Design, Simulation and Machine Learning Modelling of a 28 GHz mmWave Antenna for High-Band 5G Applications

This research concentrates on the design, computational validation, and modeling of a 28 GHz rectangular patch antenna specifically optimized for high-frequency 5G applications, utilizing machine learning methods. The proposed antenna's performance was rigorously evaluated through comparative analysis against a baseline design and existing literature, using CST Studio Suite 2019 for simulations. Rogers RT5880, a low-loss dielectric material with a permittivity of 2.2 and a loss tangent of 0.0009, was selected as the substrate to ensure minimal signal degradation in the antenna design. Annealed copper was utilized for the ground plane, while the patch itself maintained a thickness of 0.508 mm. Key performance metrics obtained through simulations include a return loss of -58.403 dB, a Voltage Standing Wave Ratio (VSWR) of 1.002, a gain of 7.36 dBi, an efficiency of 87%, and an operational bandwidth of 1.46 GHz. To enhance the design approach and mitigate the reliance on extensive physical experimentation, a predictive machine learning model was developed using multiple regression algorithms. Data generated from simulations enabled the application of advanced supervised regression techniques, including Gaussian Process Regression (GPR), Support Vector Machine (SVM), and Ensemble Decision Tree for modelling returnloss. Among these models, GPR outperformed others in predictive accuracy, demonstrating minimal error rates with an $\mathbf{R}^{\mathbf{2}}$ score of 0.9433, Mean Absolute Error (MAE) of 0.3064, Mean Squared Error (MSE) of 0.6124, and Root Mean Squared Error (RMSE) of 0.7826 and Mean Absolute Percentage Accuracy (MAPA) of 99.47% on test data. The outcome of this study underscores the antenna's strong suitability for integration into 5G millimeter-wave (mmWave) applications, offering superior performance characteristics while leveraging machine learning to streamline the design process.