Research On Intelligent Video Transmission Algorithms

Video has played an important role in people's daily life,with the rapid development of the 4G/5G technology,more and more video services emerge for generous profits.At the meanwhile,people always prefer low-latency and high-quality videos,which result in heavy traffic load to the Internet.The fluctuant network conditions make it difficult to guarantee the latency and quality requirements simultaneously for users.In conclusion,video transmission is facing these challenges: First,it is difficult to guarantee the high-quality and low latency requirements simultaneously under the unstable network conditions.Usually,we attend to one and lose another between bitrate and latency,or sacrificing the fluency of video playback.Second,video transmission suffers poor performance because of the network congestion.The interactive video is sensitive to the latency,however,as the multi-entity routing protocols,the underlay routing path is not always optimal.What's more,more and more videos delivered through the Internet take heavy traffic load on the infrastructures,and also cause huge data cost.To this end,in this article,we focus on the transmission process from the sender to the receiver,and propose some intelligent improvement approaches based on learning methods,including the rate control in the sender,the adaptive streaming in the receiver,the learning-based overlay routing strategy and the popularity-based adaptive caching algorithm.The main contributions of this work are as follows:1.End-to-end bitrate control strategies based on deep reinforcement learning.This article proposes a DRL-based rate control model,namely smart RC in the sender,which can improve 20.88% of Qo S while guaranteeing the lower latency compared with the baseline;Besides,a DRL-based adaptive bitrate algorithm for real-time streaming video is proposed,called Adaptive Streaming,which can reduce 64.72%of halt events and 91.04% of halt hold time,respectively,compared with the widely used buffer-based method.What's more,our streamlined DRL models can save62.63% inference time on mobile devices compared with the RNN-based model.2.Learning-based overlay routing.This article proposes a regret-based overlay routing algorithm COR which could improve 5% to 43% of satisfied users' ratio.What's more,we use deep neural network to predict the internet traffic and choose the best overlay path,experiment shows that the proposed model can achieve an accuracy of 90% for the overlay routing decision.3.The popularity-based adaptive caching algorithm ATW for the CDN is proposed.Based on the analysis on the mass real data set,the proposed method could save 1%to 3% of data traffic consumption,and 37.5% of the cache capacity with the same hit ratio,compared with the widely used LRU method in practice.