Primary user emulation detection using frequency domain action recognition

In this paper, we propose an approach for detecting primary user emulation attacks in cognitive radio networks based on the video processing method of action recognition. Specifically, we apply this method to analyze the FFT sequences of wireless transmissions operating across a cognitive radio network environment, as well as classify their actions in the frequency domain. Built upon the previous approach proposed by the authors, this new approach is initiated via energy detection to locate the existing wireless transmissions within a specific frequency band. The approach employs a covariance descriptor of motion-related features in the frequency domain, which is then fed into an artificial neural network for classification. The proposed approach is validated via computer simulations as well as by experimental hardware implementations using the USRP2 software-defined radio (SDR) platform. The computer simulations show that our new approach overcomes the limitations from the authors' previous approach. The hardware experiment shows that proposed approach can achieve a percentage of correct detection higher than 99% in actual wireless environments.