5G Data Collection Framework for AI Model Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current wireless communication systems face challenges in efficiently collecting and managing data for AI/ML models, particularly in 5G and 6G networks, which is essential for tasks like model training, inference, and performance monitoring.
Innovation Solution
A comprehensive data collection framework is introduced that enhances the Minimization of Drive Tests (MDT) framework by defining procedures for managing data collection sessions between base stations and user equipment (UEs). This framework includes techniques for prioritizing data collection, reducing latency, implementing buffering mechanisms, and specifying storage and indexing structures for AI/ML related data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional physical drive tests are used for network performance assessment, then comprehensive data can be collected, but it consumes significant time and resources
Solution Approach 1:
The patent creates virtual copies of network data by having UEs collect and report wireless communication data during normal operations. Instead of physical drive tests, the system copies relevant measurement data (signal strength, interference, throughput) from multiple UEs to build a comprehensive network performance model, significantly reducing time while maintaining assessment accuracy
Solution Approach 2:
The patent enables UEs to automatically collect and report their own wireless communication data during normal usage without requiring external testing equipment. The UEs self-serve as measurement nodes, continuously gathering data on signal quality, interference levels, and network performance, eliminating the need for manual drive tests
2Measurement precision
If data collection is performed continuously for AI/ML model training, then model accuracy improves, but network overhead and energy consumption increase
Solution Approach 1:
The patent implements periodic data collection where UEs report wireless communication data at specific intervals or triggered by certain events (e.g., handover, signal threshold crossing). This periodic reporting provides sufficient data for AI/ML model training while avoiding continuous transmission overhead and energy consumption
Solution Approach 2:
The patent collects data from a selective subset of UEs rather than all devices, or collects only specific relevant parameters (signal strength, interference, throughput) rather than complete data sets. This partial action approach provides adequate training data for AI/ML models while reducing overall network overhead and energy consumption
3Adaptability or versatility
If multiple parameters are collected for comprehensive AI/ML training, then model performance improves, but data management complexity increases
Solution Approach 1:
The patent segments the data collection process into distinct components: individual UEs collect specific local parameters (signal strength, interference, throughput), the network aggregates these segmented data sets, and AI/ML models process organized segments independently. This segmentation manages complexity while maintaining comprehensive multi-parameter training capability
Solution Approach 2:
The patent introduces an intermediary data management layer (network controller or edge server) that receives, standardizes, and organizes multi-parameter data from multiple UEs before feeding it to AI/ML models. This intermediary handles the complexity of data integration, parameter alignment, and quality control, allowing comprehensive parameter collection without proportional increase in overall system complexity
Data Source
AI summary
Embodiments are related to a fifth generation (5G) or sixth generation (6G) wireless communications system. A method for an access node of a wireless system comprises encoding a session start request message to start a data collection session (DCS) to collect data for a machine learning (ML) model from a user equipment (UE) by a base station of a wireless system, the session start request message including UE context information, decoding a session start response message to indicate the start of the DCS by the base station, the session start response message including DCS configuration information for the ML model, and encoding a data collection request message for the UE to collect measurements for the ML model based on the DCS configuration information for the ML model by the base station. Other embodiments are described and claimed.