Deep Learning Enabled Reliable Identity Verification and Spoofing Detection

Identity spoofing is one of the severe threats in Cyber-Physical Systems (CPS). These attacks can cause hazardous issues if they occur in Air Traffic Control (ATC) or other safety-critical systems. For example, a malicious UAV (Unmanned Aerial Vehicle) could easily impersonate a legitimate aircraft’s identifier to trespass controlled airspace or broadcast falsified information to disable the airspace operation. In this paper, we propose a joint solution of identity verification and spoofing detection in ATC with assured performances. First, we use an enhanced Deep Neural Network with a zero-bias dense layer with an interpretable mechanism. Second, based on the enhanced Deep Neural Network, we reliably verify airborne targets’ identity with the capability of detecting spoofing attacks. Third, we provide a continual learning paradigm to enable the dynamic evolution of the Deep Neural Network. Our approaches are validated using real Automatic dependent surveillance–broadcast (ADS–B) signals and can be generalized to secure other safety-critical CPS.