Design Optimization of Rectangular Microstrip Antenna Using Deep Neural Network for 3 GHz Applications in Support of SDG 9

Abstract

Objective: This study aims to investigate the effectiveness of Deep Neural Networks (DNN) for optimizing the design of a rectangular microstrip antenna operating at a target frequency of 3 GHz. The research focuses on improving antenna design efficiency by predicting antenna performance parameters based on geometric characteristics. Method: The study employed a computational simulation approach combined with machine learning techniques. A synthetic dataset consisting of 6000 antenna configurations was generated using analytical microstrip antenna equations. The dataset included geometric parameters such as dielectric constant, substrate thickness, patch width, patch length, and inset feed position. A Deep Neural Network model was trained to predict resonant frequency, return loss, and input impedance. The trained model was then used as a surrogate model to evaluate 30,000 candidate antenna designs and identify the optimal configuration. Result: The proposed model achieved high predictive accuracy with values of 0.9987 for resonant frequency prediction and 0.9988 for input impedance prediction. The optimized antenna design produced a resonant frequency of 2.996 GHz, return loss of −18.70 dB, and input impedance of 53.95 Ω, which closely match the target specifications for S-band wireless applications. Novelty: The study demonstrates that Deep Neural Networks can significantly accelerate antenna design optimization by replacing repetitive electromagnetic simulations with data-driven prediction models.