Prediction on the Urban GNSS Measurement Uncertainty Based on Deep Learning Networks With Long Short-Term Memory

The GNSS performance could be significantly degraded by the interferences in an urban canyon, such as the blockage of the direct signal and the measurement error due to reflected signals. Such interferences can hardly be predicted by statistical or physical models, making urban GNSS positioning unable to achieve satisfactory accuracy. The deep learning networks, specializing in extracting abstract representations from data, may learn the representation about the GNSS measurement quality from existing measurements, which can be employed to predict the interferences in an urban area. In this study, we proposed a deep learning network architecture combining the conventional fully connected neural networks (FCNNs) and the long short-term memory (LSTM) networks, to predict the GNSS satellite visibility and pseudorange error based on GNSS measurement-level data. The performance of the proposed deep learning networks is evaluated by real experimental data in an urban area. It can predict the satellite visibility with 80.1% accuracy and predict the pseudorange errors with an average difference of 4.9 meters to the labeled errors. Experiments are conducted to investigate what representations have been learned from data by the proposed deep learning networks. Analysis results show that the LSTM layer within the proposed networks may contain representations about the environment, which affects the prediction behavior and can associate with the real environment information.

Automated summary

GNSS performance in urban canyons can be significantly degraded by interferences, but deep learning networks can predict these interferences and improve predictive accuracy. Experimental results show that the proposed deep learning networks can accurately predict satellite visibility and pseudorange errors.