Research On Dynamic Spectrum Access Algorithms Based On Artificial Intelligence In Cognitive Wireless Networks

As wireless communication technology continues to innovate and wireless devices and mobile applications increase,the shortage of wireless spectrum resources has become an obstacle to the development of wireless communication.For this purpose,dynamic spectrum access(DSA)technology in cognitive radio(CR)has been proposed to solve this obstacle.But,designing a traditional DSA solution requires the designer to have complete knowledge of wireless communications to ensure the performance of the system.However,with the complexity of wireless networks and the innovation of wireless technology,it will be difficult for designers to build a suitable DSA solution with complete knowledge of wireless communication.Therefore,dynamic spectrum access schemes based on artificial intelligence(AI)approaches have started to be applied to alleviate the shortage of wireless spectrum resources and have gained considerable attention and research.In this thesis,a DSA scheme of joint power control based on deep reinforcement learning(DRL)is discussed for multi-user multi-channel cognitive radio network(CRN).The main work of this thesis is as follows.1.In the licensed spectrum underlay transmission mode,a CRN is constructed in which the primary users have various communication demands at different moments while the secondary users have various amounts of cached data at different moments.Considering that in this wireless network,dynamic access to licensed spectrum by secondary users will cause severe interference to the primary users.Therefore,in this thesis,a DRL-based DSA scheme with joint power control is proposed for the complex dynamic interference problem.To further reduce the slow convergence caused by the huge action space and state space,the DRL method is improved into a hierarchical DRL method in this thesis.The hierarchical DRL method implements dynamic spectrum access control and power control at the cognitive base station(CBS)and each secondary user,respectively.Finally,the simulation verifies that the proposed scheme can effectively reduce packet loss and effectively increase the convergence speed while maintaining good stability.2.To further improve the utilization of unlicensed spectrum,this thesis introduces the non-orthogonal multiple access(NOMA)technique into CRN and constructs a multiuser multi-channel NOMA cognitive wireless communication system with continuous values of user transmit power.In the unlicensed spectrum NOMA transmission mode,a DSA scheme combining two DRL methods is proposed to address the co-channel interference problem caused by multiple users sharing the spectrum in NOMA systems.First,the base station collects information about users and uses deep Q network(DQN)algorithm to complete DSA control.Then,the users accessing the channel use the deep determinacy policy gradient(DDPG)algorithm to achieve distributed continuous power control.Finally,it is verified through simulation that the proposed scheme achieves faster convergence and lower packet loss than other AI-based DSA schemes,and the continuous transmit power achieves better system performance than the discrete transmit power.