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

TL;DR

AR in 5G MEC requires tightly managed networking and computing to meet latency-sensitive QoS. The paper introduces a proof-of-concept, per-hop MUCB1 online learning scheme with static roundtrip budget splitting to jointly adapt UL/DL PRBs and edge GPU frequency, demonstrated on an OpenAirInterface testbed with OpenRTiST. Results show high QoS delivery with significantly reduced resource usage and power compared to baselines, offering a practical, neural-network-free approach to intelligent, scalable resource control for AR in MEC. The work provides reproducible setup guides and highlights the viability of monotonic MAB methods for cross-domain resource management in next-gen networks and edge computing. It paves the way for sustainable, self-programming networks by merging ML-driven control with open-source MEC infrastructure.

Abstract

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.

Figures (9)

• Figure 1: The OpenRTiST client at the user device sends frames to the server at the network's edge. The server then applies an artistic style to them and sends them back to the user. In our experiments, the client and server communicate via 5G. This figure is a modified version of Fig. 1 in openrtist.
• Figure 2: A diagram of our open-source testbed. The clocks of the two Ubuntu PCs are synchronized by PTP to accurately measure the delays incurred in UL, DL, and at the edge.
• Figure 3: From left to right: the USRP B210 SDR, the Quectel RF module, the SIM cards inserted at the Quectel module, and the SIM card reader to configure the cards. Our guide in installguide describes how to setup the testbed in Fig. \ref{['oaitestbed']} using this hardware and Ubuntu PCs.
• Figure 4: We decompose the roundtrip learning problem into hop problems by splitting the roundtrip delay budget. Each hop employs an online learning algorithm to adhere to its delay budget efficiently.
• Figure 5: For each traffic load, we depict the last $100$ allocations made by the MUCB1 to show what the algorithm eventually learned.