Denial of Service Attacks on Control Systems with Packet Loss

TL;DR

This paper analyzes how denial of service attacks impact control systems with packet loss, comparing attack strategies and their effects on system performance under detection constraints.

Contribution

It provides explicit characterization of attack costs and demonstrates that non-stationary attacks outperform IID attacks under detection constraints.

Findings

Non-stationary attacks outperform IID attacks with detection constraints

Explicit cost increase characterization for IID attack strategies

Comparison of attack strategies for UDP-like and TCP-like protocols

Abstract

The performance of control systems with packet loss as a result of an attack over the actuation communication channel is analysed. The operator is assumed to monitor the state of the channel by measuring the average number of packet losses and an attack detection criteria is established based on the statistic. The performance of the attacker is measured in terms of the increase of the linear quadratic cost function of the operator subject to a given detection constraint. Within that setting, the optimal denial of service (DoS) attack strategy is formulated for UDP-like and TCP-like communication protocols. {For both communication protocols,} DoS attack constructions that are independent and identically distributed (IID) are compared to those that are non-stationary. The main contributions of this paper are (i) explicit characterisation of the expected cost increase of the optimal attack constructions and the associated packet loss parameter for the IID case, (ii) proof, by example, that non-stationary random attacks outperform IID attacks in the presence of detection constraints.