E2 Node Federated Learning Coordination for Lower RAN Data Transfer
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
There is currently no mechanism to initiate a federated learning process within the radio access network (RAN) when the aggregator resides outside the RAN nodes, leading to inefficiencies in resource consumption and data exchange.
Innovation Solution
Implementing a method to enable federated learning operations in O-RAN by exposing the capabilities of E2 nodes as local clients to a near-real-time RIC, which acts as the central aggregator, using extended E2 interface procedures and messages to manage the federated learning process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If federated learning is implemented with aggregator outside RAN nodes, then data privacy is preserved and resource consumption is reduced, but there is no mechanism to initiate the federated learning process
Solution Approach 1:
The patent introduces an intermediary mechanism (RIC controller or O&M system) that acts as a mediator between external aggregators and RAN nodes. This intermediary receives federated learning configuration from external systems, processes it locally, and initiates the federated learning process within the RAN, thus resolving the lack of initiation mechanism while maintaining the benefits of external aggregation
2Measurement precision
If data is transferred extensively for centralized training, then model accuracy improves, but network bandwidth consumption increases and privacy is compromised
Solution Approach 1:
The patent extracts the training data from the centralized aggregation point and keeps it distributed at local RAN nodes. Only model parameters and gradients are transferred between nodes and the aggregator, not the raw training data. This extraction approach maintains model accuracy through distributed learning while dramatically reducing data transfer volume and preserving data privacy
Solution Approach 2:
The patent segments the machine learning training process into distributed local training at RAN nodes and centralized parameter aggregation. This segmentation allows each node to contribute local knowledge without sharing raw data, achieving both accuracy improvement and reduced data transfer
Data Source
AI summary
Disclosed is a method comprising receiving, from one or more E2 nodes of a radio access network, an E2 setup request message indicating a capability of the one or more E2 nodes for operating as a federated learning client in a federated learning process; determining to initiate the federated learning process for the one or more E2 nodes that indicated the capability for operating as the federated learning client; and transmitting, to the one or more E2 nodes, an RIC control request message comprising an indication to initiate the federated learning process.