5G-Enabled Distributed Intelligence Based on O-RAN for Distributed IoT Systems.

Area of Science:

• Telecommunications Engineering
• Artificial Intelligence
• Distributed Systems

Background:

• Federated learning (FL) offers decentralized, privacy-preserving model training but lacks effective deployment in Radio Access Networks (RAN).
• Emerging RAN paradigms, like Open RAN (O-RAN), enable distributed intelligence with intelligent controllers, crucial for advanced mobile networks (5G/6G).

Purpose of the Study:

• To propose and evaluate a methodology for deploying and optimizing federated learning tasks within O-RAN environments.
• To enhance distributed intelligence for 5G applications by leveraging O-RAN's capabilities.

Main Methods:

• Utilized reinforcement learning (RL) for intelligent client selection and resource allocation within O-RAN's RAN intelligent controllers (RIC).
• Implemented a network slicing approach to allocate resources for FL training based on selected clients.
• Compared the proposed FL deployment methodology against the federated averaging (FedAvg) algorithm.

Main Results:

• The proposed RL-based FL deployment in O-RAN demonstrated superior performance compared to the federated averaging (FedAvg) algorithm.
• Achieved faster convergence and required fewer communication rounds for model training.

Conclusions:

• The developed methodology effectively deploys and optimizes federated learning tasks in O-RAN, enhancing distributed intelligence for 5G.
• Reinforcement learning-based client selection and resource allocation are key to improving FL efficiency in O-RAN environments.