Diff-GNSS: Diffusion-based Pseudorange Error Estimation

2025-09-22

AI Summary

In urban environments, GNSS suffers from severe multipath and non-line-of-sight (NLOS) effects, resulting in complex, non-Gaussian pseudorange error distributions and substantial degradation in positioning accuracy. To address this, we propose the first diffusion-model-based “coarse-to-fine” pseudorange error estimation framework. First, a Mamba-based module performs rapid coarse error estimation; subsequently, a conditional denoising diffusion process—conditioned on GNSS observability quality features—refines the error modeling. We further introduce, for the first time, a per-satellite uncertainty quantification mechanism to enhance prediction reliability and interpretability. Evaluated on both public and proprietary real-world datasets, our method consistently outperforms existing state-of-the-art approaches. Moreover, its modular design is plug-and-play, enabling seamless integration into diverse GNSS positioning systems.

Technology Category

• Intelligent Robots: State Estimation
• Machine Learning: Calibration & Uncertainty Quantification
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Global Navigation Satellite Systems (GNSS) are vital for reliable urban positioning. However, multipath and non-line-of-sight reception often introduce large measurement errors that degrade accuracy. Learning-based methods for predicting and compensating pseudorange errors have gained traction, but their performance is limited by complex error distributions. To address this challenge, we propose Diff-GNSS, a coarse-to-fine GNSS measurement (pseudorange) error estimation framework that leverages a conditional diffusion model to capture such complex distributions. Firstly, a Mamba-based module performs coarse estimation to provide an initial prediction with appropriate scale and trend. Then, a conditional denoising diffusion layer refines the estimate, enabling fine-grained modeling of pseudorange errors. To suppress uncontrolled generative diversity and achieve controllable synthesis, three key features related to GNSS measurement quality are used as conditions to precisely guide the reverse denoising process. We further incorporate per-satellite uncertainty modeling within the diffusion stage to assess the reliability of the predicted errors. We have collected and publicly released a real-world dataset covering various scenes. Experiments on public and self-collected datasets show that DiffGNSS consistently outperforms state-of-the-art baselines across multiple metrics. To the best of our knowledge, this is the first application of diffusion models to pseudorange error estimation. The proposed diffusion-based refinement module is plug-and-play and can be readily integrated into existing networks to markedly improve estimation accuracy.

Problem

• Estimating pseudorange errors in GNSS systems affected by multipath interference
• Modeling complex error distributions that limit learning-based positioning methods
• Achieving controllable error estimation using conditional diffusion models

Innovation

• Coarse estimation using Mamba-based module
• Fine-grained refinement with conditional diffusion model
• Controllable synthesis using GNSS quality features