Layered multicast has been shown as a promising technique for distributing a video program to a potentially large number of heterogeneous receivers, in which the video is encoded into multiple layers that allows receivers to subscribe to a set or subset of the layers according to their processing capability and network condition. Several layered multicast approaches have been proposed, which have primarily focused on improving the receivers' perceived video playback quality in term of PSNR value. Such approaches suffer several drawbacks: 1) the encoding rate in each layer is commonly assumed to be fixed, which potentially cause significant mismatch between these fixed transmission rates and dynamic and heterogeneous bandwidth requirements from receivers; 2) given that the number of laye...[ Read more ]

Layered multicast has been shown as a promising technique for distributing a video program to a potentially large number of heterogeneous receivers, in which the video is encoded into multiple layers that allows receivers to subscribe to a set or subset of the layers according to their processing capability and network condition. Several layered multicast approaches have been proposed, which have primarily focused on improving the receivers' perceived video playback quality in term of PSNR value. Such approaches suffer several drawbacks: 1) the encoding rate in each layer is commonly assumed to be fixed, which potentially cause significant mismatch between these fixed transmission rates and dynamic and heterogeneous bandwidth requirements from receivers; 2) given that the number of layers is quite limited in practical encoders, e.g., 3-5 layers, the adaptation granularity based addition or deletion of layer(s) is very coarse; 3) more seriously, we argue in this research that improving the PSNR value itself does not always lead to better users' perceived video quality, since frequent addition and deletion of layers can cause significant quality fluctuation at receivers. The research in this thesis focuses on the fairness and stability analysis for adaptive layered video multicast.

The first part of the work studied the optimal layering and bandwidth allocation for layered video multicast. Two fundamental issues are addressed. First, what is an optimal allocation? Second, how could the optimal allocation be achieved? More explicitly, how many layers should be generated for a video session and how much bandwidth should be allocated for each layer? Our study explored the flexible and dynamic property of the state-of-the-art encoders to meet the diverse requirements from the receivers. We presented a Stability-aware Fairness Index (SFI) that is suitable for evaluating the stability and fairness in a heterogeneous environment. We then formulated the problem of minimizing the expected SFI into an optimization problem, and designed an efficient algorithm to solve this problem.

The second part of the work demonstrated that the proposed optimal bandwidth allocation scheme can be seamlessly integrated into an end-to-end adaptation protocol called SARLM (Sender-Assisted Receiver-driven Layered Multicast). This protocol is fully compatible with the current best-effort Internet infrastructure, where FIFO drop-tail routers are widely deployed and the dominant traffic is the congestion-sensitive TCP traffic. Our results conclusively showed that SARLM achieves substantial fairness and stability improvement over all existing schemes.