Detection and Analysis Against Linear Deception Attacks via Watermarking Strategy

The detection concerns associated with linear deception attacks in cyber-physical systems are addressed in this paper. We consider a general linear attack strategy that is based on arbitrary Gaussian distributions with varying means, with the cumulative error covariance serving as a performance metric for remote state estimation. A proposed detection method utilizes Kullback-Leibler (K-L) divergence, employing a watermark to protect the sensor measurements transmitted via the wireless network. This approach enables the recovery of transmitted data during normal operation, thereby ensuring the effectiveness of remote estimation. The detection performance of the scenarios with no watermarking information and limited watermarking information is analyzed. Furthermore, the optimal linear attack is reconstructed in the information disclosure scenario, and the impact on remote state estimation performance is analyzed. Finally, numerical simulations are conducted to further illustrate theoretical results.