Real-Time GPS Reference Network Carrier Phase Ambiguity Resolution

H. Sun, Kvaerner Marine Automation M.E. Cannon, University of Calgary T.E. Melgard, Kvaerner Marine Automation

BIOGRAPHY Huangqi Sun received an M.Sc. from Department of Getmatics Engineering at The University of Calgary in 1994. He also received a B.Sc. and M.Sc. in Geodesy from Wuhan Technical University of Surveying and Mapping in 1987 and 1989 respectively. He is currently working as a consultant on Wide-Area RTK system for Kvaerner Marine Automation. He is also a Ph.D candidate in the Department of Geomatics Engineering at The University of Calgary. Dr. M.E. Cannon is a Professor in Geomatics Engineering at the University of Calgary where she conducts teaching and research related to GPS and integrated GPS/INS systems. She is a Past President of the Institute of Navigation. Tor Egil Melgrd graduated in 1994 from the Norwegian Institute of Technology (NTH) with a Master degree in electrical engineering. Since then he has continued his work in the area of GPS and navigation. Currently he is employed at Kvrner Marine Automation (KMA) located near Oslo, Norway, where he is Project Manager for the development of a Wide Area RTK system for Norway. ABSTRACT GPS carrier phase ambiguity resolution has been limited to relatively short inter-receiver distances due to the rapid error decorrelation process that occurs as a function of distance. A network of reference stations can however be used to model errors as a function of user location within the network coverage area. Using this procedure, the user can correct its measurements with the error model and improve Real Time Kinematic (RTK) positioning performance. In order for this to be optimal however, network double difference carrier phase ambiguities must be resolved to integers before network error modeling. This presents a challenge for real-time operations when

the separation between reference stations is long, i.e. typically 200 km or more. This paper introduces a sequential least-squares estimation algorithm for real-time integer ambiguity resolution over such a network. Various methods for modeling the carrier phase errors within the network are analyzed and the use of network constraints for improved ambiguity resolution during the network and new satellite initialization stages are investigated and tested. Multiple ambiguity validation tests are demonstrated and the reliability of the approach is also presented.

INTRODUCTION GPS carrier phase integer ambiguity resolution over large networks is one of the major components in large GPS processing packages such as GIPSY (Lichten & Border, 1987) and the Bernese program (Beutler et al., 1994). These packages are designed for post-processing whereby the carrier phase integer ambiguities are determined together with other modeled errors (e.g. residual tropospheric effects, ionospheric error, orbit) using a data set which typically extends over several hours. With the growing requirement to offer carrier phase differential services to support Real Time Kinematic (RTK) positioning applications, the use of a network for extending the operational limits of the service has become important. For this to be effective, spatial errors in the double differenced carrier phase data must be modeled over the network (Raquet & Lachapelle, 1996; Wubbena et al. 1996; Varner & Cannon, 1997). In order to optimally estimate these errors, integer ambiguities across baselines in the network must first be determined. Real-time network integer ambiguity resolution is more complicated than for post-processing for three main reasons, namely (1) there are constraints on the computation time (i.e. the amount of time required cannot

Institute of Navigation National Technical Meeting/San Diego/January 25-27, 1999

exceed the data interval), (2) the data must be processed sequentially such that only observations before the current epoch can be considered while in post-processing the data can be treated in batch mode and, (3) operational issues such as the time a satellite is continuously tracked cannot be determined at the current epoch without knowledge of future measurements. The purpose of this paper is to present an approach which can be used for determination of the double difference integer ambiguities for a GPS network with receiver separations of up to a few hundred km. The goals for the algorithm are that it must be effective in the time required to determine integers, it must be reliable and it must be able to operate in real-time. Various L1 and L2 measurement combinations are examined which lead to different methods for modeling the carrier phase errors within the network. The use of network constraints for ambiguity resolution during the "network initialization" and subsequent "new satellite initialization" stages are investigated and tested using a data set collected on the SATREF system in Norway in September, 1997. Multiple ambiguity validation tests are demonstrated and the reliability of the approach is also presented.

x = ( AT PA) 1 AT Pl

(2)

with the corresponding Sum of Squared Residuals (SOSR) as

= ( l A x) T P((l A x) = l P l + l PAT x .
T T

(3)

A constraint can be applied to the parameters by

K x=k

(4)

where K is a matrix with zeros and ones to specify the ambiguities to be constrained, and k is a vector consisting of the actual constrained values of the ambiguities. The revised SOSR when using the constraint is then 1 = + (k K x)T (K( AT PA)1 KT )1(k K x) . If we define

(5)

S = ( K ( A T PA) 1 K T ) 1

(6)

where S is a symmetric matrix, it can be decomposed into S = CC T (7)

AMBIGUITY RESOLUTION ALGORITHM A number of ambiguity resolution approaches have been proposed for long baselines (e.g. Blewitt, 1989; Beutler, 1994). In general, these approaches implement various combinations of L1 and L2 carrier phase data to form lanes of different wavelengths to reduce the ambiguity search space. For example, widelane (i.e. L1-L2) ambiguities are usually resolved first due to its wavelength of 86.4 cm. Widelane Ambiguity Resolution The resolution of widelane ambiguities can be done using a sequential estimation approach. The search method implemented in Euler & Landau (1992) and Chen & Lachapelle (1994) are similar in that the integer search range is determined from the estimated standard deviation of the floating ambiguity states which decrease with time (i.e. as more measurements are processed). A general observation equation for this sequential approach can be expressed as

using a Cholesky decomposition. Equation (5) can therefore be rewritten as 1 = +[(k K x)T C][(k K x)T C] = + f T f where f T = (k K x) T C

(8)

(9)

This computation can be broken down by simply squaring and adding the vector elements of f. Early rejection of candidate integer ambiguities can be achieved by comparing the SOSR computed for the current set of integer ambiguity candidates with the smallest and second SOSR. If the SOSR value for the current set of integers is larger than either the smallest or second smallest then it can be rejected before too many computations are performed. L1 Ambiguity Resolution Once widelane (WL) ambiguities are solved, L1 ambiguities can be determined using the ionospheric-free combination (Ashtech, 1994; Raquet, 1998). The ionospheric-free (IF) combination can be expressed as: WL = L1 L 2 (10)

l = Ax

(1)

where l is the adjusted observation vector, x is the estimated parameter vector while A is the design matrix.

x can be computed through the following equation

 IF = L1 where L1 , L 2 f1 , f 2 WL IF

f2 L2 f1

(11)

are the L1 and L2 carrier phase measurements (cycles), are the L1 and L2 frequencies (Hz), is the widelane observation (cycles), and is the ionospheric-free observation (cycles).

Figure 1: Simple Example of a Reference Network From the network, five constraint equations can be formed, i.e. three triangles and one quadruple. Therefore, the summation of ambiguities from triangle ACD should equal to zero, e.g. N AC + N CD + N DA = 0 where is the double difference operator. Theoretically, only two of these five equations are independent since the other three can be derived from these two equations. Therefore only two constraint equations are used. Network constraints can be applied in several stages to increase the ambiguity search speed and to enhance reliability. Firstly, it can be applied to reduce the integer ambiguity search range. The same triangle (i.e. ACD) is used to illustrate this concept. The search ranges corresponding to the three baselines are N AC N AC N CD N CD and N DA N DA . (17) (18) (19) (16)

Equation (10) represents the widelane combination while Equation (11) represents the ionospheric-free signal. It should be noted that the widelane combination has a wavelength of 86.4 cm, which means that its integer ambiguity is easier to resolve than either the L1 or L2 integers which have wavelengths of 19 and 24 cm, respectively. The ionospheric-free combination does not preserve the integer ambiguity property, but it removes first-order ionospheric effects (Blewitt, 1989). From Equations (10) and (11), the ambiguity equations can be formed as N WL = N L1 N L 2 N IF f = N L1 2 N L 2 f1 (12) (13)

where N is the corresponding ambiguity term. The ionospheric-free ambiguity (Equation 13) can therefore be rewritten as: N IF f = N L1 2 ( N L1 N WL ) f1 f f2 ) N L1 + 2 N WL f1 f1

(14)

= (1

while the corresponding L1 ambiguity is then solved as N IF N L1 = f2 N WL f1 f (1 2 ) f1

Before the search process is started, we can check if there is a corresponding integer ambiguity candidate that exists from the other two baselines which fulfils the constraint given in equation (16). The ambiguity candidate that fulfills this requirement will be used for the following search for each baseline. This process is helpful to reduce the search ranges. When short and long baselines are combined in the network, the search ranges of the longer baselines will be greater than those of the shorter ones. By applying constraints, the integer search ranges of the longer baselines are reduced. This process will speed up the ambiguity resolution process after some baselines are resolved. Using the above example, if the ambiguity for baseline AC is determined, then the search ranges for this triangle will be

(15

NETWORK CONSTRAINTS It is well known that the sum of integer ambiguities around a closed loop should yield zero. As an example, a simple four station reference network is shown in Figure 1.

N AC (determined) N CD N CD and N DA N DA .

(20) (21) (22) we have to estimate five baselines, namely AC, CD, DA, CB and BA. When a fixed set of network stations are available, we can pre-determine the baselines to estimate. This decision can also be made dynamically during the estimation process. AMBIGUITY VALIDATION The residuals from the double difference measurements are still the major indicator used to validate the integer ambiguities from the search. A common approach is to test the ratio between the second smallest SOSR and the smallest SOSR. ratio = 2 1 . (25)

If the constraint equation is applied in this case, the search ranges for the other two baselines will be reduced. If the ambiguities of two baselines are resolved, the ambiguities for the third baseline can be calculated directly from the constraint equation so no search is involved. Following on with the same example, assume that the ambiguities of baselines AC and CD are resolved. The ambiguities for DA can then be calculated as N DA = N AC N CD .

(23)

After all of the ambiguities are resolved, the constraint equations can also be used to verify that the selected ambiguities are correct. If all the network ambiguities satisfy the constraint equations for the entire network, then this verification will generally guarantee the correctness of the network ambiguities. Otherwise, all the selected ambiguities of the network will be incorrect. This extreme case can only happen when the entire network is biased by the same systematic error. BASELINE SELECTION Figure 1 is used again here as an example to illustrate how the baselines are chosen for estimation. If there are only four reference stations, we only need to estimate three baselines to sufficiently determine the entire network. However, as discussed in the above section, additional baselines have to be estimated in order to apply the ambiguity constraint equations. The number of baselines used for ambiguity resolution is the same as the number of baselines required for application of the network ambiguity constraint equations. For example in Figure 1, in order to apply two constraint equations, namely N AC + N CD + N DA = 0 and N AC + N CB + N BA = 0 (24a) (24b)

For a reference network, this test becomes more reliable since positions from all the stations are held fixed. Therefore, if the integer ambiguities resolved from the search process are correct, then all the errors, including decorrelated orbit, atmospheric error, multipath and noise will be absorbed by the residuals. Another validation process is to apply the constraint equations across the entire network after the integer ambiguities from all baselines are resolved. If all integer ambiguities fulfill the constraint equations, then there is a high degree of probability that all integer ambiguities are correct. DESCRIPTION OF TEST DATA A 24-hour data set from the Norwegian SATREF network is used to verify the algorithm presented above. The data set was collected from 11 receivers in southern Norway on September 29, 1997, see Figure 2. Among the 11 receivers, five were Trimble 4000SSi and six were Ashtech Z-12. In the network, there are two pairs of stations (3 and 4, 9 and 10) which are very close to each other (only a few tens of meters apart).

64 63 Latitude (deg) 62

Reference Stations 1 2

receiver, it can show up several times the figures since three or four baselines are formed using data from this receiver.

3 61 7 60 59 58 4 6 6
Triple Diff. (cycles

0. 5

8 11 8

9 10 10 12 14

Longitude (deg)

- 5 0.

Figure 2 Reference Station Distribution There are 55 possible baselines which can be formed from the 11 reference stations. However, as discussed before, only baselines which are required to form independent constraint equations are selected for estimation. In this network, 19 baselines are selected to estimate, see Figure 3. During the selection process, shorter baselines are chosen first to minimize the effects of spatially decorrelated errors.
64 63 Latitude (deg) 62 3 61 7 60 59 58 4 6 6 5 4
1 1 44500 1 64500 1 84500 204500

GPS Tim e (s)

Figure 3 Cycle Slips On L1 for All Satellites From 19 Baselines

1

Reference Stations 1 2
Triple Diff. (cycles)

0.5

-0.5

8 11 8

9 10 10 12 14

-1 144500

164500 184500 GPS Tim e (s)

204500

Longitude (deg)

Figure 4 Cycle Slips On L2 for All Satellites From 19 Baselines

Figure 3 Baseline Selections

There were many cycle slips that occurred on low elevation satellites, especially at some particular sites. In many cases the slips are not integer values, which makes them difficult to recover. Figure 4 illustrates the fractional part of the cycle slips detected on L1 frequency while Figure 2 is the fractional part of the cycle slips for L2. Please note these two figures shows the cycle slips for all 19 baselines. Therefore, if a cycle slip occurs at one Widelane integer ambiguities are first resolved sequentially using the methodology discussed above. Figure 5 shows the time needed to fix integer ambiguities without constraints as a function of baseline length. As can be seen, the time required to fix integer ambiguities is in general correlated with the baseline length. However, for baselines from some particular stations, the time to resolution is very long. For example, all the baselines related to station number 8, it takes more than two hours

to resolve integer ambiguities. This is due to the quality of data at that station (i.e. multipath and cycle slips).
9000
9000

Time For Widelane Ambiguity Search with Constraint Equations

Time For Widelane Ambiguity Search 7200

7200

Search Time (s)

5400

Search Time (s)

5400

3600

3600

1800
1800

0 0 100 200 300 Baseline Length (Km ) 400

0 0 100 200 300 Baseline Length (Km) 400

Figure 5 Widelane Ambiguity Converge Time for 19 Baselines Without Constraints After application of the network ambiguity constraint equations, the time required to resolve integer ambiguities for the first few baselines is slightly more than without constraints. The reason is that errors from the other baselines may be distributed to these baselines through the application of constraint equations. In general the impact is not significant since the float solutions for these baselines are recomputed at each epoch. If the constraint equations affect the float solution of a certain baseline, the float solution for this baseline at next epoch will try to adjust it again. However, once integers are determined for these baselines, the constraint equations become effective. As an example, all baselines relating to station 8 required less than 1.5 hours to converge on the integers after the constraint equations were applied.

Figure 6 Widelane Ambiguity Converge Time for 19 Baselines With Constraints L1 ambiguity searching was also tested using the algorithm discussed above. Only the shorter baselines (34, 5-6 and 9-10) were resolved due to the difficulties in resolving integers with single frequency data. The lengths of these baselines varies from 11 meters to 28.7 kilometers. All other baselines could not be resolved with or without constraint equations. Other tests will be performed later to evaluate the effect of using precise orbit, other lane combinations and shorter baselines. CONCLUSIONS AND RECOMMENDATIONS This paper focused on the development and testing of an algorithm to resolve integer carrier phase ambiguities over a network of reference stations spaced 100-200 km apart. This development is done in support of generating carrier phase corrections over the network for real-time kinematic applications. Real-time integer ambiguity resolution is more difficult than in post-processing due to computational time constraints and the availability of measurements only up to the current epoch. Widelane integer ambiguities are generally easier to resolve because of its long wavelength. In order to determine L1 integer ambiguities, other phase combinations, such as the ionospheric-free observable, are required for assistance. Tests using 11 stations from the Norwegian SATREF network showed that network integer ambiguity constraint equations can reduce the ambiguity search time significantly. Once the ambiguities for some baselines are

resolved, the reduction of search time can reach up to 60 percent. Resolution of the L1-only integer ambiguities is more difficult and was only successful on the shorter baselines of the network. In order to improve the performance of integer ambiguity resolution over long baselines, more investigations are required. Other lane combinations, such as extrawidelaning will be tested. In addition, since all ambiguity searches performed here used the broadcast ephemerides, the impact of precise orbits will also be tested. Other networks that contain a denser set of reference stations will also be investigated to determine the major factors that will affect the ambiguity search for both the dual and single frequency cases.

Euler, H.J. and H. Landau (1992) Fast GPS Ambiguity Resolution O-The-Fly For the Real-Time Application, Proceedings of 6th International Geodetic Symposium on Satellite Positioning, Columbus, Ohio, 1992. Lichten, S.M and J.S. Border (1987) Strategies for High Precision GPS Orbit Determination, Journal of Geophysical Research, Vol. 92 (B12), pp.1275112762 Raquet, J. and G. Lachapelle (1996) Determination and Reduction of GPS Reference Station Multipath Using Multiple Receivers, Proceedings of ION-GPS96, pp. 673-681, Kansas City, Missouri. Raquet, J. (1998) Development of a Method for Kinematic GPS Carrier-Phase Ambiguity Resolution Using Multiple Reference Receivers, UCGE Reports, Number 20116, Department of Geomatics Engineering, The University of Calgary. Varner, C. and M.E. Cannon (1997) The Applications of Multiple Reference Stations to the Determination of Multipath and Spatially Decorrelated Errors, Proceedings of the National Technical Meeting of the Institute of Navigation, pp. 323-333, Santa Monica, California. Wubbena, G., A. Bagge, G. Seeber, V. Boder and P. Hankemeier (1996) Reducing Distance Dependent Errors for Real-Time Precise DGPS Applications by Establishing Reference Station Networks, Proceedings of 9th International Technical Meeting of Satellite Division of the Institute of Navigation, pp.1845 1852, Kansas City, Missouri.

REFERENCES Ashtech (1994) PRISM: Process Users Guide (software manual), pp. 29-30. Beutler, G., E. Brokmann, W. Gurtner, U. Hugentobler, L. Mervart and M. Rothacher (1994) Extended orbit modelling techniques at the CODE processing Center of the International GPS Service for Geodynamics (IGS): Theory and Initial Results, Manuscripta Geodaetica, Vol. 19, pp. 367-386. Blewitt, G. (1994) Carrier phase Ambiguity Resolution for the Global Positioning System Applied to Geodetic Baselines up to 2000 km, Journal of Geophysical Research, Vol. 94, No. B8. Chen, D. and G. Lachapelle (1995) A Comparsion of the FASF and Least Squares Search Algorithms for Onthe-Fly Ambiguity Resolution, Navigation, Journal of the Institute of Navigation, 42(2), pp.371-390.