From Classification to Optimization: Slicing and Resource Management with TRACTOR

TL;DR

The paper addresses the challenge of dynamically provisioning O-RAN network slices under diverse traffic patterns by introducing TRACTOR+, an end-to-end framework that combines a publicly available O-RAN KPI dataset with ML-based traffic classification (CNN and Transformer) and a reinforcement-learning (RL) PRB allocator. It demonstrates real-world viability by collecting 447 minutes of traffic from real users, replaying it in a Colosseum-based digital twin, and deploying an xApp that achieves high offline and online classification accuracy, while an iterative train-test-improve RL pipeline improves PRB utilization across slices. Key contributions include the open-source TRACTOR+ toolkit, a 2.83 GB KPI dataset, real-time traffic slice classification, and slice-aware PRB optimization that reduces performance variability and outperforms expert policies. The work advances practical ML-driven network control for O-RAN, enabling end-to-end adaptive slicing and resource management in realistic, open research environments.

Abstract

5G and beyond networks promise advancements in bandwidth, latency, and connectivity. The Open Radio Access Network (O-RAN) framework enhances flexibility through network slicing and closed-loop RAN control. Central to this evolution is integrating machine learning (ML) for dynamic network control. This paper presents a framework to optimize O-RAN operation. First, we build and share a robust O-RAN dataset from real-world traffic captured across diverse locations and mobility scenarios, replicated within a full-stack srsRAN-based O-RAN system using the Colosseum RF emulator. This dataset supports ML training and deployment. We then introduce a traffic classification approach leveraging various ML models, demonstrating rapid training, testing, and refinement to improve accuracy. With up to 99% offline accuracy and 92% online accuracy for specific slices, our framework adapts efficiently to different models and network conditions. Finally, we present a physical resource block (PRB) assignment optimization strategy using reinforcement learning to refine resource allocation. Our learned policy achieves a mean performance score (0.631), surpassing a manually configured expert policy (0.609) and a random baseline (0.588), demonstrating improved PRB utilization. More importantly, our approach exhibits lower variability, with the Coefficient of Variation (CV) reduced by up to an order of magnitude in three out of four cases, ensuring more consistent performance. Our contributions, including open-source tools and datasets, accelerate O-RAN and ML-driven network control research.

Figures (8)

• Figure 1: O-RAN system used in $\text{TRACTOR}$+ consisting of multiple UEs, a gNB, the near-RT RIC, and an ML based xApp used for traffic classification and PRB optimization.
• Figure 2: Process to create an O-RAN compliant dataset.
• Figure 3: CNN model used for traffic classification in the TRACTOR framework.
• Figure 4: Illustration of the train-test-improve pipeline, an iterative RL-based learning process. At each epoch, new policies are trained, evaluated using Bellman error, and the best-performing policy is deployed for data collection in the next round, continuously refining performance.
• Figure 5: Average offline CNN-based classifier performance for different input slice sizes.