Optimal Controller Structure Reduction for Decentralized Control

This paper addresses the design of decentralized controllers for linear discrete-time systems. We consider state feedback control for networks of scalar interconnected systems. The structure of the decentralized controller is not specified in advance but included into the controller design. Decentralized controllers are designed based on local measurement with minimal number of additional measurement links between the subsystems and the controllers. We formulate this problem as one of maximizing the degree of decentralization subject to a given error performance in terms of the H∞-norm between the system controlled by a centralized controller and the system controlled by the decentralized controller. For the resulting non-convex optimization problem, numerically tractable convex relaxations are provided and an example shows the effectiveness of this approach.