Automated Design Method Based on Boosting Algorithms for Improving the Radiation Performance of Microstrip Antenna Arrays

This paper presents an automated design methodology to improve the radiation performance of microstrip antenna arrays using boosting-based machine learning (ML) algorithms in the X-band frequency range. The proposed approach replaces computationally expensive full-wave simulations with an ML-driven framework trained on a large dataset of wide-angle impedance matching (WAIM) and microstrip antenna structures. To address various design requirements, two types of microstrip antennas are incorporated into the framework. The behavior of both antenna configurations is predicted with only one network achieved by adding preprocessing and postprocessing modules to the method. This reduces the number of trained networks while maintaining the prediction accuracy. Training networks for WAIM and antennas involve four different boosting algorithms: AdaBoost, Gradient Boosting (GB), Extreme Gradient Boosting (XGB), and Light Gradient Boosting (LightGB). Among the evaluated boosting algorithms, LightGB achieved the highest prediction accuracy for both WAIM and antenna models. Two design examples are investigated to demonstrate the framework’s capability in extending the microstrip array scanning range. The results confirm no grating lobes and improved gain at extreme scanning angles across the frequency range. Compared to traditional full-wave solvers, the ML-based method significantly reduces the order of computation time from several hours to seconds while minimizing hardware resource requirements. This automated method offers an efficient framework for designing wide-angle microstrip arrays and expanding their applications without requiring designer expertise.