Distributed Learning for Reliable and Timely Communication in 6G Industrial Subnetworks

Emerging 6G industrial networks envision autonomous in-X subnetworks to support efficient and cost-effective short range, localized connectivity for autonomous control operations. Supporting timely transmission of event-driven, critical control traffic is challenging in such networks is challenging due to limited radio resources, dynamic device activity, and high mobility. In this paper, we propose a distributed, learning-based random access protocol that establishes implicit inter-subnetwork coordination to minimize the collision probability and improves timely delivery. Each subnetwork independently learns and selects access configurations based on a contention signature signal broadcast by a central access point, enabling adaptive, collision-aware access under dynamic traffic and mobility conditions. The proposed approach features lightweight neural models and online training, making it suitable for deployment in constrained industrial subnetworks. Simulation results show that our method significantly improves the probability of timely packet delivery compared to baseline methods, particularly in dense and high-load scenarios. For instance, our proposed method achieves 21% gain in the probability of timely packet delivery compared to a classical Multi-Armed Bandit (MAB) for an industrial setting of 60 subnetworks and 5 radio channels.