Design and machine learning-based optimization of a graphene-driven funnel shaped THz MIMO antenna for 6G applications

This research study investigates several techniques, such as simulation and an RLC equivalent circuit model, to evaluate antenna performance. The key novelty of this work lies in the integration of supervised machine learning–assisted optimization with a graphene-based THz MIMO antenna, enabling rapid performance prediction and validation while achieving a rare combination of wide bandwidth, high gain, high efficiency, and excellent MIMO isolation. The manuscript evolves through a systematic design process, progressively optimizing the THz antenna’s impedance matching, bandwidth, and radiation efficiency. The design transitions from a basic structure to an advanced configuration with strategic slotting and decoupling, ultimately achieving superior performance for MIMO applications. The design process begins with a single-element graphene patch on a low-loss quartz substrate, which is geometrically evolved through iterative slotting, including a central ground-symbol slot and box-bracket slots, to achieve an wide impedance bandwidth. After that, this single element is then configured into a two-port MIMO system in a side-by-side (0°) arrangement with compact dimensions of 240.02 × 125.556 μm². The proposed MIMO THz antenna offers wideband operation (5.00–9.48 THz), high gain (15.94 dB), and excellent efficiency (92.69%), making it ideal for 6G and THz communication. The proposed MIMO THz antenna employs a graphene wall strategically placed between the radiating elements to reduce mutual coupling. This decoupling structure enhances isolation and ensures efficient independent operation of the MIMO ports, improving overall performance for next-generation THz communication systems. This graphene wall-assisted decoupling mechanism effectively suppresses surface-wave coupling and is further supported by a validated RLC equivalent circuit model, providing physical insight into the antenna behavior. The proposed MIMO THz antenna demonstrates excellent performance with an ECC below 0.000064, a DG of 9.9997, a CCL under 0.31 bps/Hz, and a TARC below − 8 dB. Supported by machine learning, the Extra Trees Regressor achieves 97.76% accuracy in predicting antenna gain. With its wide bandwidth, high gain, efficiency, and superior isolation, this antenna is ideal for next-generation high-speed THz communication, sensing, and imaging.