Research On Platform Independent Adaptive Streaming Media Transmission Based On Reinforcement Learning

With the rapid development of the Internet industry and the continuous improvement of network bandwidth,streaming media services are increasingly welcomed by Internet users.How to make the user community get better service is the development goal of streaming media technology.Because the available bandwidth of the network is limited or unstable,the downloading and playing process of the streaming media needs to be adaptively controlled according to the requirements of the network environment and the users themselves,so as to ensure that the user can obtain the maximum quality of experience(QoE).This paper studies the dynamic adaptive streaming media technology based on HTTP and uses the method of deep reinforcement learning to optimize the QoE problem.The contributions of this paper mainly include:(1)Considering the influence of streaming media slice quality,rate switching smoothness and interruption time to construct QoE problem model,this paper uses deep reinforcement learning method to avoid the problem of inaccurate prediction,taking into account the impact of the current code rate selection strategy on the future,and obtaining the optimal code rate selection strategy for a long period of time.(2)Considering the scenario where the CDN and P2 P hybrid network structure is used as the streaming media server,the QoE optimization problem modeling between the client and multiple servers is improved in this scenario,taking into account the delay caused by the switch between client and multiple node servers.(3)Considering the deployment problem of deep neural network with limited computing and storage resources,the deep neural network pruning technology is studied,and a weighted redundancy approximation calculation method is proposed,which guarantees the output accuracy.Minimizing the number of neural network weights can greatly reduce the computational overhead and storage overhead of the neural network model.This paper implements a platform-independent adaptive streaming media player based on the open source streaming media playback framework.The player combines the abovementioned reinforcement learning and neural network pruning technology to adaptively select streaming media server and the slice rate on resource-constrained clients,greatly improving the quality of the user experience.