Precise Positioning with Smartphone GNSS Observations

Abstract

In recent years, the use of low-cost and ultra-low-cost receivers has been drastically increasing due to their low cost compared to the high-end geodetic receivers. In May 2016, Google revealed that users would be able to access the raw global navigation satellite system (GNSS) measurements through a new API implemented on Android 7 or later versions. This can be marked as a significant breakthrough for the GNSS community. However, further considerations are necessary to achieve accurate positioning with these mass-market devices. The objective of this research is to investigate the performance of precise point positioning (PPP) using the smartphone GNSS observations. The quality of raw GNSS smartphone observations is investigated first. Despite the availability of raw GNSS data through Google's android.location API, generating pseudorange observations remains challenging, leading to inconsistencies in observations generated by the logging applications such as GnssLogger and Geo++ RINEX Logger. The quality of raw GNSS observation logged by GnssLogger and Geo++ RINEX Logger is investigated from different aspects, including the inconsistency between the pseudorange, carrier-phase and Doppler measurements, presence of some carrier-phase observations without changes over time and its possible reasons. The results indicate that consistency between the generated pseudorange, carrier-phase and Doppler observations from Android smartphone devices was not fully met in the RINEX outputs of the GnssLogger and Geo++ RINEX Logger Apps. To address this concern, the first contribution introduces an in-house software, named UofC CSV2RINEX written in C++, to convert the CSV files into the RINEX files and to ensure efficient conversion and consistency. In the second and third contributions, the study focuses on defining the stochastic and functional models applied to PPP using the smartphone GNSS observations, respectively. The study utilizes the least-squares variance component estimation (LS-VCE) method to assess the noise characteristics of the smartphone GNSS observations. A C/N0 and elevation-dependent weighting model is developed, and the parameters are obtained by applying the LS-VCE method to the double-differenced (DD) code and phase observations of the three smartphones, Samsung S20, Google Pixel 5 and Xiaomi Mi8. The findings indicate that there is no significant correlation between GNSS code and carrier-phase observations on the L1 frequency. Moreover, employing the derived stochastic model results in an improvement in positioning performance. The PPP method can be implemented in both combined and uncombined forms. In this research, the uncombined PPP (UPPP) model is employed. Three representations of UPPP models are examined to evaluate their performance in terms of statistical reliability and positioning accuracy. They are defined as follows: (1) UPPP1: Estimating separate receiver clocks for each frequency, (2) UPPP2: Estimating one receiver clock and one receiver differential code bias (DCB) and (3) UPPP3: Estimating one receiver clock and one receiver DCB (another representation). The minimum detectable bias (MDB) values, as an internal reliability indicator, demonstrate nearly similar results across all three models for code observations, due to their inherent noise characteristics. However, carrier-phase observations reveal smaller MDBs, with UPPP1 identified as the least reliable model. Regarding external reliability, the total positioning error (TPE) values indicate slightly better performance for UPPP3 compared to UPPP1 and UPPP2. Overall, UPPP3 demonstrates the best horizontal and vertical positioning accuracy in both static and kinematic tests. Finally, the fourth contribution of this study introduces the height-constrained UPPP model, which incorporates height information into the observation equations as the weighted constraints. While previous PPP smartphone positioning studies have mainly used GNSS observations from the smartphone's API, adding supplementary data such as height can improve accuracy and stability. This integration increases the degree of freedom and strengthens the geometric relationship between the receiver and satellites. Using both pedestrian walking and vehicular datasets, the height-constrained UPPP model demonstrates better performance compared to the UPPP model without height constraints.