Learning-based NLOS Detection and Uncertainty Prediction of GNSS Observations with Transformer-Enhanced LSTM Network

TL;DR

This work tackles GNSS localization in challenging urban environments where NLOS and multipath distort observations. It introduces a transformer-enhanced LSTM network that exploits spatio-temporal features from GNSS observations to detect NLOS receptions and predict pseudorange errors, with an attention mechanism across satellites and Bi-LSTM temporal processing. Evaluations on Aachen and Hong Kong datasets, plus cross-dataset and out-of-distribution tests, show improved NLOS detection (higher precision/recall) and better trajectory consistency when NLOS observations are excluded in state estimation. The approach also demonstrates robust generalization and benefits from feature diversity, with ablations confirming the value of attention and temporal modeling. The authors release code and Aachen data to support further research, and plan online integration into GNSS fault-exclusion workflows for real-time localization.

Abstract

The global navigation satellite systems (GNSS) play a vital role in transport systems for accurate and consistent vehicle localization. However, GNSS observations can be distorted due to multipath effects and non-line-of-sight (NLOS) receptions in challenging environments such as urban canyons. In such cases, traditional methods to classify and exclude faulty GNSS observations may fail, leading to unreliable state estimation and unsafe system operations. This work proposes a deep-learning-based method to detect NLOS receptions and predict GNSS pseudorange errors by analyzing GNSS observations as a spatio-temporal modeling problem. Compared to previous works, we construct a transformer-like attention mechanism to enhance the long short-term memory (LSTM) networks, improving model performance and generalization. For the training and evaluation of the proposed network, we used labeled datasets from the cities of Hong Kong and Aachen. We also introduce a dataset generation process to label the GNSS observations using lidar maps. In experimental studies, we compare the proposed network with a deep-learning-based model and classical machine-learning models. Furthermore, we conduct ablation studies of our network components and integrate the NLOS detection with data out-of-distribution in a state estimator. As a result, our network presents improved precision and recall ratios compared to other models. Additionally, we show that the proposed method avoids trajectory divergence in real-world vehicle localization by classifying and excluding NLOS observations.

Figures (6)

• Figure 1: Proposed labeling framework
• Figure 2: Sky plots from two scenes: a) urban tree-rich area, b) urban. The received GPS satellites are marked in orange.
• Figure 3: Proposed Neural Network Architecture.
• Figure 4: Predicted and ground-truth pseudorange errors with the model lstm_base and our network.
• Figure 5: Feature Permutation Importance of different models.