AI/ML Configuration Feedback for 5G UE Performance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current AI/ML operations in 5G networks face challenges in managing performance issues caused by high resource consumption and data generation, leading to overheating and processing problems in user equipment (UE), which can degrade radio operations and user experience.
Innovation Solution
Implementing a system that monitors performance issues in UE due to AI/ML operations and adjusts data collection and training processes by reducing data volume, resolution, or suspending training, and providing fallback measurement configurations to prioritize regular radio operations over AI/ML tasks, using a state machine to manage AI/ML operation states and detect performance degradation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If AI/ML operations are performed in the terminal to collect and process data for model training, then the intelligence and predictive capability of the network are improved, but the terminal experiences performance degradation due to high resource consumption and overheating
Solution Approach 1:
The patent extracts the AI/ML model training operation from the terminal and relocates it to the network side (gNB). The terminal only performs lightweight data collection and transmission, while the computationally intensive model training is performed by the network server, thereby reducing terminal resource consumption and overheating while maintaining AI/ML functionality.
Solution Approach 2:
The patent introduces a network-side AI/ML model as an intermediary that processes terminal data remotely. Instead of the terminal directly performing complex model training, the terminal interacts with the network AI/ML model through standardized interfaces, allowing intelligent processing without burdening the terminal resources.
2Measurement precision
If the terminal collects massive quantity of data for AI/ML model training, then the model training accuracy is improved, but the terminal suffers from overheating and connection failures
Solution Approach 1:
The patent extracts the data processing and model training functions from the terminal to the network side. The terminal collects and transmits raw measurement data, while the network performs data processing, feature extraction, and model training, thereby maintaining high model accuracy without subjecting the terminal to excessive processing loads that cause overheating.
Solution Approach 2:
The patent implements selective data collection where the terminal collects only the necessary measurement data (RSRP, RSRQ, SINR, location information) required for AI/ML model training, rather than collecting all possible data. This partial action approach maintains sufficient model training accuracy while reducing terminal resource consumption and preventing overheating.
3Speed
If the terminal performs AI/ML model training locally, then the response time for mobility optimization decisions is reduced, but the terminal resource consumption increases significantly
Solution Approach 1:
The patent extracts the energy-intensive model training function from the terminal to the network side. The terminal performs only lightweight data collection and transmission, while the network performs computationally intensive model training and generates optimization decisions, thereby maintaining fast response times without significant terminal energy consumption.
Solution Approach 2:
The patent implements preliminary model training at the network side using historical data and accumulated measurements. The trained model is then deployed for real-time inference, allowing the system to make fast mobility optimization decisions without requiring the terminal to perform intensive training operations at the time of decision-making.
Data Source
AI summary
Apparatus comprising:one or more processors, and memory storing instructions that, when executed by the one or more processors, cause the apparatus to perform:monitoring whether a terminal suffers a performance issue due to an artificial intelligence/machine learning operation performed by the terminal;performing an action related to the artificial intelligence/machine learning operation to remove or reduce the performance issue if the terminal suffers the performance issue due to the artificial intelligence/machine learning operation.