ADS-B Spoofing Detection Based On Deep Learning

Automatic Dependent Surveillance-Broadcast(ADS-B)is widely used as a new type of air traffic management surveillance technology.With the help of the Global Navigation Satellite System(GNSS),flight information is periodically broadcast to other aircraft and ground supervision systems through the air data link.This technology has high positioning accuracy and low cost,but the open technical architecture makes it extremely vulnerable to certain deceptive interference and seriously disrupts air traffic order.With the rapid development of deep learning technology,it has been widely used in the fields of image,speech recognition and natural language processing,and it also showed its huge advantages in the field of signal processing including signal modulation recognition,radar image recognition,and channel resource allocation.Therefore,this article will mainly study ADS-B deceptive interference detection methods based on deep learning.The main work of the thesis is as follows:First,the principle of ADS-B system and its signal characteristics are introduced,and some common deception methods are introduced.In addition,the application of deep learning in the field of signal processing is introduced,and the advantages and disadvantages of deep learning-based detection methods are analyzed,which lays a foundation for the subsequent research on the use of deep learning to detect deceptive interference methods.Second,a method for extracting features and detecting spoof interference using improved AlexNet is proposed.This method aims at the characteristics that the Doppler frequency deviation of the real ADS-B signal is consistent with the change of the reported position,and uses the improved AlexNet perform training and prediction.Finally,the trained neural network model is used to identify real signals and false signals.Compared with the existing methods,this method only requires a single data source.The steps are simple,and a higher recognition accuracy is obtained when the track length is shorter.Third,a method for directly detecting ADS-B time-domain signals using improved Inception-ResNet is proposed.This method uses the powerful capabilities of extracting feature of deep Inception-ResNet networks to directly train by using ADS-B signals in the time domain as training samples.The steps of this method are extremely simple,realizing end-to-end operation.Compared with the former method,although this method takes longer to train and the accuracy rate decreases slightly,its algorithm complexity is greatly reduced and it is easier to implement in engineering.