Machine learning-assisted performance prediction of graphene-silicon twin-port band-notched wideband antenna for THz 6G communication systems.

A graphene-silicon-based twin-port terahertz (THz) antenna is proposed and investigated in this work. The antenna employs an aperture-coupled asymmetric ring dielectric resonator configuration to achieve wideband operation in the THz regime. Circular metallic rings integrated with the printed feed structure introduce a band-notch characteristic between 2.75 THz and 3.15 THz for interference suppression. A graphene coating is incorporated to provide frequency tunability, while DGS is utilized to reduce mutual coupling between the antenna ports. The proposed antenna operates efficiently over the 2.2-2.65 THz and 3.3-3.7 THz frequency bands with isolation levels below - 25 dB. The antenna also exhibits stable radiation characteristics, low envelope correlation coefficient (ECC), high diversity gain (DG), low channel capacity loss (CCL), and acceptable total active reflection coefficient (TARC), confirming its suitability for THz MIMO communication systems. Furthermore, deep neural network (DNN) and random forest (RF) ML approaches are employed to predict the |S11| characteristics of the antenna using a large parametric dataset generated through full-wave electromagnetic simulations. The ML models are evaluated using MAE, MSE, RMSE, Variance Score, and R² metrics, demonstrating strong agreement between predicted and simulated results. The proposed antenna provides a compact, tunable, and ML-assisted THz communication solution suitable for future intelligent 6G wireless applications.