Resource allocation and adaptive scheduling for scalable video streaming

The obvious recent advances in areas such as video compression and network architectures allow for the deployment of novel video distribution applications. These have the potential to provide ubiquitous media access to end users. In recent years, applications based on audio and video streaming have turned out to be immensely popular and the Internet has become the most widely used vector for media content distribution, due to its high availability and connectivity. However, the nature of the Internet infrastructure is not adapted to the specific characteristics of multimedia traffic, which presents a certain tolerance to losses, but strict delay and high bandwidth requirements. In this thesis, our goal is to improve the efficiency of media delivery over the existing network architecture. In order to do so we consider the delivery of scalable video in three main delivery scenarios, namely one-to-one client server architectures, one-to-many broadcasting architectures, and many-to-one distributed streaming architectures. First, we propose a distributed media-friendly rate allocation algorithm for the delivery of both finely and coarsely scalable video streams. Unlike existing solutions, our algorithm explicitly takes the characteristics of media streams into consideration. As a result, it provides rate allocations that better fit the heterogeneous characteristics of media streams. We outline an implementation that is robust to random feedback delays and that permits a scalable deployment of the algorithm. The rate allocation that is computed by our algorithm achieves network stability and high bandwidth utilization. It moreover allows to maximize the average received quality for all streams that are delivered in the network. While considering the transmission of coarsely layered streams, we derive conditions on the encoding rates of the video layers. These conditions depend on the allowed end-to-end delay and on the rate allocation algorithm that controls the sending rates. They allow us to take full advantage of the allocated transmission rates. Second, we investigate the problem of jointly addressing the needs of multiple receivers that consume different versions of a layered media stream in a broadcasting scenario. We provide optimal scheduling algorithms that jointly optimize the playback delay and the buffer occupancy at all of these receivers when the used channel is known. Furthermore we analyze low complexity heuristics based optimization techniques, which provide close to optimal results when only limited channel knowledge is available. Finally, we explore the possibility to exploit the inherent network diversity that is provided by the Internet infrastructure. In particular, we consider media delivery schemes where multiple senders are available for the transmission of a scalable video stream to a single client. Such an architecture is referred to as a distributed streaming architecture. It has the benefit of aggregating multiple unreliable channels into a single more robust channel with high availability. Through the use of Fountain codes, we are able to transform the distributed streaming problem into a rate allocation problem of lower complexity. The solution to this problem is shown to depend not only on the average packet loss rate, but also on the average length of packet loss bursts that are observed on each of the available channels. The coding scheme that we suggest enables our system to adapt the streamed content to the network characteristics, as well as to the needs of the receiving client.