Optimal linear filter design for process state and packet loss estimation in networked control systems.

Considering the effect of packet losses on the behavior of networked systems, this work is concerned with estimation of the packet loss occurrences in the input channels possibly together with the system state. For this purpose, the commonly used Markov chain model of the successive packet loss occurrences is transformed to a linear recursive model in which the packet loss occurrence variables appear as new state variables. Two methods are proposed for combining the recursive packet loss model with the plant model to obtain an overall model for the whole networked control system (NCS). In the first method, a state space model of the plant is used which allows for simultaneous estimation of the packet loss occurrences and the plant state. In the second method, an input-output model of the plant is employed which allows for estimating only the packet loss occurrences. Both the zero and the hold packet loss handling strategies are considered and stability of the filters is analyzed. The proposed methods are compared with some existing results during an example to show their advantages.

Simultaneous estimation of state and packet-loss occurrences in networked control systems.

The information on data packet loss occurrences in a networked control system (NCS) is useful for improving the control performance and detecting communication failures or cyber-attacks. This study considers the simultaneous estimation of the plant state and packet loss occurrences at each time step. After formulation of the problem, two solutions are proposed. In the first one, an input-output representation of the NCS model is used to design a recursive filter for estimation of the packet loss occurrences. This estimation is then used for state estimation through Kalman filtering. In the second solution, a state-space model of NCS is used to design an estimator for both the plant state and the packet loss occurrences, which employs a Kalman filter. The effectiveness of the proposed solutions is shown during an example, and comparisons are made with another solution based on the interacting multiple model approach to estimation.