Three-layer Bayesian model based spectrum sensing to detect malicious attacks in cognitive radio networks

Owing to the open nature of cooperative cognitive radio networks (CRNs), security becomes a critical topic to consider. In order to acquire more spectrum resources, malicious secondary users (SUs) always launch various attacks. Among these attacks, spectrum sensing data falsification (SSDF) attack is a typical one. To cope with SSDF attacks, this paper proposes a three-layer Bayesian model. History data is processed through three layers, namely processing layer, integrating layer and inferring layer. Processing layer is modeled by hidden Markov model (HMM), which uses original data to train parameters and then provide trained emission distributions to the second layer. Within integrating layer, on the basis of different algorithms, emission distributions are processed to obtain the reputation values, balance values and specificity values of different SUs. By using different thresholds, these continuous values can be made discrete and then transferred to inferring layer. In the third layer, a Bayesian network (BN) is built to calculate the safety probabilities of SUs via using the discrete values as evidence. From simulation results, the proposed system is useful to defend against different types of malicious users, especially in low-SNR situations.