Analysis of a Machine Learning-Assisted Perturbed Patch Antenna with Circular DGS for 6G

A compact high-frequency patch antenna with a novel flower-shaped radiating element and a circular DGS (Defected Ground Structure) is proposed for 6G and satellite communication. The radiating structure comprises two orthogonal biconvex patches with boundary-to-boundary separation equal to the guided wavelength at 15 GHz. The design attains a peak gain of 8.81 dB, a return loss of –34.388 dB, and an impedance bandwidth of 6.6 GHz, validated by fabrication on an FR4 substrate and full-wave simulations using Ansys HFSS. A mathematical derivation based on Bessel's function accurately predicts resonant frequencies, aligning closely with simulation results. In addition, a machine learning framework is developed to model S-parameter behaviour, with the Random Forest Regressor demonstrating the highest predictive accuracy (R² = 0.9901). The combination of electromagnetic simulation, analytical modelling, and data-driven prediction provides a robust design methodology for next-generation compact antennas.