Detecting Arbitrary Attacks Using Continuous Secured Side Information in Wireless Networks

TL;DR

This paper proposes a detection scheme for Byzantine attacks in Gaussian two-hop relay networks using secured observations, proving asymptotic errorless detection under a non-manipulable channel condition without secret keys.

Contribution

It introduces a novel attack detection method for continuous channels that guarantees asymptotic correctness under broad attack models, requiring no pre-shared secrets.

Findings

Detection scheme achieves asymptotic errorless performance.

Non-zero channel coefficients ensure network non-manipulability.

The scheme applies to arbitrary, dependent attack distributions.

Abstract

This paper focuses on Byzantine attack detection for Gaussian two-hop one-way relay network, where an amplify-and-forward relay may conduct Byzantine attacks by forwarding altered symbols to the destination. For facilitating attack detection, we utilize the openness of wireless medium to make the destination observe some secured signals that are not attacked. Then, a detection scheme is developed for the destination by using its secured observations to statistically check other observations from the relay. On the other hand, notice the Gaussian channel is continuous, which allows the possible Byzantine attacks to be conducted within continuous alphabet(s). The existing work on discrete channel is not applicable for investigating the performance of the proposed scheme. The main contribution of this paper is to prove that if and only if the wireless relay network satisfies a non-manipulable channel condition, the proposed detection scheme achieves asymptotic errorless performance against arbitrary attacks that allow the stochastic distributions of altered symbols to vary arbitrarily and depend on each other. No pre-shared secret or secret transmission is needed for the detection. Furthermore, we also prove that the relay network is non-manipulable as long as all channel coefficients are non-zero, which is not essential restrict for many practical systems.