Analysis and Modeling of Downlink Traffic in Cloud- Rendering Architectures for Augmented Reality

One way to enable augmented reality (AR) on light-weight glasses, which are limited in battery capacity and computational power, is to shift intense operations from the device to the cloud. Instead of rendering on the device, the cloud could take over this task and stream the results to the device. However, this comes with challenges, such as low latency and high data rates, on the wireless network, especially when multiple users have to be served simultaneously. In order to derive the network requirements and to assess network performance, appropriate traffic models are necessary. Therefore, we study the video traffic in the downlink of such a cloud-rendering architecture, as it accounts for the major part. Furthermore, we expect the cloud-rendered AR video traffic to differ significantly from other video traffic due to latency-critical compression, the sparsity of content per video frame, the dependency between content and human movements, as well as the massive resolution requirements. For this paper we create and analyze a video data set, based on which we propose a realistic traffic model.