Research On Intelligent Resource Allocation Technology For 5G Optical-wireless Converged Access Network

To support dense access and diverse application scenarios,improve user service experience,and meet high-capacity high-speed access,optical-wireless converged access networks have become one of the important research directions.Meanwhile,with the emerging services of interactive and secure real-time communication needs,the access network will continue to evolve toward intelligence.By introducing machine learning to optimize resources and secure communication for multi-service scenarios in the access network,it can reduce computational complexity and better adapt to the collaborative provisioning of diverse heterogeneous resources in time-varying scenarios.Therefore,this thesis focuses on intelligent resource scheduling and secure communication methods in Device-to-Device(D2D)scenarios and Intelligent Reflective Surface(IRS)assisted Non-Orthogonal Multiple-Access(NOMA)scenarios for optical-wireless converged access networks.For the D2 D scenario,millimeter wave in non-line-of-sight(NLo S)channel leads to serious fading problem,this thesis proposed a hybrid band transmission scheme based on Convolutional Neural Network(CNN).The scheme builds and trains a suitable CNN model to distinguish the D2 D link as Lo S/NLo S by the power angle spectrum of the received signal.If the D2 D link belongs to the Lo S,the millimeter wave band with high gain and beam concentration is used for transmission,and vice versa,the microwave with lower absorption loss is used.The results show that the proposed scheme can flexibly adapt to the link transformation and combine the advantages of both millimeter wave and microwave bands,which can effectively improve the signal-to-noise ratio compared with the single-band transmission and becomes more prominent as the D2 D distance increases.For the IRS-assisted NOMA scenario,the wireless access network users are vulnerable to eavesdropping security risks.This thesis proposed a machine learningbased interference management approach to enhance secure communication.This scheme shields secure user signals from eavesdropping by maximizing inter-user interference.The optimization process of the proposed scheme involves a non-convex optimization problem,and this paper proposes to construct a two-step machine learning-based network that uses the original channel as input to optimize the phase shift value of the IRS,and the output is then used to optimize the beamforming matrix of the base station.The results show that the proposed scheme can effectively reduce the eavesdropping rate to achieve secure communication,and effectively reduce the computation time,which provides an important reference for the real-time configuration of the beamforming matrix.