Design of 5G Dual Band Antenna and Optimization using Machine Learning Techniques

Objectives: Exploring Machine Learning (ML) application in electromagnetic domain is need of the hour nowadays. This article presents a dual-band antenna for 5G mobile and Wi-Max applications at 2.2 GHz, 3.6 GHz and its optimization using Polynomial Regression (PR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) ML techniques. Methods: Design optimization and Root Mean Square Error (RSME) are obtained. Hyper parameter tuning is applied for better model performance. Antenna design dimensions are obtained from XGBoost techniques which is the better performing model. Findings: A novel microstrip antenna of (0.25𝜆0 × 0.32𝜆0)mm2 is proposed on FR-4 substrate using insert line feed. Single unit Complementary Split Ring Resonator (CSRR) is incorporated on the ground plane for dual-band characteristics. The proposed antenna has a return loss of -20.37dB at 2.2GHz and -32.27dB at 3.6 GHz, with a maximum gain of 4.06dBi at 3.6GHz. Full wave High-Frequency Structural Simulator (HFSS) is used for data generation for ML models. After hyper parameter tuning XGBoost model is giving the better performance in case of RSME. Observed values in case of train and test case are [(0.007, 0.068), (0.016, 0.340), (0.761, 4.290), (1.057, 3.575)] & [(0.005, 0.026), (0.010, 0.091), (0.525, 2.626), (0.648, 2.544)] for operating frequency, return loss target variables respectively with and without parametric tuning. Novelty: The dimensions obtained from XGBoost model are used for fabrication, and performance comparison of the proposed, ML-based simulated and measured antenna along with various antenna parameters is presented. Keywords: HFSS, Microstrip Patch Antenna; Machine Learning Algorithm; Polynomial Regression; Random Forest; Extreme Gradient Boost (XG-Boost)