Dual-band circularly polarized dielectric resonator antenna with wide impedance bandwidth and machine learning-assisted performance prediction

• Dual-band circular polarization: The proposed antenna successfully achieves left-hand circular polarization (LHCP) in two well-separated frequency bands, unlike several earlier designs that either support a single CP band or linear polarization. • Wide axial ratio bandwidth (ARBW): The ARBW values of 60.4 % and 17.4 % at the two operating bands demonstrate strong polarization stability and are notably wider than most previously reported dual-band CP DRAs. • Broad impedance bandwidth (IBW): With an impedance bandwidth of 111.1 %, the proposed design maintains stable matching across a wide frequency range, outperforming many conventional DRA configurations. • High gain and radiation efficiency: The antenna achieves a peak gain of 12.8 dBic with 95.07 % radiation efficiency, indicating efficient energy conversion and low dielectric and conductor losses. • Stable polarization purity: The measured LHCP–RHCP discrimination exceeds 20 dB at both resonant bands, confirming excellent polarization purity and minimal cross-polarization leakage. • Compact and fabrication-friendly structure: The use of a stacked DRA with an elliptical defected ground structure (DGS) provides a wideband CP response without complex feed networks or parasitic elements, ensuring ease of fabrication and reproducibility. This paper presents a compact dual-band dielectric resonator antenna (DRA) with circular polarization reconfigurability, optimized for UAV platforms, satellite communications, and emerging 6G networks. An elliptical ground slot enhances impedance bandwidth, while an asymmetric feeding structure excites orthogonal modes for robust circular polarization. The design achieves 111.1 % impedance bandwidth, dual axial ratio bandwidths of 60.4 % and 17.4 %, a peak gain of 12.8 dBic, and 95.07 % radiation efficiency. Alongside full-wave simulations, machine learning (ML) models—Artificial Neural Networks (ANN), XGBoost, and Random Forest—are employed to predict S11 and AR across 3–15 GHz. The models attain high accuracy (R2 > 0.93 for S11, R2 > 0.96 for AR) with low error values, validating their ability to capture complex electromagnetic behavior. This hybrid design and ML-assisted approach accelerate optimization, reduce simulation time, and provide reliable predictive capability. The proposed DRA’s combination of wideband performance, polarization agility, and ML integration makes it a strong candidate for compact, high-performance communication systems in UAVs, satellites, and next-generation wireless infrastructure.