AI-Assisted UE Positioning Model Selection in NG-RAN
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Direct AI/ML-based positioning methods in wireless communication systems face challenges such as increased signaling overhead, complexity, poor model generalization, backward compatibility issues, and high computational resource requirements, making it difficult to integrate into existing NR specifications effectively.
Innovation Solution
Introduce a direct AI positioning related function at the RAN or LMF entity to manage AI/ML model indication, configuration, and selection, allowing adaptive selection between AI and non-AI methods based on parameters like processing capability, accuracy level, and resource availability, with defined signaling procedures for interaction between UE, LMF, and gNB.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If direct AI/ML-based positioning methods are implemented, then positioning accuracy is improved, but signaling overhead and system complexity increase
Solution Approach 1:
The patent segments the positioning system into multiple functional components: AI model configuration module, model selection module, positioning estimation module, and fallback mechanism. Each component handles specific tasks independently, reducing overall system complexity while maintaining positioning accuracy through specialized functional blocks.
Solution Approach 2:
The patent introduces an intermediary configuration mechanism that manages AI model deployment and selection without requiring complex real-time negotiations between network entities. The configuration parameters and model information are pre-established and exchanged through standardized signaling procedures, acting as an intermediary layer that simplifies interactions.
2Measurement precision
If direct AI/ML-based positioning methods are implemented, then positioning accuracy is improved, but processing resources and computational power requirements increase
Solution Approach 1:
The patent implements dynamic model selection where the system adaptively chooses between different AI models or fallback to non-AI positioning methods based on current processing resource availability, accuracy requirements, and operational conditions. This dynamic adjustment optimizes the balance between positioning accuracy and computational resource consumption.
Solution Approach 2:
The patent changes operational parameters such as model complexity, model selection criteria, and processing intensity based on available computational resources. The system can adjust the AI model configuration parameters to match the processing capabilities of the deployed entity, reducing computational burden while maintaining acceptable positioning accuracy.
3Measurement precision
If direct AI/ML-based positioning methods are implemented, then positioning accuracy is improved, but backward compatibility with existing NR positioning methods deteriorates
Solution Approach 1:
The patent designs a universal positioning framework that can accommodate both AI-based and non-AI positioning methods within the same system. The framework supports multiple positioning approaches and allows seamless switching between them, ensuring backward compatibility with existing NR positioning methods while enabling advanced AI-based positioning when available.
Solution Approach 2:
The patent implements fallback mechanisms and backup positioning methods that are prepared in advance. If AI-based positioning fails or is unavailable, the system can smoothly transition to traditional non-AI positioning methods, cushioning against compatibility issues and ensuring continuous positioning service for legacy devices and scenarios.
4Measurement precision
If direct AI/ML-based positioning methods are implemented, then positioning accuracy is improved, but model generalization capability deteriorates when changing UE scenarios
Solution Approach 1:
The patent implements dynamic model adaptation where the AI positioning model can be reconfigured or replaced based on the specific UE scenario, environment conditions, and service requirements. The system supports multiple pre-trained models and can select or switch between them to maintain generalization capability across different scenarios while preserving positioning accuracy.
Solution Approach 2:
The patent adjusts model parameters and configuration settings based on the operational scenario to improve generalization. By changing parameters such as model architecture, input features, and processing conditions to match the current scenario, the system maintains high positioning accuracy across diverse UE situations without requiring complete model retraining.
Data Source
AI summary
This disclosure proposed a method, which introduces a new AI, based positioning related function at a NG-RAN node to allow a UE to select between applying an AI and non-AI positioning model for UE location estimation and/or to allow the UE to select an appropriate AI model among different AI positioning models based on an indication provide by the UE. The method furtherly defines the measures based on which the AI or non-AI model can be selected for UE positioning estimation and defines the related signaling procedures that allow proper interaction between different NR positioning related node nodes.