How dia-shows turn into video flows: Adapting scalable video communication to heterogeneous network conditions in real-time

Abstract

Video conferencing over IP (VCoIP) is a major trend in current Internet communication and has particularly spread to the mobile realm. In this environment, users face the problem of heterogeneous and fluctuating network conditions. A promising solution to this issue is the scalable video coding (SVC). It allows an adaptation of the video stream to the available bandwidth, but requires a reliable bandwidth estimation. Adaptation times for conversational video at fluctuating network conditions are critical, and a fast strategy for bandwidth estimation is needed to avoid congestion. In this work, we analyse the capabilities of the sender and the receiver to adapt the video coding to changing network conditions. We derive an early congestion indicator at the sender side based on the jitter variation. For receivers, we use sustained goodput to extract a feasible scaling. In thorough evaluations that include real-world 3G networks, we reveal a faster congestion detection at the sender that are also more robust but less accurate than probing at the receiver.