[Paper Review] Traffic Characteristics of Virtual Reality over Edge-enabled Wi-Fi Networks

Seyedmohammad Salehi, Abdullah M. Alnajim+4 more|arXiv (Cornell University)|Nov 18, 2020

Image and Video Quality Assessment13 references4 citations

TL;DR

This paper investigates the traffic characteristics of virtual reality (VR) applications over edge-enabled Wi-Fi networks, analyzing downlink video frame transmission and uplink tracking information under real-world conditions. It reveals that H.265 encoding reduces video frame size by 49% compared to H.264, improves frame rate compliance, and enables better QoS support in dense environments, while dynamic bitrate and tracking interval adjustments are critical for balancing performance and network load.

ABSTRACT

Virtual reality (VR) is becoming prevalent with a plethora of applications in education, healthcare, entertainment, etc. To increase the user mobility, and to reduce the energy consumption and production cost of VR head mounted displays (HMDs), wireless VR with edge-computing has been the focus of both industry and academia. However, transferring large video frames of VR applications with their stringent Quality of Service (QoS) requirements over wireless network requires innovations and optimizations across different network layers. In order to develop efficient architectures, protocols and scheduling mechanisms, the traffic characteristics of various types of VR applications are required. In this paper, we first compute the theoretical throughput requirements of an ideal VR experience as well as a popular VR HMD. We then examine the traffic characteristics of a set of VR applications using an edge-enabled Wi-Fi network. Our results reveal interesting findings that can be considered in developing new optimizations, protocols, access mechanisms and scheduling algorithms.

Motivation & Objective

• To understand the traffic characteristics of VR applications in edge-enabled Wi-Fi networks to support efficient protocol and scheduling design.
• To evaluate the throughput and delay requirements of VR applications under realistic edge-computing and Wi-Fi 802.11ax conditions.
• To identify key performance bottlenecks in downlink video delivery and uplink tracking information transmission for immersive VR experiences.
• To assess the impact of video encoding (H.264 vs. H.265) and target bitrate on network load and user experience quality.
• To explore dynamic adaptation of tracking information frequency and video bitrate to balance QoS and network efficiency.

Proposed method

• The authors conducted real-world experiments using an edge-enabled Wi-Fi network with multiple VR HMDs and edge servers to collect traffic traces.
• They evaluated various VR games (e.g., Rec Room, Job Simulator) at different target bitrates (15–60 Mbps) and video codecs (H.264, H.265).
• The study measured end-to-end delay components, including transmission, queuing, and rendering delays, using time-stamped APDUs.
• Tracking information (TI) packets were analyzed for frequency, delay, and contention impact on uplink performance.
• The researchers compared H.264 and H.265 encoding in terms of frame size, encoding time, and frame rate compliance.
• They evaluated the role of IEEE 802.11ax features such as OFDMA and MU-MIMO in reducing contention and improving QoS for VR traffic.

Experimental results

Research questions

• RQ1 How do downlink video frame sizes and uplink tracking information rates vary across different VR applications and encoding standards?
• RQ2 What is the impact of target bitrate and video codec (H.264 vs. H.265) on frame size, encoding delay, and frame rate compliance?
• RQ3 How does the frequency of tracking information transmission affect uplink delay and channel contention in dense Wi-Fi environments?
• RQ4 To what extent can IEEE 802.11ax features like OFDMA and MU-MIMO mitigate contention and improve QoS for VR traffic?
• RQ5 What dynamic adaptation strategies (e.g., for bitrate and tracking interval) can optimize VR performance while minimizing network load?

Key findings

• H.265 encoding reduces average video frame size by 49% compared to H.264, enabling better compliance with target frame rates.
• H.265 consistently achieves lower rendering delays than H.264, even though H.264 has lower HMD rendering delays.
• At 30 Mbps target bitrate, increasing to 60 Mbps provides no noticeable improvement in Rec Room, indicating diminishing returns for higher bitrates.
• H.264 fails to meet the 72 fps target frame rate on average, as encoding takes longer than 13.88 ms per frame.
• The network experienced near-zero packet loss due to high SNR, MAC-layer retransmissions, and FEC, but this may not hold in dense or crowded environments.
• Dynamic adjustment of tracking information frequency and target bitrate is essential to balance QoS and network efficiency, especially in high-density deployments.