Robust Multimedia Coding: Information Theory and Practical Architectures Abstract Robust Multimedia Coding: Information Theory and Practical Architectures

Robust Multimedia Coding: Information Theory and Practical Architectures by Rohit Puri Doctor of Philosophy in Engineering Electrical Engineering and Computer Sciences University of California, Berkeley Professor Kannan Ramchandran, Chair The ever increasing proportion of multimedia content over the Internet has fueled tremendous research activity in the field of robust multimedia transmission over best effort networks. In this thesis, we present fundamental theoretical performance limits as well as practical design principles for this problem. Posing the problem as a generalized Multiple Description (MD) coding problem, we present an achievable rate distortion region that is the current state-of-the-art in information theory. As a precursor to the MD problem, we first consider the rate-distortion performance of (n, k) Maximum Distance Separable (MDS) codes in a joint source-channel coding framework. These results are used to obtain novel rate regions for the MD problem. The theoretical coding techniques employed here point to connections, previously unexplored, between the problems of distributed source coding and MD source coding. We also propose novel practical code constructions for the (n, k) source channel codes based on scalar quantizers and linear trellis codes. These constructions have applications in distributed coding scenarios like sensor networks and they also shed insight into novel constructive ways of doing MD coding. Further, leveraging the features of distributed source coding, we present PRISM (Power-efficient, Robust, hIgh-compression Syndrome based Multimedia coding), a novel video encoding approach that represents a fundamental paradigm shift from the conventional video coding methods. While the latter comprises of a computationally intensive encoder and a lightweight decoder, PRISM moves the entire computational burden from the encoder to the decoder without sacrificing the compression performance and offering the feature of robustness additionally. PRISM is well-matched to applications involving uplink