Paired Decoder-Output Responses Under Burst H.264 Network Abstraction Layer Unit Loss
Abstract
Burst-erasure multimedia communication has a measurement gap: bitstream-level loss counts cannot distinguish missing decoder outputs from degraded surviving frames. We pair decoder-output availability with survivor fidelity and recover source display positions through content-identity hashing of compressed payloads. Across 270 analyzed streams (90 main high-rate and 180 rate-matched H. 264 streams) spanning heldout contents, erased video-coding-layer-unit rate predicts availability more consistently than survivor fidelity (mean leave-one-video-out $R_{\text{OOS}}^{2}>0.93$ in both profiles; rate-matched fidelity 0.007-0.070). At a ten-percent severity-balanced selection budget, availability- and fidelity-based rankings share only three of nine streams. Because the two rankings disagree at the decision-relevant tail, video-quality assessment and transport resource managers should retain availability and fidelity as separate operational inputs before application-specific scalarization.