Optimal Observers for Decentralized Control

A method is developed for computing optimal observers for linear decentralized feedback control systems. Necessary conditions are derived for the decentralized output-feedback control to optimally approximate a given state-feedback control. The observer gains that satisfy the necessary conditions are found using a numerical gradient-type optimization. A control structure where decentralized observers share control-input information is also considered. For this structure, the optimal observers can be computed directly from a set of decoupled algebraic Riccati equations. Centralized control is treated as a special case, resulting in a simple and novel derivation of the known results for computing optimal centralized observers.