Adaptive Bandwidth Prediction and Smoothing Glitches in Low-Latency Live Streaming

Abstract

HTTP adaptive streaming (HAS) technologies such as dynamic adaptive streaming over HTTP (DASH) and common media application format (CMAF) are now used extensively to deliver live streaming services to large numbers of viewers. However, in dynamic networks, inaccurate bandwidth prediction may result in the wrong request of bitrate, and short-term network fluctuations may produce glitches, causing unnecessary bitrate switching, thereby degrading clients' Quality of Experience (QoE). To tackle this, we propose adaptive bandwidth prediction and smoothing glitches in low-latency live streaming (called APSG) in this article. Concretely, firstly, the size of random bandwidth fluctuations is exploited as the weight of exponentially weighted moving average (EWMA) for adaptive bandwidth prediction; in addition to bandwidth prediction and buffer occupancy, glitches phenomena under a stable network environment are taken into account to enhance the viewing experience of clients. Finally, experimental results show that compared to traditional ABR algorithms under a stable network environment, APSG could reduce the number of bitrate switches and latency by up to 72.6% and 27.3%, respectively; under a dynamic network environment, APSG could reduce the number of bitrate switches and latency by up to 53.8% and 23.6%, respectively.