Achieving secure spectrum sensing in presence of malicious attacks utilizing unsupervised machine learning

In this paper, we focus on the problem of how to realize secure sensing in a comprehensive hostile cognitive radio network where both PUEA and SSDF are present, while most existing research regarding secure sensing treats Primary User Emulation Attack (PUEA) and Spectrum Sensing Data Falsification (SSDF) separately. The problem arises from the coexistence of PUEA and SSDF that even if PUEA is successfully identified, its neighboring honest users would unintentionally send contaminated sensing reports, and thus leading to unsafe sensing performance. An ideal and straightforward solution is to directly exclude those attacked secondary users from the sensing cooperation process. However, this scheme requires perfect information such as the attacking strength, geographical locations of secondary users, etc. To alleviate this requirement, a secure sensing algorithm is proposed in this paper, which identifies attacked secondary users by incorporating the idea of unsupervised machine learning, without prior information on either the attack or secondary users. Further, to account for identification error and different reliabilities of SUs, an identity value is assigned and adaptively updated for each SU. Simulations show that the proposed algorithm has better performance than conventional secure sensing algorithm in presence of both PUEA and SSDF, and remains robust even in a highly stressed environment where a significant portion of the network is attacked. It is also demonstrated that, when the sensing SNR increases, the performance of the proposed algorithm asymptotically approaches the ideal solution where the attacked users are perfectly removed.