AI Positioning Model Using Sparse Reference Signals in 5G
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional positioning methods in 5G-advanced face challenges in overcoming synchronization errors using reference signals, necessitating the need for improved dataset collection to enhance positioning accuracy.
Innovation Solution
Implementing an AI/ML-based positioning model that utilizes a multi-RTT mechanism to generate datasets with varying reference signal densities for training and monitoring stages, allowing for accurate location determination of terminal devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If reference signals with high density are used for positioning, then positioning accuracy is improved, but signaling overhead and resource consumption increase
Solution Approach 1:
The patent applies preliminary action by training the positioning model offline using high-density reference signal datasets before deployment. The trained model is then stored in the terminal device, allowing it to perform accurate positioning inference using sparser online reference signals, thus avoiding the need for continuous high-density signal transmission.
Solution Approach 2:
The patent creates a virtual copy of the high-density reference signal environment through synthetic data generation during the offline training phase. By simulating various positioning scenarios with dense reference signals, the model learns to recognize positioning patterns without requiring actual continuous high-density signal transmission during operation.
2Measurement precision
If AI/ML-based positioning model is implemented, then positioning accuracy is enhanced, but device complexity and processing requirements increase
Solution Approach 1:
The patent segments the positioning system into two distinct parts: an offline training phase that handles complex model development, and an online inference phase that handles simple model execution. This segmentation allows the complex AI/ML model to be developed and optimized separately, then deployed as a streamlined solution in the terminal device.
Solution Approach 2:
The patent introduces an intermediary offline training server that handles the computationally intensive tasks of data processing and model training. This intermediary processes the complex operations remotely, then delivers the trained model to the terminal device, reducing the processing burden on the end device while maintaining high positioning accuracy.
3Reliability
If multi-RTT mechanism is used for dataset collection, then synchronization errors are addressed, but measurement and detection difficulty increases
Solution Approach 1:
The patent merges multiple RTT measurements from different reference signals into a unified dataset for model training. By combining these measurements and processing them through the trained AI/ML model, the system collectively addresses synchronization errors across multiple signals, reducing the overall complexity of handling individual complex measurements.
Data Source
AI summary
Example embodiments of the present disclosure relate to methods, devices, and medium for communication. If a positioning model is triggered, the communication device obtains a first dataset, where the first dataset is generated based at least on a plurality of reference signals with a first reference signal resource. The communication device trains the positioning model based on the first dataset. The communication device obtains a second dataset based at least on a plurality of reference signals with a second reference signal resource, where a density of the second reference signal resource is sparser than a density of the first reference signal resource. The communication device monitors the trained positioning model based on the second dataset.