Robust Low-Latency Voice and Video Communication
over Best-Effort Networks

Yi Liang
Information Systems Laboratory
Department of Electrical Engineering

Wednesday, March 12, 2003
1:00 PM, CIS-X Auditorium
(Refreshments at 12:45 PM)

Abstract

The quality of service limitation of today's best-effort networks poses a major challenge for low-latency multimedia communication. Excessive delay, packet loss, variations in throughput, and high delay jitter all impair the performance of communication. In this work, we address all these challenges from the client side, the transport, and source coding.

On the client side, a new playout scheduling scheme is proposed to improve the trade-off between buffering delay and packet loss for real-time voice communication. In this scheme the network delay is estimated from past statistics and the playout time of the media packet is adaptively adjusted in a highly dynamic way. Proper reconstruction of continuous output audio is achieved by scaling individual media packets using a time-scale modification technique based on the Waveform Similarity Overlap-Add algorithm. Results of subjective listening tests show that this operation does not impair audio quality. Simulation results based on Internet measurements show that the trade-off between buffering delay and late loss can be significantly improved.

On the transport, we further improve the performance of communication by exploiting diversity of multiple transmission channels, where the source media are coded into multiple redundant streams and sent over independent network paths. The playout time of the received media packets is determined based on an adaptive multi-stream playout scheduling technique that uses a Lagrangian cost function to trade off delay and loss. Experiments demonstrate further gains of reduced latency and distortion, resulting from path diversity.

Compared to audio, compressed video exhibits much stronger dependency across packets. Today's Internet video streaming systems employ buffering and retransmission to guarantee the correct reception of each packet, which leads to high latency in media delivery. In this work, we present an efficient low-latency streaming system that does not require retransmission of lost packets. A source coding scheme that dynamically manages the dependency across packets is proposed using optimal reference picture selection. The optimal selection of the reference is achieved within a rate-distortion framework and is adapted to the channel conditions, which minimizes the expected end-to-end distortion under the given rate constraint. An improved trade-off between compression efficiency and error-resilience has been achieved. The increased error-resilience eliminates the need of retransmission, which makes it possible to reduce the delivery latency from 5-15 seconds to a few hundred milliseconds, with similar video quality maintained.