Distributed cooperative spectrum sensing based on reinforcement learning in cognitive radio networks

Spectrum sensing is an initial task for the successful operation of cognitive radio networks (CRN). During cooperative spectrum sensing, malicious secondary user (SU) may report false sensing data which would degrade the final aggregated sensing outcome. In this paper, we propose a distributed cooperative spectrum sensing (CSS) method based on reinforcement learning (RL) to remove data fusion between users with different reputations in CRN. This method regards each SU as an agent, which is selected from the adjacent nodes of CRN participating in CSS. The reputation value is used as reward to ensure that the agent tends to merge with high reputation nodes. The conformance fusion is adopted to promote consensus of the whole network, while it's also compared with the decision threshold to complete CSS. Simulation results show that the proposed method can identify malicious users effectively. As a result, the whole CRN based on RL is more intelligent and stable.