Clustering-Assisted Observation Domain Optimization for GNSS Multi-Fault Detection and Correction

With the rise of autonomous and semi-autonomous vehicles, effective fault detection and mitigation (FDM) methods have become essential in meeting the integrity requirements for precise and reliable Global Navigation Satellite System (GNSS)-based positioning. The existing GNSS-only FDM methods are ineffective in scenarios involving multiple faulty observations due to either their theoretical model limitations or high computational costs. In this paper, a novel GNSS multi-fault detection and correction method is developed that achieves a balance between computational complexity and performance. The proposed method incorporates an Expectation Maximization (EM) framework to jointly estimate an approximate maximum likelihood of states and latent model parameters in the presence of observation outliers, i.e., faults. However, the EM algorithm is known for its high computational complexity. To reduce the computational complexity of EM, an importance sampling step based on unsupervised clustering is introduced. As demonstrated by the results and analysis herein, the proposed method outperforms the existing Least-squares Residuals method, achieving an average improvement of up to 48% in positioning accuracy. Additionally, the computational complexity of the proposed method is an order of magnitude lower than the state-of-the-art Solution Separation method. The method enhances positioning reliability at a lower computational cost and does not require additional infrastructure, hence, can be readily integrated into standalone real-time GNSS applications