Page 107 - 卫星导航2021年第1-2合期

Shinghal and Bisnath Satell Navig (2021) 2:10 Page 10 of 17

Table 5 PPP 2D RMS positioning accuracy using diferent Measurement prediction
stochastic models for smartphone reset experiment, DOY 146, A measurement prediction technique has been devised
2019 to predict missing measurements to increase position-
Time 2D RMS error (m) C/ 2D RMS error (m) Percentage ing solution availability. As mentioned earlier, it was
Interval N -based weighting Elevation-based diference extremely difcult to tune the noise parameters for the
0
(min) weighting (%) highly variable smartphone raw GNSS measurements in
0–20 1.7 2.8 64.7 realistic environments in an EKF flter. Accordingly, the
20–40 5.2 5.3 1.9 EKF flter could not used for measurement prediction.
40–60 4.1 5.1 24.4 Hence, a separate measurement prediction technique
had to be devised. Various real-time extrapolation
and estimated Doppler prediction techniques were
tested; however, they were discarded for a simple linear
Te RMS horizontal errors for the 20-min segment for
two weighting strategies are highlighted in Table 5. extrapolator, as it provides lower prediction error for
Tese results depict the efectiveness of a C/N -based flling data-gaps in low to medium multipath environ-
0
technique due to signifcantly reduced initialization error ments. For example, the estimated Doppler prediction
on each reset and lower RMS error. Overall, the mean technique (Li et al. 2019) is limited by a lack of knowl-
RMS error was 30% lower for the C/N -based weighting edge of dynamics without the aid of an Inertial Meas-
0
strategy for the three segments. urement Unit (IMU). Te current research focuses on
Te post-ft residuals for the three weighing models are GNSS-only processing. Te logged Doppler measure-
compared in Table 6. ments show large variability and gaps, as can be seen in
Te C/N -based model outperformed the other two Fig. 12 and therefore cannot be used for prediction.
0
models in terms of residual magnitude, as there is a For satellite G32, there are no L1 carrier-phase meas-
decrease in residual magnitude with increasing C/N urements depicted by the jumps in the logged Dop-
0
values. No such dependence can be observed for the pler measurements, even though the satellite had a
residuals and the elevation angle as seen in Fig. 11, where mean elevation of above 70° and a C/N value averag-
0
the C1 post-ft residuals for three satellites have plotted ing 35 dB·Hz. Te lack of L1 carrier-phase measure-
against C/N and elevation angle. ments in such a scenario could be attributed to the
0
low-cost antenna. For satellite E30, the L5 Doppler

Table 6 Post-ft residual RMS for diferent stochastic models for Xiaomi MI 8, DOY 146, 2019

Scenario Post-ft C1 (m) Post-ft L1 (cm) Post-ft C5 (m) Post-ft L5 (cm)
Static stochastic model 13.8 32.7 2.3 25.8
C/N -based stochastic model 4.3 6.0 1.9 4.3
0
Elevation-based stochastic model 5.0 9.0 2.1 6.1

Fig. 11 Variation of PPP C1 post-ft residuals with C/N and elevation angle for Xiaomi MI 8, DOY 146, 2019
0