Improving Precise Smartphone GNSS with Robust Dynamics, Adaptive Stochastics, and Cycle Slip Repair

Abstract

The proliferation of Global Navigation Satellite System (GNSS) chipsets in mass-market smartphones has democratized access to positioning data, however, attaining decimeter-level precision remains a challenge due to the limitations of low-cost hardware. Smartphone observations are characterized by low Carrier-to-Noise density (𝐶/𝑁!), high susceptibility to multipath, and frequent loss of phase lock. Standard positioning algorithms, originally designed for geodetic-grade receivers, struggle in this environment as they rely on heuristic dynamic models, static stochastic weighting, and rigid integer ambiguity resolution strategies that discard corrupted phase data. This thesis proposes a comprehensive framework to enable precise smartphone positioning by addressing three critical failure points: erratic user dynamics, environmental volatility, and carrier-phase discontinuity. First, a Robust Dynamics model is developed using a Doppler-Based Prediction (DBP) technique. By using Doppler-derived velocity for state propagation, this method automates the estimation of process noise (Q), allowing the filter to instantaneously adapt to unconstrained user motion. Experimental validation demonstrates that DBP reduces horizontal positioning errors by approximately 57% in kinematic scenarios compared to standard models. Second, an Adaptive Stochastic model is implemented using Variance Component Estimation (VCE). This Adaptive Kalman Filter (AKF) learns the true measurement noise (R) in real-time, effectively down-weighting multipath-affected signals without discarding them. This approach yields horizontal accuracy improvements of 35% in static and 27% in vehicular environments compared to traditional elevation and 𝐶/𝑁! based weighting. Finally, the research introduces a hierarchical Cycle Slip Detection and Repair (CSDR) framework. Moving beyond the standard "detect-and-reset" paradigm, which destroys filter convergence, this method employs a hybrid detection scheme and a stochastic repair engine. Using Partial Ambiguity Resolution (PAR) with a Fixed Failure-Rate Ratio Test (FF-RT), cycle slips are estimated as integers and restored into the filter with associated variance. Validation on Google Pixel 4 datasets confirms that this stochastic repair strategy maintains phase continuity, reducing vertical positioning RMSE by approximately 45% compared to standard reset methods. Collectively, these contributions demonstrate that high-precision smartphone positioning is achievable not by filtering out noisy data, but by rigorously modeling its dynamic and stochastic characteristics.