual-port UWB MIMO antenna

• ML-assisted design keeps ECC < 0.05 and enables real-time tuning for adaptive UWB MIMO links. • Gradient-boosting model predicts S-parameters, cutting full-wave simulation time. • Compact CP UWB MIMO DRA with a shared U-shaped resonator and DGS for high isolation. • Port isolation better than –41 dB via orthogonal fields and C-shaped DR optimization. • 123 % impedance bandwidth and 1.9 GHz axial-ratio bandwidth (6.75–8.65 GHz). Accurate prediction of scattering parameters (S-parameters) is essential in modern antenna design to ensure optimal impedance matching, minimize mutual coupling, and enable efficient integration with RF transceiver systems. However, traditional full-wave electromagnetic simulations are time-consuming and computationally intensive, particularly for complex and reconfigurable multi-port antennas. To overcome this challenge, this work incorporates a machine learning-based approach—specifically, a gradient boosting regression model—to predict the antenna's S-parameters directly from simulation data. This method significantly reduces design time and provides efficient tuning of impedance in wireless communication systems. This study presents a compact, circularly polarized (CP), dual-port MIMO dielectric resonator antenna (DRA) designed for UWB applications. The antenna employs a U-shaped dielectric resonator excited by two microstrip lines and a shared radiator decoupled using a defected ground structure (DGS), achieving exceptional isolation of up to 41 dB without additional decoupling elements. Circular polarization is enabled by orthogonal electric fields generated through horizontal and vertical slots in the ground plane. Further optimization is achieved by reshaping the rectangular DR into a U-shaped structure, enhancing both isolation and 3-dB axial ratio bandwidth. A fabricated prototype confirms an impedance bandwidth of 123% and an axial ratio bandwidth of 1.9 GHz (6.75–8.65 GHz). Diversity performance attributes including an envelope correlation coefficient (ECC) below 0.05 demonstrate suitability for UWB MIMO systems. The integrated ML model effectively predicts the antenna’s S-parameters, supporting fast and accurate performance evaluation.