Application-aware MAC scheduling for XR over EPON-based 6G-backhaul

Abstract

The 3rd Generation Partnership Project (3GPP) has recognized eXtended Reality (XR) as a key use case for 5G and beyond. However, supporting XR services over such networks is challenging due to their stringent latency, reliability, and data rate requirements. Conventional Dynamic Bandwidth Allocation (DBA) protocols in Ethernet Passive Optical Networks (EPONs) fail to efficiently support the stringent delay-reliability requirements of data-intensive, bursty XR traffic, resulting in poor network utilization. To overcome these limitations, this paper proposes an application-aware cross-layer scheduling philosophy that links Media Access Control (MAC)-layer bandwidth allocation with application-layer dynamics. From the application perspective, variations in XR frame content arise from human activity, making consecutive frames highly correlated. Exploiting this correlation, the proposed “Application-aware MAC” scheduling framework employs AI-based frame prediction at the edge server to enhance delay reliability. To minimize prediction error, the original inter-arrival pattern of XR frames must be preserved. However, EPON scheduling introduces differential delays that distort this pattern. To address this, a novel MAC scheduling scheme is designed that coordinates with play-off buffers at the edge server to manage these differential delays and reconstruct the original inter-arrival pattern, thereby enabling accurate prediction and delay-reliable transmission. Simulation results show that the proposed method achieves up to a ten-fold improvement in XR user supportability compared to the DiffServ approach and about a three-fold gain over a Greedy Earliest-Deadline-First scheduler for a data rate and frame rate of 60 Mbps and 60 fps, respectively, with only negligible prediction error at the application layer. © 2025 Elsevier Inc.