BeiDou Positioning Using Graph Transformer for Urban Multipath Noise
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing satellite positioning methods, such as GNSS, face significant challenges in complex urban environments due to electromagnetic interference, signal blockage, and multipath noise, leading to low positioning accuracy and unstable results.
Innovation Solution
A high-precision BeiDou satellite positioning method that combines Kalman filtering with graph-based data modeling and Transformer models with graph-structure perception modules to improve positioning accuracy by leveraging satellite measurement data and spatial correlations between satellites.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data-driven deep learning methods are used to address unmodeled noise, then the ability to handle complex random noise is improved, but the requirement for large volume of training data increases and positioning accuracy decreases when data samples are limited
Solution Approach 1:
The patent pre-processes satellite measurement data by constructing a sky satellite graph that encodes spatial relationships between satellites before feeding it to the Transformer model. This preliminary structuring of data allows the model to learn from smaller datasets more effectively, as the spatial correlations are already organized and ready for learning without requiring massive amounts of raw training samples.
Solution Approach 2:
The patent transforms the input data representation by converting raw satellite measurements into graph-structured data with specific node and edge features. This parameter transformation allows the model to capture spatial relationships more efficiently, reducing the amount of training data needed while improving the ability to handle unmodeled noise in complex urban environments.
2Device complexity
If existing data-driven methods only consider obtained satellite measurements, then the model simplicity is maintained, but the spatial distribution information of satellites is not utilized and positioning accuracy is affected
Solution Approach 1:
The patent adds a spatial dimension to the data representation by constructing a sky satellite graph where nodes represent satellites and edges represent spatial relationships. This dimensional transformation allows the model to incorporate spatial distribution information without significantly increasing model complexity, as the graph structure naturally encodes the geometric relationships between satellites.
Solution Approach 2:
The patent introduces a graph-structured intermediate representation that bridges raw satellite measurements and the final positioning output. This intermediary graph structure serves as a mediator that organizes spatial information in a way that is easy for the Transformer model to process, improving positioning accuracy without requiring complex model architectures.
3Measurement precision
If mathematical observation model-based methods are used, then higher accuracy is achieved in open environments, but the ability to eliminate observation errors through modeling fails in urban areas with complex electromagnetic interference
Solution Approach 1:
The patent replaces traditional mathematical observation models with a data-driven Transformer model that learns patterns directly from data. This substitution allows the system to adapt to complex urban environments with electromagnetic interference and signal blockage, where predefined mathematical models fail, while still maintaining good performance in open environments through the learned representations.
Solution Approach 2:
The patent changes the fundamental approach from model-based error elimination to data-driven pattern learning. By transforming the problem into a learning task where the model adapts to different environmental conditions, the system achieves both high accuracy in open environments and adaptability in complex urban settings without relying on predefined mathematical observations.
Data Source
AI summary
Disclosed is a high-precision BeiDou satellite positioning method and system for a complex urban environment. The method includes the following steps: acquiring measurement data; solving the measurement data based on Kalman filtering before predicting to obtain an initial positioning result; performing graph-based data modeling based on the measurement data to obtain a sky satellite graph; and performing feature extraction on the sky satellite graph based on a Transformer model with a graph-structure perception module, and outputting a position correction value. By using the present disclosure, the positioning accuracy and generalization performance of a positioning model in complex scenarios can be improved. The present disclosure is widely applicable in the field of satellite positioning.