ACCoRD uses deep RL to resolve O-RAN conflicts with 30% fewer network events
New ANN-based conflict resolution agent cuts negative events in high-traffic O-RAN scenarios.
Deep Dive
Researchers Adamczyk and Kliks propose ACCoRD, a deep reinforcement learning method (PPO-Clip) for resolving control conflicts in O-RAN's Near-RT RIC. The ANN-based Conflict Resolution agent analyzes network data and conflicting decisions to infer optimal actions, then adjusts weights via batch training from network feedback. In simulations, ACCoRD significantly reduced negative network events caused by conflicting control decisions in medium and high traffic scenarios compared to rule-based approaches.
Key Points
- ACCoRD uses an Actor-Critic ANN trained with PPO-Clip to dynamically resolve conflicts between O-RAN xApps.
- In simulations, the method reduces negative network events by over 30% compared to rule-based baselines in high traffic.
- The paper introduces a new evaluation methodology for conflict resolution that accounts for both short-term and long-term network effects.
Why It Matters
ACCoRD brings adaptive, AI-driven conflict resolution to O-RAN, enabling more reliable and efficient network automation.