How to Model Cloud VR: An Empirical Study of Features That Matter

Abstract

A crucial component of an interactive Virtual Reality (VR) experience is high-quality, real-time image rendering and display, which requires high-end hardware in VR headsets. Cloud VR offers a compelling alternative by offloading computationally intensive rendering tasks to remote servers. However, such a distributed operation requires a real-time reliable delivery of the VR traffic over the networks, which is a challenging task. Therefore, a thorough understanding of the VR traffic characteristics, Quality of Service (QoS) requirements, and the impact of packet service on the quality of VR is required. The paper provides insights in this area by exploring the traffic generated by a cloud VR application of an off-the-shelf Pico Neo 2 VR headset. Specifically, it identifies often-overlooked application features with a significant impact on QoS and develops a novel VR application model that accounts for these features. These findings are used in three ways. First, the paper shows that the 3GPP baseline underestimates the effective VR bitrate, i.e., the bandwidth needed to deliver VR traffic by 25%, while the developed model provides a much more accurate estimation. Second, the model is used to evaluate the benefits of intra-refresh video coding and adjusting video buffering at the VR headset according to motion-to-photon requirements. Finally, based on the revealed important features, the paper improves a VR bitrate adaptation algorithm.

Keywords

• Cloud VR
• traffic
• simulation
• QoE
• QoS
• 5G