Dual-Band Thz MIMO Antenna Using Hybrid Defected Ground Structures, Machine Learning-Based Prediction, and Equivalent Circuit Modeling For Future Wireless Communications

Abstract

Future wireless communication systems demand compact, high performance terahertz (THz) antennas, capable of supporting ultra-high data rates, large bandwidths and increased spectral efficiency. However, the wide bandwidth, high gain and low mutual coupling in compact THz multiple-input multiple-output (MIMO) antennas are still a great challenge. In this paper, a dual-band THz MIMO antenna is proposed designed on an 8-um thick Gallium Arsenide (GaAs) substrate using ANSYS HFSS. First, a rectangular microstrip patch antenna operating at 0.7 THz is designed, and then it is extended to a two-element MIMO configuration. To improve the antenna performance and miniaturization, the ground plane is incorporated with hybrid octagonal and hexagonal split-ring slot defected ground structures (DGSs). The proposed antenna provides dual-band operation at 0.55 THz and 0.73 THz with enhanced impedance bandwidth, improved gain and reduced mutual coupling. The MIMO characteristics are evaluated in terms of isolation, envelope correlation coefficient (ECC) and diversity gain (DG), showing excellent diversity operation for THz wireless communication applications. To mitigate the computational load due to repeated full-wave electromagnetic simulations, machine learning models are constructed with a dataset of 36,000 samples obtained from parametric HFSS simulations. Important antenna performance parameters are predicted using Decision Tree, Random Forest, K-Nearest Neighbors (KNN) and XGBoost regression algorithms. The Random Forest algorithm had the best prediction performance among the investigated models with R²= 0.99986, MSE = 0.00675, RMSE = 0.08219, , and MAE = 0.04846. Furthermore, an equivalent lossy series RLC circuit model is developed in MATLAB for verification of the resonance characteristics and explanation of the antenna behavior at the circuit level. Moreover, the proposed design shows a virtual size reduction (VSR) of 21.26%, which means that the lower-band resonance at 0.55 THz is achieved with a compact physical structure. The results of the electromagnetic simulation and the predictions of the machine learning algorithm and the equivalent circuit model are in close agreement, thus validating the accuracy of the proposed methodology. The combined electromagnetic, machine learning and circuit-theoretic analysis demonstrates the effectiveness of the proposed antenna for compact THz MIMO systems and future wireless communication applications.