Forensic Outlier Detection and Penalty Analysis To Regulate Cognitive Radio Network
MetadataShow full item record
Cognitive Radio Network (CRN) has emerged as a promising solution to the pseudo scarcity of the spectral congestion problem. In a CRN, the new/Secondary Users (SUs), are allowed to access the spectrum in such a way that the existing/Primary Users (PUs) experience limited interference. As such, SUs cooperatively sense the PU signal and share the spectrum only when the PU is absent. However, such cooperation raises new concerns for security and reliability. This is because abuser SUs may report unreliable/falsified sensing data to influence the cooperative decision for personal gain (e.g., gaining unfair access to the spectrum or inflicting interference). To date, most of the CRN research has mainly focused on developing efficient cooperative sensing algorithms. Much less attention has been paid to information security issues associated with the sensing. Sufficient scientific input is required to develop proper spectrum security policy, otherwise entirely new security threats will impact CRNs' ability to become a long-term commercially-viable concept. In this thesis, we consider a CRN wherein SUs report their sensing decisions to a fusion center for cooperative spectrum sensing. We consider the presence of an abuser SU in the CRN and adopt forensic analysis of the reports for outlier identification. We first assume that the behavioral category of the outlier is known and propose optimal and suboptimal detectors specific to each category. The proposed detectors significantly outperform the state-of-the art detectors and are robust against system parameter variation. Then we consider a CRN where the presence as well as behavioral category of the outlier is unknown. We propose practical algorithms that identify outlier SU's behavioral category. In addition, to regulate the outlier SUs, we introduce the concept of penalizing the outliers proportionally to the crime. Then we propose an algorithm that calculates the punishment (upon identification of an outlier) according to its type and forces it to behave legitimately. The proposed algorithms are robust against variations in system parameters. The proposed detectors and algorithms can be used by government agencies for forensic testing in policy control methodology to regulate CRN abusers.