Abstract
The Japanese QuasiZenith Satellite System (QZSS) has recently (October 2017) reached its first 4satellite constellation. In this contribution, the standalone performance of this 4satellite QZSS constellation is assessed by means of its triplefrequency (L1 + L2 + L5) realtime kinematic (RTK) integer ambiguity resolution and precise positioning capabilities. Our analyses are carried out for data collected in Perth, Australia, and include a study of the noise characteristics of the QZSS code and phase data, particularly concerning their precision, time correlation and multipath. Our results show that while the phase observations on different frequencies are of similar precision, the code observations on different frequencies show considerably different precisions and can be ordered, from high to low, as L5, L2 and L1. As to positioning and ambiguity resolution, we demonstrate that the Position Dilution Of Precision (PDOP) and the Ambiguity Dilution Of Precision (ADOP) exhibit complementary characteristics, both of which are important for predicting precise positioning capabilities. We show that despite the large PDOPs, the ADOPs are sufficiently small to indicate (almost) instantaneous successful ambiguity resolution. This is confirmed by our empirical data analyses, demonstrating that instantaneous ambiguity resolution is feasible, despite the relatively poor 4satellite receivertosatellite positioning geometry over Australia, thus showing that already now centimeterlevel standalone QZSS positioning is possible with the current 4satellite constellation (February–March 2018).
Introduction
The QuasiZenith Satellite System (QZSS) is a Japanese satellite positioning system operated by Japan Aerospace Exploration Agency (JAXA). With the orbits specially designed to increase highelevation signals in Japan, QZSS has been developed as a GPScomplementary system to increase the availability, reliability, integrity and accuracy of positioning in the AsiaOceania region, especially in urban canyons and mountainous areas (QZSS 2018; Wu et al. 2004a). The first QZSS satellite (QZS1) is in orbit since 2010. With the second (QZS2), the third (QZS3) and the fourth QZSS satellites (QZS4) launched, respectively, in June, August and October of 2017, QZSS has now completed a foursatellite constellation. QZSS is planned to be extended to a sevensatellite system by 2023 (Murai 2015). Except for QZS3, which is a geostationary (GEO) satellite, the other three QZSS satellites are in quasizenith orbits (QZO) with an inclination of around 43 ± 4°. Having similar figureeight shaped ground tracks, the QZO satellites spend an extended time with high elevations above Japan with a separation of 8 h passing the same region (Teunissen and Montenbruck 2017). Figure 1 [left] shows the colormap indicating the percentage of a 24h period with atleast one QZSS satellite (out of QZS1, 2, 3 and 4) visible above 70° of elevation on 2 February 2018. We note that this percentage indeed reaches 100% in Japan. The middle panel in Fig. 1 illustrates the ground tracks of the four QZSS satellites where J1, J2, J3 and J7 are, respectively, the PRN numbers of QZS1, 2, 4 and 3. The colormap shown in Fig. 1 [right] indicates the percentage of a 24h period on 2 February 2018 with all the four QZSS satellites being visible simultaneously where a cutoff elevation angle of 10° is considered. It can be seen that there is an eyeshaped area in AsiaOceania region, covering a large part of Australia, where all the four QZSS satellites always lie above 10° of elevation.
The QZSS satellites send GPScompatible signals on L1 (1575.42 MHz), L2 (1227.6 MHz) and L5 (1176.45 MHz) and augmentation signals, i.e. the L1SAIF (SubmeterClass Augmentation with Integrity Function) signal and the LEX (Lband Experimental) signal. Signal analyses were performed by different studies after the first QZSS satellite was launched. Using the signals from QZS1, Nie et al. (2015) observed similar multipath/noisecombined behaviors for QZSS and GPS on each of the three frequencies after forming the multipath combination (Leick et al. 2015), while Hauschild et al. (2012) reported lower standard deviations of the noise and multipath errors on QZSS L5 than those on GPS L5. In the zenith direction, both of them amounted to around 5 cm. Based on zerobaseline tests, the standard deviations of the doubledifferenced carrierphase residuals were reported to be between 0.5 and 1 mm for QZSS L1, L2 and L5 (Quan et al. 2016), and Nadarajah et al. (2016) showed that the QZSS L5 code observations have smaller zenithreferenced standard deviation compared to the QZSS L1 code signals. A recent study by Steigenberger et al. (2018) investigates the QZSS signal power and clock performance.
Having reached a foursatellite constellation just recently, QZSS positioning performance has hitherto been studied only in combination with the other Global Navigation Satellite Systems (GNSSs). Comparative studies on GPSonly versus GPS + QZSS positioning can be found in Wu et al. (2004b) and Choi et al. (2015). Nadarajah et al. (2016) assessed multiGNSS singlefrequency realtime kinematic (RTK) positioning and attitude determination performance through combining QZSS L1/L5 signals with their counterparts from GPS, Galileo and IRNSS. RTK results of QZSS in combination with GPS, Galileo and Beidou were presented in, e.g. Odolinski et al. (2015), Quan et al. (2016) and Odolinski and Teunissen (2017).
With four operational QZSS satellites available in orbit, it is now possible to carry out positioning based on standalone QZSS. In this contribution, using the signals from all the four QZSS satellites on the three frequencies L1, L2 and L5, a first assessment of standalone QZSS ambiguity resolution performance and positioning accuracy is presented. After introducing the processing strategy, analysis of signal precision is performed for L1, L2 and L5. The impact of multipath on both the code and phase observations is discussed and the zenithreferenced variance matrices are presented before and after multipath mitigation. Next, considering a shortbaseline RTK model, the singleepoch ambiguity resolution performance is investigated based on the QZSS triplefrequency observations, through ambiguity dilution of precision (ADOP) and integer bootstrapped success rate. It is thereby shown that instantaneous successful ambiguity resolution is feasible using both the multipathuncorrected and mitigated observations. Afterward, the corresponding positioning accuracies are analyzed and compared for ambiguityfloat and fixed cases under singleepoch RTK model without and with multipath mitigation. We further extend our evaluations to a longer baseline for which the differential atmospheric delays are not negligible. A summary of conclusions is provided at the end.
Processing strategy
For baselines within a few kilometers, it is assumed that the ionospheric delays are significantly reduced by forming betweenreceiver differences of the GNSS signals (Goad 1998). The doubledifferenced (DD) observedminuscomputed code and phase observations transmitted by m satellites on f frequencies to a pair of receivers at a single epoch can be formulated as (Teunissen and Montenbruck 2017)
where \({E}\left[ \cdot \right]\) represents the expectation operator, and \(\otimes\) denotes the Kronecker product (Henderson et al. 1983). The DD code and phase observations are, respectively, represented by \(p={\left[ {p_{1}^{{T}}, \ldots ,p_{{f}}^{{T}}} \right]^{T}}\) and \(\phi ={\left[ {\phi _{1}^{{T}}, \ldots ,\phi _{{f}}^{{T}}} \right]^{T}}\), where \({p_j}\) and \({\phi _j}\,\left( {j=1, \ldots ,f} \right)\) contain the \((m  1)\) DD code and phase observations on frequency \(j\). The symbols \({e_f}\) and \({I_{m  1}}\) stand for a vector of ones with the length of \(f\) and identity matrix with the size of \((m  1)\), respectively. The \(f \times f\) diagonal matrix \(\Lambda\) has as its diagonal entries \({\lambda _j}\), the wavelength of frequency \(j\,(j=1, \ldots ,f)\). The \((m  1)\)vector \(\widetilde {\rho }\) captures the nondispersive parameters including the DD geometric ranges \(\rho\) and DD slant tropospheric delays \(\tau\), and \(a\) denotes the \(f(m  1)\)vector of DD ambiguities. The dry part of the slant tropospheric delays can a priori be corrected using Saastamoinen model (Saastamoinen 1972), and the remaining residual tropospheric delays (wet part) \(\Delta \tau\) can be parametrized in the scalar Zenith Tropospheric Delay (ZTD) \(\Delta {\tau _{\text{Z}}}\). One can further linearize the link between the DD geometric ranges and the baseline components to achieve
with \(\Delta \rho\) the DD geometric range increments, \(\Delta x\) the 3vector of baseline increments, \({u^s}\,(s=1, \ldots ,m)\) the receivertosatellite unit vectors, \({g^s}\,(s=1, \ldots ,m)\) the satellite elevationdependent mapping function which maps the slant tropospheric delays from all satellites to ZTD (Ifadis 1986), and \(D_{{\text{m}}}^{{\text{T}}}=\left[ {  {e_{m  1}},~~{I_{m  1}}} \right]~\) is the differencing operator forming the betweensatellite differences. Note in (2), the elevation angle from both receivers of the baseline to a given satellite is assumed to be the same. We also remark that the differential ZTD in (2) can be assumed negligible for short baselines, i.e. \(\Delta {\tau _{\text{Z}}} \approx 0\).
The variance matrix of the DD observations in (1) can be formed as
where \({D}\left[ \cdot \right]\) denotes the dispersion operator, and the variance matrices of the undifferenced code and phase observations \({Q_{pp}}\) and \({Q_{\phi \phi }}\) read
where the weighting element \({w^s}\) to the observations of the satellite \(s\) is calculated using the elevationdependent exponential function (Euler and Goad 1991):
in which \({\beta ^s}\) denotes the elevation angle in degrees from the rover to satellite \(s\). The \(f \times f\) matrices \({C_{{pp}}}\) and \({C_{\phi \phi }}\) are, respectively, variance matrices of the code and phase observable types in zenith.
Measurement setup
In this study, the processing is performed using 1 Hz code and phase QZSS data on L1, L2 and L5, collected by two baselines of different lengths in Perth, Australia. The first baseline, of around 9 m length, is formed by the stations CUAA and CUBB which are equipped with JAVAD TRE_G3TH DELTA receivers (Fig. 2). While the second baseline, of around 8 km length, is formed by the stations CUCC and UWA0 which are, respectively, equipped with a JAVAD TRE_G3TH DELTA receiver and a Septentrio PolaRx5 receiver (Fig. 3). Stations CUAA, CUBB and CUCC are connected to TRM59800.00 SCIS antennas, and station UWA0 is connected to a JAVRINGANT_DM SCIS antenna. The cut off elevation angle for all the data analysis in this contribution is set to be 10° and the broadcast ephemeris is used for the processing. Table 1 summarizes the information about the datasets in use.
Stochastic properties
RTK positioning based on (1)–(3) requires information about the structure of \({C_{{\text{pp}}}}\) in (4) and \({C_{\phi \phi }}\) in (5). Our aim here is to find representative values for the entries of these matrices. To do so, we first form the singleepoch, singlebaseline DD code and phase residuals as follows
where the \((m  1)\)vector of DD geometric ranges \(\rho\) is computed with the ground truth of the baseline coordinates and the broadcast satellite ephemeris. The \(f(m  1)\)vector \(a\) contains the reference integer ambiguities obtained from the strong baselineknown model. The residuals \({e_{p}}\) and \({e_\phi }\), in addition to the noise of the DD code and phase observations, also capture any mismodeled effects like, e.g. differential atmospheric delays and multipath. Therefore, one can write
in which \({m_{p}}\) and \({m_\phi }\) denote the mismodeled effects on the DD code and phase observations respectively. Assuming that the geometry vector and ambiguity vector are deterministic, \({n_{p}}\) and \({n_\phi }\), respectively, denote the DD code and phase observations noise—a combined effect of the transmitted signals quality and the receiver architecture.
To estimate the entries of \({C_{{pp}}}\) and \({C_{\phi \phi }}\), we apply the leastsquares variance component estimation (LSVCE) method (AmiriSimkooei et al. 2009; Teunissen and AmiriSimkooei 2008) to the residuals \({e_{p}}\) and \({e_\phi }\) computed for the baselines CUAA–CUBB and CUCC–UWA0. As such, our estimations will capture the combined effects of the transmitted signals quality, the receiver architecture and any mismodeled effects like multipath. The multipath effect on the QZSS observations can be mitigated to a large extent through a procedure explained in the following.
Multipath
The QZSS satellite geometry repeats after one sidereal day. In addition, the surrounding environment of the static stations in use remains almost unchanged over time. Therefore, it is expected that the multipath pattern repeats after one sidereal day. As multipath corrections, one may then use the DD residuals in (8) corresponding with the same satellite geometry obtained after/before one sidereal day (Radovanovic 2000; Choi et al. 2004; Zaminpardaz et al. 2017; Zaminpardaz and Teunissen 2017). Subtracting these multipath corrections from the DD observations will largely mitigate the unwanted multipath effect, i.e. \({m_{p}} \to 0\) and \({m_\phi } \to 0\). Note, however, that subtracting the DD residuals of one single day from our dataset to diminish multipath would also increase the observations noise by a factor of \(\sqrt 2\), i.e. \({n_{p}} \to \sqrt 2 \times {n_{p}}\) and \({n_\phi } \to \sqrt 2 \times {n_\phi }\). To reduce such increase in the noise level of observations, one can make use of less noisy multipath corrections generated through averaging over the DD residuals of multiple days corresponding with the same satellite geometry. If the data of d days are used in this averaging, then the increase in the observations noise would be a factor of \(\sqrt {1+\frac{1}{d}}\). The multipath corrections then get less noisy if the data of more days become involved in its computation.
Multipath is the main source of time correlation for the shortbaseline data. Therefore, to evaluate the efficiency of the multipath correction method explained above, we compute the time correlation of the QZSS signals collected by CUAA–CUBB on DOY 71 of 2018 before and after applying multipath corrections. The multipath corrections are taken from the corresponding observations on DOY 72 of 2018 (one single day). Figure 4 illustrates the time correlation of the QZSS code and phase observations on L1, L2 and L5 as functions of the time differences which are computed using the LSVCE method. It can be seen that the significant time correlation of all the signals are dramatically reduced to negligible values upon applying the multipath corrections.
Variance component estimation
Table 2 provides information about the zenithreferenced variance/covariance of QZSS undifferenced code and phase observations on all the three QZSS frequencies, which are the entries of \({C_{{pp}}}\) and \({C_{\phi \phi }}\). Two types of estimations are given for each parameter, specified as MPuncorrected and MPmitigated which are, respectively, computed based on the multipathuncorrected and multipathmitigated QZSS observations. The results are provided for both the CUAA–CUBB and CUCC–UWA0 baselines, by taking the average of the singleepoch LSVCE estimations over the time intervals considered. Estimations for CUAA–CUBB are based on the data collected on DOY 71 of 2018 which are corrected for multipath using the observations on DOY 72 of 2018 (1 day), whereas estimations for CUCC–UWA0 are based on data collected on DOY 70 of 2018 which are corrected for multipath using the observations on DOY 71 of 2018 (1 day). The differences between the values corresponding with CUAA–CUBB and CUCC–UWA0 stem from several factors amongst which different receiver types and different multipath environments.
For both the baselines in use, our estimations confirm that the code–code and code–phase covariances corresponding with different frequency pairs are negligible, and therefore, we have skipped them in Table 2. We, therefore, consider \({C_{{pp}}}\) to be diagonal. However, as shown in Table 2, the phase–phase covariances are significant (w.r.t. phase precision) and cannot be neglected. With this in mind, in our following analyses, we make use of the fullypopulated \({C_{\phi \phi }}\). We remark that for the CUCC–UWA0 baseline, as the residuals in (8) also contain nonnegligible differential ZTDs, we also modeled the impact of ZTD when conducting LSVCE estimation. Also, note that the multipathmitigated estimations capture the noise level increase due to daydifferencing. In other words, assuming that multipath would completely vanish through the mentioned daydifferencing, then the multipathmitigated results in Table 2 divided by \(\sqrt 2\) would represent the dispersion of the QZSS data in a multipathfree environment. With these in mind, we make two comparisons; once between multipathuncorrected results and multipathmitigated ones, and once between multipathuncorrected results and multipathfree ones.
Comparing the results in the third (fourth) column of Table 2 with those in the fifth (sixth) column, we note that applying 1day multipath corrections, although mitigates multipath to a large extent (see Fig. 4), increases the standard deviation of L1 and L2 code observations. Showing the best code precision, L5code signal improves in precision upon applying multipath corrections. The degradation of L1 and L2code precision can be circumvented if more precise multipath corrections are provided through involving the data of more than 1 day. For example, applying 6day multipath corrections to the data of CUAA–CUBB baseline, our estimations for the L1, L2 and L5code standard deviation amount to 0.315, 0.242 and 0.103 m, respectively. The standard deviation of phase observations does not show significant dependency on frequency and multipath corrections.
Our second comparison is made between multipathuncorrected standard deviations and multipathfree ones which, as previously mentioned, are obtained by dividing the results in the fifth and sixth columns of Table 2 by \(\sqrt 2\). By doing so, one can find out, when the observations are contaminated by multipath, that the code precision estimations degrade by a factor of around 1.2 for L1 and L2 signals, and by a factor of around 2.2 for the L5 signal. Multipath also deteriorates the phase precision estimations by a factor of around 1.2.
RTK performance
In this section, we analyze the capability of the current (February–March 2018) QZSS constellation, for instantaneous RTK positioning application. In doing so, we employ the QZSS observations on L1, L2 and L5 in their original form as well as their multipathmitigated form. The data is collected at the rate of 1 Hz on DOY 69 of 2018 for CUAA–CUBB, and on DOY 72 of 2018 for CUCC–UWA0. The 1day multipath corrections are provided as the corresponding data recorded on DOYs 70 and 73 of 2018, respectively. Figure 5 shows the 24h skyplot view of the four QZSS satellites (J1, J2, J3 and J7) in Perth with the cut off elevation of 10°. In the following, we discuss the RTK results first for the CUAA–CUBB baseline, and then for the CUCC–UWA0 baseline.
CUAA–CUBB baseline
In this subsection, we present the standalone QZSS RTK results based on the data of the shortbaseline CUAA–CUBB for which the differential atmospheric delays can be neglected, i.e. \(\Delta {\tau _{\text{Z}}} \approx 0\) in (2). Therefore, the only unknown parameters to be estimated are \(\Delta x\) and \(a\). The current QZSS constellation provides the users at Perth with four satellites being simultaneously visible (with the cut off angle of 10°) only for 50% of a 24h period (see Fig. 1). However, even if four satellites are visible, they do not always form a geometry which is desirable from a positioning point of view. In this case, the position dilution of precision (PDOP) reaches extremely large values. To circumvent these extreme PDOP values, our evaluations will be carried out over the time intervals with four visible satellites considering different thresholds for PDOP computed as
with \(G=  {[{u^1},{u^2}, \ldots ,{u^m}]^{T}}\), \({\text{tr}}\{ .\} ~\) the trace operator and \({P_{{{\text{D}}_{m}}}}={D_{m}}{\left( {D_{{m}}^{{T}}{W^{  1}}{D_{m}}} \right)^{  1}}D_{{m}}^{{T}}{W^{  1}}\) the orthogonal projector. The singleepoch PDOP time series at Perth during DOY 69 of 2018 with the cut off angle of 10° is depicted in Fig. 6 employing the criterion of PDOP < 100.
Ambiguity resolution
Realization of RTK positioning relies on successful phase ambiguity resolution. In this subsection, we evaluate the triplefrequency QZSS ambiguity resolution performance using two measures, i.e. ambiguity dilution of precision (ADOP) and Integer Bootstrapped (IB) success rate—probability of correct integer estimation. As introduced by Teunissen (1997), ADOP is an easytocompute scalar diagnostic, which measures the model strength for successful ambiguity resolution. With the observational model in (1), ADOP is computed as
where \({Q_{\widehat {a}\widehat {a}}}\) denotes the variance matrix of the float ambiguities and \(\left \cdot \right\) stands for the determinant. Figure 7 shows the singleepoch ADOP time series of the QZSS triplefrequency multipathmitigated data collected by the baseline CUAA–CUBB on DOY 69 of 2018. The multipath corrections are provided as the corresponding data on DOY 70 of 2018. The time points satisfy the criterion of PDOP < 100. According to Odijk and Teunissen (2008), low ADOP values indicate high ambiguity success rates. As a rule of thumb, the integer leastsquares (ILS) ambiguity success rate becomes larger than 99.9% when the ADOP value reduces to below 0.12 cycles. Therefore, since the ADOP value in Fig. 7 does not exceed 0.08, a high ambiguity resolution success rate is expected. Here, we remark that the ADOP time series has a different signature from that of the PDOP time series in Fig. 6. For example, just after sample 20000, while PDOP increases dramatically, ADOP shows a decreasing behavior. This demonstrates importantly that the PDOP and ADOP characteristics can be quite distinct and that one, therefore, should not confuse a poor PDOP with poor ambiguity resolution capabilities and correspondingly poor ambiguityfixed positioning performance. Hence, a positioningwise good geometry (i.e. small PDOP) is not necessarily needed for successful ambiguity resolution (Zaminpardarz et al. 2017).
The formal IB success rate, for the ambiguity \(f(m  1)\)vector in (1), is given as (Teunissen 1998)
with
where \({\sigma _{{{\widehat {z}}_{iI}}}}\) is the conditional standard deviation of the \(i{\text{th}}\) decorrelated ambiguity with \(~i=1, \ldots ,f\left( {m  1} \right)\) and \(I=1, \ldots ,i  1\). Note that IB success rate is the sharpest lower bound to the ILS success rate, which itself has the highest success rate of all admissible integer estimators (Teunissen 1999; Verhagen and Teunissen 2014).
Table 3, for the short baseline CUAA–CUBB based on the QZSS triple frequency data on DOY 69 of 2018, gives the empirical and formal IB success rates. The results are provided based on both the MPuncorrected and the MPmitigated data (based on 1day multipath corrections). The formal values are computed by taking an average of the formal IB success rate (12) of all the selected processing epochs, whereas the empirical IB success rate is computed as.
To judge whether a DD ambiguity vector is correctly fixed, its corresponding IB solution is compared with the reference integer DD ambiguity vector computed based on the ILS solution of the baselineknown model. The results in Table 3 are classified based on a PDOP threshold. The reason for such classification is that for the positioning results, coming in the next subsection, we consider these thresholds for PDOP value. One should, however, bear in mind that the ambiguity resolution performance is not characterized through PDOP (Zaminpardaz et al. 2017). In Table 3, both the multipathuncorrected and the multipathmitigated data show a high level of success rate around 96 and 100%, respectively, revealing that instantaneous RTK is feasible using standalone QZSS if the data on all the three frequencies are employed. Also, the formal and empirical ambiguity success rates are well in agreement with each other, confirming the consistency between model and data.
Positioning performance
Having analyzed the ambiguity resolution performance of the standalone QZSS triplefrequency shortbaseline model, here we investigate the corresponding ambiguityfloat and fixed positioning accuracies. Considering PDOP < 100, the deviations of the coordinate estimates from the ground truth are plotted in Fig. 8. Three types of solutions are shown in each panel; float solutions as gray dots, correctlyfixed solutions as green dots and wronglyfixed solutions as red dots. It can be observed that the float solutions (gray), except for some intervals, match well with their corresponding 95% formal confidence intervals (blue curves). The inconsistencies between the formal and empirical float solutions can be attributed to the fact that the multipath corrections do not remove the whole multipath effect, but they capture the lowfrequency multipath components to a large extent and partly the highfrequency multipath components. To elaborate more on this, we focus on the inconsistency showing up during the interval [9000, 13,000]. Our analysis shows that the presence of this inconsistency can be traced back to the highfrequency multipath residuals in the DD observations corresponding with the satellite pair J4–J3. Figure 9, illustrates the DD code residuals on L5 (cf. 8) corresponding with J4–J3 over the interval [9000, 15,000]. The top panel shows these residuals on DOYs 69 (blue) and 70 (purple) of 2018, while the bottom panel shows the difference of these residuals. It can be seen that, over the period [9000, 13,000], once the data are corrected for multipath, while the trend of the multipath is largely eliminated, the highfrequency multipath residuals are not mitigated, thus affecting the corresponding positioning results.
Table 4 lists the singleepoch empirical and formal standard deviations of CUAA–CUBB baseline components before and after multipath mitigation for both the ambiguityfloat and fixed cases, considering different thresholds for PDOP. The ambiguityfixed results are computed based on the correctlyfixed solutions. The formal outcomes are obtained from taking the average of all the singleepoch leastsquares baseline variance matrices, whereas the empirical outcomes are obtained from the differences between the estimated baseline and the available ground truth of the CUAA–CUBB baseline. It can be observed that the meterlevel coordinate errors in the ambiguityfloat case are reduced to cmlevel after fixing ambiguities in all three directions. In both the ambiguityfloat and fixed cases, the empirical and formal results match well with each other, validating the stochastic model used for the processing.
From Table 4, it can be seen that multipath mitigation has indeed improved the precision of float solutions, but marginally degraded that of fixed solutions. However, one should keep in mind that these fixed solutions would only be available in case the ambiguities can successfully be fixed, and as Table 3 shows, the probability of successful ambiguity resolution gets higher when the data is corrected for multipath. To explain the impact of multipath corrections on baseline solutions precision, we first give the singleepoch baseline variance matrix for both the ambiguityfloat case \({Q_{\widehat {x}\widehat {x}}}\) as well as the ambiguityfixed case \(Q_{\check{x}\check{x}}\)
with \(\overline {u} =\left( {{{\mathop \sum \nolimits_{{s=1}}^{m} {w^s}{u^s}} \mathord{\left/ {\vphantom {{\mathop \sum \nolimits_{{s=1}}^{m} {w^s}{u^s}} {\mathop \sum \limits_{{s=1}}^{m} {w^s}}}} \right. \kern0pt} {\mathop \sum \limits_{{s=1}}^{m} {w^s}}}} \right)\) being the weightedaverage of the lines of sight. The difference between the baseline variance matrices corresponding with multipathuncorrected and mitigated data lies in the difference between the entries of the corresponding matrices \({C_{{pp}}}\) and \({C_{\phi \phi }}\) (see Table 2). Also, note that the multipathmitigated results in Table 4 are obtained based on 1day multipath corrections. As a result, the ratio of the term \({1 \mathord{\left/ {\vphantom {1 {e_{{f}}^{{T}}C_{{{pp}}}^{{  1}}~{e_{f}}}}} \right. \kern0pt} {e_{{f}}^{{T}}C_{{{pp}}}^{{  1}}~{e_{f}}}}\) for multipathuncorrected case to that of the multipathmitigated case amounts to around 1.68, and \(\sqrt {1.68} \approx 1.3\) is then the factor by which the ambiguityfloat positioning precision improves upon applying multipath corrections.
Now, we consider the ambiguityfixed case where the ratio of the term \({1 \mathord{\left/ {\vphantom {1 {e_{{f}}^{{T}}\left( {C_{{{pp}}}^{{  1}}+C_{{\phi \phi }}^{{  1}}} \right)~{e_{f}}}}} \right. \kern0pt} {e_{{f}}^{{T}}\left( {C_{{{pp}}}^{{  1}}+C_{{\phi \phi }}^{{  1}}} \right)~{e_{f}}}}\) for the multipathuncorrected case to that of the multipathmitigated case amounts to around 0.63. Therefore, it is expected that ambiguityfixed positioning precision degrades upon applying multipath corrections by almost a factor of \(1/\sqrt {0.63} \approx 1.26\). This degradation, however, can be circumvented if one, instead of 1day multipath corrections, makes use of less noisy multipath corrections. Based on 6day multipath corrections, our evaluations confirm an instantaneous ambiguity resolution success rate more than 99.9%, and the consequent ambiguityfixed results have the same precision as those based on multipathuncorrected observations.
Among the three baseline components, the poorest precision is obtained for the East component while the best precision is achieved for the North component which can be explained by the behavior of the relative QZSS satellite geometry in Perth as follows. Here, as an example, we take the ambiguityfloat case with the original data for which the singleepoch baseline variance matrix is given in (15). Figure 10 shows, over the observational period with PDOP < 100, the absolute values of the East, North and Height components of \(\sqrt {{w^s}} \left( {{u^s}  \overline {u} } \right)\). The larger such component is, the better this component can be estimated. Therefore, according to Fig. 10, one can indeed expect the precision of the baseline components can be ranked in an ascending order as East (red), Height (blue) and North (green) component.
CUCC–UWA0 baseline
So far, we have investigated the QZSS performance in shortbaseline RTK positioning. In this section, we extend our evaluations to a longer baseline, i.e. CUCC–UWA0 of around 8 km length (Fig. 3). For this longer baseline, using the QZSS data on three frequencies, we checked the significance of the differential ionospheric and tropospheric delays. Our evaluations showed that while the impact of atmospheric residuals on the noisy code observations can still be neglected, the differential tropospheric delays impact on the precise phase observations can reach significant levels compared to the phase precision. Therefore, when parametrizing \(\widetilde {\rho }\) in (1) for the observations of the CUCC–UWA0 baseline, in addition to \(\Delta x\), one needs to take into account \(\Delta {\tau _{\text{Z}}}\) as well.
Ambiguity resolution
Note, since we are working with the observations of four QZSS satellites, that parametrization of vector \(\widetilde {\rho }\) according to (2) would not bear additional redundancies for the singleepoch observational model in (1), implying that ADOP and ambiguity resolution performance would remain invariant w.r.t. such parametrization. Therefore, here, we assess the ambiguity resolution performance without further parametrizing \(\widetilde {\rho }\). Figure 11 shows the singleepoch ADOP time series of the QZSS triplefrequency multipathmitigated data collected by the baseline CUCC–UWA0 on DOY 72 of 2018. The time points satisfy the criterion of four satellites being visible simultaneously. Since the ADOP value in Fig. 11 does not exceed 0.08, a high ambiguity resolution success rate is expected. Our results confirm an instantaneous ambiguity resolution success rate of 99.65%.
Positioning performance
As was mentioned earlier, \(\widetilde {\rho }\) for the observations of the CUCC–UWA0 baseline contains significant residual tropospheric delays. However, with ZTD included as an unknown parameter, instantaneous RTK will require the observations of at least five satellites simultaneously. This in tandem with the fact that the data of only four QZSS satellites are available make singleepoch positioning infeasible for CUCC–UWA0. To tackle this issue, instead of the measurements of a single epoch, we incorporate the data of two epochs assuming that the baseline coordinates remain invariant. To realize a sufficient change of receiver–satellite geometry, these two epochs are chosen as follows.

Check the number of visible satellites on a given day and store the indicator of the epochs with four satellites in \(\Omega\);

If the size of \(\Omega\), denoted by \(k\), is odd, make it even by removing one element;

Define \({\Omega _1}\) containing the first \(\frac{k}{2}\) elements of \(\Omega\), and \({\Omega _2}\) containing the last \(\frac{k}{2}\) elements of \(\Omega\);

For twoepoch positioning, take one set of DD observations at an epoch from \({\Omega _1}\) and one set at an epoch from \({\Omega _2}\).
To assess the performance of twoepoch positioning explained above, use is made of the QZSS triplefrequency multipathmitigated data collected by the baseline CUCC–UWA0 on DOY 72 of 2018. The multipath corrections are given by the corresponding data on DOY 73 of 2018. Through the above procedure, the maximum and minimum time differences between the epochs in the epoch pairs considered are, respectively, 09:28:31 and 14:27:19. As was discussed in the previous subsection, the corresponding ambiguities can successfully be resolved to their integer values already based on singleepoch multipathcorrected observations. Hence, based on a stronger model formed by the observations of two epochs with the assumption of baseline components being timeconstant, ambiguities must successfully be fixed. In the following, we, therefore, present the results corresponding with the multipathcorrected data.
The deviations of the ambiguityfloat (gray) and fixed (green) coordinate estimates from the ground truth are plotted in Fig. 12. It can be observed that the float solutions (gray) match well with their corresponding 95% formal confidence intervals (blue curves). In this figure, we also note some sudden changes in the positioning accuracy (see blue curves). This happens as a result of sudden changes in the time difference between the two epochs in use, thereby having significant changes in satellite geometry. For example, the standard deviation drop from data number 13,570 to 13,571 is due to that fact that the time difference between the two epochs in use increases (from 9 h 34 min 25 s to 13 h 1 min 46).
Table 5 provides the precision of baseline float and fixed solutions. To compare the results of this table with those in Table 4 (for e.g. PDOP < 100), we divide the standard deviations in Table 4 by \(\sqrt 2\) to achieve the approximate twoepoch positioning precision for the shortbaseline case. One can then realize that the inclusion of ZTD as an unknown parameter will deteriorate the precision of the baseline components. The larger deterioration in the vertical direction can be explained by the strong correlation between the ZTDs and the receiver height component (Rothacher and Beutler 1998).
Summary and conclusions
In this contribution, for the foursatellite QZSS constellation as a standalone system, we provided an initial assessment of L1 + L2 + L5 RTK performance in the form of integer ambiguity resolution success rate and positioning accuracy. We analyzed the noise characteristics of the QZSS code and phase observations collected by two baselines in Perth, Australia, i.e. CUAA–CUBB with identical receiver types and a length of 9 m and CUCC–UWA0 with mixed receiver types and a length of 8 km. Our results were presented in the form of zenithreferenced variance–covariance matrices and time correlations.
It was shown that through applying multipath corrections, the significant time correlation of all the QZSS code and phase observations become negligible. Zenithreferenced standard deviations of the code observations were the largest for the L1 signal and the lowest for the L5 signal. The phase observations on different frequencies, however, had almost the same precision. Applying multipath corrections reduced the L1, L2 and L5 code standard deviation by, respectively, around 15, 20 and 50%. Our estimations of covariances between different observation types implied that the code observations on different frequencies can be considered uncorrelated with each other and also with the phase observations, both before and after multipath corrections. However, significant covariances were observed between the phase observations on different frequencies. The standard deviation of the phase observations on each frequency had very low sensitivity w.r.t multipath correction.
Having investigated the characteristics of the QZSS signals, we then presented the standalone QZSS triplefrequency RTK performance. In doing so, we first considered the shortbaseline CUAA–CUBB for which the differential atmospheric delays were negligible. In Perth, Australia, the current foursatellite constellation of QZSS provides PDOP values lower than 100 for more than 10 h per day during which, ADOP values are lower than 0.08 for triplefrequency case indicating a high ambiguity success rate. It was shown that the instantaneous integer bootstrapped (IB) success rate is around 96% and 100% for the original and multipathmitigated data, respectively, implying that the instantaneous RTK is feasible based on standalone QZSS observations.
In both the ambiguityfloat and fixed cases, the empirical and formal results matched well with each other, confirming the consistency of the model and data. Meterlevel coordinate errors in the ambiguityfloat case were reduced to cmlevel after fixing ambiguities in all three directions. The multipath mitigation improved the float solutions precision, and only marginally degrades the fixed solution accuracy which was due to using noisy multipath corrections. It was shown that such degradation can be prevented using less noisy multipath corrections computed based on the data of multiple days. The baseline components in terms of precision can be ranked in ascending order as East, Height and North. This was explained as a result of specific QZSS satellites geometry at Perth. Our illustrations also demonstrated that PDOP and ADOP characteristics can be completely different and that one, therefore, should not confuse a poor PDOP with poor ambiguity resolution capabilities. To predict ambiguityfixed positioning performance, one should therefore, first consult the ADOP before considering the PDOP.
Finally, the standalone QZSS positioning performance was assessed for the longer baseline CUCC–UWA0 for which the impact of differential tropospheric delays on the precise phase observations can reach significant levels compared to phase precision. As such, the observational model for this baseline included, in addition to baseline components and DD ambiguities, the zenith tropospheric delay (ZTD) as well. In order for all the unknown parameters to be estimable, we made use of the data of two epochs assuming that the baseline coordinates are timeinvariant. Based on multipathmitigated data, it was shown that the ambiguity resolution success rate is equal to 100%. Comparing the ambiguityresolved positioning performance of CUCC–UWA0 with that of shortbaseline CUAA–CUBB, the precision of the baseline components gets poorer, particularly in the vertical direction, when including ZTD as an unknown parameter.
References
AmiriSimkooei AR, Teunissen PJG, Tiberius CCJM (2009) Application of leastsquares variance component estimation to GPS observables. J Surv Eng 135(4):149–160. https://doi.org/10.1061/(ASCE)07339453(2009)135:4(149)
Choi K, Bilich A, Larson KM, Axelrad P (2004) Modified sidereal filtering: implications for highrate GPS positioning. Geophys Res Lett 31(22):L22608
Choi BK, Cho CH, Cho JH (2015) Evaluation of point positioning using the global positioning system and the QuasiZenith satellite system as measured from South Korea. J Astron Space Sci 32(4):403–409. https://doi.org/10.5140/JASS.2015.32.4.403
Euler HJ, Goad CC (1991) On optimal filtering of GPS dual frequency observations without using orbit information. Bull Geod 65(2):130–143. https://doi.org/10.1007/BF00806368
Goad CC (1998) Short distance GPS models (Chap. 11). In: Teunissen PJG, Kleusberg A (eds) GPS for geodesy, 2nd edn. Springer, Berlin, pp 457–481
Google Earch (2018) Google Earth imagery (November 5, 2017). Google Earch 7.0.3.8542. Perth, Australia. 31\(^\circ\)59'26.61''S, 115\(^\circ\)51'22.64''E, Eye alt 10.91 km. @ 2018 Google. https://www.google.com/earth/. Accessed 16 July 2018
Hauschild A, Steigenberger P, RodriguezSolano C (2012) Signal, orbit and attitude analysis of Japan’s first QZSS satellite Michibiki. GPS Solut 16(1):127–133. https://doi.org/10.1007/s1029101102455
Henderson HV, Pukelsheim F, Searle SR (1983) On the history of the Kronecker product. Linear Multilinear Algebra 14(2):113–120. https://doi.org/10.1080/03081088308817548
Ifadis II (1986) The atmospheric delay of radio waves: modeling the elevation dependence on a global scale. Technical Report 38L, Chalmers University of Technology, Göteborg, Sweden
Leick A, Rapoport L, Tatarnikov D (2015) GPS satellite surveying. 4th edn. Wiley, Hoboken, ISBN: 9781118675571
Murai Y (2015) Project Overview of The QuasiZenith Satellite System. In: Proceedings of the 28th International Technical Meeting of The Satellite Division of the Institute of Navigation (ION GNSS+ 2015), Tampa, Florida, September 2015, pp. 1291–1332
Nadarajah N, Khodabandeh A, Teunissen PJG (2016) Assessing the IRNSS L5signal in combination with GPS, Galileo, and QZSS L5/E5asignals for positioning and navigation. GPS Solut 20(2):289–297. https://doi.org/10.1007/s1029101504508
Nie X, Zheng F, Lou Y, Zheng J, Li Z (2015) An analysis of the QZSS signal based on the data of IGS. Sun J et al (eds.) China satellite navigation conference (CSNC) 2015. Proceedings: vol II, lecture notes in electrical engineering 341. https://doi.org/10.1007/9783662466353_27
Odijk D, Teunissen PJG (2008) ADOP in closed form for a hierarchy of multifrequency singlebaseline GNSS models. J Geodesy 82:473–492. https://doi.org/10.1007/s0019000701972
Odolinski R, Teunissen PJG (2017) Lowcost, 4system, precise GNSS positioning: a GPS, Galileo, BDS and QZSS ionosphereweighted RTK analysis. Meas Sci Technol 28(12):125801. https://doi.org/10.1088/13616501/aa92eb
Odolinski R, Teunissen PJG, Odijk D (2015) Combined BDS, Galileo, QZSS and GPS singlefrequency RTK. GPS Solut 19(1):151–163. https://doi.org/10.1007/s1029101403766
Quan Y, Lau L, Roberts GW, Meng X (2016) Measurement signal quality assessment on all available and new signals of multiGNSS (GPS, GLONASS, Galileo, BDS, and QZSS) with real data. J Navig 69(2):313–334. https://doi.org/10.1017/S0373463315000624
QZSS (2018) Overview of the QuasiZenith satellite system (QZSS). Cabinet Office, National Space Policy Secretariat. http://qzss.go.jp/en/overview/services/sv01_what.html. Accessed April 2018
Radovanovic RS (2000) High accuracy deformation monitoring via multipath mitigation by daytoday correlation analysis. Proceedings of ION GPS 2000, Salt Lake City, UT, September 19–22, pp 35–44
Rothacher M, Beutler G (1998) The role of GPS in the study of global change. Phys Chem Earth 23(9–10):1029–1040. https://doi.org/10.1016/S00791946(98)001438
Saastamoinen J (1972) Atmospheric correction for troposphere and stratosphere in radio ranging of satellites. In: Henriksen SW, Mancini A, Chovitz BH (eds) The use of artificial satellites for geodesy, vol 15. American Geophysical Union, Washington, DC, pp 247–251 (Geophysics Monograph Series)
Steigenberger P, Thoelert S, Hauschild A, Montenbruck O, Langley RB (2018) QZS3 and QZS4 join the QuasiZenith satellite system. GPS world, innovation column. http://gpsworld.com/innovationqzs3andqzs4jointhequasizenithsatellitesystem/. Accessed 20 Mar 2018
Teunissen PJG (1997) A canonical theory for short GPS baselines. Part IV: precision versus reliability. J Geodesy 71(9):513–525
Teunissen PJG (1998) Success probability of integer GPS ambiguity rounding and bootstrapping. J Geodesy 72(10):606–612. https://doi.org/10.1007/s001900050199
Teunissen PJG (1999) An optimality property of the integer leastsquares estimator. J Geodesy 73(11):587–593. https://doi.org/10.1007/s001900050269
Teunissen PJG, AmiriSimkooei AR (2008) Leastsquares variance component estimation. J Geodesy 82(2):65–82. https://doi.org/10.1007/s001900070157x
Teunissen PJG, Montenbruck O (eds) (2017) Springer handbook of global navigation satellite systems. Springer International Publishing, Switzerland. https://doi.org/10.1007/9783319429281
Verhagen S, Teunissen PJG (2014) Ambiguity resolution performance with GPS and BeiDou for LEO formation flying. Adv Space Res 54(5):830–839. https://doi.org/10.1016/j.asr.2013.03.007
Wu F, Kubo N, Yasuda A (2004a) A study on GPS augmentation using Japanese QuasiZenith Satellite System. PLANS 2004. Monterey, CA, USA, April, 2004. https://doi.org/10.1109/PLANS.2004.1309016
Wu F, Nobuaki K, Yasuda A (2004b) Performance analysis of GPS augmentation using Japanese QuasiZenith Satellite System. Earth Planets Space 56(1):25–37. https://doi.org/10.1186/BF03352488
Zaminpardaz S, Teunissen PJG (2017) Analysis of Galileo IOV + FOC signals and E5 RTK performance. GPS Solut 21(4):1855–1870. https://doi.org/10.1007/s1029101706599
Zaminpardaz S, Teunissen PJG, Nadarajah N (2017) GLONASS CDMA L3 ambiguity resolution and positioning. GPS Solut 21(2):535–549. https://doi.org/10.1007/s102910160544y
Author information
Affiliations
Corresponding author
Rights and permissions
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
About this article
Cite this article
Zaminpardaz, S., Wang, K. & Teunissen, P.J.G. Australiafirst highprecision positioning results with new Japanese QZSS regional satellite system. GPS Solut 22, 101 (2018). https://doi.org/10.1007/s1029101807635
Received:
Accepted:
Published:
Keywords
 QZSS
 Stochastic properties
 Multipath
 Time correlation
 Ambiguity resolution
 RTK