Multi-Timescale QoE Provisioning for Adaptive Video Streaming in Heterogeneous Deployments

摘要

This paper considers an optimization problem that maximizes an aggregate utility, formulated as the weighted geometric mean of the "in-context" suitability of a set of radio access technologies (RATs), to support adaptive video streaming subject to the existence of legacy data transfers. Motivated by the unfeasibility of solving the formulated problem centrally when the various RATs are loosely integrated (i.e., at core network (CN) level), a hybrid (i.e., network-assisted user-driven) strategy is devised to approximate its optimum solution. Unlike previous hybrid approaches, the proposed methodology exploits network assistance to ensure a friendly co-existence between adaptive video streaming clients and legacy users. It operates on different timescales, where the fastest timescale operation is performed on the video clients according to a policy that is tuned by the network on slower timescales. The user operation on the fastest timescale (i.e., tens of ms) enables to adapt video streaming depending on the perceived quality-of-experience (QoE) and local components of the context (e.g., remaining credit and battery level). A small-cell tuning on a slower timescale (i.e., hundreds of ms) enables to preempt the resources used by legacy users based on the operating conditions (e.g., load and type of scheduler). Finally, a tuning performed by the network on the slowest timescale (i.e., few seconds) offloads legacy data transfers to unlicensed bands whenever the amount of interference on licensed bands reaches critical levels, which helps to sustain good QoE for all video clients. A cost-benefit analysis reveals that the proposed methodology performs closely to its centralized counterpart with much less control overhead on the radio interface.