Title: Co-design for Security and Performance: Geometric Tools

Abstract: When control systems are endowed with model-based anomaly detectors, the scale of what an attacker is able to accomplish is constrained if an attack is to remain stealthy to the detector. This paper develops a novel framework based on geometric sums for computing precise ellipsoidal bounds on the set of states reachable due to constrained stealthy attacks. We show that this approach provides much more exact ellipsoidal bounds than alternative methods based on convex optimization. The increased tightness of these bounds enables a thorough investigation into the inherent trade-off between closed-loop performance (measured based on the $\|H\|_2$ gain) and security (measured based on the reachable set of the attack) through the design of the controller and observer gain matrices in estimate-based feedback on linear time-invariant control systems.