Research On Machine Learning Based Spectrum Sensing Policy In Cognitive Radio

With the rapid development of wireless communication technology and the large scale application of Internet of Things, wireless spectrum must be allocated to more and more devices. The existing fixed spectrum allocation strategy can not meet the increasing demand of spectrum.Cognitive radio as an effective technology to improve the utilization of spectrum, has gained a lot of attention from both domestic and foreign counterparts and research institutions. Spectrum sensing is the basis and key of the cognitive process, and also the focus of cognitive radio technology research. As the secondary users can effectively reduce the sensing performance loss caused by multipath fading and hidden nodes through collaboration, the study of cooperative spectrum sensing has important theoretical and practical significance. Based on a large number of references, this thesis studied and improved the reinforcement learning based cooperative spectrum sensing strategy for the disadvantages of secondary users location correlation and sensing energy consumption in cooperative spectrum sensing. Through a large number of simulations, the rationality and effectiveness of the improved algorithm were verified. The main work and innovation of this thesis are as follows:1. This thesis researched and summarized the common technologies in cognitive radio networks, deeply researched the key technologies in spectrum sensing, and elaborated the cooperative spectrum sensing from the aspects of definition, structure and fusion rules.2. The reinforcement learning based cooperative spectrum sensing strategy has been improved in this thesis. The key of the existing sensing strategy was to pick out the subbands and secondary users whose Q values were enhanced through reinforcement learning. In this thesis,when the location correlation of secondary users was large and shadow effect or multipath fading occurred simultaneously, the sensing performance was greatly reduced by the existed policy. To solve this problem, the location information has been used to further filter the secondary users. The improved algorithm has been compared with the algorithm which did not consider the location correlation. A lot of simulations has shown that when the secondary user location was correlated and shadow or multipath fading appeared, the proposed algorithm improved the detection probability of the system, the throughput performance has been increased by about 15%, and the robustness of the system has also been improved.3. When the sensing performance of collaborative spectrum sensing been improved, the energy consumption of system is increased. To solve this problem, this thesis introduced a greedy algorithm, which achieved the goal of reducing the sensing time loss and achieved the highest sensing efficiency while satisfying the detection probability required by the target scenario by balancing the pseudo-random deployment and the reinforcement learning based sensing deployment. A lot of simulations has been done, and the performance of the system has been analyzed from the perspectives of sensing time, detection probability and throughput. It has been proved that the proposed algorithm was reasonable and effective.