Harvest and Jam: Optimal Self-Sustainable Jamming Attacks against Remote State Estimation

TL;DR

This paper studies a self-sustainable jamming attack on remote state estimation, optimizing energy use to maximize estimation errors under different knowledge scenarios, with proven optimal policies and algorithms.

Contribution

It formulates the attacker's power allocation as an MDP, proves the existence of optimal policies, and develops convergent algorithms with structural properties for both knowledge cases.

Findings

Optimal attack policies are derived for both knowledge scenarios.

Algorithms converge to the optimal policy over time.

Structural properties of policies enable faster computation.

Abstract

This paper considers the optimal power allocation of a jamming attacker against remote state estimation. The attacker is self-sustainable and can harvest energy from the environment to launch attacks. The objective is to carefully allocate its attack power to maximize the estimation error at the fusion center. Regarding the attacker's knowledge of the system, two cases are discussed: (i) perfect channel knowledge and (ii) unknown channel model. For both cases, we formulate the problem as a Markov decision process (MDP) and prove the existence of an optimal deterministic and stationary policy. Moreover, for both cases, we develop algorithms to compute the allocation policy and demonstrate that the proposed algorithms for both cases converge to the optimal policy as time goes to infinity. Additionally, the optimal policy exhibits certain structural properties that can be leveraged to accelerate both algorithms. Numerical examples are given to illustrate the main results.