Performance enhancement of THz antenna with dielectric resonator and prediction bandwidth using machine learning approach for 6G applications

Abstract

The need for ultra-low latency and ultra-wideband in 6G applications requires efficient solutions for dielectric resonator antenna design. This paper presents the results of using a machine learning approach to improve the performance of a dielectric resonator antenna operating in the terahertz frequency band. The antenna design uses a polyimide substrate material with a compact size of 58 × 73 µm 2 . A silicon dielectric resonator in the shape of a rectangle ( \(\:{\epsilon\:}_{r}\) of 11.9) is placed on the patch to improve its performance. The antenna demonstrated a wide bandwidth of 4.3936 THz, a maximum gain of 6.02 dBi at 4.59 THz and an efficiency of 77.19%, exhibiting excellent radiation patterns and excellent correspondence. In addition, many machine learning methods were used to predict the bandwidth based on different antenna characteristics. Various measures were used to examine machine learning models, such as variance score, R 2 , mean square error (RMSE), mean absolute error (MAE) and mean square error (MSE). Out of the five models tested, the K-Nearest Neighbor (KNN) model had the lowest error values (MAE of 0.000808, MSE of 0.000002, and RMSE of 0.00135) and the best R 2 , reaching 0.9647 in bandwidth prediction. The suggested terahertz dielectric resonator antenna, confirmed by electromagnetic simulations and machine learning-based predictions, is a viable option for 6G terahertz applications due to its outstanding impedance and radiation matching performance.