Research On Dynamic Spectrum Allocation Based On Machine Learning

Spectrum allocation is a key technology in cognitive radio.Through power control and channel allocation for cognitive users,they can access authorized frequency bands and successfully reuse "spectrum holes" without affecting the quality of service of the primary user.Effectively improve the utilization rate of the spectrum.Considering that the level of integration of software and hardware continues to increase,the types of devices and the ways to access channels are changing with each passing day.the requirements for cognitive radio technology to deal with multiple problems continue to increase,but traditional spectrum allocation models no longer have this condition,based on Existing spectrum allocation technology.this paper combines the deep reinforcement learning method with spectrum allocation technology,and proposes two dynamic spectrum allocation algorithms based on deep Q network.The main work of this paper is as follows:Chapter 3 analyzes the problems of multi-user and multi-state access methods and joint power control.Based on the advantages of deep Q networks,research a spectrum allocation algorithm with state selection.First,the status of the primary user is divided into two types according to the signal-to-noise ratio threshold.For the different status of the primary user,the secondary user will share spectrum resources with the primary user in Underlay or Amplifyand-Forward mode.Subsequently,in order to improve the success rate of spectrum sharing and the transmission rate of primary and secondary users,secondary users learn effective power control strategies to enable them to still meet their respective signal-to-interference and noise ratio thresholds after accessing the channel,thereby avoiding This affects the quality of service for users.Finally,the experimental part shows that the allocation algorithm can effectively improve the utilization of spectrum resources in a multi-user and multi-state scenario.Chapter 4 analyzes the problem of multi-user channel allocation and joint power control in the case of multiple networks crossing,and research a spectrum allocation algorithm based on dueling Q networks.According to the location information and intersection of each cell,the entire area is divided into three sub-areas.First,the channels in each area are serialized,and then deep reinforcement learning is used to enable the central manager to learn the spectrum allocation strategy of the secondary user in a dynamic environment,so that the primary user and the secondary user can successfully share the spectrum.Finally,in the experimental part,it is proved that the method is robust in dealing with the spectrum allocation problem of multiple cognitive networks,and effectively improves the performance of the system.