BML3: Accurate Bandwidth Measurement for QoE Optimization in Low Latency Live Streaming

HTTP adaptive streaming (HAS) has become the most popular system for delivering videos over the Internet. When it comes to live video, it is hard to deliver an actual, real, and interactive live streaming experience. The latency problem is especially noticeable when video is distributed using conventional HAS techniques. Meanwhile, with the emergence of common media application format (CMAF) and chunked transfer encoding (CTE), HAS can deliver low-latency live streaming without sacrificing encoder efficiency. While using CMAF/CTE can achieve a lower latency by allowing a media segment to be generated and delivered at the same time, conventional adaptive bitrate (ABR) techniques suffer from inaccuracies in bandwidth measurements due to the presence of idle periods between the chunks that arise because of variable network conditions and the encoder speed. The bandwidth measurement issue results in wrong ABR decisions and, therefore, a low viewer experience. To bridge this gap, this paper presents BML
³
, a robust bandwidth measurement solution for low-latency live streaming scenarios. BML
³
implements three steps, namely, chunk boundary identification, chunk filtering, and segment bandwidth smoothing, to enhance the viewer QoE performance while achieving a near-second camera-to-display latency. We confirm the effectiveness of BML
³
through trace-driven live streaming experiments, with our results showcasing a minimum of 38% enhancement in bandwidth measurement accuracy and a notable 23% to 46% improvement in average QoE across various LLL-based ABR schemes, compared to its counterparts.