Machine Learning-Aided Optimization of a Slot–Parasitic Rectangular Microstrip Patch Antenna for Sub-6 GHz Applications

Abstract

This paper proposes a machine learning-assisted methodology for the optimization of a compact dual-band microstrip patch antenna operating in the Sub- 6 GHz range. A fully parameterized reference antenna is analyzed using fullwave electromagnetic simulations. A sensitivity-based analysis is conducted to identify the most influential geometrical parameters, leading to a reduced six-dimensional design space. A dataset of 1000 antenna samples is generated using CST Microwave Studio, from which the resonant frequencies and -10 dB impedance bandwidths are automatically extracted. A Random Forest regression model is trained as a surrogate to predict antenna performance with high accuracy and is embedded into a multi-objective NSGA-II optimization framework to maximize impedance bandwidth while maintaining stable resonances around 3.5 GHz and 5.8 GHz. Selected Pareto-optimal designs are validated through full-wave simulations, showing strong agreement with surrogate predictions. The optimized antenna exhibits a significant bandwidth improvement compared to the reference design, demonstrating the effectiveness of the proposed approach for efficient antenna optimization.