A method for improving user experience quality of live video system

Abstract

The invention discloses a method for improving the user experience quality of a live video system, and aims to improve the user experience quality in the live video system and reduce time delay whileensuring the video watching quality. Different from existing documents, the method has the advantages that the calculation resources and the wireless spectrum resources which can be obtained are modeled into a random process, and the random process is closer to a real wireless communication environment. A dynamic system is modeled into a Markov decision process, and as the action space and the state space are continuous and multi-dimensional, a traditional reinforcement learning algorithm such as a deep Q learning network and a strategy gradient is relatively low in efficiency and relatively high in difficulty when the problems are processed. Therefore, the invention provides an online Actor-by combining a video stream transcoding strategy, a user scheduling strategy and a resource allocation method. According to the Critic reinforcement learning algorithm, qualification traces are introduced into the Actor part and the Critic part, and the learning process is accelerated. Theoreticalsimulation proves that the performance is obviously better than that of a deep Q learning network, and the convergence speed is faster than that of a strategy gradient algorithm.

Description

technical field [0001] The invention relates to the technical field of the fifth generation wireless communication, in particular to a method for improving user experience quality of a live video system. Background technique [0002] Live video has great application value in rescue, route navigation and entertainment. The video stream can be divided into multiple versions according to the quality. The version of the video transmission is affected by several factors: ① the bandwidth limitation of the wireless channel environment; ② different user preferences; ③ the video format supported by the mobile device. Using video transcoding technology in the downlink can convert live video into several versions with different resolutions. However, there are some problems in the existing live video system. [0003] Problem 1: The burden of the core network is large and the problem of delay. Due to the large amount of calculations involved in video transcoding, the current common pr...

AI Technical Summary

Problems solved by technology

However, there are some problems in the existing live video system

[0003] Problem 1: The burden of the core network and the problem of delay

[0004] Problem 2: Low efficiency of resource optimization

Due to the dynamic nature of wireless networks, it becomes difficult to efficiently utilize all available wireless spectrum resources and computing resources at the edge of the network

[0005] Question 3: The one-sidedness of the research on the Quality of Experience (QoE) function

However, both video quality and latency are very important to users. ①Sacrifice latency for high-quality video. Although the video is high-definition, it often pauses during user experience, which seriously affects the quality of user experience; ②Sacrifice video quality for low latency. Although the video is smooth during user experience, it is not clear, which will seriously affect the quality of user experience.

[0006] Problem 4: Traditional reinforcement learning algorithms have poor performance and low efficiency

However, the problem with the current Actor-Critic algorithm is that the qualification trace is only introduced in the Critic part, and the learning efficiency is low

Images