A GNSS Interference Detection Method Utilizing Multi-Feature Transfer Learning from ADS-B Data

The global navigation satellite system （GNSS） serves as a critical foundation for modern aviation systems； however， it is highly vulnerable to radio frequency interference， which can result in flight diversions， go-arounds， or aborted approaches， posing serious risks to aviation safety. automatic dependent surveillance-broadcast （ADS-B）， which depends on GNSS for acquiring aircraft position information， is similarly affected when GNSS is subject to radio frequency interference， thereby compromising the availability of ADS-B. Detecting GNSS interference detection based on ADS-B data has become a feasible solution. To overcome the limitations of existing GNSS interference detection models， such as incompatibility with multiple ADS-B versions and inadequate adaptability to China’s specific operational environment， this study focuses on analyzing ADS-B data collected during GNSS interference events. This research investigates the characteristics under interference conditions， including trajectory fluctuations and variations in navigation quality indicators. By incorporating a sliding window technique， statistical features are dynamically computed， and the feature dimensions are expanded to more comprehensively and accurately capture the impact of interference. A novel GNSS interference detection method is proposed， integrating a long short-term memory-autoencoder （LSTM-AE） with a domain adversarial neural network （DANN）. The LSTM-AE extracts features from ADS-B data across different versions and maps them into a unified feature space， ensuring consistent feature representations. The DANN network is subsequently utilized to detect GNSS interference in DO-260A/B version ADS-B data （source domain）， while leveraging DANN’s transfer learning capability to adapt to DO-260 version ADS-B data （target domain）， thereby enabling efficient cross-version detection. Experimental results indicate that the proposed LSTM-AE-DANN model achieves superior detection performance and robust adaptability across both DO-260 and DO-260A/B version ADS-B datasets. This approach is particularly well suited for China’s aviation system requirements and holds substantial practical value.