Video Transmission Algorithm With Model Predictive Control Optimization

With the development of technology,people’s demand for high quality video is grad-ually increasing.However,video streaming transmission is highly sensitive to network,client and server states,and is susceptible to conditions such as transmission link loss,network latency,and drastic bandwidth fluctuations,which in turn affect the bitrate and user’s quality of experience.In the current video transmission process,the Adaptive Bitrate(ABR)algorithm is generally used to optimize the Quality of Experience(QoE).Video is typically encoded on the server side into several alternative compression versions,each video chunk repre-senting the original video but with different quality,target bitrate or resolution.The client can choose a different quality bitrate depending on the state of the network at the time.The adaptive bitrate algorithm will switch to different quality video encoding versions halfway through when the client needs it,improving the user’s QoE.The ABR algorithm breaks the video into chunks,typically lasting 2-6 seconds each,and encodes each ver-sion of each video chunk independently,so that existing video chunks can be replaced without accessing any other blocks.With the ABR algorithm,the client can switch the next chunk to a different quality when one block is about to finish playing.The different replacements are often referred to as different ”bitrates”.By switching video chunks,the video bitrate adaption is completed.This paper analyzes how to use model predictive control and Bayesian neural net-works to accomplish bitrate adaptation in streaming video transmission.There are two main innovations in the paper as follows:1.We propose a pre-computed model predictive control algorithm that converts the problem of choosing the optimal video bitrate into a stochastic optimal control problem under restricted conditions.The algorithm improves the robustness of traditional bitrate adaptive algorithm.For each possible network state,the optimal parameter configuration for a given ABR algorithm is calculated offline in advance,reducing the frequency of playback stutter events.Experiments demonstrate the superiority of the method,with an overall QoE improvement of at least 3.2%.2.We design a Bayesian neural network-based throughput predictor,which im-proves the traditional throughput point estimation model and uses confidence intervals to assess the accuracy of throughput prediction.The prediction accuracy of the Bayesian neural network-based throughput prediction model can be perceived by the prediction uncertainty.The algorithm reduces user’s QoE fluctuations due to network environment changes.The new throughput predictor is experimentally proven to be effective in im-proving user’s QoE,with an overall QoE improvement of at least 5.2%.