Detecting and Counteracting Statistical Attacks in Cooperative Spectrum Sensing

In this paper we propose a novel Bayesian method to improve the robustness of cooperative spectrum sensing against misbehaving secondary users, which may send wrong sensing reports in order to artificially increase or reduce the throughput of a cognitive network. We adopt a statistical attack model in which every malicious node is characterized by a certain probability of attack. The key features of the proposed method are: (i) combined spectrum sensing, identification of malicious users, and estimation of their attack probabilities; (ii) use of belief propagation on factor graphs to efficiently solve the Bayesian estimation problem. Our analysis shows that the proposed joint estimation approach outperforms traditional cooperation schemes based on exclusion of the unreliable nodes from the spectrum sensing process, and that it nearly achieves the performance of an ideal maximum likelihood estimation if attack probabilities remain constant over a sufficient number of sensing time slots. Results illustrate that belief propagation applied to the considered problem is robust with respect to different network parameters (e.g., numbers of reliable and malicious nodes, attack probability values, sensing duration). Finally, spectrum sensing estimates obtained via belief propagation are proved to be consistent on average for arbitrary graph size.