Instant Active Positioning with One LEO Satellite

NADAV LEVANON
Tel Aviv University, Tel Aviv, Israel
Received December 1997; Revised June 1999

ABSTRACT: Autonomous position determination in the Globalstar satellite communication system is discussed.
The two-way communication between a user terminal on the earths surface and a single low earth orbit (LEO) satellite makes it possible to derive range and range-rate and determine the user terminal position instantly. Expected accuracy is presented, and a simple direct solution is given.

INTRODUCTION Globalstar w 1x is a satellite communication system designed to provide voice and low-rate data communication to user terminals UTs. on the earth. The satellite constellation includes 48 low earth orbit LEO. satellites at a height of 1400 km, arranged in eight orbits with an inclination of 52 deg. The Globalstar satellite acts as a bent-pipe relay between a regional gateway GW. and the UT. The orbit of the satellite is known accurately. Primarily for operational reasons, before a UT phone call is connected, the system needs to know the UT position with a horizontal accuracy of better than 10 km. Connection of the call cannot be delayed for more than about 3 s, and a call should go through even if the UT sees only one Globalstar satellite. Hence there is a need for instant coarse positioning using the systems own satellite and signals w 2x . While two-satellite, two-dimensional 2-D., active positioning is found in other operational systems w 3, 4x , single-satellite instant positioning is unique. This paper therefore concentrates on this aspect of Globalstar positioning. Single-satellite positioning is based on measurements of range and range rate. The range between the satellite and the UT is derived from a round-trip delay measurement, from which the known GW-tosatellite leg is removed. Globalstar uses a code division multiple access CDMA. concept, which utilizes a wideband spread-spectrum signal, as in GPS. Hence the delay measurement resolution is relatively high. Range rate is derived from Doppler measurements. In a passive satellite Doppler naviNAVIGATION : Journal of The Institute of Navigation Vol. 46, No . 2, Summer 1999 Printed in the U.S.A.

gation system, such as TRANSIT w 5x , the true Doppler cannot be separated from the UT oscillator frequency offset; hence only Doppler differences rather than absolute Doppler. can be measured. When the UT is active receives and transmits. and uses the same master oscillator in its receiver and transmitter, two Doppler measurements one at the satellite in practice at the GW. and one at the UT and communicated to the GW. provide enough information to separate the true Doppler from the UT frequency offset see Appendix A.. This paper derives the expected accuracy of 2-D positioning based on one range and one range-rate measurement to a single satellite. In the third dimension, the UT is assumed to be on the earths surface. The horizontal error resulting from an elevation error is also derived. Finally, a simple direct solution is presented that can be used by itself or to provide a rst estimate for an iterative positioning algorithm.

POSITIONING BASED ON RANGE AND RANGE RATE The positioned UT is located at one of the two intersections between three surfaces see Figure 1.: the range sphere, the range-rate Doppler. cone, and the earths surface. The range sphere is centered at the satellite antenna, which is also where the apex of the cone is located. The cone axis of symmetry is the velocity vector. Delay range. and Doppler range-rate. errors cause an error in the determined UT position. The cross-track positioning error is magnied dramatically when the intersection is near the satellite subtrack on the earths surface. An error in the UT assumed height above 87

and the range rate by

t. s R

yv y y vt.

'x

q y y vt. q h y H .

2.

Fig. 1 Three Surfaces Dening the UT Position

the reference earth surface causes further crosstrack positioning error, which also increases as the UT gets closer to the subtrack. The performances of such a positioning system can be analytically derived for a simplied model that assumes a at earth and a straight-line orbit. The validity of this assumption is demonstrated by comparing analytic results using the at earth model with numerical results using a spherical earth model. The satellite-UT geometry in a at earth model is depicted in Figure 2. At t s 0, the satellite is at the 0, 0, H. coordinate and moving in the y direction with a velocity v. Both H and v are known accurately. The stationary UT is at x, y, h.. A at reference earth surface is assumed, represented by the z s 0 plane. The range to the UT is given by R t. s

Our simple analysis assumes that the range and range rate, derived from the delay and Doppler measurements, are free from bias errors, but suffer from relatively small, independent, random errors with zero mean and standard deviations of R and R , respectively. In reality, bias errors do exist. In Globalstars single-satellite positioning, random measurement errors dominate the position error. The bias errors become a dominant factor in twosatellite positioning. Yet it should be noted that even though the error contour maps in this paper were derived assuming random, zero-mean measurement errors, areas with large sensitivity to random measurement errors are also sensitive to bias measurement errors. The UT location x, y. is determined on the reference surface z s 0 plane., assuming erroneously that h s 0. Later it is shown that a deviation of the true elevation from the assumed elevation zero. results in a bias error in the estimated cross-track x. coordinate of the UT. LOCATION ERROR AS FUNCTION OF MEASUREMENT ERROR The random errors in range and range rate cause a UT position error. The position error is separated into cross-track and along-track components, with standard deviations of x and y , respectively. The transformation of errors from the measurement domain to the geometrical domain is done for t s 0, and requires the partial derivatives of the measurements with respect to the location parameters. At

'x

q y y vt. q h y H .

1.

Fig. 2 Satellite-UT Geometry

88

Navigation

Summer 1999

t s 0, the partial derivatives are given by R x R y s s x R y R yxyv R3 v R y2 R2 3. 4.

tion would result in nonzero off-diagonal elements.. Its inverse, also a diagonal matrix, is termed the weight matrix: 1 Ws
2 R

0 1 R
2

8.

R
x

0 5.

s s

R
y

1y

6.

The error variance of the estimated UT coordinates, x and y, can be obtained from H and W w 6x :
2 x s H T WH. 1 , 1 2 y s H T WH. 2 , 2 y1 y1

9. 10.

They are arranged as a matrix of partial derivatives: R Hs x R y

Symbolic solutions of equations 9. and 10. yield 7.

R
x

R
y

R2 y 2 x2 v R2 v
2 2 R

2 2 R

q 1y

y2 R2

2 R

11.

The errors in range and range rate are assumed to be independent. This assumption is well justied in Globalstar, where the two measurements delay and Doppler are nearly uncorrelated. The error covariance matrix is therefore diagonal. Correla-

2 y

y2 R2

2 R

12.

Figures 3 and 4 are contour maps of the crosstrack error, x , and the along-track error, y , re-

Fig. 3 Cross-Track Location Error (in kilometers)

Vol. 46, No. 2

Levanon: Instant Active Positioning with One LEO Satellite

89

Fig. 4 Along-Track Location Error (in kilometers)

spectively, in kilometers, for the following typical scenario: H s 1400 km, v s 7 kmrs, R s 30 m, and R s 3 mrs. The most prominent feature is the large cross-track error in positioning UTs located near the satellite subtrack cross-track s 0.. This feature implies large geometric dilution of precision GDOP. near the subtrack. The poor GDOP stems from the fact that on the subtrack, the circle drawn on the earths surface by the intersection with the range sphere Figure 1. is tangential to the hyperbola drawn on the earths surface by the intersection with the range-rate cone. Any small error in range or range rate will cause a large cross-track error in the calculated location. As the UT location gets farther away from the subtrack, the GDOP singularity is replaced by another problem ambiguity. Single-satellite positioning based on one pair of range and range-rate measurements always yields two symmetrical solutions on the two sides of the subtrack Figure 1.. When the true UT location is far from the subtrack, the two solutions are far from each other. In that case, the true solution can be identied using information from several satellite antenna beams. When the true UT location is close to the subtrack, the 90

two solutions are close to each other and may both fall within the illumination area of the same antenna beam. In that case, the ambiguity cannot be resolved. Figures 3 and 4 are plots of the analytic results of equations 11. and 12., respectively, which were derived using a at earth model. To demonstrate their validity for the real earth, a numerical error analysis was performed using a spherical earth model and a circular orbit. The result for the crosstrack error is presented in Figure 5. The excellent agreement between Figures 3 and 5 and similar agreement, not plotted, with respect to the alongtrack error. proves that the use of a at earth model is justied, and that equations 11. and 12. apply to more realistic earth models. ELEVATION-INDUCED ERROR Equations 1. and 2. indicate that in both the range and the range-rate measurement, the UTs elevation, h, and cross-track coordinate, x, are coupled together in the expression x 2 q h y H. 2 . Hence an error, h, in the assumed elevation must create an error, x, in the cross-track coordinate. Summer 1999

Navigation

Fig. 5 Cross-Track Location Error (in kilometers) Derived Numerically Using Spherical Earth Model

To keep the expression unchanged, the following relationship must hold x s Hyh x h f H x h 13.

graphic map to be a measurement with error standard deviation E . The partial derivative and weight matrix are increased in size to 3 = 3: R x Hs R y R h

According to equation 13., when the cross-track distance is approximately equal to the satellite height, the elevation-induced cross-track error is equal to the elevation error. Clearly the error increases as the UT gets closer to the subtrack. In the actual Globalstar positioning algorithm, the horizontal position is rst solved using sea-level elevation. Then the terrain elevation for the solved position is obtained from a topographic map and used instead of sea level, and an improved position is obtained. Unless the terrain is very rough, convergence is expected after two or three iterations. The elevation-induced horizontal positioning error can be incorporated in the error contour maps by declaring the elevation obtained from the topo-

R
x h x 1
2 R

R
y h y 0 1
2 R

R
h h h 0 0 1
2 E

14.

Ws

0 0

15.

Vol. 46, No. 2

Levanon: Instant Active Positioning with One LEO Satellite

91

where the additional elements are R h s s hyH R y h y H . yv R3 h y s 0, h h s1 16.

R
h h x

17. 18.

s 0,

hence Figure 4 holds for the 3-D case as well. The cross-track error equation 19.. differs from equation 11. in the additional term predicted by equation 13.. The slightly degraded cross-track error contour map assuming an elevation measurement error standard deviation of E s 200 m is presented in Figure 6. There is no need to plot the standard deviation of the error in the estimated height since it must everywhere be equal to E , as indicated by equation 21.. OTHER ERRORS As the various error contour maps indicate, with the expected random measurement error, typical positioning errors are between 0.5 and 10 km excluding the area near the satellite subtrack.. This performance level can be considered coarse positioning, but it meets the systems needs. Other error sources, such as ionospheric effects and ephemeris errors, that are considered in more accurate positioning systems can be neglected in our error analysis. The main remaining source of error is UT velocity. UT velocity generates additional Doppler, which effectively increases the range-rate error. Because we are using a single Doppler measurement, the fact that it is a bias rather than a random error makes no difference, and the effect of

The cross-track, along-track, and height error variances are derived from the three diagonal elements of the matrix H T WH.y1 , yielding R2 y 2 x2 v y2 R2
2

2 x

2 2 R

q 1y

2 R q

h y H. 2 x2

2 E

19.
2 y s

R2 v

2 q 2 R

y2 R2

2 R

20. 21.

2 2 h s E

Note that the along-track error expression equation 20.. is identical to the 2-D case equation 12..;

Fig. 6 Cross-Track Location Error with Contribution from Elevation Error (in kilometers)

92

Navigation

Summer 1999

UT velocity can be incorporated into the error analysis by increasing the range-rate random error. POSITIONING ALGORITHM The operational positioning algorithm applies to all possible scenarios, including the two- and three-satellite cases. The latter are overdetermined systems, which invites a least-squares solution. The general algorithm thus uses an iterative weighted least-squares solution. Such an algorithm requires an initial estimate of the UT location. A simple approach to obtaining a good rst estimate is to solve explicitly the minimal set of two equations measurements.: the range and range rate to a single satellite. One explicit solution is described in Appendix B. In a single-satellite situation, the explicit solution can replace the iterative solution. CONCLUSIONS Autonomous positioning in Globalstar may necessitate determining the UT position when only one satellite is available. The random error analysis presented in this paper demonstrates that with the expected measurement accuracy, coarse position can be determined instantly for UTs everywhere except near the satellite subtrack. ACKNOWLEDGMENT The author gratefully acknowledges the support of Qualcomm Inc., San Diego, California.

APPENDIX A SEPARATING DOPPLER FROM UT OSCILLATOR OFFSET To determine the satellite-UT range rate, it is necessary to separate the UT oscillator offset from the Doppler shift in the satellite-UT leg.. In an unpublished report, Steven A. Kremm suggested a method that applies when the UT uses the same local oscillator for both transmit and receive. The method is explained with the help of Figure A.1 and the index below. It is assumed that the Doppler shifts in the GW-satellite leg are perfectly known and removed.

s range rate of the satellite-to-UT leg R C s propagation velocity speed of light. f F s forward link nominal carrier frequency 2500 MHz. f R s reverse link nominal carrier frequency 1600 MHz.
f off f0 s normalized frequency offset of UTs oscillator

Two measured frequencies are available at the GW: the reported measurement from the UT

f me as, UT s f F y

R
C

f off f0

A-1.

Fig. A.1 Relationship Between Frequencies Used to Separate Doppler from Frequency Offset

Vol. 46, No. 2

Levanon: Instant Active Positioning with One LEO Satellite

93

and the measurement performed at the GW itself

are known: the position coordinates s s w x s ys z s x

f me as, GW s f R y

R
C

f off f0

B-3.

/
/ /

A-2. and the velocities v s w vx vy vz x

Adding and subtracting equations A-1. and A-2. yields both the UT offset and the range rate: f off f0 1 2 f meas, GW fR f meas, UT fF q

B-4.

The radius of the satellite orbit is s

A-3.

2 2 rs2 s sT s s x 2 s q ys q z s

B-5.

sy R

C 2

f me as, GW fR

f meas, UT fF

The two available measurements are range A-4. R s <s y u < s

'x

y x . q ys y y . q z s y z .

B-6. APPENDIX B DIRECT SOLUTION The direct solution assumes a smooth ellipsoid earth model and uses earth-centered, earth-xed Cartesian coordinates. There are many references to direct solutions for geolocation using time-difference or frequency-difference measurement w 7x . However, the direct solution becomes very simple when absolute measurements are available. and range rate s y u. T v r 1 r x s y x . vx q ys y y . vy q z s y z . vz B-7.

s R
s

Outline Denitions The UT Cartesian coordinates earth-centered, earth-xed. are u s w xyzx

Dene xs As v x bs 1 2 ys vy B-8.

B-1.

? rs2 q rc2 y R2 @

B-9.

Initially, the UT is assumed to be on a sphere of radius rc whose value is arbitrarily selected between the equatorial earth radius r E and the polar earth radius r P .. That assumption yields the rst equation: rc2 su usx qy qz
T 2 2 2

s x svx q ysvy q z svz y RR B-10. c s sT v y RR

B-2.

Six satellite parameters at the measurement epoch

The z coordinates of the UTs true and mirror solutions are obtained from

z1, 2 s 94

q "

' q .

y 1 q 2 q 2 . 2 q 2 y rc2 .

1 q 2 q 2 Navigation

B-12.

Summer 1999

and the remaining coordinates of the two solutions are given by x 1 , 2 s y z1, 2 y1 , 2 s y z1, 2 B-13. B-14.

the center of the earth:

rc2

s sin2 q

/
rE

rp

B-20. cos 2

Resolving which of the two is the true solution requires additional information. That information can be obtained, for example, from the satellite antenna beam through which the UT signal was received. Explanation The range and range-rate equations B-6. and B-7. can be rewritten as w s vx T u s w b cx T B-15.

where b and c are as dened in equations B-9. and B-10.. From equation B-15., x and y can be expressed as functions of z: x s y z y s y z B-16. B-17.

This local earth radius for the approximate UT location. is used in equation B-9., and the direct solution is repeated to yield more accurate Cartesian coordinates of the UT location. When the direct solution is used to obtain a rst estimate for the iterative solution, the accuracy achieved thus far is sufcient. If the direct solution is used by itself, an elevation correction is required. To obtain the elevation, the Cartesian coordinates must be converted into longitude and geodetic latitude w 8x , the surface elevation for that location is obtained from a digital terrain map, the distance to the earth center is adjusted accordingly, and the direct solution is repeated.

REFERENCES
1. Hirsheld, E., The Globalstar System, Applied Microwave & Wireless, Summer 1995, pp. 26 41. 2. Levanon, N., Quick Position Determination Using 1 or 2 LEO Satellites, IEEE Transactions on Aerospace and Electronic Systems, Vol. AES-34, No. 3, July 1998, pp. 736 754. 3. Ames, W. G., A Description of Qualcomm Automatic Satellite Position Reporting QASPR . for Mobile Communications, Proceedings of the 2nd International Mobile Satellite Conference, Ottawa, Ontario, Canada, June 1990, pp. 285 290. 4. Colcy, J. N., G. Hall, and R. Steinhauser, Euteltracs: The European Mobile Satellite Service. Electronics & Communication Engineering Journal, April 1995, pp. 81 88. 5. Parkinson, B. W., T. Stansell, R. Beard, and K. Gromov, A History of Satellite Navigation, NAVIGATION, Journal of The Institute of Navigation, Vol. 42, No. 1, 1995, pp. 109 164. 6. Sorenson, H. W., Parameter Estimation, Principles and Problems, New York: Marcel Dekker Inc., 1980, Ch. 2. 7. Ho, K. C. and Y. Y. Chan, Geolocation of a Known Altitude Object from TDOA and FDOA Measurements, IEEE Transactions on Aerospace and Electronic Systems, Vol. AES-33, No. 3, July 1997, pp. 770 783. 8. Olson, D. K., Converting Earth-Centered, Earth-Fixed Coordinates to Geodetic Coordinates, IEEE Transactions on Aerospace and Electronic Systems, Vol. 32, No. 1, Jan. 1996, pp. 473 476.

where , , , and are as dened in equation B-11.. Inserting equations B-16. and B-17. in equation B-2. yields a quadratic equation of z y z . 2 q y z . 2 q z 2 s rc2 B-18.

whose two solutions are given by equation B-12. above. Using the two solutions for z back in equations B-16. and B-17. yields the corresponding two solutions for x and y, as given in equations B-13. and B-14. above. Ellipsoid Earth Model The rst improvement to the above simple solution is to assume that the UT is on an oblate ellipsoid that has a polar radius r P and an equatorial radius r E . From the Cartesian coordinates derived assuming a spherical earth, we nd sine and cosine terms of the approximate geocentric latitude : sin s z rc ; cos s

'x

q y2 rc

B-19.

Using the ellipsoid model and the approximate geocentric latitude yields a more accurate distance to

Vol. 46, No. 2

Levanon: Instant Active Positioning with One LEO Satellite

95