Towards End-to-End GPS Localization with Neural Pseudorange Correction

• DOI: 10.48550/arxiv.2401.10685
• Publication Date: 2024-01-01
• Authors: Xu Weng; Keck Voon Ling; Haochen Liu; Kun Cao
• Keywords: Pseudorange; End-to-end principle

Pseudorange errors are the root cause of localization inaccuracy in GPS. Previous data-driven methods regress and eliminate pseudorange using handcrafted intermediate labels. Unlike them, we propose an end-to-end GPS framework, E2E-PrNet, to train a neural network for correction (PrNet) directly final task loss calculated with ground truth receiver states. The gradients respect learnable parameters backpropagated through differentiable nonlinear least squares optimizer PrNet. feasibility is verified data collected by Android phones, showing that E2E-PrNet outperforms state-of-the-art methods.