Design And Implementation For An Adaptive Live Streaming System

Live streaming applications are becoming “killer” applications in recent years with the improvement of the Internet infrastructure and the wide spread of mobile devices.The upcoming challenge is how to provide users with good QoE(quality of experience)with respect to network environment and terminal devices changes.Adaptive streaming is designed to automatically adjust the transmitted stream to make the most of available bandwidth to provide users with stream of the highest quality,while maintaining playout fluency.This thesis takes a thorough research on adaptation algorithms and finds that most of these algorithms are based on bandwidth estimations.The problem is that bandwidth estimation is inaccurate when network jitters and should not be considered the only factor for adaptation decisions.This thesis proposes a new algorithm that combines bandwidth estimation technique and buffer level variation technique together.The main contribution of this thesis include the following aspects:(1)Researches are taken on popular adaptation algorithms.Bandwidth is considered the primary factor for adaptive streaming,and is adopted by most adaptation algorithms.But bandwidth estimation is inaccurate especially when network jitters.Client buffer level variation can reflect bandwidth drops but not bandwidth improvement.Other techniques like rate adaptation and selective frame drops are lack of scalability.(2)Researches are taken on live streaming protocols and playout techniques.RTSP is mostly adopted in time-delay sensitive area.But RTSP-based applications require NAT.RTMP depends on Adobe platforms.HTTP-based applications is becoming a trend due to its simplicity and scalability and the emergence of HTML5.(3)Design and implement a live streaming application with HTML5 and Media Source Extensions based on DASH standard.This application requires that streaming server transcodes original stream into multi-bit rate streams and transmits data via HTTP protocol.Our new algorithm is tested under this application.Experiments show that our application can work properly,and our new algorithm can increase the fluency of playout.It can also decrease the need for frequent bandwidth estimation compared with Adobe HDS and other bandwidth estimation based applications.