Precise positioning using smartphones has been a topic of interest especially after Google decided to provide raw GNSS measurement through their Android platform. Currently, the greatest limitations in precise positioning with smartphone Global Navigation Satellite System (GNSS) sensors are the quality and availability of satellite-to-smartphone ranging measurements. Many papers have assessed the quality of GNSS pseudorange and carrier-phase measurements in various environments. In addition, there is growing research in the inclusion of a priori information to model signal blockage, multipath, etc. In this contribution, numerical estimation of actual range errors in smartphone GNSS precise positioning in realistic environments is performed using a geodetic receiver as a reference. The range errors are analyzed under various environments and by placing smartphones on car dashboards and roofs. The distribution of range errors and their correlation to prefit residuals is studied in detail. In addition, a comparison of range errors between different constellations is provided, aiming to provide insight into the quantitative understanding of measurement behavior. This information can be used to further improve measurement quality control, and optimize stochastic modeling and position estimation processes.

The ubiquity of smartphones catalyzes myriad smartphone-based Internet of Things (IoT) applications, amongst which smartphone positioning which utilizes global navigation satellite system (GNSS) observations to provide spatial information plays a crucial role. However, noisy smartphone GNSS measurements prevent decimeter-level positioning performance in user environments. Recovering the integer property of GNSS carrier phase measurement ambiguities shows great potential in achieving high-accuracy positioning solutions. However, inaccurate ambiguity estimates and short signal wavelengths are the main barriers to successful ambiguity resolution (AR). Therefore, a partial AR with wide-lane (WL) ambiguities and an automatic ambiguity hold strategy is proposed. Simulated results show that, for single-epoch WL AR, even with one WL ambiguity correctly fixed, the positioning solution can be improved by 3 cm. With actual static and kinematic datasets, the proposed algorithm is evaluated and validated. Static results show an improvement of 83% in horizontal position when the WL AR approach is applied, and positioning accuracies can reach 6.8, 2.9, and 11.5 cm in the E, N, and U direction components, respectively. For kinematic data collected in highly variable realistic driving environments, the time series of positioning errors of WL AR solutions exhibit less variation than float solutions. And with fixed WL ambiguities, solutions can be improved to varying degrees, ranging from several centimeters to up to 8 dm depending on the environment. The largest improvement of 8 dm is observed for 95th percentile horizontal positioning errors under a suburban environment.

The global navigation satellite system (GNSS) positioning performance is widely recognized to degrade in challenging environments due to complex observation uncertainty and variability. In GNSS positioning framework, the stochastic model plays a crucial role in achieving unbiased parameter estimation by correctly capturing statistical observation characteristics. However, classical empirical stochastic models, such as elevation-dependent and carrier-to-noise ratio (C/N0)-based weighting schemes, are deemed inadequate for kinematic positioning services. The environmental context, which comprises terminal space and received signal characteristics, possesses the potential to promote quantifying observation uncertainty. To address this issue, we propose to construct context-aware adaptive GNSS stochastic models that integrate context information. Firstly, leveraging a substantial urban vehicle GNSS dataset, code residuals are accurately extracted using high-accuracy reference trajectory, with significant context features extracted and selected. Secondly, a temporal neural network (TNN) is developed for urban scenario recognition, including open sky, urban canyon, boulevard, and under viaduct. Additionally, context-aware stochastic functions are formulated through correlation analysis and function fitting between C/N0 and median code residuals. Finally, the context-aware stochastic model can be adaptively configured with context prediction results yield by TNN. Experimental results show the context detection model achieves an high-confidence accuracy of 95.37%. Furthermore, the proposed method is evaluated and compared with the classical weighting schemes using code-based single point positioning (SPP). Remarkably, the context-aware stochastic model surpasses the classical C/N0-based stochastic model in terms of continuity and accuracy, representing improvements of 17.60 and 22.39% for horizontal and vertical components, respectively, thereby highlighting its remarkable environment adaptability.