Broad Banding Techniques for Microstrip Patch Antennas using Machine Learning Optimization Approaches

Abstract

This study suggests a data-driven approach towards bandwidth optimization in U-slot microstrip patch antennas through machine learning and simulated data. The objective is to curtail dependence upon full-wave solvers without compromising the design accuracy. A dataset was artificially created using reasonable slot dimension and bandwidth values. A Random Forest Regressor was learned to forecast bandwidth based on slot width and length, and a Genetic Algorithm was used to find the best configuration. Model accuracy was tested on error metrics and plotted using 3D surface and 2D line plots. The regression model achieved a mean absolute error of 0.025 GHz and R²=0.277, effectively capturing bandwidth trends. The optimization yielded a best-performing slot configuration of 6.62 mm length and 2.86 mm width, with a predicted bandwidth of 0.6686 GHz. Visual results confirmed the presence of a nonlinear, resonant relationship in the design space. This work introduces a simulation-free design framework using ML and evolutionary optimization, enabling efficient, low-cost antenna design. It is especially valuable in academic or prototyping scenarios lacking simulation access.