Optimizing Selective ARQ for H.264 Live Streaming: A Novel Method for Predicting Loss-Impact in Real Time

This work proposes a quality-oriented, real-time capable prioritization technique for media units of H.264/AVC video streams. The derivation of estimates is based on the analysis of the macroblock partitioning, the spatial extents of temporal dependencies, and the length and strength of prediction chains existing among macroblocks, thus incorporating the expected impact of error propagation. It is demonstrated how the prioritization scheme can be beneficially integrated into live streaming systems which are characterized by tight timing constraints, with the focus on content-aware selective automatic repeat request mechanisms. Additionally, it is shown how potentially limited feedback can be used to adapt the estimation process to leverage prediction preciseness. The approach is compared against existing techniques in terms of practicability and efficiency, and tested under independent and bursty loss conditions in a wired and a wireless test setup. Moreover, the performance is examined when low-latency and constant bitrate video settings are enforced by using x264's novel encoding feature periodic-intra-refresh. Results of both experiments and simulations indicate that the proposed technique outperforms all reference techniques in nearly all test cases, and that the video quality can be further improved by incorporating receiver feedback.