Research On Resource Allocation Method Of 5G Heterogeneous Cloud Wireless Access Networks

5G heterogeneous cloud wireless access network(H-CRAN)using D2D(device-to-device)communication can greatly improve network performance,which is an important technology in 5G communication application.However,in the communication process,remote radio users and D2Ds(collectively known as non-macro cellular users)need to reuse the communication resources of macro cellular users,which will cause interference to macro cellular users.Therefore,how to reasonably allocate resources in H-CRAN to suppress interference has become an urgent problem to be solved.This thesis proposes different resource allocation methods for different network scenarios to increase system throughput.The main contents are as follows:(1)For the resource allocation problem when the number of non-macrocell users in H-CRAN is less than or equal to the number of macrocell users,a resource allocation algorithm based on matching theory and in line with Karldor-Hicks efficiency is proposed.Firstly,the network model is constructed and the bidirectional preference lists of non-macrocell users and macrocell users are established.Secondly,the Gale-Shapley algorithm in matching theory is used to get the initial match.Finally,on the basis of the initial matching,the resource exchange strategy that conforms to the Karldor-Hicks efficiency is utilized to adjust the initial matching to get the final resource allocation results.(2)For the resource allocation problem when the number of non-macrocell users in H-CRAN is greater than the number of macrocell users,a centralized resource allocation algorithm based on double deep Q network is proposed.Firstly,state space,action space and reward function are defined according to the network model.Then some samples are obtained and stored in the experience pool by using the e-greedy exploratory development strategy.In order to accelerate the training speed of neural network,random priority sampling and importance sampling are employed to obtain samples from experience pool.In order to avoid dimension disaster and overestimation problem,double neural network is used for training.Finally,according to the trained neural network model,the resource allocation results in different states are obtained.(3)The simulation experiments are designed,and the validity of the proposed method is verified by comparing with the relevant methods.