A detection-estimation approach to filtering with intermittent observations with generally correlated packet dropouts

This paper is concerned with the problem of state estimation for the class of linear discrete-time Gaussian systems with intermittent observations due to packet losses. This is a common case in networked control systems, where the state of a remote plant is estimated from measurements carried through a lossy network. We assume that the receiver does not know the sequence of packet dropouts. This is typical, e.g., in wireless sensor networks or in networks that cannot rely on protocols that provide information on packet loss. Moreover, we assume that the sequence of packet dropouts is correlated, thus subsuming both the cases of independent dropouts and dropouts modeled as a Markov chain. We propose a detection-estimation approach to the problem of state estimation. The estimator consists of two stages: the first is a nonlinear optimal detector, which decides if a packet dropout has occurred, and the second is a time-varying Kalman filter, which is fed with both the observations and the decisions from the first stage. The overall estimator has finite memory and the tradeoff between performance and computational complexity can be easily controlled. As a case study, we derive the decision rule in closed form in the case of dropout sequence modeled as a Markov chain. Simulation results highlight the effectiveness of the proposed approach, which outperforms the linear recursive estimator of Hadidi and Schwartz.