Dynamic adaptive video streaming with minimal buffer sizes [article]

Yongtao Shuai, Universität Des Saarlandes, Universität Des Saarlandes
2018
A B S T R A C T Recently, adaptive streaming has been widely adopted in video streaming services to improve the Quality-of-Experience (QoE) of video delivery over the Internet. However, state-of-the-art bitrate adaptation achieves satisfactory performance only with extensive buffering of several tens of seconds. This leads to high playback latency in video delivery, which is undesirable especially in the context of live content with a low upper bound on the latency. Therefore, this thesis aims
more » ... t pushing the application of adaptive streaming to its limit with respect to the buffer size, which is the dominant factor of the streaming latency. In this work, we first address the minimum buffering size required in adaptive streaming, which provides us with guidelines to determine a reasonable low latency for streaming systems. Then, we tackle the fundamental challenge of achieving such a low-latency streaming by developing a novel adaptation algorithm that stabilizes buffer dynamics despite a small buffer size. We also present advanced improvements by designing a novel adaptation architecture with low-delay feedback for the bitrate selection and optimizing the underlying transport layer to offer efficient realtime streaming. Experimental evaluations demonstrate that our approach achieves superior QoE in adaptive video streaming, especially in the particularly challenging case of low-latency streaming. iii v The proposed algorithm is the most suitable adaptation for low-latency streaming compared to other state-of-the-art algorithms. It achieves the best performance with a buffering size as small as a single segment duration. The majority of bitrate adaptations do not explicitly take the network delay into consideration during their design. However, the network delay is critical to lowlatency streaming. First, it introduces a significant delay for client feedback and thus limits the effectiveness of the adaptation. Second, it incurs the reception delay of consecutive segments, which directly influence the overall latency budget for the streaming. In our third contribution, we present an adaptive streaming architecture designed to address this issue. The key advantage of our architecture is a serverside adaptation based on the throughput and buffer information, which provides a low-delay feedback for the video bitrate selection and near-zero values for the reception delay. Furthermore, we optimize the underlying transport layer of the streaming architecture in order to support predictable delay variation and a high throughput utilization for video streaming. With these refinements, our approach exhibits advanced improvements in terms of user-perceived video quality. vi Silent gratitude -Xiaomei
doi:10.22028/d291-27373 fatcat:hqv5uziqc5g25fsw62je2t72oy