A Decentralized Framework for AI-based Resource Allocation in Open RAN

Modern mobile networks, particularly within the Open-RAN (O-RAN) architecture, increasingly rely on Artificial Intelligence for dynamic network slicing. However, conventional approaches often depend on a centralized AI model, which introduces significant challenges related to scalability, reliability as a single point of failure, and data privacy. This project proposes and implements a novel decentralized framework, to address these limitations using Federated Learning and Deep Learning. The system features autonomous DL agents, implemented in Pyt1orch, that reside at the network edge, making real-time resource allocation decisions based on local data. These agents collaboratively train a global intelligence model, orchestrated by a central FL server, by sharing only their learned model parameters, thus preserving data privacy. Through a custom-built simulation environment, this work demonstrates that the federated system successfully learns to manage network resources, achieving performance comparable to a centralized model while exhibiting graceful degradation and enhanced scalability.