Though technologies for video compression and client-server based video distribution have matured, applications fueled by recent development of communication networks pose new challenges. In this dissertation, we study two classes of video applications.

The first class is low-latency video streaming over lossy networks. Today's dominant codecs based on motion-compensated predictive coding, such as MPEG, can compress videos efficiently. However, the decoded video quality is susceptible to transmission packet losses, which occur frequently over both the Internet and cellular networks. In order to enable robust low-latency video transmission, we adopt a video coding framework based on information-theoretical principles of distributed source coding. We present the theoretical foundation of a distributed source coding based video codec as well as the practical implementations. Extensive simulations demonstrate the superiority of the proposed codec over conventional robustness-enhancing methods, such as intra refresh and forward error correction codes.

The second class is video distribution over peer-to-peer (P2P) overlay networks. Compared to the traditional client-server architecture, P2P technologies can offer tremendous scalability and greatly reduce server cost. Due to the bandwidth asymmetry experienced by Internet users, however, P2P systems are oftentimes bottlenecked by users' limited upload bandwidth. We propose to ease the bottleneck by utilizing Internet users with spare upload capacity, whom we term helpers. We present a light-weight helper protocol that is backwards-compatible with the popular BitTorrent protocol and analyze the steady-state system performance. We verify the efficiency and effectiveness of the proposed protocol and the accuracy of the analysis through extensive simulations. We further extend the philosophy for live video streaming. We use a simple fluid-level analysis to guide our system design. We demonstrate that the simple analysis provides a good estimate of system performance and verify that the proposed system can efficiently utilize helpers' upload bandwidth even with high peer churning through extensive simulations.





Download Full History