AI-Driven Electromagnetic Design and Performance Prediction of Microstrip Antennas

Author

Eduardo Javier Vazquez, The University of Texas Rio Grande Valley

Date of Award

12-1-2025

Document Type

Thesis

Degree Name

Master of Science in Engineering (MSE)

Department

Electrical Engineering

First Advisor

Yong Zhou

Second Advisor

Jingao Yang

Third Advisor

Nantakan Wongkasem

Abstract

This thesis investigates the application of Deep Learning to automate and accelerate microstrip antenna inverse design. The initial investigation was to predict microstrip antenna performance from its geometry parametric input, and found out that forward prediction with adopted geometric representation results in ill-posed scenario, high ambiguity and unstable mapping. The study later mostly focuses on the inverse prediction by machine learning from S11 parameter input to predict antenna patch geometry parameters instead.

A dataset of 5,000 ANSYS HFSS simulated antennas (later filtered to 4,136 valid samples) was generated using cubic spine -described geometry profiles. Multiple neural architectures were adopted for training, and the final regression model achieved rapid convergence with RMSE decreasing from around 8 to 3, demonstrating a successful learning of underlying structure-to-performance relationships. However, further investigation indicates that the one-by-one physical comparisons revealed geometry reconstruction remains partially inaccurate despite good accuracy and precision performance of global machine learning metrics—highlighting the fundamental ill-posed nature of the inverse electromagnetic design problem.

This work contributes a validated framework for simulation-driven dataset generation, forward/inverse model experimentation, and performance analysis for AI-based antenna synthesis. Results show Deep Learning can approximate electromagnetic behavior, but unique geometry recovery from S-parameter remains non-deterministic, opening pathways for improved encoding, additional physical constraints, and hybrid optimization with electromagnetic solvers in future research.

Comments

Copyright 2025 Eduardo Javier Vazquez. All Rights Reserved. https://proquest.com/docview/3292478754

Recommended Citation

Vazquez, E. J. (2025). AI-Driven Electromagnetic Design and Performance Prediction of Microstrip Antennas [Master's thesis, The University of Texas Rio Grande Valley]. ScholarWorks @ UTRGV. https://scholarworks.utrgv.edu/etd/1832