| As video applications nowadays become prevalent on Internet, it becomes critical to stream video over diverse networks in an efficient way with guaranteed quality of service at the same time. Existing approaches for video multicast may not fit well into the fast developing and highly heterogeneous networking environments in terms of scalability, adaptivity, quality of service, infrastructure support and deployment availability. Multiple description with forward error correction (MDFEC) coding performs scalable coding and provides operational distortion-rate optimality for video streaming in a networking circumstance with random packet losses. This thesis investigates bandwidth allocation strategies using MDFEC for heterogeneous video multicast where multiple videos of various characteristics are streamed to receivers with different path bandwidth and loss statistics.;Firstly we consider a problem of assigning diverse receivers, who request to watch a video lively, into a number of groups within each of which receivers receive an MDFEC-coded bitstream of the requested video, in such a way that average distortion is optimized. We show that the optimal solution can be easily calculated in the case that the distortion-rate function is a linear function, and develop a dynamic programming algorithm to calculate the solution in the case that the distortion-rate function is a convex function. Then we extend the problem by including the server bandwidth constraint, and develop time-efficient algorithms. Next we utilize statistical multiplexing and MDFEC together for efficiently assigning server bandwidth to multiple videos being streamed by the server, with consideration of the dynamically changing complexity of videos and characteristics of path bandwidth and packet losses. Besides, we design an approach for performance control on different clusters of users by introducing a weighting factor into the MDFEC coding algorithm, and show that the performance difference of high-end and low-end users can be controlled monotonically by the weighting factor. |
