LQG control for distributed systems over TCP-like erasure channels

This paper is concerned with control applications over lossy data network. Sensor data is transmitted to an estimation-control unit over a network and control commands are issued to subsystems over the same network. Sensor and control packets may be randomly lost according to a Bernoulli process. In this context the discrete-time Linear Quadratic Gaussian (LQG) optimal control problem is considered. In B. Sinopoli, et al., (Aug. 2003) a complete analysis was carried out for the case the network is composed of a single sensor and control channel. Here a nontrivial generalization to the case of sensor and actuator networks with p distinct sensor channels and m control channels is presented. It has been proven that the separation principle still holds for all protocols where packets are acknowledged by the receiver (e.g. TCP-like protocols). Moreover it has been pointed out for the first time that the optimal LQG control is a linear function of the state that explicitly depends on the command channels lost probabilities. Such a dependence is not present in pre-existing literature, since the amplitude of each control input has to be weighted by the loss probability associated to its own channel. This is not observed in the single channel case. In the infinite horizon case stability conditions on the arrival are derived. Their computation requires the use of linear matrix inequalities (LMIs).