Receiver Clock Prediction-Aided GNSS Positioning Using Factor Graph Optimization With at Least Three Satellites Visible

Global navigation satellite system (GNSS) receiver clock modeling methods are extensively used to improve positioning accuracy and continuity. However, traditional receiver clock modeling-aided navigation solvers often fail to provide reliable positioning results in complex wireless environments, particularly under limited satellite visibility conditions. This limitation arises from two primary factors: the inadequate accuracy of receiver clock parameter modeling and the insufficient robustness of conventional navigation solvers in challenging operational scenarios. To address these issues, we propose a two-stage, data-driven, dual attention mechanism-enhanced temporal convolutional network (DAME-TCN) that improves the precision of receiver clock modeling. Additionally, we integrate this enhanced network into a factor graph optimization (FGO)-based navigation solver, ensuring robust and accurate positioning performance, even in highly complex wireless environments. The perceptual field size of the temporal convolution network (TCN) was adjusted according to feature lengths to improve memory capacity. Self-attention and soft-weighted attention mechanisms were used to improve the modeling capacity and prediction performance of the TCN for periodicity. Furthermore, the outcomes of the predictor are employed to impose constraints on the receiver clock state within the FGO framework. The clock prediction and positioning performance of DAME-TCN-based receivers were validated using data from the International GNSS Service, and an open-source localization dataset for urban canyon environments. The results demonstrate that the DAME-TCN can effectively predict the receiver clock and improve positioning accuracy and continuity when at least three satellites are visible.