An adaptive, perception-driven error spreading scheme in continuous media streaming

For transmission of continuous media (CM) streams such as audio and video over the Internet, a critical issue is that periodic network overloads cause bursty packet losses. Studies on human perception of audio and video streams have shown that bursty losses have the most annoying affects. Hence, addressing this issue is critical for multimedia applications such as Internet telephony, videoconferencing, distance learning, etc. Classical error handling schemes like retransmission and forward error recovery have the undesirable effects of (a) introducing timing variations, which is unacceptable for isochronous traffic, and (b) using up valuable bandwidth, potentially exacerbating the problem. This paper introduces a new concept called error spreading, which is a transformation technique that takes an input sequence of packets (from an audio or video stream) and permutes them before transmission. The packets are then un-permuted at the receiver before delivery to the application. The purpose is to spread out bursty network errors, in order to achieve better perceptual quality of the transmitted stream. Analysis has been done to determine the provable lower bound on bursty errors tolerable by this class of protocols. An algorithm to generate the optimal permutation for a given network loss rate is presented. While our previous work had focused on streams with no inter-frame dependencies, e.g. MJPEG encoded video, in this paper the technique is generalized to streams with inter-frame dependencies, e.g. MPEG encoded video.