RAN-Terminated PDU Sessions for Federated Learning Data Exchange
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing communication networks have a limited scope of distributed and federated learning, primarily confined to the application layer, and inefficiently communicate unstructured AI-related data between radio access networks and devices, leading to suboptimal performance and computational inefficiencies.
Innovation Solution
Enhanced protocol data unit sessions are established with a session manager component to facilitate communication between radio access networks and devices, allowing unstructured data exchange and expanding the scope of distributed and federated learning, including AI-related data, by terminating at the RAN rather than the core network, and utilizing data radio bearers for high-quality service flows.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If unstructured AI-related data is communicated between radio access network and devices using existing protocol structures, then communication can be maintained with standard interfaces, but communication efficiency and performance are suboptimal
Solution Approach 1:
The patent introduces a new PDU session type parameter specifically designed for AI-related data communication. This parameter change enables the network to distinguish and handle AI data flows differently from traditional data, allowing for optimized resource allocation, QoS parameters, and routing decisions that improve communication efficiency for unstructured AI data without requiring complete protocol redesign.
2Adaptability or versatility
If federated learning is confined to the application layer, then implementation is simpler with existing infrastructure, but the scope and effectiveness of distributed learning are limited
Solution Approach 1:
The patent extends federated learning from the application layer into the network layer by introducing dedicated PDU sessions and network-level support mechanisms. This dimensional expansion allows learning operations to leverage network resources, QoS policies, and routing capabilities that were previously unavailable, thereby significantly broadening the scope and effectiveness of distributed learning while maintaining manageable complexity through standardized interface extensions.
3Speed
If PDU sessions terminate at the core network, then network control and management are centralized, but communication latency and inefficiency increase for RAN-local AI processing
Solution Approach 1:
The patent segments the PDU session termination point based on data type and service requirement. Traditional data sessions continue to terminate at the core network, while AI-related data sessions are configured to terminate at the RAN. This segmentation allows AI processing to occur locally at the RAN with reduced latency, while core network functions maintain oversight through session management signaling, thus improving speed without overwhelming RAN complexity.
Data Source
AI summary
RAN, and distributed and federated learning, can be enhanced. Session manager can initiate establishing PDU session between UE and RAN. PDU session is associated with PDU session type corresponding to type value associated with RAN to indicate PDU session can terminate at RAN. Using DRB associated with PDU session, RAN and UE can communicate unstructured data to each other. RAN can comprise global AI component comprising global AI model. UEs can comprise local AI components comprising local AI models. Global AI component can train global AI model based on respective first AI-related data generated by local AI models and received from respective UEs. Global AI component can train global AI model based on respective first AI-related data, and trained global AI model can generate second AI-related data. RAN can take action based on second AI-related data or can communicate second AI-related data to UE to update local AI model.