Deep Learning-Based Modeling and Inverse Design for Vector Graphics-Based Planar Antenna Structures

In this paper, we present an inverse design methodology for vector graphics-based patch structures of planar antennas, utilizing deep convolutional neural network (CNN) modeling to enable rapid and accurate prediction of antenna <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$S_{11}$</tex> - parameters. Unlike previous machine learning approaches that rely on fixed template geometries with limited degrees of freedom, our method generalizes to vector graphics-based patch structures, expanding the design space beyond what is achievable with traditional optimized antennas from finite libraries. By replacing complex and time-consuming electromagnetic simulations with a machine learning-based approach, we propose an inverse design process enhanced by a genetic algorithm. This methodology is demonstrated through simulation and measurement results of double and triple band planar antennas designed across various frequency ranges, showcasing the potential of this approach for innovative antenna design.