Angle Only Tracking in Presence of Unknown Additive False Data Injection Attack

Often, while transmitting measurements from a sensor over a communication channel, the channel may be susceptible to cyber-attacks. Thus, the received measurements are corrupted with false data which complicate the task of state estimation. This paper focuses on the problem of bearings-only tracking, where two sensors receive bearing measurements of a target and transmit them to an estimation center over a wireless communication channel. However, the channel is vulnerable to false data injection attacks, and the statistics of these attacks are unknown. Here, a method of estimating the target kinematics has been proposed under additive false data attack considering the attack statistics follow normal inverse Wishart distribution. The variational Bayesian approximation is used to factorise the joint distribution of the target state, the attack mean, and the covariance to the product of the individual probability density functions. Minimizing the Kullback-Leibler divergence between the joint density function and the factored density functions, the target state, along with the attack mean and the covariance, are estimated. The approach is implemented using traditional filtering techniques, and the estimation accuracy is compared with the filters where the attack statistics are assumed to be known.