The Research On Access Selection Mechanism For Wireless Network Slcing

With the rapid development of mobile communication networks and Internet technologies,traditional mobile communication networks have become increasingly difficult to cope with the growth of traffic in the network and changes in service demands.In the research of the next-generation mobile communication network represented by 5G,Software Defined Networks(SDN)combined with Network Function Virtualization(NFV)technology is used to improve network transmission performance and meet the rapidly growing demand for business services.Recently,end-to-end network slicing based on SDN and NFV technologies is considered as a promising technology to satisfy the drastically increasing requirements of diversified services in future mobile networks.The proposed network slicing technology provides technical support for meeting diverse business scenarios(eg,automotive,mobile broadband or the Internet).A slice consists of a series of custom logical network functions that support the communication service needs of a particular use case or business model.Specifically,the operator's physical network is divided into multiple virtual end-to-end(E2E)networks,each slice being logically isolated.Each slice has special handling in terms of performance(such as latency,throughput,etc.)or functionality(such as resiliency,security,etc.).With SDN and NFV technologies,network managers can dynamically schedule and deploy network slice related functions for plug-and-play.In the mobile network,users need to select the appropriate slice to access,and its performance will directly affect whether the user's service quality can be guaranteed,and determine the utilization efficiency of system resources.In the new slice-based mobile network,the user association problem is very different from the problem in the traditional cellular network.The problem of optimal matching between mobile users and access sites in traditional networks has become an optimization matching problem between mobile users,network slices and access sites in the access slice selection problem.Therefore,in this paper,we study the wireless access selection mechanism for end-to-end network slicing from the perspective of optimal matching among users,BSs,and slices.The main research contents are as follows:Based on the slice-based mobile network,this paper adopts a performance index called Satisfaction Degree(SD)as the basis for slice selection.With the goal of maximizing SD in the system,the theoretical optimization model of slice selection is established,and through theoretical analysis,it is proved that it is NP-hard.Furthermore,we use Genetic Algorithm(GA)to approximate the optimization model,and based on this,we design the corresponding slice selection algorithm.Through simulation experiments,the effectiveness of our proposed GA algorithm is verified.The numerical results show that compared with the typical access selection algorithm based on RSS or greedy algorithm in traditional networks,GA algorithm can enable users to obtain better access and transmission performance.In addition,the network state is constantly changing in consideration of the arrival of users and the randomness of access conditions,as well as the mobility of users in the network.Therefore,we need to make user's handoff decision according to the user's movement,as well as the changes of the network environment and service load.The user's slice selection is dynamically adjusted to achieve optimized network transmission performance.In this study,we model the handoff problem of network slices as a Markov decision process(MDP).Applying reinforcement learning theory,we propose an intelligent handoff algorithm based on deep reinforcement learning.Through simulation experiments,the effectiveness of our proposed handoff algorithm is verified.The numerical results show that the algorithm can achieve better transmission performance by increasing a small number of handoffs.Finally,we studied the multi-user handoff problem in the network scenario of dense users.Considering the dynamic state of the network and the competition of multiple users for limited network resources,the access selection of one user will affect the available resources in the network,which in turn affects the access selection results of other users.In this study,we model the handoff problem of multi-user as a multi-person random game problem.Based on the theory of multi-agent Reinforcement Learning(MARL),we propose a distributed multi-agent reinforcement learning algorithm.Through simulation experiments,the effectiveness of our proposed handoff algorithm is verified.The numerical results show that the algorithm can improve the user's transmission rate within a reasonable range of handoff times.