Adaptive, Best-Effort, Delivery of Live Audio and Video Accr

This videotape is a demonstration of a transport protocol developed by the authors for the transmission of live audio and video streams. The goal of this work has been to understand the complexity of supporting applications such as desktop video conferenci

ADAPTIVE, BEST-EFFORT DELIVERY OF LIVE AUDIO AND VIDEO

ACROSS PACKET-SWITCHED NETWORKS

Video Abstract

Kevin JeffayDonald L. Stone

University of North Carolina at Chapel Hill

Department of Computer ScienceChapel Hill, NC 27599-3175{jeffay,stone}@cs.unc.edu

INTRODUCTION

This videotape is a demonstration of a transport protocol devel-oped by the authors for the transmission of live audio and videostreams. The goal of this work has been to understand the com-plexity of supporting applications such as desktop video confer-encing when the network does not support real-time communica-tion. We believe this problem is important because such networkswill likely exist for the foreseeable future, hence the problemsaddressed by this work are fundamental in delivering continuousmedia in real-time across the “last mile” to the desktop.

Our protocol is a “best effort” protocol that attempts to amelioratethe effect of three basic phenomena: jitter, congestion, and packetloss, to provide low latency, synchronized audio and video com-munications [3]. This goal is realized through four transport anddisplay mechanisms, and a real-time implementation of thesemechanisms that integrates operating system services (e.g.,scheduling and resource allocation, and device management) withnetwork communication services (e.g., transport protocols), andwith application code (e.g., display routines). The fourmechanisms are: a facility for varying synchronization betweenaudio and video to achieve continuous audio in the face of jitter, anetwork congestion monitoring mechanism that is used to controlmedia latency, a queueingmechanism at the senderthat is used to maximizethroughput with out un-necessarily increasing la-tency, and a forward errorcorrection mechanism fortransmitting audio framesmultiple times to amelio-rate the effects of packetloss in the network.A key difficulty in evalu-ating our work has beenthe lack of metrics forcomparing two given me-dia transmission and play-out scenarios. For exam-ple, performance mea-sures such as end-to-endlatency, frame transmis-sion and playout rates,gap-rates, intermedia syn-chronization differential,etc., are relatively easy tocompute, but difficult torelate. For example, if

scheme A results in lower end-to-end latency than scheme B, butB provides a lower gap-rate than A, which has performed better?We do not provide any answers to this dilemma. Instead, wesimply demonstrate, through the use of our protocol, thequalitative effects of varying and trading off performanceparameters such as lip synchronization and gap-rate.

This videotape attempts to (1) demonstrate the quality of theaudio/video streams delivered via our protocol on congestednetworks, and (2) give viewers a qualitative feel for the effects ofvarying various so-called quality-of-service parameters such asnumber of discontinuities (e.g., gap rate), end-to-end latency, lipsync, and throughput.

DESCRIPTION OF VIDEOTAPE

Three demonstrations of transmitting digital audio and videoacross interconnected local-area networks are presented. The firstillustrates the latency inherent in our video conferencing system.End-to-end latency is one of the most important performanceparameters for a videoconferencing system as latency can severelyimpair and impede interaction between conference participants [2,8]. At present there is some agreement that an end-to-end latency

of no more than 250 ms.is acceptable [1]. In thebest case, our system iscapable of delivering syn-chronized audio and videostreams with an end-to-end latency of approxi-mately 170 ms. In thefirst demonstration weillustrate the effect of thislatency by comparing oursystem with an analogconferencing system(with no latency). Weshow a split screen withanalog video in one halfand digital video in theother half (Figure 1). Thedigital video is shownafter having beenacquired by a workstation,compressed, transmittedover an idle network,received by a secondworkstation, decomp

Figure 1: Demonstration of latency differential betweenressed, and displayed. Itanalog (upper left) and digital (lower right) systems.

In: Proc. of the Second ACM International Conference on Multimedia, San Francisco, CA, October 1994, ACM Press, pp. 487-488.

This videotape is a demonstration of a transport protocol developed by the authors for the transmission of live audio and video streams. The goal of this work has been to understand the complexity of supporting applications such as desktop video conferenci

takes approximately 170 ms. for a video frame to propagate fromthe camera to the display [3].

The second demonstration illustrates the effect of varying thesynchronization between the audio and video streams. Asdescribed in [3], a useful technique for ameliorating the effect ofnetwork congestion is to purposely play audio and video out ofexact synchronization; specifically, to play audio frames ahead (intime) of their corresponding video frames. Although thistechnique has proved effective in improving quantitative measuresof video conference performance, playing audio “ahead” of videois unnatural. In nature the speed of sound is several orders ofmagnitude slower than the speed of light and hence whenever weview noise-emitting scenes from a distance, we perceive the visualinformation before the corresponding sonic information. Humansare therefore more tolerant of audio “behind” video. Our systemassumes (somewhat arbitrarily although motivated by [7]) thatusers will tolerate a synchronization differential of at least 100 ms.The second demonstrations varies the degree to which audio isplayed ahead of vide …… 此处隐藏：8216字，全部文档内容请下载后查看。喜欢就下载吧 ……