Stand-alone, double differenced, and residual-based noise analysis of smartphone raw GNSS observations

After Google’s release of Android Nougat (Android 7) in 2016, raw GNSS observations of Android smartphones have been freely accessible to the GNSS research community. Using smartphone raw GNSS measurements, numerous theories and practical methods were developed to achieve high precision in positioning, navigation, timing, and GNSS remote sensing. However, due to the chipset and built-in antenna quality, smartphone observations have more powerful noises compared to the geodetic grade GNSS receivers. Besides, frequent interruptions and cycle slips are present in smartphone observations. Therefore, it is difficult to achieve high positioning accuracy with smartphones, since it requires more sophisticated data processing methods. Smartphone stochastic modeling can be included in these methods since the variation of observation noise follows behaviors that are difficult to represent as a function of satellite elevation angle, while carrier-to-noise ratio (C/N0) representation provides a more suitable weighting scheme e.g., for smartphone Precise Point Positioning (PPP).
In addition to the current developments, analyzing the time-correlation behavior of smartphone observation noise can help to develop resilient algorithms for positioning, such as the detection of cycle slips, outliers, and adaptive filtering methods. In this contribution, we present noise analysis of smartphone raw GNSS measurements from different perspectives, namely stand-alone, double-differenced, and based on PPP residuals. To eliminate systematic effects such as the multipath effect from smartphone observations, we use a Kalman Filter algorithm and we compare it with Least Squares Harmonic Estimation (LS-HE). To investigate the time-correlation property of filtered observations, we use autocorrelation and Allan Deviation (AD) methods. Results showed that even if the strong periodicities of the multipath effect are filtered, there are residual multipath effects that remain in smartphone observations and residuals. Furthermore, AD analysis showed that smartphone observations/residuals contain white noise and time-correlated noise that exhibits similar characteristics to a Gauss–Markov process. Using these insights, our study discusses the influence of the weighting scheme and time-correlated errors on smartphone PPP and recently developed adaptive filtering techniques with the goal of improving PPP performance. We aim to provide a foundation for further advancements in modeling and mitigating multipath and time-correlated smartphone observation noises.