Active Learning-Based Sampling for Surrogate Modeling of Microstrip Bandpass Filters

The increasing complexity of modern communication systems has led to unprecedented demands on the design of RF and microwave components, necessitating the development of efficient optimization design methods. Recently, machine learning-based surrogate models have emerged as a significant research direction, enabling rapid prediction of device characteristics once trained. In this paper, an Active Learning (AL)-based framework is developed for modeling a microstrip bandpass filter, and the impact of different sampling methods on the accuracy of deep neural networks (DNNs) is investigated. Our results show that surrogate models trained using AL achieve significant improvements in prediction accuracy, as measured by root mean square error (RMSE) and the coefficient of determination (R2), compared to traditional Latin Hypercube Sampling (LHS) method.