F. Boavida et al. (Eds.): WWIC 2007, LNCS 4517, pp. 165 – 176, 2007. © Springer-Verlag Berlin Heidelberg 2007 A Quality Adaptation Scheme for Internet Video Streams Panagiotis Papadimitriou and Vassilis Tsaoussidis Demokritos University, Electrical & Computer Engineering Department, 12 Vas. Sofias Street, Xanthi, 67100, Greece {ppapadim, vtsaousi}@ee.duth.gr Abstract. We propose a layered quality adaptation scheme for video streams to smooth the short-term oscillations induced by Additive Increase Multiplicative Decrease (AIMD) mechanisms, and eventually refine the perceptual video qual- ity. The layered scheme utilizes receiver buffering, adapting the video quality along with long-term variations in the available bandwidth. The allocation of a new layer is based on explicit criteria that consider the available bandwidth, as well as the amount of buffering at the receiver. Consequently, the adaptation mechanism prevents wasteful layer changes that have an adverse effect on user- perceived quality. In the sequel, we concentrate on the interactions of the layered approach with Scalable Streaming Video Protocol (SSVP). Exploiting performance measures related to the perceived quality of rate-adaptive video streams, we quantify the combination of SSVP rate control and receiver- buffered layered adaptation. 1 Introduction An increasing demand for multimedia data delivery coupled with reliance on best- effort networks, such as the Internet, has spurred interest in rate-adaptive multimedia streams. Video streaming, in particular, is comparatively intolerant to delay and varia- tions of throughput and delay. Unlike bulk-data transfers, video delivery requires a minimum and continuous bandwidth guarantee. Rate adaptive video streams offer the clients the benefit of being resilient to changing network conditions and allow for a large number of streams to concurrently share network resources. Video streams can be adaptive, since user-perceived Quality of Service (QoS) is often satisfactory over a range of stream compression levels. Although this adaptivity is limited (i.e. multime- dia streams have minimum subscription levels, below which service quality is unac- ceptable), they have the capability of adjusting their subscription levels in response to congestion, much as elastic flows do. Today’s Internet is governed by the rules of Additive Increase Multiplicative De- crease (AIMD) [2], which effectively contribute to its stability. Essentially, the goal of such algorithms is to prevent applications from either overloading or under-utilizing the available network resources. Although Transmission Control Protocol (TCP) provides reliable and efficient services for bulk-data transfers, several design issues ren- der the protocol a less attractive solution for multimedia applications. More precisely, the process of probing for bandwidth and reacting to the observed congestion causes oscillations to the achievable transmission rate. Furthermore, TCP occasionally intro- duces arbitrary delays, since it enforces reliability and in-order delivery. In response to