• Open RAN uses AI xApps/rApps for dynamic optimization but independent ICP adjustments can cause conflicts. • Conflicts (direct, indirect, implicit) destabilize network and degrade KPIs, making rule‑based management impractical at scale. • Introduces GenC, a synthetic conflict generation framework that produces large, labeled, imbalanced datasets for training. • Classification pipeline uses GNNs, Bi‑LSTM, and SMOTE‑enhanced GNNs; SMOTE‑GNN shows best robustness on imbalanced data. • Experiments on synthetic (5‑50 xApps) and ns3‑oran simulations show AI methods 3.2× faster than rule‑based with near‑perfect accuracy. • Framework effectively resolves Energy Saving/Mobility Robustness Optimization conflicts and scales to large xApp environments, enabling autonomous conflict management.
Article Summaries:
- Researchers have developed an AI‑driven framework to detect, classify, and mitigate conflicts among xApps in Open Radio Access Networks (RAN). The system introduces GenC, a synthetic conflict‑generation tool that produces large, labeled datasets with realistic class imbalance. Using graph neural networks (GNNs), Bi‑LSTM, and a SMOTE‑enhanced GNN, the authors achieve near‑perfect accuracy while classifying conflicts faster-up to 3.2× quicker than rule‑based methods. Validation on synthetic and ns3‑ORAN simulations, including Dublin‑topology scenarios, demonstrates robust performance in energy‑saving and mobility‑robustness conflict cases. The approach scales to many xApps, enabling autonomous, self‑optimizing conflict management essential for resilient, low‑latency 6G networks.
Sources: