IP video streaming with fine-grained TCP-friendly rate adaptation

Video streaming over the Internet is a challenging task since the Internet is a shared environment offering only best effort service. That is, it offers no quality of service and no guarantee of resources in term of (1) bandwidth, (2) transfer delay, (3) delay variation (jitter), and (4) packet losses. Then, network stability and traffic fairness become critical issues. To solve these problems some source rate control and adaptation should be introduced for UDP traffic as well, in such a way that this traffic becomes TCP-compatible "TCP-friendly". In this article we propose an adaptive streaming framework for unicast MPEG-4 streams over TCP/IP networks. Based on Audio-Visual Content (AVOs) classification and network congestion feedback, video sources dynamically adds and drops MPEG-4 AVO to the streamed multiplex to conform to the TCP-Friendly Rate Control (TFRC) mechanism. Using a content classification model, TFRC automatically adjusts the number of AVOs to be streamed to adapt to network congestion while given much attention to the quality of the service perceived by the end-user. To achieve such a dynamic output rate and video quality adjustment, MPEG-4 AVOs are classified and multiplexed according to both application-level QoS parameters and AVOs semantic descriptors. AVOs requiring same QoS from the network are automatically classified and mapped to one of the available IP DiffServ PHB (Per Hop Behaviors). Performance evaluation shows that transmitted video gracefully adapts to network bandwidth variations while optimizing user perceived quality.