Congestion control and error concealment for real-time video transmission over lossy packet networks

The recent revolution in computer and network technologies has made real-time video transmission over the Internet a reality. At the same time, advancement in wireless communications has extended the reach of the Internet to more users through wireless connections. Due to the lossy nature of the Internet and wireless links and the delay requirement of real-time applications, packets are often lost at a video receiver causing degradation of the received video quality. On the other hand, in a network with both wired and wireless links, there are two types of packets loss: loss caused by congestion and loss caused by bit errors in wireless channels. Sending rate should be reduced to alleviate network over-crowdedness at congestion losses, while this action is often not appropriate or necessary at wireless losses. For an application to achieve good performance by reacting differently to different losses, it needs to be able to differentiate between the two types of losses.; This dissertation addresses two issues in the transmission of real-time video over lossy packet networks: error concealment in the application layer and end-to-end congestion and wireless loss differentiation in the transport layer.; At the application layer, we study how to best conceal errors caused by packet loss. First, we propose two new error concealment methods for video decoding, which often provide better performance than several other schemes commonly used. Next, to achieve better performance, we propose a new idea of adaptively choosing error concealment methods by using a decision tree. The decision tree picks a good concealment method based on the characteristics of the missing area, its neighbors, and the overall video sequence. Our tree approach consistently provides lower distortion in the concealed areas than does the use of any single method.; At the transport layer, we study end-to-end congestion and wireless loss differentiation algorithms (LDAs) for use with congestion-sensitive video transport protocols over networks with both wired and wireless links. Compared to previous work, we evaluate LDAs under more realistic situations, including two distinct network topologies, as well as various different competing traffic scenarios. We also propose two new LDAs. Both provide relatively consistent and good performance in terms of throughput and fairness across different types of network conditions.