Data Collection Framework Without RAN Awareness
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current data collection methods for AI-ML model training in wireless networks require RAN awareness, which limits their practicality due to constraints in data availability, storage, computational capacity, and compilation capabilities on user equipment (UE).
Innovation Solution
A general framework for data collection that operates without RAN awareness, where an Over-The-Top (OTT) server initiates data collection requests to UEs through the application layer, and UEs collect and deliver AI-ML related data using a delivery tunnel established between the UE and the UE server.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If RAN awareness data collection method is used, then data collection can be performed with network support, but device complexity and network overhead increase
Solution Approach 1:
The patent extracts the data collection functionality from the RAN and relocates it to the UE side. The UE autonomously collects AI/ML training data based on local configuration without requiring RAN awareness or network involvement in the actual data collection process, thereby reducing system complexity while maintaining data collection capability
Solution Approach 2:
The patent introduces an OTT server as an intermediary between the UE and the eventual data usage point. The OTT server receives data collection requests, sends configuration to UEs, and collects training data without requiring RAN awareness, thus simplifying the network architecture while enabling remote model training
2Measurement precision
If UE-side model training is performed, then model accuracy can be improved, but data availability and storage capacity on UE are insufficient
Solution Approach 1:
The patent shifts the data collection scope from purely local UE resources to a network-wide dimension. By enabling UEs to collect and report AI/ML training data based on OTT server configuration, the system accesses data across multiple UEs and network locations, effectively expanding the available training data volume beyond individual UE constraints
3Adaptability or versatility
If comprehensive data collection is performed, then model generalization improves, but data delivery overhead and network consumption increase
Solution Approach 1:
The patent implements selective data collection where UEs only gather and report AI/ML training data that matches the specific configuration criteria provided by the OTT server. This partial action approach collects sufficient data for model generalization while avoiding unnecessary data transmission, thereby reducing network overhead and energy consumption
Data Source
AI summary
Apparatus and methods are provided for data collection general framework for AI-ML model training. In one novel aspect, the UE server sends the data collection request/triggering to UE through the application layer without network involvement with the general frame for data collection including data collection triggering, data collection configuration, measurements, and data delivery. In one embodiment, the assistance information is received from the RAN node with unicast or groupcast. In another embodiment, the UE delivers assistance information with the collected AI-ML model related data to deliver to the UE server. In one novel aspect, the UE server initiates a data collection request destined to the one or more UEs and receives one or more data collection responses and AI-ML model related data and assistance information through a data tunnel connected with the wireless network.