Machine learning based optimization of tunable high gain THz MIMO antenna for wireless applications

Abstract The terahertz (THz) spectrum enables ultra-high data rate communication for defense, biomedical, and space applications. This paper presents a graphene-based dual-port tunable MIMO antenna designed for efficient THz operation. Each port comprises three patches with a 50 Ω offset feed on polyimide ( ϵ r = 3.5) and alumina ( ϵ r = 9.4) substrates, forming a stepped resonator structure. The proposed antenna achieves a wide 5.5–10 THz bandwidth, 12 dBi peak gain, and < −20 dB isolation, with reduced mutual coupling via circular patch-edge cuts. To enhance and predict antenna performance, machine learning regression models—Linear Regression, Gaussian Process Regression (GPR), and Support Vector Regression (SVR)—are implemented using MATLAB to estimate the reflection coefficient ( S 11 ) from design parameters. The GPR and SVR model exhibits the lowest mean square error and the highest correlation with simulated results, enabling data-driven parameter optimization and rapid performance prediction. Comprehensive performance metrics including Channel Capacity Loss (CCL), Total Active Reflection Coefficient (TARC), Mean Effective Gain (MEG), and Envelope Correlation Coefficient (ECC) confirm the antenna’s suitability for THz MIMO applications with excellent diversity performance. The integration of EM simulation and machine learning thus offers a powerful framework for adaptive, high-efficiency THz antenna design.