Secure distributed spectrum sensing in cognitive radio networks using multi-armed bandits

A solution to security of distributed spectrum sensing in cognitive radio networks is provided. Malicious nodes report the channel as busy, when they detect it to be idle, to be able to exclusively use it. However, when they detect the channel as busy, they do not have incentives to misreport and send actual result to the fusion center making it nontrivial to combat spectrum sensing data falsification attack. To neutralize the effect of misreporting sensed spectrum state, this problem is modeled as a binomial multi-armed bandit process using randomized probability matching and Bayesian methods. In each round the fusion center selects subset of most rewarding nodes to use their reports in the fusion rule, while trying to identify and single out misreporting nodes. Results show this technique sifts malicious reports from truthful ones with false alarm or misdetection with high confidence by removing uncertainties in derived patterns of reports.