A Network-Adapative SVC Streaming Strategy with SVM-Based Bandwidth Prediction

Abstract

This paper presents an online streaming strategy for SVC (Scalable Video Coding).The crucial technique adopted is the combination of network bandwidth prediction and rating mechanism for different streaming options. An SVM (Support Vector Machine) algorithm is used for prediction of network bandwidth variation, with which the system then evaluates every strategy that can be made at that time. The experiment shows this strategy not only cut down video playing back interruption times, but also improves overall video quality users perceive. Index Terms—Bandwidth prediction, support vector machine, SVC streaming. I. I NTRODUCTION The techniques for video content delivery have evolved from UDP (User Datagram Protocol) to HTTP (Hypertext Transfer Protocol) based methods. HTTP based solutions have many advantages including easy deployment and better utilization of web cache infrastructure. Furthermore, they are Firewall and NAT (Network Address Translate) friendly. Thus many mainstream video providers, such as YouTube, deliver contents via HTTP. The providers also allow users to choose from multiple versions with different quality. To do this, videos will be encoded into different and independent files with much redundant information. So space waste is one shortcoming in mainstream approaches. Another defect is although the mechanism for choosing different video versions is provided; users are not usually making the right choice and suffer from playback interruption or low quality of downloaded videos.