This paper presents a data-driven approach for terahertz (THz) antenna optimization targeting next-generation 6G communication systems by integrating machine learning-based prediction with electromagnetic simulation. In the existing method, a Support Vector Machine (SVM) model is used for antenna parameter prediction, achieving a Mean Square Error (MSE) of 323.8380 and an R² score of 0.0723, indicating limited prediction accuracy for complex antenna characteristics. To address these limitations, a 1D Convolutional Neural Network (1D CNN) is proposed for improved modeling of nonlinear relationships in antenna design parameters. The 1D CNN model achieves an MSE of 336.1580 and an R² score of 0.0126. Although the standalone CNN shows a slight increase in error metrics compared to SVM, it provides enhanced capability in automatic feature extraction and better adaptability to high-frequency THz data. Based on the prediction output, the optimized antenna parameter (W1 = 22.6274) is obtained and utilized for antenna design in High Frequency Structure Simulator (HFSS). The designed antenna operates at approximately 0.28 THz and demonstrates a maximum gain of 2.8 dB, with a minimum return loss of approximately −40 dB and VSWR values close to 1, indicating excellent impedance matching and efficient radiation characteristics. The proposed approach effectively links prediction and simulation, reducing design complexity and enabling efficient development of THz antennas for 6G communication applications. Thus, the proposed deep learning-based approach offers better adaptability for complex antenna optimization compared to traditional SVM methods.
K.Rojamani et al. (Thu,) studied this question.