Improving Robustness of DASH Against Unpredictable Network Variations

Bo Wang, Mingwei Xu, Fengyuan Ren, Jianping Wu

Abstract: Most video players use adaptive bitrate (ABR) algorithms to provide good quality-of-experience (QoE) in dynamic network conditions. To deal with the adaptation challenges, many ABR algorithms select bitrate by optimizing a defined QoE function. Within the framework, various algorithms mainly differ in how the optimization problem is solved, including prediction-based approaches and learn-based approaches. However, these algorithms suffer from limited performance in the current popular mobile streaming which has limited resources and rapidly changing link rates. Existing machine-learning approaches face deployment difficulties on mobile devices, and prediction-based approaches that rely on throughput prediction experience large buffer occupancy variations in cellular networks, resulting in rebuffering frequently. To provide a robust and lightweight ABR algorithm for mobile streaming, this work improves the robustness of prediction-based scheme against unpredictable network variations and develops RBC (Robust Bitrate Controller) algorithm. Rather than optimizing QoE over the entire buffer capacity, RBC creates buffer margins to absorb the impact of throughput jitters and solves QoE maximization on the narrowed buffer range. The amount of buffer margin is dynamically adjusted based on the real-time throughput fluctuation to ensure sufficient de-jitter space. For online lightweight deployment, RBC provides a closed-form solution of the desired bitrate with small computation complexity by using adaptive control approach. Trace-driven experiments and real-world tests show that RBC effectively reduces the playback freezing and gains an improvement in overall QoE.