Intelligent Resource Scheduling In The Ultra-Dense Networks

The ever-increasing traffic has forced the fifth-generation(5G)mobile communication system to be ultra-dense networks(UDNs).Driven by emerging services,resource scheduling encounters novel challenges.On one hand,due to the proliferation of base stations and users,large-scale and highly-complex resource scheduling optimization problem always leads to a nearly-optimal solution instead of a truly-optimal one.On the other hand,due to diverse services,traditional resource scheduling schemes,which are specific for partial traffic types,cannot well satisfy users' quality-of-service(Qo S)demands in terms of reliability,throughput,and delay,etc..In addition,current resource scheduling is separately achieved by user scheduling module and resource allocation module,and usually focuses on optimizing a single module,which obtain sub-optimality instead of global-optimality and cause gain loss because of the mismatch between modules.Therefore,this thesis studies intelligent resource scheduling in the UDNs and introduces intelligence to solve the above issues.Firstly,this thesis proposes an integrated and intelligent resource scheduling framework.Given the current resource scheduling process,the proposed framework can intelligently model and solve resource scheduling problem in the UDNs,realizing automatic radio resource management.In addition,the proposed framework integrates user scheduling and resource allocation,by which the gain loss caused by module separation can be reduced.Secondly,based on the proposed framework,this thesis proposes an integrated and intelligent resource scheduling(IIRS)algorithm.IIRS algorithm is achieved by deep reinforcement learning(DRL)and considers users' heterogeneous Qo S demands,aiming at maximizing user satisfaction.User satisfaction is defined as a function of multiple Qo S metrics such as guaranteed bit rate(GBR),delay,and packet loss rate(PLR)to flexibly meet Qo S requirements.Further,to deal with the problems of slow convergence and poor performance of IIRS algorithm,this thesis proposes an enhanced IIRS(E-IIRS)algorithm.By introducing main network(Main Net)and target network(Target Net),experience replay and prioritized sweep,and heuristic mechanism,the convergence rate can be accelerated and the performance can be promoted.Finally,this thesis simulates the proposed algorithms and the existing algorithms in a system-level platform constructed by Python and Simpy.When low traffic,IIRS algorithm achieves the dissatisfaction reduction ratio about 91% compared with Q algorithm;E-IIRS algorithm achieves the dissatisfaction reduction ratio about 87%?10%,and 32% compared with IIRS algorithm,non-integrated resource scheduling(NIRS)algorithm,and max channel/interference(Max C/I)algorithm.Therefore,IIRS algorithm and E-IIRS algorithm can achieve efficient and intelligent resource scheduling in the UDNs for better user experience.