In May 2016, the availability of GNSS raw measurements on smart devices was announced by Google with the release of Android 7. It means that developers can access carrier-phase and pseudorange measurements and decode navigation messages for the first time from mass-market Android-devices. In this paper, an improved Hatch filter algorithm, i.e., Three-Thresholds and Single-Difference Hatch filter (TT-SD Hatch filter), is proposed for sub-meter single point positioning with raw GNSS measurements on Android devices without any augmentation correction input, where the carrier-phase smoothed pseudorange window width adaptively varies according to the three-threshold detection for ionospheric cumulative errors, cycle slips and outliers. In the mean time, it can also eliminate the inconsistency of receiver clock bias between pseudorange and carrier-phase by inter-satellite difference. To eliminate the effects of frequent smoothing window resets, we combine TT-SD Hatch filter and Kalman filter for both time update and measurement update. The feasibility of the improved TT-SD Hatch filter method is then verified using static and kinematic experiments with a Nexus 9 Android tablet. The result of the static experiment demonstrates that the position RMS of TT-SD Hatch filter is about 0.6 and 0.8 m in the horizontal and vertical components, respectively. It is about 2 and 1.6 m less than the GNSS chipset solutions, and about 10 and 10 m less than the classical Hatch filter solution, respectively. Moreover, the TT-SD Hatch filter can accurately detect the cycle slips and outliers, and reset the smoothed window in time. It thus avoids the smoothing failure of Hatch filter when a large cycle-slip or an outlier occurs in the observations. Meanwhile, with the aid of the Kalman filter, TT-SD Hatch filter can keep continuously positioning at the sub-meter level. The result of the kinematic experiment demonstrates that the TT-SD Hatch filter solution can converge after a few minutes, and the 2D error is about 0.9 m, which is about 64%, 89%, and 92% smaller than that of the chipset solution, the traditional Hatch filter solution and standard single point solution, respectively. Finally, the TT-SD Hatch filter solution can recover a continuous driving track in this kinematic test.

Precise orbit knowledge is a fundamental requirement for low Earth orbit (LEO) satellites. High‐precision non‐gravitational force modeling directly improves the overall quality of LEO precise orbit determination (POD). To address the potential systematic errors in solar radiation pressure (SRP), we introduce observed radiation data and modeled physical effects to describe the real in‐flight environment of satellites. Time‐dependent solar irradiance data and a highly physical shadow model are considered for SRP modeling. We develop an advanced thermal reradiation model for satellite solar panels. A set of improved non‐gravitational force models is performed for LEO POD, and we discuss the benefits of the enhanced dynamic models on orbit quality and dependence on empirical parameters. The Gravity Recovery and Climate Experiment Follow‐On (GRACE‐FO), Jason‐3, and Haiyang‐2B missions are selected for the POD process. Estimated empirical acceleration and scale parameters and independent satellite laser ranging (SLR) are used to validate the final orbit solutions. The magnitude of empirical acceleration estimated in POD is reduced by 19% with the enhanced dynamic modeling, and the estimated scale factor for the SRP converges to stable and reasonable level. Furthermore, the steady‐state temperature model used in thermal reradiation can effectively reduce mismodeled effects in the SRP, and the systematic linear dependency revealed by the SLR residuals is significantly reduced for the GRACE‐C and Jason‐3 satellites, with improvements of approximately 61% and 49%, respectively. Overall, advances are made in the explicit modeling of non‐gravitational forces to pursue superior satellite orbits, suggesting a more dynamic orbit solution.