ML/GA-based performance optimization of PBG-enhanced THz microstrip patch antennas on PTFE-SWCNT.

This study presents the design and optimization of a terahertz (THz) microstrip patch antenna enhanced with photonic bandgap (PBG) structures. The antenna is implemented on a Polytetrafluoroethylene (PTFE) substrate with Single-Wall Carbon Nanotube (SWCNT) conductors, leveraging the substrate's low loss tangent and stable permittivity together with the high conductivity of SWCNTs to improve radiation performance. Key physical parameters, including air gap side, lattice constant, and substrate thickness, were varied using CST simulations to generate a comprehensive dataset. Four machines learning models Linear Regression, K-Nearest Neighbors, Decision Trees, and Neural Networks were trained, with the neural network achieving the best predictive accuracy (R² > 0.94) and very low errors across bandwidth (± 0.05 GHz), gain (± 0.1 dBi), efficiency (< 0.5%), and return loss (0.4 dB). Optimization through a genetic algorithm identified the optimal geometry (Y = 60 μm, D = 80 μm, h = 85 μm), yielding 36.8 GHz bandwidth, 9.4 dBi gain, 93.7% efficiency, and - 26.1 dB return loss. Specific Absorption Rate (SAR) analysis confirmed safety compliance, with a maximum value of 1.4 W/kg under FCC limits. By integrating electromagnetic simulation, machine learning, and evolutionary optimization, the proposed approach provides a faster and more accurate design methodology. Owing to its compactness, efficiency, and material flexibility, the antenna shows strong potential for non-invasive medical imaging, biosensing, and wearable health-monitoring in the THz domain.