Efficient Dependency Tracking in Packetised Media Streams

Scheduling and error control mechanisms for robust delivery of media streams over packet networks rely on distortion metrics to optimally allocate resources and protect streams front uncontrolled quality degradation. Current distortion metrics are accurate, but the actual distortion values are expensive to obtain. Therefore, distortion models often assume fixed dependency patterns and neglect fragmentation issues. While this decreases runtime complexity, it also limits the application of such models to special stream classes and network environments. In response, we present a practical, efficient and format-independent framework to reason about dependencies in media streams. Based on correlation analysis we show that the estimations made by our framework match traditional distortion metrics for a number of H.264 encoded streams. Performance benchmarks indicate, that our framework is applicable at very-low computational overheads.