Virtual reality application modeling

This research direction tackles the critical challenge of enabling high-quality wireless Virtual Reality (VR) experiences, a technology poised to transform numerous aspects of daily life, from immersive gaming and education to professional training and social interaction in emerging metaverses.

As VR applications demand ever-higher resolutions and frame rates to prevent user discomfort, the required computing power increasingly shifts from headsets to remote cloud servers, a paradigm known as Cloud VR. This approach, however, places immense strain on wireless networks, requiring real-time delivery of high-bitrate video streams under extremely low latency constraints. In this context, accurate modeling of VR traffic in network simulations is not merely an academic exercise but a fundamental necessity; it is the cornerstone for designing efficient network protocols, developing robust Quality of Service (QoS) mechanisms, and ultimately ensuring a seamless and immersive user experience in real-world deployments.

In our study, we investigated the traffic characteristics and Quality of Service (QoS) requirements of cloud Virtual Reality (VR) applications based on real-world measurements of the Pico Neo 2 headset. We developed a novel and accurate VR application model that incorporates often-overlooked features like stereo video streams, inter-frame time variance, and realistic frame size correlations from encoder traces. Our analysis reveals that the baseline 3GPP model of VR traffic underestimates the effective bandwidth needed for cloud VR by 25%, while our model provides a much more precise estimation. We demonstrated that employing intra-refresh video encoding significantly reduces the peak-to-average frame size ratio, lowering the effective bitrate by 30% and improving visual quality (PSNR) by nearly 2 dB. Furthermore, we determined that an initial buffer depth of three frames at the headset optimally balances the trade-off between motion-to-photon delay and frame loss in both 5G and Wi-Fi networks. Finally, we enhanced a bitrate adaptation algorithm, EVeREst-Intra, which dynamically adjusts video quality to changing network conditions, outperforming existing solutions in terms of system goodput and network capacity. This research provides crucial insights for accurately modeling VR traffic and developing effective QoS optimization strategies for next-generation wireless networks.

List of relevant papers: