This talk will present the ongoing activities at Ericsson Research in Massy, France related to RAN automation. After a general presentation, we will discuss in detail two specific use cases:
(a) the Mobility Robustness Optimization for intra-frequency handover; we learn to automatically tune the Cell Individual Offset of each cell boundary using offline Reinforcement Learning without exploration, by exploiting already collected data.
(b) the frequency scheduling of users to sub-bands supporting MU-MIMO concurrent transmission. This is a challenging large-action and large-state space problem which we tackle using online action decomposition methods inspired my multi-agent reinforcement learning (MARL) and Graph Neural Networks (GNNs).
In both use cases we show that AI-methods outperform traditional rule-based solutions and highlight the potential benefits from AI in RAN automation of future 6G networks.
References:
[MRO] “Offline Reinforcement Learning for Mobility Robustness Optimization”, P. Alizadeh, A. Giovanidis, P. Ramachandra, V. Koutsoukis, O. Arouk, in European Wireless 2025 (https://arxiv.org/abs/2506.22793)
[MU-MIMO Frequency scheduling] A. Giovanidis, M. Leconte, S. Aroua, T. Kvernvik and D. Sandberg, “Online Frequency Scheduling by Learning Parallel Actions,” 2024 3rd International Conference on 6G Networking (6GNet), Paris, France, 2024, pp. 153-160, doi: 10.1109/6GNet63182.2024.10765674.