A Service-Oriented Intelligent Underwater Networking Framework for Autonomous Marine Research Infrastructures

  The ocean plays a central role in regulating Earth’s climate system, driving the global carbon cycle, and sustaining marine ecosystems. However, substantial data gaps persist in deep and remote ocean regions due to extreme operating conditions, limited underwater acoustic communication capabilities, and the high cost of long-term deployment and maintenance.With the ENVRI community’s growing demand for long-term, distributed, and autonomous observations, current networking architectures—typically centralized, strictly synchronized, and statically configured—are increasingly inadequate to support next-generation marine observatory research infrastructures.

  We propose an intelligent underwater communication and collaborative observation networking framework to support autonomous operation of marine environmental research infrastructures , with a focus on unmanned underwater cluster observation scenarios. The framework elevates the communication network from a passive data-transfer layer to an intrinsic infrastructure capability, enabling distributed underwater observing units to self-organize and operate collaboratively under long propagation delays and limited local information.

  From a system-design perspective, the framework introduces a multi-segment, multi-orthogonal resource-block time–frequency structure, and formulates underwater link scheduling as a conflict-constrained Maximum Weighted Independent Set (MWIS) problem. Link weights jointly capture mission load, information freshness, historical resource utilization, and node-level credibility, thereby reflecting fairness and stability requirements under long-term operation. In contrast to conventional multi-round contention-based or centralized scheduling schemes, we develop a distributed, asynchronous, and consensus-oriented scheduling mechanism: lightweight contention is performed only at transmitters, while receivers act as local consensus anchors to enable conflict-free selection. This design supports concurrent scheduling of multiple links across multiple resource blocks within a single control cycle.

  To improve nodes’ awareness of local conflict structures and traffic dynamics, we incorporate graph neural networks (GNNs) as cognitive components to compute link priority scores on locally constructed conflict subgraphs. This enables an approximation of global scheduling relevance without requiring global topology knowledge or centralized control. 

  Simulation studies and underwater acoustic sensor-network experiments conducted in realistic marine environments show that the proposed framework outperforms conventional approaches in clustered underwater communication scenarios. It effectively prevents individual observation nodes from monopolizing communication resources, enables conflict-free data exchange among unmanned underwater clusters, and improves fairness and operational stability under long-term deployment conditions. Overall, the framework provides a scalable, autonomous, and service-oriented communication and collaborative observation capability for marine environmental research infrastructures (RIs). It can operate in conjunction with advanced sensors, autonomous observation platforms, and cloud-based data services, supporting long-term observations of marine carbon cycling, ecological change, and climate-driven processes.