Multi-stream and Multi-path Audio Transmission
==============================================

Yi Liang

Proposal
--------

The major challenge for real time voice communication has been meeting the
strict quality requirements when transmitting data over packet networks,
which provide services on a best-effort basis. In this project we propose
using multi-stream to transmit real-time audio data over IP networks.

The idea is based on the fact that audio streams usually consume very little
bandwidth compared to other types of traffic, while the real-time nature of
the communication imposes stringent requirements on the end-to-end quality
of delivery, such as low latency, low loss rate and reasonable variation of
the playout rate.

Under this scheme, audio is encoded and sent in at least two streams, with
certain amount of redundancy and/or overlap across packets in different
streams. In case of packet erasure, lost audio can be recovered from the
redundant information carried by other good packets. Audio packets can also
be encoded scalably using multiple description coding (MDC) [1][2].
In case of packet loss, the quality of received audio can be degraded 
gracefully instead of losing the packets completely.

We plan to compare the overlapped multi-stream scheme with FEC, an error
protection scheme which is well studied [3][4]. We are
going to compare these two schemes in the aspects of loss recovery
capability, delay reduction, and the scalability of speech quality.

Transmitting multiple audio streams using path diversity provides even
stronger protection against packet erasure and improves the QoS in many
ways. Packets from source to destination are sent explicitly over different
links via relay servers at different geographical locations. Benefit from
multi-path includes averaged network behavior, isolated burst loss and
smaller outage probability [5]. Most importantly, by selecting and
switching to the optimal path dynamically, the end-to-end delay can be
reduced compared to using either path, which is quite significant for
real-time communication.

With path diversity, we receive and play out the packets from the link with
the lowest cost (e.g. delay). While this selection is updated dynamically as
the network condition changes. This process also reduces the variation of
the playout rate. The seamless switching between streams can be realized by
adaptive playout scheduling technique [6].

In this work, we plan to build the network topology simulating typical
Internet topology in ns, and study packet loss and delay behavior. On this
test bed, the performance of multi-stream vs. FEC can be compared under the
same conditions. Another major task of the study based on this simulation
scenario is to find out how much gain, in terms of QoS parameters, we can
obtain from the multi-stream and multi-path schemes.

In order to use path diversity, we also plan to design an adaptive playout
scheduling algorithm which is suitable for playing out multi-path streams.

Also, multi-path measurement over the Internet [7] using relay servers 
in different geographical locations is on the schedule.

Work plan
---------

Week 1: 
- build the topology model in ns
- study what topology simulates the situation in the real world

Week 2: 
- continue with ns simulation 
- modeling of the simulation results.

Week 3: 
- adaptive playout scheduling algorithm 
- MDC

(some work in weeks 1-3 will be carried out in parallel)

Week 4: 
- compare the performance of multi-stream and FEC.

Week 5: 
- analyze results, conclude and write reports.

References
----------

[1] Wenqing Jiang and A. Ortega,
 ``Multiple description speech coding for robust communication over
  lossy packet networks,''
 in International Conference on Multimedia and Expo, Aug. 2000,
  vol. 1, pp. 444--7,
 New York, NY, USA.

[2] G. Kubin and W.B. Kleijn,
 ``Multiple-description coding (MDC) of speech with an invertible
  auditory model,''
 in 1999 IEEE Workshop on Speech Coding Proceedings, June 1999,
  pp. 81--3,
 Porvoo, Finland.

[3] J.-C. Bolot and A Vega-Garcia,
 ``Control mechanisms for packet audio in the internet,''
 in Proceedings of IEEE INFOCOM '96, Mar. 1996, vol. 1, pp.
  232--9.

[4] J-C. Bolot, S. Fosse-Parisis, and D. Towsley,
 ``Adaptive FEC-based error control for internet telephony,''
 in Proceedings of IEEE INFOCOM '99, Mar. 1999, vol. 3, pp.
  1453--1460.

[5] John G. Apostolopoulos,
 ``Reliable video communication over loss packet networks using
  multiple state encoding and path diversity,''
 in Proceedings visual communication and image processing, Jan.
  2001,
 To appear.

[6] Yi J. Liang, Nikolaus Faerber, and Bernd Girod,
 ``Adaptive playout scheduling using time-scale modification in packet
  voice communications,''
 in 2001 IEEE International Conference on Acoustics, Speech, and
  Signal Processing. Proceedings ICASSP01, May 2001,
 To appear.

[7] S. Savage, A. Collins, E. Hoffman, J. Snell, and T. Anderson,
 ``The end-to-end effects of internet path selection,''
 in Proceedings of the ACM SIGCOMM, Oct. 1999, vol. 29, pp.
  289--99.