Tolerating adversaries in the estimation of network parameters from noisy data: A nonlinear filtering approach

Estimating network parameters from noisy data is a hard problem that can be made even more difficult by the presence of a malicious adversary who may corrupt the measurement process by capturing a trusted node or perturbing data externally. The adversary may have complete knowledge of the networking protocols that rely on the parameter estimates and may adjust its effect on the system to push protocols into incorrect operating regimes. This work focuses on studying how an adversary may impact the estimation of link quality (LQ) of a communications link. We propose a nonlinear filtering solution that simultaneously tracks both the quality of a link and the state of the adversary, tracking the latter to tolerate better the corruption in tracking the former. We provide empirical results while considering several types of adversarial perturbation, including ones that falsely report the LQ measurements or jam a link. Extensions of these analytical techniques and empirical results show how assumptions about symmetry between the LQ of each direction of a bidirectional link can improve adversary tracking and, in turn, LQ estimation.