Multi-way Multicast Speech for Multimedia Conferencing over Heterogenous Shared Packet Networks (RAT - Robust Audio Tool) EPSRC MNA Project #GR/K772780
Principal investigator: Dr. Vicky Hardman, Co-investigator: Dr. Angela Sasse, both of the Department of Computer Science, University College London.
This document describes the background to, and the research done by the EPSRC RAT project in the department of Computer Science, University College London.
The aim of this research was to provide multi-way multicast speech encoding and transmission for use in multimedia conferencing over heterogeneous shared packet networks, such as the Mbone (multicast backbone for the Internet). A further pre-requisite was that the system be developed for a range of heterogeneous platforms, with no extra specialist hardware, since this would enable most users who already have suitable network connection to benefit from this research.
Despite current efforts to improve the underlying infrastructure of networks, the growth in the use of shared packet networks (especially the most successful - the Internet) is exponential, and existing performance problems (such as packet loss) have persisted to date. Shared packet multicast networks are however ideal for multimedia systems, because multi-way transmission and integration of services are easier than with a network based on a conventional circuit-switched model.
Audio transmission over packet networks is currently the only truly real-time service; video is still essentially slow-scan. Speech is the most important component of a multimedia conferencing system to users, and in order to provide the requisite quality, low end-to-end delay, and packet loss protection must be catered for.
The research in the RAT project is novel because the combination of the shared heterogeneous packet network (high packet loss, with inconsistent network support for voice across the network), and the heterogeneous platform with no need for a real time multi-tasking operating system make this a new problem, where existing research is not directly applicable.
Research Undertaken in the RAT Project
The research objectives of the RAT project and their outcome are:
1. Implementation of a publicly available audio tool for heterogeneous multicast packet networks, and heterogeneous multi-tasking operating system platforms.
RAT is used internationally, and widely recognised to be one of the best audio tools. It also used as a research platform by BT, HP, Nortel, CERN and many academic institutions.
2. Development and assessment of scalable redundancy for the reconstruction of multi-way speech over multicast packet networks.
UCL CS is one of the original proponents of redundancy for packet loss repair, which has led to an Internet draft standard in its own right, as well as numerous papers.Our work has also strongly influenced other IETF standards. Colin Perkins, the RA on the RAT project is now co-chair of the (Audio and Video Transport) AVT group of the IETF.
3. Development and implementation of speech coding algorithms that adapt to both the varying network conditions, and the processor load.
We developed an adaptive algorithm to intelligently protect RAT from OS servicing anomalies. Our research showed us that a complexity adaptive version of a speech processing algorithm would not be as important in this area as one which dealt in a more intelligent way with heterogeneity in bandwidth and loss rates. We developed a mechanism to do this, called Self-Organised Transcoders.
4. Development and implementation of synchronisation and co-ordination services for multimedia applications on generally available multi-tasking operating systems.
The Multicast Bus (MBUS) was developed in co-ordination between UCL CS and Bremen University. We hope this work will become an IETF standard.The lip-sync application that uses the conference bus was developed by UCL CS.
The RAT software is available from:
A list of the publications and full report are available from:
Document converted from word 8 by
MSWordView (mswordview 0.5.1)
MSWordView written by Caolan McNamara