Year: 2023
Type: dissertation
Abstract: The revolution of Extended Reality (XR) has already started and is rapidly expanding as technology advances. Announcements such as Meta’s Metaverse have boosted the general interest in XR technologies, producing novel use cases. With the advent of the fifth generation of cellular networks (5G), XR technologies are expected to improve significantly by offloading heavy computational processes from the XR Head Mounted Display (HMD) to an edge server. XR offloading can rapidly boost XR technologies by considerably reducing the burden on the XR hardware, while improving the overall user experience by enabling smoother graphics and more realistic interactions. Overall, the combination of XR and 5G has the potential to revolutionize the way we interact with technology and experience the world around us. However, XR offloading is a complex task that requires state-of-the-art tools and solutions, as well as an advanced wireless network that can meet the demanding throughput, latency, and reliability requirements of XR. The definition of these requirements strongly depends on the use case and particular XR offloading implementations. Therefore, it is crucial to perform a thorough Key Performance Indicators (KPIs) analysis to ensure a successful design of any XR offloading solution. Additionally, distributed XR implementations can be intrincated systems with multiple processes running on different devices or virtual instances. All these agents must be well-handled and synchronized to achieve XR real-time requirements and ensure the expected user experience, guaranteeing a low processing overhead. XR offloading requires a carefully designed architecture which complies with the required KPIs while efficiently synchronizing and handling multiple heterogeneous devices. Offloading XR has become an essential use case for 5G and beyond 5G technologies. However, testing distributed XR implementations requires access to advanced 5G deployments that are often unavailable to most XR application developers. Conversely, the development of 5G technologies requires constant feedback from potential applications and use cases. Unfortunately, most 5G providers, engineers, or researchers lack access to cutting-edge XR hardware or applications, which can hinder the fast implementation and improvement of 5G’s most advanced features. Both technology fields require ongoing input and continuous development from each other to fully realize their potential. As a result, XR and 5G researchers and developers must have access to the necessary tools and knowledge to ensure the rapid and satisfactory development of both technology fields. In this thesis, we focus on these challenges providing knowledge, tools and solutiond towards the implementation of advanced offloading technologies, opening the door to more immersive, comfortable and accessible XR technologies. Our contributions to the field of XR offloading include a detailed study and description of the necessary network throughput and latency KPIs for XR offloading, an architecture for low latency XR offloading and our full end to end XR offloading implementation ready for a commercial XR HMD. Besides, we also present a set of tools which can facilitate the joint development of 5G networks and XR offloading technologies: our 5G RAN real-time emulator and a multi-scenario XR IP traffic dataset. Firstly, in this thesis, we thoroughly examine and explain the KPIs that are required to achieve the expected Quality of Experience (QoE) and enhanced immersiveness in XR offloading solutions. Our analysis focuses on individual XR algorithms, rather than potential use cases. Additionally, we provide an initial description of feasible 5G deployments that could fulfill some of the proposed KPIs for different offloading scenarios. We also present our low latency muti-modal XR offloading architecture, which has already been tested on a commercial XR device and advanced 5G deployments, such as millimeter-wave (mmW) technologies. Besides, we describe our full endto- end complex XR offloading system which relies on our offloading architecture to provide low latency communication between a commercial XR device and a server running a Machine Learning (ML) algorithm. To the best of our knowledge, this is one of the first successful XR offloading implementations for complex ML algorithms in a commercial device. With the goal of providing XR developers and researchers access to complex 5G deployments and accelerating the development of future XR technologies, we present FikoRE, our 5G RAN real-time emulator. FikoRE has been specifically designed not only to model the network with sufficient accuracy but also to support the emulation of a massive number of users and actual IP throughput. As FikoRE can handle actual IP traffic above 1 Gbps, it can directly be used to test distributed XR solutions. As we describe in the thesis, its emulation capabilities make FikoRE a potential candidate to become a reference testbed for distributed XR developers and researchers. Finally, we used our XR offloading tools to generate an XR IP traffic dataset which can accelerate the development of 5G technologies by providing a straightforward manner for testing novel 5G solutions using realistic XR data. This dataset is generated for two relevant XR offloading scenarios: split rendering, in which the rendering step is moved to an edge server, and heavy ML algorithm offloading. Besides, we derive the corresponding IP traffic models from the captured data, which can be used to generate realistic XR IP traffic. We also present the validation experiments performed on the derived models and their results.
Author: González Morín, Diego
Language: English
Topic: IoT and Edge/Fog Computing
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7056050888
