A Proof of Concept Resource Management Scheme for Augmented Reality Applications in 5G Systems

Augmented reality applications are bitrate intensive, delay-sensitive, and computationally demanding. To support them, mobile edge computing systems need to carefully manage both their networking and computing resources. To this end, we present a proof of concept resource management scheme that adapts the bandwidth at the base station and the GPU frequency at the edge to efficiently fulfill roundtrip delay constrains. Resource adaptation is performed using a Multi-Armed Bandit algorithm that accounts for the monotonic relationship between allocated resources and performance. We evaluate our scheme by experimentation on an OpenAirInterface 5G testbed where the considered application is OpenRTiST. The results indicate that our resource management scheme can substantially reduce both bandwidth usage and power consumption while delivering high quality of service. Overall, this work demonstrates that intelligent resource control can potentially establish systems that are not only more efficient but also more sustainable.

💡 Research Summary

This paper presents a proof-of-concept resource management scheme designed to efficiently support Augmented Reality (AR) applications within 5G Mobile Edge Computing (MEC) systems. AR applications are characterized by their high bitrate demands, strict latency sensitivity, and significant computational requirements. To address these challenges, the proposed scheme jointly manages both networking resources (bandwidth at the Base Station) and computing resources (GPU frequency at the edge server) to meet end-to-end roundtrip delay constraints while minimizing resource usage and power consumption.

The core of the scheme involves decomposing the overall roundtrip delay budget for an AR frame into three sub-budgets corresponding to the uplink (UL) transmission, edge processing, and downlink (DL) transmission hops. At each hop, an independent online learning agent, implemented using a Multi-Armed Bandit (MAB) algorithm, dynamically adjusts the allocated resources. A key insight incorporated into the MAB design is the monotonic relationship between allocated resources and performance (i.e., more resources lead to lower delay), which significantly accelerates learning and improves policy efficiency. This approach avoids the need for complex, pre-trained neural network models that suffer from the sim-to-real gap, favoring instead a lightweight and fast-adapting online learning method.

The proposed system is implemented and evaluated on a real-world, open-source testbed. The testbed integrates OpenAirInterface 5G software stack running on a USRP B210 Software-Defined Radio (acting as the gNodeB and core network), a Quectel RM500Q-GL module as the User Equipment (UE), and an NVIDIA GeForce GTX 1650 GPU for edge processing. The target AR application is OpenRTiST, which performs neural style transfer on video frames in real-time. The researchers conducted extensive experiments (over 20 hours across five scenarios) comparing their MAB-based scheme against several baseline policies, including static allocation and threshold-based reactive methods.

The experimental results demonstrate that the intelligent resource management scheme successfully maintains a high Quality of Service (QoS) delivery ratio, ensuring that frame delays remain within the target budget. Crucially, it achieves this while substantially reducing the average amount of Physical Resource Blocks (PRBs) used in both uplink and downlink and lowering the GPU operating frequency, leading to significant savings in both bandwidth utilization and system power consumption. This work provides a tangible implementation of cross-domain (networking and computing) resource automation for MEC, showcasing the potential of machine learning for creating more efficient and sustainable next-generation networks. All code and detailed setup guides have been made publicly available on GitHub to ensure reproducibility.