Hierarchical AI orchestration for multi-domain integrated communication and sensing over optical transport networks
Abstract
This paper investigates the role of optical transport networks as a key enabler for integrated sensing and communication (ISAC) services in emerging 6G infrastructures. Unlike conventional communication-centric mobile networks, ISAC introduces stringent requirements for transporting, synchronizing, and processing high-volume sensing data while simultaneously satisfying communication quality-of-service (QoS) guarantees. These requirements necessitate coordinated orchestration across the radio access network (RAN), transport, and compute domains. We consider an open-radio access network (O-RAN)-compliant architecture supporting integrated communication and sensing services. To efficiently orchestrate the underlying radio, optical transport, and edge/cloud compute resources, we introduce a hierarchical artificial intelligence (AI) framework based on reinforcement learning (RL), where domain-specific AI agents manage the RAN, the optical transport network, and the edge/cloud compute resources under the supervision of a central intent-driven controller. The proposed framework dynamically adapts network operation to time-varying communication traffic and sensing requirements, including jointly optimizing resource allocation and switching between centralized sensing and federated learning (FL) modes to balance sensing accuracy and transport resource utilization. The approach is experimentally validated through a hybrid electronic–optical transport network integrating RAN and compute elements. Experimental testbed results, together with large-scale simulation results, demonstrate that the proposed AI-driven orchestration can significantly improve transport resource efficiency while maintaining sensing accuracy and communication service guarantees.