Rapid Static Positioning Using GPS and GLONASS S. C. SCHAER 1, E. BROCKMANN 1, M. MEINDL 2, G. BEUTLER Abstract For rapid static positioning, the combined analysis of all observed GNSS becomes more and more important in view of the steadily increasing number of active satellites (early in 2009 exceeding 50 GPS+GLONASS satellites). Successful ambiguity fixing for all involved GNSS is the key for this application. First results of an implementation of a “GNSS-capable” LAMBDA ambiguity resolution scheme into a project version of the Bernese Software are presented. 1 Implementation of a “GNSS-capable” LAMBDA ambiguity resolution scheme The initiation of GLONASS ambiguity resolution on an operational basis as part of the regional GNSS data analysis procedures as accomplished at swisstopo’s Permanent Network Analysis Center (PNAC) was an important step towards the best possible use of GNSS observation data [Schaer et al. 2007]. In particular for (additionally generated) GLONASS-only solutions, GLONASS-extended ambiguity fixing results in a clear quality improvement. The evaluated GNSS ambiguity resolution strategies of the Bernese Software are, 2 however, only applicable to static positioning with observation time spans of, let us say, longer than one hour. For so-called rapid static positioning, with typical observation time spans of maximally a few minutes, the Bernese Software Version 5.0 offers the “general search” strategy for GPS ambiguity resolution [Dach et al. 2007], a generalization of the Fast Ambiguity Resolution Approach (FARA) developed by Frei [1991]. The corresponding software realization does not allow it to keep all involved GNSS ambiguity parameters on the single-difference (SD) level, however, a basic requirement we believe to be of greatest importance for the success of correct GLONASS ambiguity resolution strategies [Schaer et al. 2007]. Therefore, we decided to develop a “GNSScapable” ambiguity resolution scheme using another, well established integer ambiguity resolution method, namely the LAMBDA (Least-squares AMBiguity Decorrelation Adjustment) method as developed by Teunissen [1995] and de Jonge and Tiberius [1996]. Figure 1 illustrates the three LAMBDA ambiguity resolution schemes we implemented for test purposes. Figure 1: Illustration of the implemented “GPS/GLONASS-capable” LAMBDA ambiguity resolution schemes for an assumed observation scenario involving 4 GPS (G) and 2 GLONASS (R) satellites: separate LAMBDA resolution steps 1 Swiss Federal Office of Topography swisstopo, Geodesy Division, Seftigenstrasse 264, CH-3084 Wabern, Switzerland, Phone: ++41 31 963 21 11, Fax: ++41 31 963 24 59, E-mail: stefan.schaer@swisstopo.ch, Web site: www.swisstopo.ch 2 Astronomical Institute of the University of Berne (AIUB), Sidlerstrasse 5, CH-3012 Bern, Switzerland for GPS and subsequently GLONASS (top, called “GPS/GLONASS”); common LAMBDA resolution step with (middle, called “mixed-GNSS”) and without (bottom, called “separate-GNSS”) intersystem double-difference ambiguity fixing. A dual-frequency observation scenario is assumed with 4 GPS and 2 GLONASS satellites. For this particular observation scenario a total of 12 single-difference ambiguity parameters have to be dealt with. The following two principles are common to all illustrated ambiguity resolution schemes: – – A set (or subset) of single-difference ambiguity parameters is examined by the LAMBDA method (on the basis of the real-valued estimates and their complete variance-covariance information coming from an initially computed ambiguity-float leastsquares parameter adjustment). Each examined set is identified by a solid bar in Figure 1. Properly selected differences of LAMBDAproposed SD integers are fixed (finally resulting in an integer ambiguity fixing on the doubledifference level). These differences are identified by horizontal squared brackets in Figure 1. Obviously, we apply the LAMBDA method to SD information, not to double-difference (DD) information as it is traditionally done (and this is correct in the GPS-only case). Ambiguity fixing, however, is furthermore done on the level of double differences. The implementation scheme in Figure 1 (top) includes two separate LAMBDA ambiguity resolution steps: one applied to the GPS-related ambiguity parameters and one to the GLONASS-related ambiguity parameters. An extra inversion of the remaining, thus “GPS-ambiguity-fixed” normal equation matrix is performed to achieve an improvement of the formal errors of the internally built GLONASS DD ambiguity estimates. Consequently, GPS may be seen as the master observation system for the first implementation scheme. We denote it in the following with “GPS/GLONASS.” The two other implementation schemes include just one common LAMBDA ambiguity resolution step for both GNSS observation systems, where all appearing SD ambiguity parameters are examined at one time by the LAMBDA method. The decisive point is that the second implementation scheme in Figure 1 (middle) does explicitly include integer fixing of intersystem ambiguity parameters (between GPS and GLONASS), whereas the third scheme (shown at the bottom of Figure 1) does not include this feature. Following this basic difference, we simply use “mixed-GNSS” and “separate-GNSS” to label the second and the third implementation scheme. Let us now explain why we treat all involved GNSS ambiguity parameters on the SD, not on the DD level: For GLONASS ambiguity resolution, different frequencies and consequently different carrier wavelengths between pairs of satellites are crucial. The bias introduced in the difference of two SD ambiguities is proportional to the initialization bias of the involved Page 2 SD ambiguities and, moreover, proportional to the frequency difference between the pairs of satellites considered. According to [Habrich 1999], GLONASS SD ambiguity parameters have to be initialized with an uncertainty of better than 285 cycles to keep the singledifference bias term below 0.1 cycles (more precisely for a frequency channel difference i–j of 1). This basic requirement is illustrated by Figure 2. We conclude that manipulation of GNSS (particularly GLONASS) ambiguity parameters on the DD level is dangerous, if not forbidden (due to the initialization uncertainty). It should be mentioned that we impose an a priori constraint (of nominally 200 meters) on each SD ambiguity parameter in order to avoid singularities (to be expected for the ambiguity-float solution). After successful GLONASS ambiguity fixing, all remaining GNSS SD ambiguity parameters finally become regular (actually determined) in the reduced normal equation matrix. They directly reflect the occurring, initially hidden SD ambiguity initialization bias values with a comparably small formal uncertainty (typically a fraction of a cycle). Note that these bias values depend on the receiver clock synchronization and therefore are de facto a function of the GNSS pseudorange biases. In this context, it is worth to specify the total number of remaining GNSS SD ambiguity parameters to be expected for the computation of the ambiguity-fixed solution: namely four for the “GPS/GLONASS,” two for the “mixed-GNSS,” and again four for the “separate-GNSS” ambiguity resolution scheme, or generally speaking, the initial number of SD ambiguity parameters minus the total number of fixed integers (in agreement with the illustration in Figure 1). To be more precise, two SD ambiguity parameters (one parameter per frequency) finally absorb potentially significant, non-negligible biases concerning SD ambiguity initialization (see Figure 2) and the two Figure 2: Illustration of the single-difference (SD) ambiguity bias term, as introduced in [Habrich 1999], to be proportional to the single-difference ambiguity initialization bias (and proportional to the GLONASS frequency channel difference i–j). For implementation into a project version of the Bernese Software Version 5.0, we used specifically the “LAMBDA4” Fortran-coded module (implying “direct computation of ZT”) as provided by the Delft University of Technology [de Jonge and Tiberius 1996]. Nevertheless, it should be possible to adapt the main principles and results discussed in this paper to any ambiguity resolution method suited for rapid static positioning. This includes also the “general search” method as previously addressed. 2 Test data set and processing options The baseline we considered for testing is shown in Figure 3, including details of the AGNES GNSS receiver network of Switzerland—which was enhanced from GPS to GNSS mid of 2007 [Ineichen et al. 2007]. It connects the Trimble NetR5 receivers (each associated with a Trimble Zephyr GNSS Geodetic II antenna) at ZIM2 (Zimmerwald) and HUTT (Huttwil). With a length of approximately 41 km, the selected baseline is relatively long for a “rapid-static” baseline. By this intended circumstance, additional requirements in terms of adequate modeling of the anticipated ionosphere delays are created (described below). information produced by CODE was used in conjunction with stochastic ionosphere parameters (SIPs) set up for each epoch and each GNSS satellite involved, imposing elevation-dependent a priori constraints of nominally 1 cm SD delay on L1 phase at zenith. No troposphere zenith path delay (ZPD) parameters were estimated because of the short observation sessions. We just relied on a standard troposphere model. 3 Discussion of the results 3.1 Baseline vector repeatability Figure 4 shows a particular time series of the ZIM2HUTT baseline vector computed on the basis of fiveminute GPS/GLONASS observation periods consecutively selected over the specified 24-hour period (September 27, 2008). The results shown in Figure 4 (and in the next two figures) are generally referred to an elevation mask angle of 15º. They may be accepted as evidence that the first introduced “GPS/GLONASS” LAMBDA ambiguity resolution scheme, adopted in this case, is working properly. We used the corresponding daily RINEX observation files for September 27, 2008 (containing data sampled at intervals of 30 seconds). An automated processing procedure was set up using the Bernese Processing Engine (BPE) and the ambiguity-fixed baseline vector was computed for – 288 consecutive observation periods, covering five minutes (eleven epochs) each, – elevation mask angles of 3º, 5º, 10º, 15º, etc. – the three introduced GNSS LAMBDA resolution schemes (using GPS and GLONASS)—plus a LAMBDA-based reference solution using GPS observation data only. Figure 3: Detail of the AGNES GNSS receiver network. The indicated baseline connecting the Trimble NetR5 receivers at ZIM2 (Zimmerwald) and HUTT (Huttwil) was considered for validation of the GNSS-capable LAMBDA ambiguity resolution implementation. The GPS-only reference solutions were also computed in complete accordance with our GNSS LAMBDA implementation (basically relying on a SD-level ambiguity parameter treatment). We used the consistent orbit information of the final orbit product of the GNSS analysis product line generated of the Center for Orbit Determination in Europe (CODE) for the International GNSS Service (IGS) [Dach et al. 2009]. Furthermore, we adopted an elevation-dependent observation weighting to the phase data. The weighting function is a reciprocal sine function, 1/sin(e)2, and assumed to be identical for the two frequencies and the two GNSS systems. For ionosphere modeling, global ionosphere map (GIM) Page 3 In Figure 4 (bottom) we included the difference of a ZIM2-HUTT ZPD reference signal (as extracted from swisstopo’s regular AGNES daily processing results and moreover multiplied by a factor of 3). The high correlation between the Up component residual signal and the ZPD reference signal shows the effect of the neglected differential troposphere variations in the rapid static positioning. Nevertheless, we emphasize again that the estimation of ZPD troposphere parameters for rapid static positioning (and accordingly for very short observation sessions) would be inappropriate, if not counterproductive. The ZIM2-HUTT baseline vector repeatabilities were computed for the GPS-only case and for different GPS/GLONASS-combined cases, performing the “GPS/GLONASS” and the “mixed-GNSS” LAMBDA implementation. They are evaluated statistically in Figure 5. The standard deviation values turned out to be minimal for the “GPS/GLONASS” case, consistently for all three components. Compared to the GPS-only case, the improvement may reach values up to 15%. The latter value is, by the way, relatively consistent to the value reported by Ineichen et al. [2008], where the quality of GPS-only and GNSS kinematic positioning results was compared. GPS/GLONASS analysis, respectively. These values must be considered as rather hypothetical in view of the circumstance that accurately known ZPD troposphere estimates are in practice hardly available (in particular for rapid static positioning). The general results of Figure 5, however, remain unchanged by these “troposphere-reduced” values. Figure 6 shows the total number of observed GNSS satellites as a function of time using an elevation mask angle of 15º (generally used in this subsection). The number of available GPS satellites varies between 5 and 10, that of GLONASS between 1 and 6. The total number of observed GNSS satellites did never fall below 8 and reached occasionally 14. Note that these numbers are valid for September 27, 2008, a day for which a total of 32 GPS plus 14 GLONASS active satellites is made available in CODE’s GNSS orbit product line. It should also be noted that the tracking data of one particular GLONASS satellite (R09) is commonly considered as unsuitable and therefore excluded from POD at CODE. 16 14 12 Standard deviation (mm) Figure 4: ZIM2-HUTT baseline vector repeatability on the basis of GPS/GLONASS observation data collected on September 27, 2008, computed for 288 consecutive five-minute observation spans (using elevation mask angle of 15º). The bottom subfigure showing the Up component includes the difference of the ZIM2-HUTT ZPD reference signal (multiplied by a factor of 3). 10 8 6 4 2 0 The assessment of the performance for the two “GNSS” special cases, “mixed-GNSS” and “separateGNSS,” is interesting. Whereas the “separate-GNSS” case did not reveal any difference with respect to the “GPS/GLONASS” case (this case is therefore not explicitly shown in Figure 5), we observed a significantly degraded quality of the “mixed-GNSS” case, which, for the East component, turned out to be even worse than that of the GPS-only case. This degradation in quality is a clear indication that intersystem DD ambiguity parameters are definitely not of an integer nature. We conclude that no attempt should be made to resolve GPS–GLONASS DD ambiguities to integers [Schaer et al. 2007]. In the context of GPS–GLONASS phase biases, it is a must to specify that we generally assume that the variability of these biases has to be on a level comparable to the phase measurement noise. Schaer et al. [2007] give examples (considering an old GNSS receiver model) where this essential assumption is occasionally not met. For the sake of completeness, we give the reduction for the standard deviation of the Up component when daily processed ZPD troposphere estimates (as indicated at the bottom of Figure 4) were taken into account. We obtained 8.9 mm for the GPS-only and 7.8 mm for the Page 4 North East Up GPS-only 6.1 4.5 14.1 GPS/GLONASS 5.2 4.3 12.8 Mixed-GNSS 5.9 4.8 13.3 Figure 5: Statistics of ZIM2-HUTT baseline vector repeatability results, for September 27, 2008 (elevation mask angle: 15º). GPS-only, “GPS/GLONASS,” and “mixed-GNSS” (with intersystem DD ambiguity fixing) results are compared. Figure 6: Number of GNSS satellites as observed on the ZIM2-HUTT baseline during consecutive fiveminute observation spans, for September 27, 2008 (elevation mask angle: 15º). Dashed lines indicate the daily mean numbers for each GNSS system. 3.2 Availability analysis based on maximum elevation mask angle The elevation mask angle was mentioned in Section 2 as an important processing option. It is used here to simulate a gradually increasing restricted satellite visibility. We checked the gathered analysis parameters for each GNSS LAMBDA evaluation (at intervals of 5º) in order to find the maximum elevation mask angle, where a particular GNSS LAMBDA evaluation was just able to provide a correct and adequately accurate baseline vector determination. The following analysis parameters were consulted to check whether a particular solution is acceptable or not: the root-meansquare (rms) error of unit weight of the final leastsquares parameter adjustment, the absolute deviation (from a long-term average) and the formal accuracy for each of the three components. Figure 7 shows the accumulated maximum elevation mask angle values as a function of time for two cases: GPS-only and GPS/GLONASS LAMBDA ambiguity resolution. The circles plotted, indicating the enhancement from GPS-only to GPS/GLONASS, are evidence of the advantage of the “GPS/GLONASS” LAMBDA application. It is really remarkable that the GPS/GLONASS case is in no case inferior to the GPSonly case (the resulting enhancement is always positive). In the GPS/GLONASS application, we were able to produce adequate baseline vectors several times using an elevation mask angle of up to 55º. The daily mean values for the assessed maximum elevation mask angle (sufficient to get a successful GNSS LAMBDA solution) were found to be 32.8º (GPS-only) and 37.7º (GPS/GLONASS). This corresponds, statistically speaking, to an average availability improvement of around 5º (at that time the GNSS constellation, consisting of simply 32+14=46 satellites, was far from being complete). Note that the addressed daily mean values are indicated with dashed lines in Figure 7. Figure 7: Maximum elevation mask angle with successful GPS-only and “GPS/GLONASS” LAMBDA ambiguity resolution, respectively, as determined for the ZIM2-HUTT baseline, for September 27, 2008. Dashed lines indicate daily mean values. The total number of observed GPS (not GPS+ GLONASS) satellites above the maximum elevation mask angle (see Figure 7) is shown in Figure 8, underlying again the clear advantage of the GPS/GLONASS-combined application. In the GPSonly case, the minimally accepted number of satellites is sporadically at the level of 4 (conforming to the theoretical minimum number), in the majority of cases at 5, and, on the average, at a level of about 5.4 GPS satellites. In contrast to the GPS-only case, the GPS/GLONASS application is able to provide adequate GNSS LAMBDA solutions on the basis of, sometimes only 3 GPS satellites, relatively often 4, and, on the average, about 4.6 GPS satellites. Although the GLONASS constellation was accessible through a comparably small number of active satellites at that time, we may draw the conclusion that making use of the GLONASS observation system in a consistent way is strongly advisable to significantly improve the solution availability and reliability for rapid static positioning. The latter statement, however, can only be justified, if the required GPS/GLONASS ambiguity resolution process is performed in a proper form. If, e.g., GPS– GLONASS intersystem ambiguity parameters are included in the ambiguity resolution process, the situation may be judged differently. The lack of the “mixed-GNSS” LAMBDA method is obvious in Figure 9, where the enhancement concerning maximum elevation mask angle is compared for the two previously introduced “GNSS” cases with respect to the reference values of the “GPS/GLONASS” LAMBDA ambiguity resolution (shown in Figure 7). As opposed to the “separate-GNSS” case, where the resulting availability values are practically identical to the “GPS/GLONASS” reference values, the “mixedGNSS” case turned out to be unacceptable. Page 5 unjustified, arbitrary integer resolution of intersystem phase biases. It should be emphasized that all these schemes may be easily adapted to schemes coping with dual-frequency, or single-frequency ambiguity parameters from multiple GNSS systems (involving three or more GNSS systems). Concerning our implementation, the following aspect is essential: generally all GNSS ambiguity parameters are kept on the level of single (not double) differences. Ambiguity fixing, however, is rigorously done on the level of double differences. Figure 8: Number of observed GPS satellites above the maximum elevation mask angle, as determined for the ZIM2-HUTT baseline, for September 27, 2008. Dashed lines indicate the daily mean numbers. Figure 9: Enhancement concerning maximum elevation mask angle for the “separate-GNSS” case (top) and the “mixed-GNSS” case (bottom) compared to the reference values of the “GPS/GLONASS” LAMBDA ambiguity resolution (shown in Figure 7). 4 Summary and conclusions We developed three different ambiguity resolution schemes for adequately handling the GPS and GLONASS ambiguity parameters in the “rapid static positioning” mode (where a few minutes or less of data are supposed). We used the well established LAMBDA method for managing the ultimate task of ambiguity resolution to integers. We denoted these LAMBDAbased ambiguity resolution schemes—as illustrated in Figure 1—with: “GPS/GLONASS” (without), “mixedGNSS” (with), and “separate-GNSS” (again without intersystem integer fixing). The scheme of “mixedGNSS” was added in anticipation of problems due to Page 6 All our GNSS ambiguity resolution schemes were implemented into a project version of the Bernese Software (using the LAMBDA4 Fortran module). Thereafter, a dedicated test procedure was set up for validation on the basis of a RINEX daily data set, in particular for September 27, 2008, and the baseline connecting ZIM2 (Zimmerwald) and HUTT (Huttwil), both occupied by a Trimble NetR5 receiver of the AGNES GNSS receiver network of Switzerland. With a length of approximately 41 km, the baseline is relatively long, a circumstance asking for careful modeling of the anticipated ionosphere delays (realized by the additional estimation of tightly constrained parameters absorbing these ionosphere delays). We computed a set of ambiguity-fixed baseline vectors by scanning (a) the time domain (at intervals of 5 minutes, (b) a varying elevation mask angle (at intervals of 5º), and ultimately (c) four different application cases: “GPS/GLONASS,” “mixed-GNSS,” “separate-GNSS,” and in addition a consistently handled GPS-only case, conforming to a LAMBDAbased reference solution without GLONASS contribution. The dimension (a) was used to test the operational reliability of each GNSS-capable LAMBDA ambiguity resolution scheme, (b) to assess the availability (the varying elevation mask angle was used to simulate a gradually more and more restricted satellite visibility), and finally the dimension (c) was dedicated to assess the general performance of the slightly differing GNSS ambiguity resolution schemes. The repeatabilities obtained with the various time series of baseline determinations clearly indicate that the implemented GNSS-capable ambiguity resolution scheme (relying on the LAMBDA method) may be applied in the GPS-only as well as in the “GPS/GLONASS” case. Furthermore, the performance of the “GPS/GLONASS” application turned out to be significantly better in terms of both baseline vector accuracy and solution availability (compared to the GPS-only case). The repeatability (standard deviation) values specific to “GPS/GLONASS” could be reduced by up to about 15%, a reduction rate that is consistent to those reported by Ineichen et al. [2008] and Dach et al. [2009] for ambiguity-fixed single-epoch solutions. The improvement of the solution availability, expressed by an increment in the elevation mask angle, was assessed to be about 5º. This value, however, is just a daily average (valid for a comparably poor ratio of 32/14 GPS/GLONASS satellites). In extreme cases, we were able to generate suitable “GPS/GLONASS” LAMBDA solutions using elevation mask angles of up to 55º. The solution availability is mainly a function of the number of available (and actually observed) GNSS satellites. The application “GPS/GLONASS” is in no case inferior to that of GPS-only (the resulting enhancement is always positive or equal to zero). This is more than remarkable. number of satellites), but crucial in case of a severely restricted satellite visibility (with a total number of observed satellites close to the mathematically allowed minimum). – In the context of the previous item, the following qualitative characteristic is essential—as long as intersystem GPS–GLONASS double-difference observation equations are built: such (obviously existing) intersystem phase biases must be constant roughly on the level of the phase observation noise (over the observation period). – The auxiliary “separate-GNSS” LAMBDA scheme generated nearly identical results to those of our primary “GPS/GLONASS” scheme (both without intersystem integer fixing). – Our development is transferable without restriction to a more general “multi-GNSS” ambiguity resolution scheme (consistently applicable to three or more GNSS). Let us finally highlight the basic findings of our dualGNSS (GPS/GLONASS) and single-GNSS (GPS) LAMBDA-based ambiguity resolution experiments: – – – – The performance of the dedicated application of GPS and GLONASS (following our primary “GPS/GLONASS” LAMBDA implementation scheme) turned out to be significantly better in terms of both baseline vector accuracy and solution availability (compared to the consistently executed application of GPS only). The use of observation data of the GLONASS system is extremely helpful, in particular under difficult observation conditions (with restricted satellite visibility). The performance improvements are expected to become even more pronounced with the expected decreasing ratio of GPS to GLONASS satellites (thanks to the continuously growing GLONASS constellation, but also due to the potential drop in the number of healthy GPS satellites caused by the delay in the Block IIF program). Handling of the GNSS ambiguity parameters on the single-difference, not double-difference level is considered to be of greatest importance for a proper treatment of the GLONASS singledifference bias term, which is directly caused by slightly different carrier frequencies for each GLONASS satellite—and which is definitively crucial when finally forming (double) differences from GLONASS single-difference ambiguities (see, e.g., [Habrich 1999], or [Schaer et al. 2007]). This technicality is transferable to any other GNSS ambiguity resolution strategy and therefore generally adopted in the Bernese Software. We successfully processed GNSS observation data from diverse “receiver-mixed” baselines. These (randomly accomplished) applications confirm that our GNSS-capable ambiguity resolution scheme is self-calibrating and, moreover, able to cope with very significant GLONASS-relevant ambiguity initialization bias values (of a few of hundreds of cycles to be reckoned with in the case of a pair of mixed receiver models). The same is true for random tests with considerably shorter observation time spans (with 30 seconds, or two epochs). Intersystem GPS–GLONASS double-difference phase biases should not be fixed to integers. The assumption of the integer nature of these biases is obviously not valid. The impact of ignoring this fact is rather marginal in case of a good observation scenario (with a relatively large Strict resolution of all involved ambiguity parameters to integers is not appropriate for rapid static positioning using multiple GNSS (two or more GNSS). This, however, is an assumption for the LAMBDA method. Two of the three developed LAMBDA-based ambiguity resolution schemes (namely “GPS/ GLONASS” and “separate-GNSS”) may actually be regarded as a technically feasible bypass (avoiding integer fixing of intersystem ambiguity parameters). The use of (adequately downweighted) observations from low elevation angles (as we recommend it) is another strong argument for leaving the general rule of fixing all involved ambiguity parameters. Our results suggest that each ambiguity resolution method should be extended to comply with the formulated requirement. The “general search” ambiguity resolution method (traditionally used in the ernese GPS Software) might be easily extended in this direction. References Dach R., U. Hugentobler, P. Fridez, M. Meindl (Eds.) (2007): Bernese GPS Software Version 5.0. Documentation, Astronomical Institute, University of Bern, January 2007. Dach R., E. Brockmann, S. Schaer, G. Beutler, M. Meindl, L. Prange, H. Bock, A. Jäggi, L. Ostini (2009). GNSS Processing at CODE: Status Report. Journal of Geodesy, 83, 353–365. Frei E. (1991): Rapid Differential Positioning with the Global Positioning System (GPS). Geodätischgeophysikalische Arbeiten in der Schweiz, Volume 44. Habrich H. (1999): Geodetic Applications of the Global Navigation Satellite System (GLONASS) and of GLONASS/GPS. Ph.D. thesis, Astronomical Institute, University of Berne. Ineichen D., E. Brockmann, S. Schaer (2007): Enhancing the Swiss Permanent GPS Network Page 7 (AGNES) for GLONASS. In: Torres, J.A. and H. Hornik (Eds): Subcommision for the European Reference Frame (EUREF), London, 2007. Ineichen D., E. Brockmann, S. Schaer (2008): Processing Combined GPS/GLONASS Data at swisstopo’s Local Analysis Center. In: Torres, J.A. and H. Hornik (Eds): Subcommision for the European Reference Frame (EUREF), Brussels, 2008. Jonge de P.J. and C.C.J.M. Tiberius (1996): The LAMBDA method for integer ambiguity estimation: implementation aspects. Publications of the Delft Geodetic Computing Centre, LGR Series, No. 12, August 1996. Schaer S.C., E. Brockmann, D. Ineichen (2007): Inclusion of GLONASS for EPN Analysis at CODE/swisstopo. In: Torres, J.A. and H. Hornik (Eds): Subcommision for the European Reference Frame (EUREF), London, 2007. Teunissen P.J.G. (1995): The least-squares ambiguity decorrelation adjustment: a method for fast GPS integer ambiguity estimation. Journal of Geodesy, 70, 65–82. Page 8