Joint User Association And Power Allocation For Hybrid VLC/RF Networks

With the development of related technologies in the fields of internet and communication,the global data volume is growing explosively due to the increasing demand of users.However,compared with the continuous growth of data volume,the available spectrum resources are very limited,and traditional Radio Frequency(RF)communication technology is facing increasingly serious resource bottlenecks and interference problems.In this situation,Visible Light Communication(VLC)emerged due to its various performance advantages and is considered a powerful supplement to traditional RF communication technology.In the VLC/RF heterogeneous network consisting of these two technologies,both technical advantages can be fully utilized to overcome RF spectrum resource limitations,provide high-speed and reliable data transmission,and provide good indoor coverage with broad indoor application prospects.Specifically,in VLC/RF heterogeneous networks,due to the unique properties of VLC technology,interference between users is more severe compared to RF networks.To improve network performance and user experience,optimization of user association and power allocation is required.User association specifies which AP each user is connected to,while power allocation assigns VLC and RF resources,and the two are interdependent and mutually influential.This dissertation discusses the joint solution of user association and power allocation problems in the network,and designs algorithms from multiple aspects such as user access,channel state,and cell interference to maximize network performance and improve user experience.The main contents and research results are summarized as follows.Firstly,this dissertation selects and models the indoor network architecture of the VLC/RF heterogeneous network,including the system composition of the network,the working principle and channel transmission characteristics of LED,and the channel transmission characteristics of RF.Simulation results show the illumination distribution of indoor VLC,and then the signal-to-noise ratio and rate distribution of VLC and RF on the indoor reception plane are simulated respectively under the condition of satisfying the indoor illumination standard,which verifies the performance advantages of the two technologies and provides network parameters and evaluation criteria for the algorithm design in the following sections.Then,this dissertation proposes an algorithm based on unsupervised deep neural networks(DNNs)to solve the problem of joint user association and power allocation in VLC/RF heterogeneous networks.The algorithm combines multiple neural layers and trains large datasets to learn the mapping functions of user association and power allocation.The neural network takes channel state information as input and outputs the associated access points for each user and the power allocation for each access point,with the goal of maximizing the network’s total rate.Simulation results show the convergence characteristics of the algorithm and the variation of the algorithm under different numbers of users and room sizes.Finally,this dissertation proposes a Multiagent Deep Reinforcement Learning(MDRL)algorithm to jointly solve the user association and power allocation problem in VLC/RF heterogeneous networks.While the DNN algorithm heavily relies on the quantity and quality of training data and has poor generalization capability,the MDRL algorithm can learn through simulation and trial-anderror,requiring less data and allowing for flexible adjustments and optimizations as network complexity increases.The proposed algorithm adopts centralized training with decentralized decision-making,where a central controller merges the state and action spaces of each agent and computes the global reward during the learning process,enabling the iterative learning of the best policy through multiple interactions between users and the environment to maximize the overall system throughput.Simulation results demonstrate the learning process of the algorithm and validate its superior performance over comparison algorithms in varying numbers of users and room sizes.