Plant and buffer state estimation for networked predictive control over multiple erasure channels.

This paper investigates optimal estimation for networked control systems composed of linear, time-variant, and multi-input multi-output (MIMO) plants that communicate with the controller over multiple independent erasure channels. In the controller-to-plant link, we cope with data losses by considering a networked predictive control strategy in which the controller not only sends the current control inputs to the actuators, but also sequences of predicted control inputs at every time step. These are stored in buffers located at the corresponding actuator side, with the aim of being used in case of data loss. Additionally, for losses occurring in the plant-to-controller link, we assume that the controller contains a data-dropout compensation mechanism capable of replacing the missing measurements with their optimal estimates. It is assumed that acknowledgment packages are sent from the plant to the controller, to indicate if the previous sequence was received or not. For this setup, we compute optimal time-varying estimates of the plant and buffer state, and for the lost measurements as well. Finally, we provide the optimal stationary estimator, which serves as a simpler alternative to the time-varying one.