Adaptive Streaming Media QoS/QoE Management Based On Reinforcement Learning

Quality of Service(QoS)and Quality of Experience(QoE)are important metrics to measure the performance of network and service.With the rapid development of short videos,ultra-high-definition videos and other services,adaptive streaming media requires support for rich perceptual experiences,leading to higher consumption of network resources.In this case,adaptive streaming media QoS/QoE management faces two challenges.On the one hand,the existing adaptive bit rate(ABR)algorithms face difficulty in effectively adapting to frequent fluctuations in wireless network bandwidth.This limitation makes them prone to bitrate decision errors,resulting in a decline in user QoE.On the other hand,the best effort service is unable to meet the diverse QoE requirements of users.The core of these two challenges lies in adopting more flexible and intelligent QoS/QoE management schemes.Based on this,this thesis focuses on the distinct QoS/QoE management requirements at the application layer and network layer.It investigates optimization schemes for adaptive streaming media transmission as well as end-to-end differentiated QoE guarantee strategies,and proposes specific solutions to address these issues.The innovation points of this thesis are as follows:1.From the perspective of adapting service to network status,a cross-layer adaptive streaming transmission scheme based on reinforcement learning(RL)is proposed.To address the problem of decreased QoE in adaptive streaming media caused by inaccurate bandwidth prediction and bitrate selection errors,this thesis introduces low-level network information to get accurate bandwidth prediction,combined with deep reinforcement learning(DRL)to make reasonable bit rate decisions,thus achieving optimal end-to-end QoE.Based on the ns3,a simulation of cross-layer adaptive streaming transmission in wireless networks is built to validate and evaluate the objective QoE metrics of the aforementioned scheme.The experimental results show that compared with the best ABR algorithm at this stage,the reinforcement learning-based cross-layer ABR algorithm improves the average QoE by 4.2%.2.From the perspective of network adaptation for business needs,a cross-layer QoS/QoE differentiation management scheme based on reinforcement learning is proposed.In order to guarantee differentiated QoE requirements of users,this thesis provides upper-layer business state information to the underlying network and dynamically adjusts network QoS parameters to meet the target QoE.This scheme achieves end-to-end differentiated QoE assurance for adaptive streaming media.A QoE enforcement algorithm based on DRL is designed to address the highly nonlinear relationship between QoS and QoE,and architecture simulation and performance evaluation are conducted based on ns3.The experimental results indicate that the QoE enforcement algorithm based on DRL can effectively achieve QoE guarantee at different QoE levels.