Research On Edge-assisted Video Streaming Quality Optimization

The main idea of STIN(Satellite-Terrestrial Integrated Network)is to integrate the satellite network and terrestrial network,and it has the advantage of wide service coverage.However,due to the limited bandwidth and expensive price of satellite communication,it is difficult and not economical to provide high-resolution video streaming service with high QoE(Quality of Experience).Thus,the key problem is how to provide high-quality video service to users while saving the bandwidth of the satellite communication link.To solve this problem,this paper introduces the MEC(Mobile Edge Computing)technology to the STIN,and research on the edge-assisted SR(Super-Resolution)-based video streaming optimization.More specifically,we first propose and develop an SR-based video streaming system in STIN.Then,to solve the problem in proposed system,we design a Reinforcement Learning(RL)-based adaption algorithm for SR-based video streaming.The main work of this paper includes:(1)The SR-based video streaming system in STINDue to the limited bandwidth and expensive price of satellite backhaul link,an edge-assisted SR-based video streaming system is designed and realized.There are two challenges during developing the system—the compatibility and the real-time SR processing.In order to guarantee the architecture is compatible with the conventional MPEG-DASH standard,we design the video request procedure based on the reverse proxy technology,so the SR-based MEC server can be deployed conveniently without any modifications for the client side and video server side;To realize the real-time SR-based video streaming,we use multi-process and multi-GPU method,which can let users watch the SR video without rebuffering.Finally,a real-world system is built and tested.The test results show that our system can realize the function of SR-based video streaming and have high performance in video quality and super-resolution processing speed.(2)RL-based adaption algorithm for SR-based video streamingBased on the proposed SR-based video streaming system,to improve the bad QoE caused by the dynamic RAN(Radio Access Network)bandwidth and heterogeneous computational capacity of MEC servers,this paper proposes a RL-based adaption algorithm for SR-based video streaming.To handle the tradeoff between the high operational efficiency and low decision-execution delay,a pipelined SR processing mechanism is presented.To solve the QoE optimization problem,this paper proposed a deep RL-based adaption algorithm for SR-based video streaming.The algorithm can dynamically choose a proper SR reconstruction scale factor to get optimal QoE.The simulation results shows that the proposed scheme outperforms baseline schemes by 13%-63%in terms of QoE metric.