Research On Spectrum Sensing And Aggregation Algorithm Based On Deep Reinforcement Learning

The rapid development of wireless communication technology and the explosive growth of wireless communication equipments lead to the increasing demand for spectrum resources.However,the limited spectrum resources cannot be fully utilized under the traditional fixed allocation policy.So cognitive radio technology came into being to alleviate the problem of spectrum shortage by improving spectrum efficiency.In recent years,the research of reinforcement learning theory in artificial intelligence has made great progress.The researchers applied this theory to the field of cognitive radio networks.Great breakthroughs have been made in spectrum allocation and power control problems.With deep reinforcement learning as a tool,this thesis achieves the following work for spectrum allocation in cognitive wireless networks:Firstly,the research background and significance of spectrum allocation based on reinforcement learning from multiple perspectives,and the current research status of this topic at home and abroad are summarized.Secondly,the basic concepts and key technologies of cognitive radio,including spectrum sensing technology and spectrum aggregation technology are elaborated,and the basic theory and common algorithms of reinforcement learning are also introduced.Thirdly,the spectrum access problem in single cognitive user scenarios,considering the user's bandwidth requirements and the spectrum aggregation ability,is studied in detail.A dynamic spectrum sensing and aggregation solution is proposed,and the greedy strategy,reinforcement learning method and deep reinforcement learning method are adopted to achieve it.The performance of the three algorithms is compared from three aspects,including decision-making accuracy,robustness and time complexity.Fourth,the spectrum allocation problem under multiple user scenarios is studied.Coordinated spectrum utilization in multi-user scenarios is realized by deep reinforcement learning algorithm considering the user's bandwidth requirements and the priority level,and the environmental instability problem under the multi-user reinforcement learning situation is addressed.Finally,the performance of the algorithm is discussed from the average reward feedback and each user channel quality.In this thesis,deep reinforcement learning tools are applied in the field of spectrum allocation to find a more efficient and flexible adaptive spectrum allocation method.Some results have been achieved in a variety of scenarios,which verifies the feasibility of deep reinforcement learning in spectrum allocation.The future research direction can focus on the perspective of algorithm and scene,design more applicable algorithm to improve performance,establish a more concrete scene model to fit the reality,and finally achieve spectrum allocation with high performance and high efficiency.