Deflection-based Attack Detection for Network Systems

This paper considers a deflection-based detector for random attacks compromising the inputs of a network system, using measurements from nodes non-collocated with the input nodes. We derive the decision rule of our deflection-based detector, and characterize its performance as a function of the edge weights, attack and noise statistics, and the locations of input and output nodes. In the asymptotic measurement regime, we show that the detector's performance is governed by the singular values of the system's transfer function matrix. Finally, for a given input and output node locations, we numerically solve an optimization problem to find the optimal network edge weights that maximize the detector's performance. Numerical examples are presented to validate the theoretical results.