MULTIPLE DESCRIPTION CODING FOR INTERNET VIDEO STREAMING Manuela Pereira * , Marc Antonini, Michel Barlaud I3S laboratory of CNRS, University of Nice-Sophia Antipolis Bˆ atiment Algorithme/Euclides, 2000 route des Lucioles - 06903 Sophia Antipolis Cedex, France {pereira, am, barlaud}@i3s.unice.fr ABSTRACT We present a system for video streaming well adapted to the unpredictable and varying nature of Internet. The proposed system uses a superposition of several Multiple Description Coding (MDC) schemes, each with N =2 descriptions, to reach rate scalability and adaptability to varying chan- nel conditions. Each MDC (N =2 descriptions), that we will call base MDC has a bit rate and a redundancy asso- ciated. The superposition of several base MDC results on a MDC scheme for N> 2 descriptions with different bit rates, allowing rate scalability, for different redundancies. The proposed scheme is well adapted to varying channel conditions. In the proposed method multiple descriptions are gener- ated by the coder and downloaded in the server, leaving to the server the only task to choose sending out the right de- scription at the right time depending of channel conditions (bandwidth and loss rate). 1. INTRODUCTION The use of video streaming over Internet knew an enormous increase in the past few years being the design of Internet video streaming a challenging task due to the unpredictable and varying nature of network conditions. Internet only of- fer best-effort service, so there is no guarantees on: band- width, delay jitter or loss rates. So, the video streaming system must be able to adapt video rate to available band- width and must be robust to packet losses. Conventional ap- proaches for dealing with packet loss for static data, such as retransmission may not be possible in streaming context due to real time nature of the content. Thus, additional mecha- nism are needed to provide streaming media delivery over packet networks. In streaming video the client performs a demand to a server that transmits media packets over a network that serves fairly several clients. The server can implement intelligent transport mechanisms, by sending out the right packets at * On leave from Universidade da Beira Interior, Portugal. Research par- tially supported by PRAXIS XXI grant SFRH/BD/1234/2000 the right time, but the amount of computation that it can per- form for each media stream is very limited due to the large number of stream to be served simultaneously. Then, the task to compress video signal is left to the encoder. There- fore this task as to be done without the a priori knowledge of the channel conditions (bandwidth and loss rates). This is why representations that allow rate scalability must be adapted to varying network throughput without requiring computation at the media server. Multiple redundant rep- resentations are an easy way to achieve this task and will be used in the present work. 1.1. Prior work on video streaming Almost all present approaches use Forward Error Correc- tion (FEC) and/or Automatic Repeat reQuest (ARQ). For instance, in [1] they use a two-layer scalable video coder combined with unequal error protection. In [2] they use a hybrid FEC/ARQ approach known as incremental redun- dancy. In this work they compute the rate distortion op- timized transmission policy using the Iterative Sensitivity Adjustment (ISA) algorithm introduced in [3]. The ISA algorithm involves estimation of the probability that a sin- gle packet will be communicated in a error as a function of the expected redundancy, or cost, used to communicate the packet. The [4] work is an extension of the work in [3]. It in- corporates models for packet path diversity. Also in [5] the authors exploits path diversity. In [6] is used a media layer representation for transmission over current heterogeneous networks. The authors propose a framework for scalable streaming media delivery, that involves a scheduling algo- rithm called expected runtime Distortion Based Scheduling (EDBS) which decides the order in which packets should be transmitted in order to improve client playback quality in the presence of channel losses. In [7] they use adaptive media playout to reduce the delay introduced by the client buffer. They consider retransmission of lost media packets. To minimize retransmission in [8] the authors use pre-stored representations of certain frames at the server such that the chosen representation only uses previous frames, as refer- ence, received with very hight probability. 0-7803-7750-8/03/$17.00 ©2003 IEEE. ICIP 2003