Resilient 6G Optical-Wireless Convergence with Federated, Hierarchical Reinforcement Learning, and Triple Connectivity

Abstract

The sixth generation (6G) mobile network demands exceptional resilience to support mission-critical applications such as remote surgery, autonomous driving, and industrial automation. Conventional architectures typically separate the wireless access and optical transport domains, which limits coordination and increases vulnerability to cascading failures. This study proposes a cross-domain resilient architecture that integrates federated learning (FL) for optical path prediction with hierarchical reinforcement learning (HRL) for radio resource allocation, coordinated through a virtualised intelligence plane. A continuous-time Markov chain (CTMC) model is developed to characterise network resilience using coverage and throughput metrics, with analytically derived transition probabilities. To mitigate link failures, a load-balancing triple-connectivity (LBTC) scheme based on a correlated failure model (Clayton copula) is introduced, allowing each user to maintain three simultaneous connections to the nearest radio units. The outage probability analysis explicitly incorporates the correlated failure model, ensuring consistency between the theoretical framework and simulation results. Extensive OMNeT++ simulations, aligned with the O-RAN functional split option 7.2 (3GPP TR 38.801) and SD-EON optical transport, demonstrate that the proposed framework reduces end-to-end latency by 33% for URLLC services, increases throughput by 17% for mixed traffic, decreases outage probability by 62%, and reduces recovery time from 8 s to 2 s compared to state-of-the-art baselines, while maintaining acceptable quality of service under failure conditions.