Detecting Byzantine attacks in Cognitive Radio Networks : A two-layered approach using Hidden Markov Model and machine learning

Abstract

Cognitive Radio Networks (CRN) are gaining popularity due to their ability to harness the TV whitespace. CRNs are expected to usher in a new wireless technology to cater to the ever growing population of wireless mobile devices while the current ISM range of wireless technologies is increasingly becoming insufficient. CRNs uses the principle of collaborative spectrum sensing (CSS) where unlicensed users, called Secondary Users (SU) keep sensing a licensed band belonging to the incumbent user called the Primary User (PU). Such a collaborative sensing mechanism enables an increased detection accuracy of the incumbent’s presence (or absence) as all the SUs send their local sensing reports to a Fusion Centre (FC), where these sensing reports are aggregated to arrive at a final sensing decision. However, this collaborative sensing introduces vulnerabilities which can be used to carry out an attack called the Byzantine Attack (a.k.a. Spectrum Sensing Data Falsification (SSDF) attack). We present a two-layer model framework to classify Byzantine attackers in a CRN. The first layer, Processing layer, uses Hidden Markov Model (HMM) to get a probabilistic relationship between the PU’s states and the SUs’ sensing reports. This generates the required dataset for the next layer. The second layer, Decision layer, uses several ML algorithms to classify the SUs into Byzantine attackers and normal SUs. Extensive simulation results confirm that the learning classifiers perform well across various testing parameters. Finally, a comparison analysis of the proposed method with an existing non-ML technique shows that the ML approach is more robust especially under high presence of malicious users.