Adaptive Best Integer Equivariant Estimator for Smartphone RTK

Ananya Vishwanath, Hongzhou Yang

Peer Reviewed

Abstract:	Achieving reliable RTK performance on smartphones is difficult as measurement quality varies significantly across environments, influencing both residual behaviour and ambiguity resolution. This paper presents an adaptive Best Integer Equivariant (BIE) framework that links environment classification with distribution-aware ambiguity weighting. A GNSS dataset spanning highway, suburban, and downtown conditions is constructed using visibility, SNR, and geometry-based features, and a lightweight LSTM classifier trained on this dataset provides per-epoch environment labels. The classifier achieves 97.6–99.1% accuracy across the three environments in training and 89% accuracy on an independent suburban–highway test trajectory. Using these predicted environments, standardised ambiguity residuals are modelled with Gaussian, contaminated-normal, and Student-t distributions, and the fitted parameters define environment-specific weighting functions within the BIE estimator. Evaluation is performed on a raw observation collected using a Google Pixel 7 Pro on a suburban to highway driving segment. The adaptive BIE solution limits the larger excursions seen in both float and LAMBDA-fixed processing and achieves the lowest overall positioning error, improving 2D and 3D RMSE by 14.5% and 9.8% relative to the float solution, and by 3.6% and 10.3% relative to the fixed solution. These results demonstrate that coupling lightweight environment recognition with distribution-specific ambiguity modelling can provide reliable positioning solutions in varied environments.
Published in:	Proceedings of the 2026 International Technical Meeting of The Institute of Navigation January 26 - 29, 2026 Hyatt Regency Orange County Anaheim, California
Pages:	527 - 539
Cite this article:	Vishwanath, Ananya, Yang, Hongzhou, "Adaptive Best Integer Equivariant Estimator for Smartphone RTK," Proceedings of the 2026 International Technical Meeting of The Institute of Navigation, Anaheim, California, January 2026, pp. 527-539. https://doi.org/10.33012/2026.20561
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In