ConvGRU-MHM: a CNN GRU-enhanced MHM for mitigating GNSS multipath

Abstract

In high-precision global navigation satellite system (GNSS) short-baseline positioning, multipath is the main source of errors. If the station environment is quasi-static, repeat periods of satellites can be utilized to generate time- or space-dependent multipath models to mitigate multipaths. However, two general problems are associated with the multipath models constructed based on satellite mechanics: (1) an accuracy decrease occurs when the above models are applied to multipath mitigation over a long time-span; (2) when constructing the spatial and temporal grids of the satellite-based spatially dependent multipath model, it is challenging to balance computational efficiency and spatial resolution. We propose a convolutional neural network-gated recurrent unit enhanced multipath hemispherical map (ConvGRU-MHM) in the observational domain to address these problems. The proposed method directly mines the deep features of elevation, azimuth angle, and multipath and the mapping relationship between these to establish a real-time prediction model. The predicted multipath is obtained and returned to the observation equation for multipath mitigation when the real-time position of the satellite is placed in the pre-trained model. We compared the multipath mitigation performance of sidereal filtering (SF) and a multipath hemispherical map (MHM) with that of the ConvGRU-MHM to demonstrate the advantages of the proposed method. The experimental results are as follows: (1) in the short time-span (first 20 days), the mean accuracy improvements of the ConvGRU-MHM in the E/N/U direction performed better than those of the SF and MHM; and (2) in the long-term time (after 50 days), the mean accuracy improvements of the ConvGRU-MHM in the E/N/U direction are higher than that of the SF and MHM by 10–20%. As a lightweight model, the ConvGRU-MHM can effectively improve the measurement accuracy of GNSS real-time monitoring in fields, such as deformation monitoring and seismic research.

Deep Learning CNN-GRU Method for GNSS Deformation Monitoring Prediction

“…Meanwhile, the GRU model represents an improved model of the RNN, adept at capturing long-term dependencies in the dataset [35]. Previous studies have applied CNN for GNSS residual processing [36] and classification [37], GRU for landslide prediction [38] and multipath modeling [39], as well as CNN-GRU for forecasting wind power [40], particulate matter concentrations [41], and the pressure of a concrete dam [42]. The CNN-GRU model integrates the characteristics of CNN and GRU, which has better a performance in predictions of GNSS deformation monitoring.…”