Date of Defense
10-11-2025 6:00 PM
Location
F1-1164
Document Type
Thesis Defense
Degree Name
Master of Electrical Engineering (MEE)
College
COE
Department
Electrical and Communication Engineering
First Advisor
Prof. Mousa Hussein
Abstract
Microstrip patch antennas (MPAs) rely on precise impedance matching for efficient power transfer between the antenna and feed line. This is often achieved using a number of different techniques, one of which is the quarter-wavelength transformer (QWT). However, optimizing the width of the QWT presents significant computational and analytical challenges due to the unknown antenna impedance and the absence of explicit design relationships. This thesis aims to overcome these limitations by developing and comparatively evaluating artificial intelligence (AI) models for QWT width optimization. Methods involved Random Forest (RF) and a novel Probabilistic Deep Neural Network (PDNN) on a custom dataset of approximately 20,000 antenna designs simulated in CST Microwave Studio. Results indicate the PDNN excelled at bandwidth prediction (R² ≈ 0.70) and achieved moderate accuracy for QWT width (R² ≈ 0.35), though it struggled with frequency prediction (negative R²). RF analysis provided crucial feature importances, validating physical insights, such as patch length influencing frequency and patch width affecting bandwidth. This work significantly contributes by demonstrating AI's ability to augment traditional design workflows and provide interpretive insights into complex parameter interdependencies, thereby filling the gap in direct, comparative AI optimization for specific matching network components
Included in
MICROSTRIP ANTENNA DESIGN BASED ON AI AND MACHINE LEARNING
F1-1164
Microstrip patch antennas (MPAs) rely on precise impedance matching for efficient power transfer between the antenna and feed line. This is often achieved using a number of different techniques, one of which is the quarter-wavelength transformer (QWT). However, optimizing the width of the QWT presents significant computational and analytical challenges due to the unknown antenna impedance and the absence of explicit design relationships. This thesis aims to overcome these limitations by developing and comparatively evaluating artificial intelligence (AI) models for QWT width optimization. Methods involved Random Forest (RF) and a novel Probabilistic Deep Neural Network (PDNN) on a custom dataset of approximately 20,000 antenna designs simulated in CST Microwave Studio. Results indicate the PDNN excelled at bandwidth prediction (R² ≈ 0.70) and achieved moderate accuracy for QWT width (R² ≈ 0.35), though it struggled with frequency prediction (negative R²). RF analysis provided crucial feature importances, validating physical insights, such as patch length influencing frequency and patch width affecting bandwidth. This work significantly contributes by demonstrating AI's ability to augment traditional design workflows and provide interpretive insights into complex parameter interdependencies, thereby filling the gap in direct, comparative AI optimization for specific matching network components