Resilient Sensor Placement for Kalman Filtering in Diffusion Networks
Primary Investigator:
Shreyas Sundaram
Lintao Ye, Sandip Roy and Shreyas Sundaram
Abstract
We consider networked systems where there is a source node affected by an unknown (stochastic) input stream. We first study the problem of how a system designer should optimally place sensors on the nodes of the network in order to best estimate the states of the system, under a sensor placement budget constraint. In particular, the goal is to minimize the mean square estimation error of the system states generated by the corresponding Kalman filter. We provide an optimal and computationally efficient sensor placement strategy for the system designer in such scenarios. Next, we consider the perspective of an adversary whose goal is to remove some of the sensors placed by the system designer under an attack budget constraint. Using our insights for the sensor placement problem, we provide an optimal sensor removal strategy for the attacker. Finally, we study the problem of resilient sensor placement, where the system designer is aware of the potential attacks from the adversary and seeks to find an optimal sensor placement that will yield the minimum mean square estimation error after the adversary’s actions. We show that the resilient sensor placement problem is NP-hard in general. We then provide an algorithm based on dynamical programming to solve this problem.