Prediction of microstrip patch antenna S11 using XGBoost algorithm

Abstract

With the rapid advancement of wireless communication technologies, microstrip antennas, as core components of modern RF systems, face increasingly stringent performance requirements and escalating design complexity[1]. Traditional design workflows rely on full-wave electromagnetic simulation tools such as HFSS, employing extensive parametric sweeps combined with engineering experience for iterative optimization. This "trial-and-error" design paradigm suffers from substantial computational cost and prolonged design cycles when confronting multi-parameter coupling and nonlinear responses. To address these challenges, this paper proposes an efficient microstrip antenna optimization method based on the XGBoost gradient boosting regression algorithm. A dataset comprising 479 samples was constructed through HFSS parametric simulation, with patch length (a2), patch width (a4), and feed width (Wf) as input features, and resonant frequency and return loss (S11) as prediction targets. Experimental results demonstrate that the XGBoost model achieves an R² of 0.9968 and RMSE of 0.011 GHz for resonant frequency prediction, and an R² of 0.9234 with RMSE of 1.97 dB for S11 prediction. Furthermore, a trilinear interpolation model on the complete parameter grid achieves exact prediction (R² = 1.000), corroborating the smoothness of the electromagnetic response in parameter space[2].