Improving user experience in low-latency adaptive streaming by stabilizing buffer dynamics

Abstract:

State-of-the-art rate adaptation is not suitable for low-latency dynamic streaming due to a lack of explicit stabilization of client buffer dynamics. Specifically, when the client buffer is at its maximum level, a biased throughput feedback is induced that leads to suboptimal adaptation decisions. We propose a rate adaptation algorithm that minimizes buffer deviation from the desired level, and implement the algorithm within a server-based architecture to avoid delayed feedback. We evaluate the performance of our solution using video quality metrics derived from subjective user tests and compare it against two well-known streaming architectures, DASH VLC plugin and QAC. The results show that our solution improves user experience in dynamic streaming with buffering delays as low as the chunk duration. Furthermore, we explore the impact of the server-based architecture on the streaming performance of the proposed solution by experimental analysis. Within a client-based streaming architecture, our rate adaptation algorithm also achieves at least 65% lower impairment of playback stalls.