Carrier phase bias estimation of geometryfree linear combination of GNSS signals for ionospheric TEC modeling
 854 Downloads
Abstract
The ionosphere can be modeled and studied using multifrequency GNSS signals and their geometryfree linear combination. Therefore, a number of GNSSderived ionospheric models have been developed and applied in a broad range of applications. However, due to the complexity of estimating the carrier phase ambiguities, most of these models are based on lowaccuracy carrier phase smoothed pseudorange data. This, in turn, critically limits their accuracy and applicability. Therefore, we present a new methodology of estimating the phase bias of the scaled L1 and L2 carrier phase difference which is a function of the ambiguities, the ionospheric delay, and hardware delays. This methodology is suitable for ionospheric modeling at regional and continental scales. In addition, we present its evaluation under varying ionospheric conditions. The test results show that the carrier phase bias of geometryfree linear combination can be estimated with a very high accuracy, which consequently allows for calculating ionospheric TEC with the uncertainty lower than 1.0 TECU. This high accuracy makes the resulting ionosphere model suitable for improving GNSS positioning for highprecision applications in geosciences.
Keywords
GNSS Carrier phase bias Ionosphere Ionospheric delay TECIntroduction
The ionospheric delay is one of the most dominant error sources in global navigation satellite systems (GNSS) positioning. Thus, reliable modeling of the ionospheric is one of the most challenging aspects of precise GNSS positioning (Leick et al. 2015; Wielgosz 2011) and GNSSbased geodetic and geodynamic studies (Bosy 2005). Currently, most of the available global, regional, and local ionosphere models are based on carrier phase smoothed pseudorange data, which presents low accuracy and requires strong smoothing to estimate the total electron content (TEC) (Brunini et al. 2004; KrypiakGregorczyk et al. 2013; Alizadeh et al. 2015). The ionospheric delay obtained from smoothed pseudoranges has an accuracy of several TEC units. One TECU = 10^{16} el/m^{2}, and it is equivalent to 0.162 m of L1 signal delay. This is one of the reasons why most of the research centers use spherical harmonics expansion (SHE) or other heavily smoothing functions for TEC parameterization in their global and regional solutions (Schaer 1999; HernándezPajares et al. 2009; Schmidt et al. 2011). As a result, the obtained ionospheric maps are characterized by a low spatial resolution of several degrees and temporal resolution of 5–120 min. The widely used International GNSS Service (IGS) global model has a spatial resolution of 5.0° × 2.5° and temporal resolution of 2 h. The most recent studies show that the absolute TEC accuracy is rather low and amounts to 4–5 TECU (HernandezPajares et al. 2017; RoviraGarcia et al. 2016). Due to the very dynamic changes of the ionosphere, taking place not only during adverse solar activity events and magnetic storms, the accuracy and resolution offered by the current global and regional models are often not satisfactory.
The continuity of GNSS technology and its increasing precision make it one of the most efficient ways for TEC monitoring and modeling. On the other hand, the accurate knowledge of the state of the ionosphere plays a key role in determining a position using GNSS observations. This is due to the important influence of the ionospheric delays on determining carrier phase ambiguities. Therefore, the development of highaccuracy models with higher spatial and temporal resolution is required to support carrier phase ambiguity resolution (Wielgosz et al. 2005; Charoenkalunyuta and Satirapod 2014; RoviraGarcia et al. 2016).
Achieving better accuracy than existing models may be possible through the use of the undifferenced carrier phase observations and their geometryfree linear combination to derive TEC estimates. Initial results are already provided in KrypiakGregorczyk et al. (2017), where we demonstrated that such a model was suitable to support ambiguity resolution in static relative positioning. Here, we focus on a new methodology that estimates the carrier phase bias of the geometryfree linear combination of GNSSscaled carrier phase observations, i.e., the difference L1–L2, which is a function of the respective scaled ambiguities, the ionospheric delay, and hardware delays. This methodology is suitable for ionosphere modeling at regional and continental scales. Another goal is to validate the accuracy of the estimated phase bias of this function over a daytoday boundary discontinuities.
Methodology

Step 1 Estimation of the carrier phase bias.

Step 2 TEC calculation at the ionospheric piercing points (IPPs).

Step 3 TEC interpolation to form a regular grid (TEC map).
Therefore, the accuracy of the resulting TEC maps depends primarily on the accuracy of the estimated carrier phase bias for each continuous data arc. In other words, the accurate carrier phase bias is the prerequisite for any accurate TEC model. In this section, we provide basics on the ionosphere modeling together with our approach for solving the unknown carrier phase bias carried out in step (1).
GNSSTEC estimation
Estimation of the carrier phase bias
The design matrix [A] consists of two groups of parameters: (1) epochdependent parameters [A_{ I }] representing the ionosphere and (2) constants [A_{ B }] representing the carrier phase biases for each continuous L_{GF} arc, while L is the misclosures vector. We assume that all observables have the same precision and are uncorrelated; therefore, the weight matrix is the identity matrix. The system consists of a very large number of observations, and our internal tests showed that applying, e.g., elevationdependent weighting scheme did not improve the results. Also note that due to spatial correlation of the ionosphere, the function parametrizing the ionosphere does need to have a high spatial resolution. Its errors average out over time when using a 24h data set, and the resulting \( B_{\text{iGF}}^{\text{k}} \) is estimated with a high accuracy as seen in subsequent sections. Note that the selected function is only used to estimate the carrier phase bias, and the function parameters are treated as nuisance ones. As mentioned above, the final TEC maps are obtained in step (3) and result from the interpolation of vertical TEC calculated in step (2), and the latter, in turn, is computed based on the estimated bias of step 1. Note that any interpolation function may be used in step (3).
For the ionosphere modeling, a single layer model (SLM) with its associated mapping function is used (Schaer 1999; Shagimuratov et al. 2002). Due to the errors associated with a simple SLM mapping function, it is recommended to increase the elevation cutoff angle for GNSS observations to 20°–30°.
Regional ionosphere modeling at UWM in Olsztyn
The proposed approach for estimating the carrier phase bias \( B_{\text{iGF}}^{\text{k}} \) has already been used in the processing of a regional European ionospheric model developed at the University of Warmia and Mazury in Olsztyn (UWM). Its initial performance for Central Europe has been demonstrated in KrypiakGregorczyk et al. (2014). Our approach is based solely on precise undifferenced dualfrequency carrier phase data and the function \( L_{\text{GF}} \).
In the present study, observations from more than 200 GPS + GLONASS ground network stations were used. In particular, data from the EUREF Permanent Network (EPN) and European Position Determination System (EUPOS) stations were used to estimate the carrier phase bias (Bosy et al. 2007; Bruyninx et al. 2011). In the presented example, carrier phase biases for each continuous observation arcs were estimated together with parameters of the SHE functions. For the data preprocessing, which includes data cleaning and cycle slip detection, a 30s sampling interval was used. The cycle slips detection problem is treated very carefully here, as any undetected slips affect the estimated carrier phase bias. Our detection approach is based on screening single and double \( L_{\text{GF}} \) time differences, similar to the method proposed by Cai et al. (2013). Any jump in double time differences over ± 0.1 m is treated as a cycle slip occurrence. In such a case, a new bias parameter is set up. Note that we do not try to repair any detected cycle slips. Also note, since the data comes from highquality receivers mounted at the reference station networks, and the elevation cutoff angle is 20°–30°, cycle slip occurrences are very rare in the processed data. Any undetected slips resulting in \( L_{\text{GF}} \) jumps under ± 0.1 m may affect the estimated bias. Our numerical tests based on actual GNSS data show that for some slip combination, e.g., − 6 cycles on L1 and − 5 cycles on L2, the effective jump on L_{GF} is about 8 cm. This may pass undetected with the detection threshold of ± 0.1 m. The data adjustment results show this L_{GF} slip of 8 cm causes error in the estimated bias up to 0.035 m and 0.3 TECU in the vertical TEC. On the other hand, our data analysis shows that double time differences of the \( L_{\text{GF}} \) are below ± 0.01 m, and in case of slips the differences are usually over ± 1 m. For instance, Zhao et al. (2015) reported a standard deviation of double time difference of the \( L_{\text{GF}} \) at the level of 4 mm. Also, our method may be further improved when applying triplefrequency data to detect and repair cycle slips (Zhao et al. 2015). This is because repairing cycle slips allows for using longer data arcs for \( B_{{i{\text{GF}}}}^{k} \) estimation.
In this study, the data processing in step (1) was carried out with 1200s sampling interval. Hence, 72 epochs were processed over a 24h period. The parameterization of the presented procedure makes the separation of the slant ionospheric delays and carrier phase biases in (5) possible. The biases are estimated in the step (1) of the regional ionospheric modeling. Then in the step (2), known carrier phase biases are substituted into (5) resulting in precise slant ionospheric delays at the IPPs. The slant ionospheric delays, in turn, are required for step (3) for subsequent calculation of the ionospheric vertical TEC. However, in this contribution we focus solely on the estimation and analysis of the carrier phase biases estimated in step (1).
Accuracy analysis of carrier phase bias estimation
In order to investigate the accuracy of the carrier phase bias estimation, we validated our method by analyzing of the bias differences at the daytoday boundaries for adjacent 24h arcs. Note that the L_{GF} carrier phase bias is estimated using 24h datasets. A similar approach is used, e.g., to validate the accuracy of precise orbits of GNSS satellites (Griffiths and Ray 2009; Tegedor et al. 2014). Moreover, we also present the analysis of TEC differences calculated at adjacent epochs at 23:59:30 on the first day and at 0:00:00 on the second day of a particular satellite arc. Note that TEC at both epochs comes from different and independent bias estimates that are computed from different daily RINEX files. This is also the case for the bias comparisons at daytoday boundaries.
Statistics of the carrier phase bias differences over day boundary discontinuities (30degree elevation mask) (m)
DOY  RMS  Min  Max  Mean  <± 10 cm (%)  <± 20 cm (%) 

67/68  0.09  − 0.20  0.20  0.03  75.00  100.00 
68/69  0.07  − 0.11  0.15  0.03  89.47  100.00 
69/70  0.06  − 0.12  0.13  0.03  91.23  100.00 
70/71  0.05  − 0.10  0.13  0.02  95.92  100.00 
71/72  0.07  − 0.14  0.15  0.02  79.25  100.00 
72/73  0.07  − 0.12  0.14  0.02  78.57  100.00 
Mean  0.07  84.91  100.00 
Statistics of the carrier phase bias differences over day boundary discontinuities (20degree elevation mask) (m)
DOY  RMS  Min  Max  Mean  <± 10 cm (%)  <± 20 cm (%) 

67/68  0.10  − 0.20  0.27  0.04  70.77  92.31 
68/69  0.09  − 0.05  0.21  0.06  66.67  98.48 
69/70  0.06  − 0.12  0.13  0.02  92.86  100.00 
70/71  0.06  − 0.12  0.14  0.02  90.00  100.00 
71/72  0.07  − 0.15  0.18  0.03  78.57  100.00 
72/73  0.10  − 0.28  0.19  0.01  71.25  93.75 
Mean  0.08  78.35  97.42 
In general, the bias differences for all four stations reach the lowest values during active days and then increase during the first and the last days of the test period. In the presented examples, the residuals for station WROC, located in the center of the model area, are clearly lower than those for LAMA and WSRT, and most of them are within 5–10 cm range for the first variant, and about 10 cm for the second variant, except for PRN25, showing residuals of about 20 cm for 20° cutoff. Station AUTN located in the west achieved slightly worse results than station WROC, but clearly better than LAMA and WSRT. As one can see in the figures, in all cases the biases estimated using observations with the 30degree elevation cutoff present slightly lower differences compared to the biases estimated using a 20° cutoff. In both variants, for stations LAMA, WROC, and WSRT the bias differences over day boundaries present the highest values for PRN25. The fourth station, station AUTN, achieved almost the lowest values for PRN25 while the results for PRN29 reached 20 cm for both variants. It is observed from the figures that the differences for LAMA and WSRT are characterized by the highest values reaching 12 cm for 30degree elevation mask and 15 cm for the 20° mask. Since stations LAMA and WSRT are located in the northern part of our model, the larger differences of the bias over day boundaries for LAMA and WSRT are caused by fewer observations from this region used for the bias estimation.
Statistics concerning mean differences between the bias values over day boundaries, their minimum, maximum and the overall RMS for the two analyzed elevation cutoff angles and all 7 days of the test period and all GPS and GLONASS satellites are presented in Tables 1 and 2. In the tables, the percentiles of the bias differences falling within arbitrarily selected limits of ± 10 cm and ± 20 cm are presented. These limits correspond to 1.0 and 2.0 TECU. It should be noted that we did not notice effects depending on satellite block or constellation.
The results in Table 1 show that during all analyzed days the RMS does not exceed 10 cm. In the first variant (30°), the RMS values amounts to 9 cm at the beginning of the test period, but at the end fall to 7 cm (Table 1). For active days 68/69 and 69/70 the RMS reached only 5 cm, and it was similarly low over the next boundary (70/71). The mean absolute differences did not exceed 5 cm and for active day boundary 68/69; these values reached only 3 cm. In addition, the analyses show that the application of the presented approach results in the carrier phase bias differences in the ± 20 cm range in 100% cases for the first variant. It should be noted that for the whole analyzed period more than 84% of differences are within ± 10 cm range.
In the second variant, as it might be expected, the accuracy of the carrier phase bias estimation decreased (Table 2). However, the RMS values increased by 1 cm only compared to the first variant. A significant difference in the results from the two variants is evident in case of differences falling within ± 10 cm and ± 20 cm ranges. This metric confirms the lower accuracy of the bias estimation in case of the second variant. It should be noted that increasing the elevation cutoff (variant 1) improves the resulting bias accuracy. However, this reduces the number of usable observations from 1898 IPPs to 1717 IPPs during a single day, which is about 10%. This in turn may affect the accuracy of the derived vertical TEC maps.
It can be seen that on the stormy day the accuracy of the L_{GF} carrier phase bias estimation is significantly higher as compared to the quiet days. It should be noted, however, that the main phase of this storm was characterized by a decrease of the electron content. That may explain the higher bias estimation accuracy during the storm.
Daily mean standard deviations \( \sigma_{x} \) in meters of the estimated \( B_{{i{\text{GF}}}}^{k} \) for 20 and 30degree elevation cutoff
DOY  \( \sigma_{x} \) _{(20°)}  \( \sigma_{x} \) _{(30°)} 

67  0.06  0.06 
68  0.06  0.05 
69  0.06  0.06 
70  0.05  0.05 
71  0.04  0.05 
72  0.04  0.04 
73  0.04  0.04 
Mean  0.05  0.05 
Statistics of the vertical TEC differences over day boundary discontinuities (30degree elevation mask) (TECU)
DOY  RMS  Min  Max  Mean  < 0.5 TECU (%)  < 1 TECU (%) 

67/68  0.72  0.02  1.81  0.55  56.14  87.72 
68/69  0.54  0.03  1.33  0.46  61.40  96.49 
69/70  0.49  0.00  1.11  0.39  68.97  96.55 
70/71  0.42  0.00  1.07  0.34  76.00  98.00 
71/72  0.57  0.02  1.21  0.48  60.38  94.34 
72/73  0.58  0.00  1.11  0.47  61.40  89.47 
Mean  0.55  64.05  93.76 
Statistics of the vertical TEC differences over day boundary discontinuities (20degree elevation mask) [TECU]
DOY  RMS  Min  Max  Mean  < 0.5 TECU (%)  < 1 TECU (%) 

67/68  0.79  0.00  2.08  0.59  61.54  76.92 
68/69  0.74  0.02  1.79  0.61  46.15  81.54 
69/70  0.46  0.02  1.13  0.38  71.43  97.14 
70/71  0.45  0.00  1.13  0.36  76.27  94.92 
71/72  0.57  0.01  1.51  0.45  59.70  91.04 
72/73  0.69  0.01  1.58  0.52  61.84  81.58 
Mean  0.62  62.82  87.19 
Conclusions
We presented and validated a methodology for accurate bias estimation of the phase function L_{GF} for use in GNSSbased regional ionospheric modeling. The methodology allows for subsequent TEC calculation using exclusively highaccuracy carrier phase data. The accuracy of the estimated bias was determined by analysis of its repeatability over the daytoday boundaries for GPS and GLONASS satellites. Test results show that the bias accuracy (RMS) is at the level of 7–8 cm, i.e., below 1 TECU depending on the selected data elevation cutoff. Note that the RMS may be partially impacted by undetected small cycle slip combinations. The RMS is slightly higher than the formal standard deviations derived from the variance matrix, which amounted to 4–6 cm.
In our opinion, these results confirm the high accuracy of our method that may be applicable to computing accurate ionospheric models. Such models may be used, e.g., to analyze disturbed states of the ionosphere or to support GNSS data processing, both absolute and relative, in highaccuracy applications in geosciences.
Notes
Acknowledgements
The research is supported by Grant No. UMO2013/11/B/ST10/04709 from the Polish National Center of Science.
References
 Alizadeh MM, Schuh H, Schmidt M (2015) Ray tracing technique for global 3D modeling of ionospheric electron density using GNSS measurements. Radio Sci 50(6):539–553CrossRefGoogle Scholar
 Böhm J, Schuh H (2013) Atmospheric effects in space geodesy. Springer, Berlin. https://doi.org/10.1007/9783642369322 CrossRefGoogle Scholar
 Bosy J (2005) Data processing of local GPS networks located in a mountain area. Acta Geodyn Geomater 3(139):49–56Google Scholar
 Bosy J, Graszka W, Leonczyk M (2007) ASGEUPOS—a multifunctional precise satellite positioning system in Poland. Eur J Nav 5:2–6Google Scholar
 Brunini C, Meza A, Azpilicueta F, Zele MAV (2004) A new ionosphere monitoring technology based on GPS. Astrophys Space Sci 290(3–4):415–429CrossRefGoogle Scholar
 Bruyninx C, Baire Q, Legrand, Roosbeek F (2011) The EUREF permanent network (EPN): recent developments and key issues. Presented at EUREF 2011 symposium, Chisinau, 25–28 May 2011Google Scholar
 Cai C, Liu Z, Xia P, Dai W (2013) Cycle slip detection and repair for undifferenced GPS observations under high ionospheric activity. GPS Solut 17(2):247–260CrossRefGoogle Scholar
 Charoenkalunyuta T, Satirapod C (2014) Effect of Thai Ionospheric Maps (THIM) model on the performance of network based RTK GPS in Thailand. Surv Rev 46(334):1–6CrossRefGoogle Scholar
 Griffiths J, Ray JR (2009) On the precision and accuracy of IGS orbits. J Geod 83(3–4):277–287. https://doi.org/10.1007/s0019000802376 CrossRefGoogle Scholar
 HernandezPajares M, RomaDollase D, Krankowski A, GarciaRigo A, OrúsPerez R (2017) Methodology and consistency of slant and vertical assessments for ionospheric electron content models. J Geodesy 91(12):1–10. https://doi.org/10.1007/s001900171032z CrossRefGoogle Scholar
 HernándezPajares M, Juan JM, Sanz J, Orus R, GarciaRigo A, Feltens J, Komjathy A, Schaer S, Krankowski A (2009) The IGS VTEC maps: a reliable source of ionospheric information since 1998. J Geod 83(3–4):263–275. https://doi.org/10.1007/s0019000802661 CrossRefGoogle Scholar
 HofmannWellenhof B, Lichtenegger H, Wasle E (2008) GNSS global navigation satellite systems; GPS, GLONASS, Galileo & more. Springer, New YorkGoogle Scholar
 Hunsucker RD, Hargreaves JK (2002) The highlatitude ionosphere and its effects on radio propagation. Cambridge University Press, CambridgeCrossRefGoogle Scholar
 KrypiakGregorczyk A, Wielgosz P, Gosciewski D, Paziewski J (2013) Validation of approximation techniques for local total electron content mapping. Acta Geodyn Geomater 10(3):275–283. https://doi.org/10.13168/AGG.2013.0027 CrossRefGoogle Scholar
 KrypiakGregorczyk A, Wielgosz P, Krukowska M (2014) A new ionosphere monitoring service over the ASGEUPOS network stations. In: 9th international conference environmental engineering (9TH ICEE)—selected papers. https://doi.org/10.3846/enviro.2014.224, http://leidykla.vgtu.lt/conferences/ENVIRO_2014/Articles/5/224_Krypiak_Gregorczyk.pdf
 KrypiakGregorczyk A, Wielgosz P, Jarmołowski W (2017) A new TEC interpolation method based on the least squares collocation for high accuracy regional ionospheric maps. Meas Sci Technol 28(4):045801. https://doi.org/10.1088/13616501/aa58ae CrossRefGoogle Scholar
 Leick A, Rapoport L, Tatarnikov D (2015) GPS satellite surveying, 4th edn. Wiley, HobokenGoogle Scholar
 Lyu H, HernándezPajares M, Nohutcu M, GarcíaRigo A, Zhang H, Liu J (2018) The Barcelona Ionospheric Mapping Function (BIMF) and its application to northern midlatitudes. GPS Sol (accepted)Google Scholar
 RoviraGarcia A, Juan JM, Sanz J, GonzálezCasado G, Ibáñez D (2016) Accuracy of ionospheric models used in GNSS and SBAS: methodology and analysis. J Geod 90(3):229–240CrossRefGoogle Scholar
 Schaer S (1999) Mapping and Predicting the Earth’s Ionosphere Using the Global Positioning System. Ph.D. dissertation Astronomical Institute University of BerneGoogle Scholar
 Schmidt M, Dettmering D, Moßmer M, Wang Y, Zhang J (2011) Comparison of spherical harmonic and B spline models for the vertical total electron content. Radio Sci 46(6):RS0D11. https://doi.org/10.1029/2010RS004609 CrossRefGoogle Scholar
 Shagimuratov I, Baran LW, Wielgosz P, Yakimova GA (2002) The structure of mid and highlatitude ionosphere during September 1999 storm event obtained from GPS observations. Ann Geophys 20(6):665–671Google Scholar
 Tegedor J, Ovstedal O, Vigen E (2014) Precise orbit determination and point positioning using GPS, Glonass, Galileo and BeiDou. J Geod Sci 4(1):65–73. https://doi.org/10.2478/jogs20140008 Google Scholar
 Wielgosz P (2011) Quality assessment of GPS rapid static positioning with weighted ionospheric parameters in generalized least squares. GPS Solut 15(2):89–99. https://doi.org/10.1007/s1029101001686 CrossRefGoogle Scholar
 Wielgosz P, Kashani I, GrejnerBrzezinska DA (2005) Analysis of longrange network RTK during severe ionospheric storm. J Geod 79(9):524–531. https://doi.org/10.1007/s00190005y CrossRefGoogle Scholar
 Zhao Q, Sun B, Dai Z, Hu Z, Shi Ch, Liu J (2015) Realtime detection and repair of cycle slips in triplefrequency GNSS measurements. GPS Solut 19:381. https://doi.org/10.1007/s1029101403962 CrossRefGoogle Scholar
Copyright information
Open AccessThis 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.