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