SR in Buffer: Edge-Assisted and Neural-Enhanced VBR-Encoded Video Streaming

Neural-enhanced operations for Variable Bitrate (VBR) video remain challenging due to varying quality enhancement and bandwidth savings across video chunks. Mobile users face computational limitations, making edge servers crucial for offloading intensive tasks. However, limited GPU resources at the edge complicate super-resolution (SR) operations for all chunks. This paper proposes SR in Buffer (SIB), an edge-assisted framework that selectively performs SR on video chunks cached in users’ playback buffers, enhancing video quality and reducing playback smoothness penalties. SIB also optimizes GPU resource allocation using an efficient approximation algorithm. Experiments show SIB improves average QoE by 2.29%–21.86% over state-of-the-art ABR algorithms.