A Machine Learning GNSS Interference Detection Method based on ADS-B Multi-index Features

The modern air traffic management system relies heavily on GNSS systems, which provide multiple information for communication, navigation, and surveillance systems, GNSS interference seriously affects the accuracy, consistency, and reliability of various aviation equipment, endangering aviation safety. When GNSS interferes, Automatic Dependent Surveillance-Broadcast (ADS-B) reported data will also appear abnormal, since ADS-B data mainly depends on GNSS. Therefore, the ADS-B report, which is high-frequency and contains spatiotemporal data characteristics provides a new idea for interference detection. However, most existing methods for detecting GNSS interference sources based on ADS-B data rely on only one or two index from a single ADS-B report. The inherent uncertainty of the ADS-B data leads to a tendency for these detection methods to identify non-interfering as interfering.In this paper, we propose a new A Machine Learning GNSS interference detection method based on ADS-B multi-index features, which reduces the impact of the uncertainty of the ADS-B data itself on GNSS interference detection through two improvements. Firstly, we analyze the multiple relevant features of the interference data based on laboratory simulation experiences and actual accidents, then observes the data distribution of each ADS-B feature under normal and interfering conditions and the interrelationship between ADS-B features. The Navigation Integrity Category (NIC), Navigation Accuracy Category for Position (NACp), Source Integrity Level (SIL), messages update interval, the change rate of ground speed, the change rate of position, track angle(TA), flight level(FL), and ADS-B equipment version were finally extracted as multi-index features. Secondly, this paper uses sliding windows to construct new inputs that contain time dimension change information. This construction of input data can obtain more accurate manual annotation and enable various machine learning classifiers to be more effectively applied to ADS-B report data.To prove the effectiveness of the above two improvements, the logical regression model based on multi-index system with original inputs construction, is used as a baseline for classifier performance, it first compared with the original method with a single index, through experiments we find that the classification method based on multiple index has a better performance. Then, various multi-index machine learning methods with new inputs constructed were used to detect GNSS interference, including Recurrent Neural Network(RNN) and Long Short Term Memory(LSTM). Also, take the multi-index logistic regression model as a baseline, and finally, the experimental results are compared and discussed.