Predicting the performance of a graphene-based patch antenna using a machine learning model for terahertz...

This article presents the design of a graphene-based microstrip patch antenna, operating frequency range: (1-5) THz for terahertz (THz) applications. This paper presents simulations and a machine learning (ML) approach to characterize the performance characteristics, such as S11, Voltage Standing Wave Ratio (VSWR), gain, radiation, and total efficiencies, as well as the radiation pattern in horizontal (H, XZ) and vertical (V, XY) planes. The Computer Simulation Tool (CST) full microwave studio is used to model the antenna with dimensions of Length (L) and × Width (W): 93 μm × 113 μm, a return loss around - 40 dB, with 11 multi-band frequencies, achieving a maximum gain of 7.5 dBi at 3.2 THz. To analyze the effect of geometric parameters like length ([Formula: see text]) and width ([Formula: see text]) of the patch and graphene properties such as chemical potential ([Formula: see text]) and relaxation time (τ) on the performance characteristics of the patch antenna, three ML models are developed, such as Artificial Neural Networks (ANN), Random Forest ([Formula: see text]), and Support Vector Machine (SVM). The training data is collected for 784 simulations. The ANN architecture is built with four features in the input layer and one output layer to predict performance characteristics. The performance of the developed models is evaluated using metrics such as Mean Squared Error (MSE) and R-Squared (R²). Out of three developed models, ANN predicts the performance within 0.7 milliseconds (ms) with high accuracy, achieving an R² of 0.99 for all performance characteristics. The predicted results discuss that the regression-based predictive models can capture the nonlinear relationship between antenna geometry and electromagnetic (EM) responses. These advantages, such as faster predictions and higher prediction accuracy, make these models, especially the ANN model, a replacement for traditional EM simulations by reducing computation time. Such qualities made the proposed ML models a powerful alternative to traditional simulation tools, making these antennas useful for next-generation wireless communication systems in the THz frequency range and beyond 6G.