ML-Driven Service-Aware Radio Resource Allocation for Open RAN-Enabled 6G Networks

Abstract

Sixth-generation (6G) mobile networks are expected to support highly heterogeneous services with stringent and diverse quality-of-service (QoS) requirements, exposing the limitations of conventional throughput-centric radio access network (RAN) control mechanisms. To address these challenges, AI-native RAN architectures integrating machine learning (ML) into control loops have gained significant attention. In this paper, we propose SAIL-RAN (Service-Aware Intelligent Learning for Open RAN), an ML-driven and uncertainty-aware MAC-layer radio resource allocation framework for Open RAN-enabled 6G networks. SAIL-RAN employs ML-based service urgency prediction to proactively identify users at risk of service-level agreement (SLA) violations and dynamically adapt resource allocation via the near-realtime RAN Intelligent Controller (RIC), enabling risk-aware and QoS-driven MAC-layer scheduling. The proposed framework is implemented and evaluated using a realistic end-to-end simulation testbed integrating ns-3, ns-O-RAN, and a THz channel model. Simulation results under heterogeneous traffic conditions show that SAIL-RAN significantly reduces SLA violation probability, tail latency, and jitter compared to conventional baseline schemes, while maintaining comparable throughput and spectral efficiency. These results demonstrate the effectiveness of uncertainty-aware, learning-driven RAN control in meeting the service-centric requirements of future 6G networks.