Regional Adaptive Synthetic Sampling-Based Early Warning Algorithm for Satellite Voice Quality
Jun Chen, Yanjun Wang, Chuanhong Liu, Yang Yang · 2025
Due to the maintenance costs inherent in satellite communication systems, early warning of voice quality degradation is essential for ensuring reliable voice services. However, the performance of early warning algorithms is often hindered by imbalanced data. To solve the issue, a Regional Adaptive Synthetic sampling (RASyn) is proposed, which synthesizes adaptive weights, interpolation direction selection, and interpolation location restriction to generate effective samples. Then, we propose RASyn-Stacking early warning algorithm. Experiments demonstrate that the RASyn sampling algorithm leads to a 30.4% improvement compared to the traditional Stacking algorithm, with the RASyn-Stacking early warning algorithm reaching an F1 score of 0.5714, achieving highperformance early warning.