Generative Adversarial Scheme Based GNSS Spoofing Detection for Digital Twin Vehicular Networks

Digital twin vehicular network is an emerging architecture to realize vehicle communications. Anti-GNSS-spoofing becomes a challenging issue due to the growing automotive intelligence. However, the anti-spoofing methods are faced with several challenges: the additional cost of anti-spoofing devices, the limited computation resource within the vehicles, the lack of abnormal data, and model bias. To solve these problems, a generative adversarial scheme based anti-spoofing method is proposed for digital twin vehicular networks. The scheme consists of two deep-learning models of the generator and the detector, which generates pseudo normal data and detects spoofing. The LSTM model is introduced as the generator model, which fabricate the abnormal data with the GNSS/CAN/IMU data from Comma2k19. The DenseNet is introduced as the detector model, which make prediction on the basis of latitude, longitude, speed, steering angle and acceleration forward. The generative adversarial scheme is implemented for performance analysis, which indicates that the proposed scheme is suitable for digital twin vehicular applications.