Adversarial Robustness in Cognitive Radio Networks

TL;DR

This paper introduces a novel online algorithm, OBRDHT, for assessing the resilience of cognitive radio networks against adversarial attacks using byzantine resilience distributed hypothesis testing, with evaluations based on false alarm and miss detection probabilities.

Contribution

It proposes a new generic online algorithm, OBRDHT, for measuring network resilience in adversarial settings within cognitive radio networks, extending to other network types.

Findings

The algorithm effectively measures resilience through false alarm and miss detection probabilities.

Simulation results demonstrate the algorithm's accuracy in various sensing times.

The approach generalizes to different network configurations and attack scenarios.

Abstract

\textit{When an adversary gets access to the data sample in the adversarial robustness models and can make data-dependent changes, how has the decision maker consequently, relying deeply upon the adversarially-modified data, to make statistical inference? How can the resilience and elasticity of the network be literally justified $-$ if there exists a tool to measure the aforementioned elasticity?} The principle of byzantine resilience distributed hypothesis testing (BRDHT) is considered in this paper for cognitive radio networks (CRNs) $-$ without-loss-of-generality, something that can be extended to any type of homogeneous or heterogeneous networks $-$ while the byzantine primary user (PU) has a signal-to-noise-ratio (SNR) including the coefficient of $\frac{d\ell \big ( \theta | \mathscr{s}_0 \big )}{d\ell \big ( \theta \big )} $ which is in relation to the temporal rate of the $\alpha-$leakage as the appropriate tool to measure the aforementioned resilience. Our novel online algorithm $-$ which is named $\mathbb{OBRDHT}$ $-$ and solution are both unique and generic over which an evaluation is finally performed by simulations $-$ e.g. an evaluation of the total error as the false alarm probability in addition to the miss detection probability versus the sensing time.