Guaranteeing Differential Privacy in Distributed Fusion Estimation

With the increasing openness of networked sensor systems, information leakage has become an ever-critical issue due to the malicious eavesdropping attacks on privately available data. The distributed fusion estimates (DFEs), which perform well by combining local state estimates (LSEs) supplied by multiple sensors, imperatively require privacy preserving. Note that the eavesdropper can acquire DFEs by wiretapping fusion information from fusion center and weighting LSEs from sensors. For countering such dual eavesdropping, we introduce the idea called differential privacy from database literature. Basically, the publicly released estimates (PREs), which gather statistical information about DFEs, are considered to be disturbed with stochastic noises. The lower bounds of noise covariances can be designed with the privacy parameters and estimation systems. Furthermore, we disperse the location of noise insertion from PREs to LSEs. In this case, the eavesdropper cannot obtain optimal estimate even if it fuses local data by using original criterion, while the differential privacy is still satisfied. Finally, a target tracking system with multisensor observation is given to verify the effectiveness of the proposed methods.