Controlled Unfair Adaptive 360 VR Video Delivery over an MPTCP/QUIC Architecture
Brian Hayes, Yusun Chang, George Riley · 2018
Distributing high-quality 360 VR content to clients with modern networks using existing delivery techniques is economically challenging at scale. Despite the abundance of research in multihomed delivery and client-side adaptive bitrate (ABR) algorithms, current techniques continue to suffer from similar fundamental laws; namely the reliance on inflexible networks, inaccurate network models, and rigid control rules leading to suboptimal performance. This paper proposes a unique approach to address many of these limitations. First, through the implementation of a modified version of an MPEG initiative called Server and Network Assisted DASH (SAND) for use in multipath networks enabling opportunistic messaging between the client, server, and network. Secondly, through the optimization of ABR algorithms that were generated using reinforcement learning and finally, by active management of multipath enabled transport protocols. Testbed experimental results revealed that our technique outperformed many state-of-the- art ABR algorithms in multipath networks. The average multipath differential delay decreased by 30% or more and outperformed standard Multipath TCP (MPTCP) and Quick UDP Internet Connections (QUIC) in key quality of experience (QoE) metrics by up to 20%.