Deep-Learning-Based Localization Approach with Pseudorange for Pseudolite Systems

Abstract

The pseudolite system can broadcast navigation signals and can be used as a good supplement or substitute to the global navigation satellite systems. The pseudorange measurement is usually used to calculate the initial position. However, the pseudorange positioning is still faced with the multipath problem. To improve the positioning accuracy, a deep-learning localization method is proposed in this paper. The proposed method mainly includes the offline initial phase and the online localization phase. During the offline phase, the pseudorange data is collected and transformed into single-differenced pseudorange. The database is expanded by copying the pseudorange vectors and masking pseudorange values. Then a residual fully connected neural network (ResFCNN) is used to learn the mapping relationship between single-differenced pseudorange and location. A residual connection across fully connected layers is added in the ResFCNN to strengthen the learning ability of neural networks. During the online phase, the trained model and real-time pseudoranges are used to predict the location. Experimental results show that the root-mean-square error (RMSE) of the proposed method is 0.73 m under a multipath condition and 1.21 m with a signal absence, which are reductions in RMSE of 81 and 85%, respectively, compared to the conventional iterative least squares method.