Wireless Positioning Label Reporting for NLOS Model Generalization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
AI/ML models for wireless positioning face poor inference performance in unfamiliar environments due to poor correlation between input data characteristics and training data, particularly in harsh channel conditions with strong NLOS paths, and lack of accurate label data for model supervision.
Innovation Solution
Transmitting label data and quality-related information between network devices and terminal equipment to enable model optimization and improve generalization, using methods like PRU and GNSS for accurate data acquisition, and reporting error information, confidence levels, and environmental data to assist model training and supervision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If AI/ML models are trained on specific input data and deployed in unfamiliar environments, then positioning accuracy is improved in the training environment, but inference performance deteriorates in unfamiliar environments due to poor correlation between input data characteristics and training data
Solution Approach 1:
The patent changes the parameters of the AI/ML model by introducing environment-specific model parameters that can be adjusted based on the deployment environment. The model includes a plurality of model parameters, where at least one parameter is determined based on the environment in which the model is deployed, allowing the model to adapt to different wireless environments while maintaining positioning accuracy
Solution Approach 2:
The patent makes the model parameters dynamic rather than static. The model parameters can be changed based on environmental conditions, allowing the AI/ML model to adapt its behavior to match the characteristics of the current wireless environment, thereby improving inference performance in unfamiliar environments
2Adaptability or versatility
If channel measurement data is collected to retrain or fine-tune models, then model generalization is improved, but data collection and model retraining complexity increases
Solution Approach 1:
The patent performs preliminary action by pre-determining model parameters based on environmental characteristics before deployment. Instead of collecting data and retraining models when entering unfamiliar environments, the system has already prepared environment-specific model parameters in advance, reducing the need for complex data collection and retraining processes at deployment time
Solution Approach 2:
The patent uses copying by creating environment-specific copies of model parameters. Rather than retraining the entire model from scratch in new environments, the system copies and adapts pre-trained model parameters to match the characteristics of the new environment, simplifying the adaptation process
3Device complexity
If traditional positioning methods are used in harsh NLOS environments, then device complexity is reduced, but measurement precision deteriorates due to inability to accurately measure wireless environments with strong NLOS paths
Solution Approach 1:
The patent introduces an intermediary element - the environment-aware AI/ML model with adaptive parameters - that mediates between the simple traditional positioning methods and the complex task of accurate positioning in NLOS environments. This intermediary model processes the wireless environment characteristics and adjusts parameters to improve measurement precision while maintaining reasonable complexity
Data Source
AI summary
A device for transmitting wireless positioning information configured in a terminal equipment, includes: acquiring processor circuitry configured to acquire label data for wireless positioning and quality related information corresponding to the label data; and a transmitter configured to transmit the label data and/or the quality related information corresponding to the label data to a network device.