Research On Clustering And Resource Management Algorithm Based On H-CRAN

5G ultra-dense networking technology improves user experience,signal coverage and system capacity by densely deploying small base stations within the range of macro base stations,but at the same time,it also leads to increasingly complex topology and serious interference between base stations.In addition,in the era of resource shortage and energy conservation advocacy,the problem of spectrum resource waste caused by tidal effect cannot be ignored.Environmental changes are easy to trigger the frequent execution of algorithms,resulting in computing,signaling and other costs cannot be underestimated.Heterogeneous Cloud Radio Access Network is the basic architecture of 5G.Based on this architecture,it is convenient to realize the sharing of user data sending and receiving,channel quality and other information,as well as the unified management and scheduling of resources,which can reduce the interference between base stations.It is important to improve spectrum utilization and network throughput.In the ultra-dense heterogeneous networks,with the goal of enhancing the ability of the algorithm to adapt to environmental changes,solving the serious interference problem,and reducing the algorithm complexity and system overhead,the task of this thesis is mainly divided into two sections.In the first section,a centralized clustering algorithm of small base station based on improved K-means fitting is proposed based on the H-CRAN architecture to simplify the topology structure and reduce the algorithm complexity.In addition,the information measurement and uploading mechanism are designed to obtain the information needed in the clustering and resource allocation stages.Firstly,according to the characteristics of small base stations,a three-dimensional Euclidian coordinate containing the information of "interference degree by macro station,number of access users and number of adjacent small base stations" is defined to identify and distinguish each small base station.According to the characteristics of the cluster center,the density factor and the central factor are defined,and the two-factor coordinates composed of them contain the information of whether each small base station is suitable for becoming the cluster center.Then the traditional K-means algorithm is improved,mainly including the design of fitting function is used to fitting the double factor to coordinate the distribution of redefining the clustering center node to join principles and update the clustering center method,defines the network topology due to node to join or quit change local update method of clustering results.The performance of the algorithm was verified by simulation on Matlab platform.The results show that the algorithm can automatically determine the initial clustering center and the number of clustering based on fitting,and each region can be clearly divided.The average number of iterations is 1/2 of the traditional k-means algorithm.It is lower than the greedy-based dynamic clustering algorithm,and with the increase of the number of small base stations,the advantage of reducing the number of iterations is more obvious.In terms of throughput performance,this clustering algorithm is slightly better than the greedy-based dynamic clustering algorithm.With the increase of the number of small base stations,compared with the traditional k-means algorithm,the throughput performance is significantly improved,up to about 11.56%.In the second section of this thesis,a two-level resource allocation algorithm based on Qlearning is designed according to the topology structure after clustering of the proposed small base station clustering algorithm based on the improved K-means fitting.The target is to lower the complexity of the algorithm and increase the ability to adapt to environmental changes.Under the centralized control of micro base stations,the first-level resource allocation algorithm allocates resource block groups for each cluster and itself based on Q-learning,and RBG contains several resource blocks.The second-level resource allocation algorithm also uses the RB allocation strategy under the control of the macro station to distribute RB to the the small base station in the cluster,and further distributes resource blocks to the users accessing the macro base station and distributes resource elements to the users accessing the small base station based on the existing proportional justice algorithm.The algorithm takes RBG as the learning state and takes the resource reuse or orthogonal situation on RBG as the action to accelerate the convergence speed.According to the environmental feedback,the multiplexing or orthogonal among the acer stations and the clusters were used to balance the spectrum utilization and the system interference.Secondly,only taking the clustering center as the object of resource allocation based on Q-learning can comprehensively coordinate the interference between small base stations and the interference between small base stations and macro stations by taking clustering as the unit,thus reducing the algorithm complexity.The performance of the proposed algorithm is simulated and verified based on Matlab platform.The results show that the throughput of the proposed algorithm is close to that of the centralized algorithm based on Q-learning,but the iteration times can be reduced by 1/8 on average,compared with the distributed resource allocation algorithm based on Q-learning,the decentralized Q-learning algorithm can increase the throughput by 20.34% at most and reduce the number of iterations by about 1/4 on average.