GNSS Multipath Signal Identification Using Temporal-Spatial Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing GNSS technologies face challenges in accurately identifying multipath signals in complex urban environments due to poor generalization performance and low accuracy of AI-based methods, which are unstable and computationally intensive, limiting their practical application.
Innovation Solution
A satellite multipath signal identification method utilizing a long short-term memory (LSTM) network, transformer block, and fully connected network to extract time series and environmental features, enhancing multipath signal identification through temporal and spatial interaction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If AI methods are applied to multipath signal identification, then identification accuracy may be improved, but generalization performance is poor and computation is intensive
Solution Approach 1:
The patent segments the AI model into two independent parts: a time series LSTM network for temporal feature extraction and a spatial transformer network for environmental characterization. This segmentation allows each component to specialize in specific aspects, improving overall accuracy while enabling more efficient computation through modular processing rather than a monolithic complex model.
Solution Approach 2:
The patent introduces a new dimensional approach by processing satellite data through two separate networks operating in different dimensional spaces - the LSTM network handles temporal dimensionality while the transformer network handles spatial dimensionality. This dimensional separation allows the model to capture complex patterns without the computational burden of processing all dimensions simultaneously in a single network.
2Reliability
If hardware-based methods are used, then multipath suppression may be achieved, but hardware modification is required and application scenes are limited
Solution Approach 1:
The patent replaces hardware-based multipath suppression mechanisms with a software/AI-based identification system. Instead of modifying physical antennas or receiver hardware, the solution uses neural networks to identify and characterize multipath signals through computational processing of satellite data, thereby maintaining reliability while achieving software flexibility and broader adaptability.
3Ease of manufacture
If software-based methods are used, then hardware modification is avoided, but computation is huge and stability is poor
Solution Approach 1:
The patent divides the software-based multipath identification into two specialized modules: temporal processing via LSTM and spatial processing via transformer. This segmentation improves stability by assigning specific, well-defined tasks to each module, making the overall system more predictable and reliable compared to monolithic software approaches, while still avoiding hardware modifications.
Data Source
AI summary
The present disclosure provides a satellite multipath signal identification method based on temporality and spatial interaction. The method includes: acquiring satellite data, and dividing the satellite data into a time series dataset and a multi-satellite input dataset; building a multipath signal identification model, and inputting the time series dataset and the multi-satellite input dataset into the multipath signal identification model, where the multipath signal identification model includes a long short-term memory (LSTM) network, a transformer block, and a fully connected network; performing, by the LSTM network, feature extraction on the time series dataset to acquire a time series feature; performing, by the transformer block, feature extraction on the multi-satellite input dataset to acquire an environmental characterization; and fusing, by the fully connected network, the time series feature and the environmental characterization to acquire a multipath signal identification result.