Wireless RAN AI/ML Training With Selective UE Participation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenges of high signaling traffic load and increased device power consumption in wireless devices due to AI/ML operations in radio access networks (RAN) with mobile devices and base stations are not adequately addressed in current communication systems.
Innovation Solution
Implementing an event-triggering method to select a subset of user equipment (UE) for AI/ML model training based on pre-defined threshold values, combined with periodic participation, to manage UE involvement in global AI/ML model training efficiently, thereby reducing unnecessary signaling and power consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all UEs participate in AI/ML model training, then training accuracy is improved, but signaling traffic load increases
Solution Approach 1:
The patent divides the UE population into multiple groups or subsets, where only selected subsets participate in AI/ML model training at any given time. This segmentation approach maintains training accuracy by ensuring sufficient diverse data coverage while reducing overall signaling traffic load by limiting active participants.
Solution Approach 2:
The patent implements partial participation of UEs in AI/ML training by selecting only a subset of UEs based on specific criteria (e.g., channel conditions, data availability, device capabilities). This partial action reduces signaling overhead while maintaining adequate training quality through selective sampling.
2Adaptability or versatility
If all UEs participate in AI/ML model training, then model diversity is improved, but device power consumption increases
Solution Approach 1:
The patent segments the UE population into different groups with varying levels of participation in AI/ML training. By dividing UEs into subsets based on capabilities, channel conditions, or other criteria, the system maintains model diversity through diverse participant selection while reducing individual device power consumption through selective engagement.
Solution Approach 2:
The patent implements periodic selection of UE subsets for AI/ML training rather than continuous participation. This periodic action allows devices to alternate between training and idle states, maintaining model diversity over time while reducing average power consumption through intermittent engagement.
Data Source
AI summary
Event triggering and/or periodic approaches for artificial intelligence/machine learning (AI/ML) applications for radio access networks (RAN) are disclosed. The approaches facilitate federated learning and may include where global model located in radio access network or server base station (BS) is distributed to user equipments (UEs) after collecting local model feedback.