Leveraging O-RAN SC AI/ML Framework & Non-RT RIC for AI-Driven Network Slice QoS Optimization

The advent of Beyond 5G (B5G) networks necessitates innovative approaches for managing network resources to ensure flexibility and maintain Quality of Service (QoS). This paper presents a comprehensive study of integrating Artificial Intelligence (AI)/Machine Learning (ML) Framework (AIMLFW)s with Open Radio Access Network (O-RAN) architectures for QoS management in network slicing. We detail the development and deployment of a live O-RAN setup at Commonwealth Cyber Initiative (CCI) xG Testbed Virginia Tech in collaboration with Samsung R&D Institute Bangalore (SRI-B), which leverages AI/ML model for real-time resource management. Our methodology includes creating network slices in Software Defined Radio (SDR) based 5G network, collecting and processing slice-level data, training an ML model to predict Physical Resource Block (PRB) utilization per slice, and dynamically adjusting resource allocations to maintain optimal QoS. The results highlight the efficacy of the AIMLFW-integrated O-RAN systems for future B5G networks by demonstrating efficient performance of ML based rApp to predict PRB utilization and consequently QoS management in each slice. This work contributes to advancing the state-of-the-art in network slicing and resource management, providing a robust framework for adaptive and automated network operations.