Electromagnetic Compatibility Prediction of GPS Using Spectral Envelope and Knowledge Distillation

This work proposes a novel neural network-based framework to address the critical task of predicting the electromagnetic compatibility of satellite navigation systems. The proposed framework is mainly constituted by a teacher–student architecture known as the knowledge distillation algorithm. Specifically, the teacher model learns from a limited amount of experiment data, whereas the student model mainly learns from the simulation data in the meantime of being governed by the teacher model. Simulations and experiments were conducted and divided into training and test sets to examine the performance of the proposed framework. The results verified the validity of the proposed framework with an accuracy of 96.30% and 92.59% in predicting whether the equipment under test was sensitive or not, respectively. Moreover, the significance of teacher–student architecture is proved with the help of ablation studies, and it is shown that the knowledge distillation algorithm improves the prediction accuracy by 10% compared to employing only the student model. Overall, this method reduces the need for extensive test data and significantly enhances prediction accuracy, offering promising implications for realistic applications of the Global Positioning System in diverse electromagnetic environments.