Rate-distortion optimized packet scheduling for video streaming

Abstract

Internet video streaming places new demands on source coding and network transport algorithms. The challenge is to deliver compressed video packets before their play-out deadline, despite of varying throughput, packet delay, and loss. This problem has to be solved in a way that simultaneously maximizes the video quality at the streaming client, meets transmission rate limitations, and satisfies latency constraints. The most significant recent advance in streaming technology is the emergence of rate-distortion optimized streaming techniques that take into account packet importance and knowledge about the channel in a Lagrange rate-distortion cost function J = D + lambdaR. The research in this thesis aims to overcome the current limitations in rate-distortion optimized packet scheduling, such that these techniques can be applied to a broader range of streaming problems, and a better end-to-end performance can be achieved. In particular, first we develop a new advanced framework for rate-distortion optimized video streaming which incorporates for the first time a general description scheme for the rate-distortion characteristics of packetized compressed video in the event of lost or omitted packets, and an iterative algorithm for computing optimal packet schedules in the presence of error concealment. Then, we derive instances of our framework for several advanced video streaming architectures, applications, and coding techniques, some of which could not be previously combined with rate-distortion optimized streaming due to its initial limitations. These include streaming over multiple network paths or from multiple media servers, streaming from an intermediate network proxy, and streaming with rich acknowledgements. Finally, we assess the performance of our framework through series of systematic experiments in all of these scenarios. The experimental results are accurately predicted through analysis. In addition, we evaluate the performance of rate-distortion optimized packet scheduling for streaming over network traces of packet losses and packet delays collected in the Internet.