EPSCoR Research Fellows: NSF: Online Hierarchical Learning for Network Autonomy in Open Radio Access Networks

The lack of network infrastructure in the Midwest and EPSCoR states has widened the digital and economic divide between rural and urban America. Open radio access network (O-RAN) initiatives have gained significant momentum in revolutionizing, defining, and shaping next-generation mobile networks, including Beyond 5G and 6G. In O-RAN mobile networks, network management plays a critical role in overseeing various aspects of network infrastructure, including service orchestration in non-RT RICs and resource allocation in near-RT RICs.

Existing approaches generally rely on offline-train-online-deploy strategies using homogeneous AI/ML agents, which face challenges such as simulation-to-reality discrepancies and non-stationary learning environments in real-world, large-scale networks. The long-term vision of this project is to achieve autonomous mobile networks for 6G by designing novel AI/ML techniques to address real-world network management challenges, including, but not limited to, safety, scalability, robustness, and practicality.

The project's outcomes are expected to significantly reduce the operating expenses (OpEx) of current mobile networks, thereby facilitating the widespread deployment and cost-effective operation of mobile networks across Nebraska, the Midwest, and EPSCoR states, ultimately contributing to bridging the digital divide between rural and urban America.

This fellowship project outlines a first-of-its-kind safe zero-touch network management system by designing a new safe online hierarchical learning framework for O-RAN mobile networks. Leveraging the city-scale network infrastructure at the host institution, Iowa State University, the project will focus on three research objectives. First, the project will focus on online resource allocation at near-RT RICs using safe deep reinforcement learning.

Another focus will be on the online service orchestration at non-RT RICs through robust Bayesian learning. Additionally, the project will conduct extensive field testing and evaluation within the O-RAN mobile network at both the home and host institutions. This fellowship will generate long-lasting benefits for the home institution, the University of Nebraska-Lincoln, by strengthening the PI’s research portfolio, enhancing the existing educational platform, and fostering a collaborative research group.

This project holds the potential to revolutionize current practices in the acceptance, adoption, and deployment of AI/ML techniques for managing nationwide mobile networks within the telecommunications industry.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.