Buffer Displacement Based Online Learning Algorithm For Low Latency HTTP Adaptive Streaming
Mingyue Hao, Jinghao Yuan, Bingcong Lu, Li Song, Rong Xie, Wenjun Zhang · 2021
Live streaming has gained more and more interest in recent years, where low latency adaptive bitrate(ABR) algorithms play a vital role in providing high quality of experience(QoE). However, low latency requires a small buffer of the player, which makes it harder to achieve high bitrate while keeping a low rebuffer rate. Most proposed solutions that depend on reliable bandwidth predictions usually result in high rebuffer rates and frequent bitrate switches. In this paper, we propose a buffer displacement based online learning algorithm called BDA for low latency adaptive streaming utilizing convex optimization. Unlike traditional buffer-based ABRs that take absolute buffer lengths as input, BDA makes bitrate decisions based on the real-time buffer displacement. We evaluate BDA with four other state-of-the-art ABRs under various throughput traces and target latency settings. Compared to four other ABRs, BDA reduces bitrate switch frequency by 52.9%, increases average bitrate by 18.3% and achieves 23.1% improvement in total QoE.