BACKGROUND ART It is generally recognized that it will be many years before satellite navigation augmentation systems, such as the wide-area augmentation system (WAAS) and equivalent systems, will provide such a capability. As stated in Aviation Week, 8/18/97, pp 23,26, "nonprecision approaches are associated with 60% of the commercial airline crashes worldwide that are attributed to controlled flight into terrain (CFIT)"and"CFIT is the leading cause of fatal accidents in commercial aviation.""tI t will be at least 10-15 years before fully autonomous, airborne navigation systems with 3-D capabilities are available to replace them".

The technology required for three-dimensional autonomous integrity monitored extrapolation (3D AIME) already exists today, and has been proven. Unlike WAAS, 3D AIME will be available on a worldwide basis autonomously on the aircraft, without using any ground based navigation aids.

DISCLOSURE OF INVENTION By using the 3D AIME apparatus, non-precision approach procedures based on 2D horizontal position with constant barometric altitude steps can be replaced by 3D constant-rate- of-descent procedures. This is achieved by using the assured-integrity monitored-extrapolation (AIME) navigation apparatus described in United States Patent 5,583,774 dated December 10, 1996. The AIME navigation apparatus is also referred to in the literature as the"autonomous integrity monitored extrapolation"mechanization.

The AIME mechanization provides accurate horizontal position, with high integrity, which is necessary for accessing stored altitude terrain profile tables along the approach path.

This stored terrain profile information could be obtained from contour maps, as used to implement advanced ground proximity warning systems, or from airport surveys. These tables are used to correct radar altimeter measurements for local terrain altitude above sea level to obtain measurements hM (t) of altitude above mean sea level.

This is normally done with the help of a ground based differential GPS (DGPS) station ("Design and Flight Test of a Differential GPS/Inertial. Navigation System for Approach/Landing Guidance", Journal of the ION, Vol. 38, No. 2,1991). The differential station is normally necessary since accurate horizontal position is required in order to obtain the correct local altitude from the tables. As an example, in mountainous terrain a 200 meter error in horizontal position can easily result in a 50 meter error in altitude.

In the 3D AIME mechanization, the requirement for a differential ground station is avoided. Rather than using the measurements hm (t) directly, they are used to estimate corrections to the baro-inertial output from the IRS. This corrected output is then used for the 3D AIME vertical position output. In this way, the measurements hm (t) over the entire approach, which can extend for 5 to 10 miles, are averaged. A 200 meter error in position will

result in an altitude error of less than 20 meters, assuming that the average terrain slope in one direction during the approach is less than 10 percent. Approaches where the average slope in one direction over the entire approach exceeds 10 percent would be rare and would be known in advance.

BRIEF DESCRIPTION OF DRAWINGS FIG. 1 is an illustration of the way that radar altimeter measurements and stored terrain profiles are used to obtain measurements of altitude above mean sea level.

FIG. 2 is a diagram showing how a GPS receiver, an IRS, a radar altimeter, and stored terrain altitude profiles are processed to obtain accurate 3-dimensional position for 3D approach procedures.

FIG. 3 shows the error states, the dynamics matrix, and the observation matrices used in the 3D AIME approach Kalman filter.

BEST MODE FOR CARRYING OUT THE INVENTION The three-dimensional autonomous integrity monitored extrapolation (3D AIME) apparatus is similar in many respects to the apparatus described in United States Patent 5,583,774 dated December 10,1996 which is hereby incorporated by reference.

In the 3D AIME mechanization 1 shown in Fig. 2, the measurements hM (t) of altitude above mean sea level (see Fig. 1) are first referenced to AIME computed altitude ! (t) by adder 3 and then pre-filtered by pre-filter 5. The computed altitude hC (t) is obtained from computing unit 11 through switch 4. The pre-filtered measurements pass through switch 6 to AIME Kalman filter 7 where they are used as observations in obtaining estimates of the barometric offset at the runway and the barometric scale factor offset from the runway to the aircraft present altitude. The smoothed barometric-inertial altitude from the inertial reference system (IRS) 9 is corrected by these offsets in computing unit 11 to obtain the 3D AIME vertical position output hC (tJ.

The pilot first obtains the barometric offset at the airport by radio, as provided by the weather service. The first pre-filtered observation of present altitude during the approach is differenced with this offset in Kalman filter 7 to obtain the first estimate of the barometric scale factor offset. As each new pre-filtered observation becomes available, it is used to improve the estimates of both the barometric scale factor offset and the barometric offset at the airport runway. The estimated offset at the runway will improve in accuracy as the airport runway altitude is approached, and the scale factor error has less effect. As the runway is approached, if the estimated runway offset does not agree with the offset of the runway as obtained from the weather service by radio, within a tolerance, a missed approach is initiated.

Fig. 1 is a typical example of a 3D AIME approach where the approach is initiated at a distance of 60,000 feet (approximately 10 nautical miles) from the airport at an altitude of 3000 feet above the airport. This gives a slope of 1: 20, which corresponds to an approximate glideslope of 3 degrees. In Fig. 1, the vertical scale is exaggerated for illustrative purposes.

If the aircraft velocity were perfectly constant, the actual path would appear as a straight line. Assuming altitude measurements are taken at a typical rate of 10 Hz by radar altimeter 13, the measurements would be equally spaced as shown in the figure. The path of the aircraft could then be obtained by a two parameter least squares fit of a straight line, to determine the vertical intercept altitude at touchdown and the slope. This would have the effect of reducing the errors due to noise, both in the radar altimeter measurements, and in the stored terrain altitudes contained in memory 15.

Since the aircraft velocity is not perfectly constant, the noise effects cannot be reduced by a simple straight line least-squares fit, as just described. However, an almost equivalent filtering effect can be obtained by using the smoothed barometric altitude from IRS 9 as a reference to obtain altitude observations for Kalman filter 7.

The smoothed barometric altitude hB (t) from IRS 9 is obtained from a third order complementary filter, as specified in ARINC 704 for IRS outputs. This filtering provides an output whose steady state accuracy is that of the barometric altitude. Yet, the output follows almost instantaneous changes in altitude because of the high frequency accuracy of the inertial reference.

Since the output is smoothed barometric altitude, rather than altitude above mean sea level, a correction for the barometric offset difference is necessary. The barometric offset at the airport is provided by periodic weather reports. However, the offset also changes with altitude

above the airport. It can be assumed that this change in offset varies approximately linearly with altitude above the airport to an altitude difference of 3000 feet.

The 3D AIME mechanization 1 uses the measurements in Fig. 1 to estimate the barometric offset at the airport hBO and the change in offset at 3000 feet above the airport, dhB3000. As the actual altitude decreases to the altitude of the airport, the estimated offset at the airport should agree approximately with the offset provided by the airport weather reports. If this disagreement is not within a certain tolerance, depending on altitude above the airport, a missed approach is initiated at the missed approach point (MAP) as indicated in Fig. 1.

It is assumed that AIME has been available for the navigation phase prior to this time, with a typical horizontal protection level of 0.3 nm or better. The AIME Kalman filter used in the navigation phase will be referred to as the Navigation Kalman Filter (NKF). It is used to initialize an Approach Kalman Filter (AKF) for the approach phase which is denoted in Fig. 2 as Kalman filter 7.

The AKF will use an update interval of 10 seconds, rather than the 150 second update interval of the NKF. The AKF will have fewer states than the NKF. These are indicated in Fig.

3. The AKF states consist of two horizontal position states dX and dY, two horizontal velocity states dVx and dVy, the user clock bias dB, the user clock bias rate dBr, the barometric offset dh_Ro (t), the barometric offset scale factor dhB3000 (t), and one state for each of the satellite range bias errors dRBi (t), up to a maximum of eight, as used by the NKF.

The total number of error states in the AKF is therefore 16 instead of 24, as used in the NKF. The missing 8 states are the three navigation axis misalignments, the two horizontal accelerometer bias errors, and the three gyro bias errors. These are unnecessary, since the errors due to these states will have been balanced before the approach by the Navigation Kalman filter, assuming at least 30 minutes of operation before the approach. These states cannot become unbalanced due to process noise during the approach, because the duration of the approach is less than 5 minutes.

The AKF uses updates from the same satellites as the NKF using GPS receiver 17, but in addition it uses updates from radar altimeter 13, corrected by stored terrain altitude from memory 15. As shown in Fig. 2, the corrected horizontal position output during the approach is obtained by computing unit 11 from the un-updated inertial horizontal position supplied by IRS 9, corrected by the estimated horizontal position errors supplied by error model 19.

The error model 19 obtains error estimates utilizing the dynamics matrix shown in Fig.

3 and error state correction vector x (k) supplied at AT time intervals by Kalman filter 7, the index being an integer and iT being equal to 10 seconds. The output of the error model 19 is the error states vector x (t) supplied at At time intervals, the index t being an integer and At being equal to 0.1 second. The error states vector x (t) is obtained by integrating the differential equations described by the dynamics matrix F in Fig. 3. In matrix form these equations are <BR> <BR> dx_Fg<BR> ar where the vectors dxldt, and x are represented by column matrices with 16 elements, as indicated at the left side of Fig. 3. The dynamics matrix F is a square matrix with 16 rows and 16 columns, as indicated at the center and right side of Fig. 3. The error states x are initialized every 10 seconds by Kalman filter 7, and then integrated at a rate of 10 Hz. from the differential equations described by the dynamics matrix F.

Since the F matrix has mostly zero elements, the differential equations are very simple, and the non-zero error model coefficients are the constants 1 and-1/ With this simplified dynamics matrix, these coefficients can be obtained without inputs from the IRS 9. However, in a more general case, the coefficients would be obtained from IRS 9 as indicated in Fig. 2.

As with the NKF, these corrected positions are used along with the estimated user clock bias by computing unit 11 to determine a computed pseudorange (PR) at intervals of At. The computed pseudoranges are differenced with the measured pseudoranges from GPS receiver 17 by adder 21 to obtain the satellite measurements z at intervals of At, as shown in Fig. 2.

These differences are averaged or pre-filtered by lowpass pre-filter 23 to obtain averaged measurements at AT time intervals. As with the AIME NKF, the observation matrices involve the direction cosines exi, eyi, and ezi of the line of sight, and minus ones for the user clock bias, and the particular satellite range bias error, as indicated in Fig. 3.

The altitude measurement zh is obtained in the following way. The computed altitude above mean sea level hc (t) is obtained by computing unit 11 by applying corrections to the un- updated third order barometric-inertial loop output supplied by IRS 9. The corrections consist of the estimated barometric offset dhBO (t) and scale factor error ((h-ho)/3000) dhB3000 (t). The measured altitudes above mean sea level hvt) and the computed altitudes above mean sea level

hc (t) are differenced by adder 3 to obtain altitude measurements zh (t) at At time intervals. These differences are averaged, or pre-filtered, by lowpass pre-filter 5 to obtain averaged . measurements at AT time intervals. The observation matrix in this case is simply 1 for the barometric offset and altitude (h-ho)/3000 for the barometric scale factor, as indicated in Fig.

3.

The NKF continues to run in parallel with the AKF, without updates from the radar altimeter. This is done in case the radar altimeter measurements are determined to be unreliable, and a missed approach is initiated. The integrity of the satellite measurements is monitored in the same way as in the NKF, by using the magnitude of the long term average of the Kalman filter residuals over many update cycles as the test statistic to detect failures.

An alternative mechanization which avoids the use of a Kalman filter is also shown in Figure 2. The uncorrected inertial reference system altitude hB (t) passes through switch 4 to adder 3 which obtains the difference zh (t) of hm (t) and hB (t). Pre-filter 5 approximates zh (t) as a low order polynomial function of time. The coefficients of this polynomial are based on a least-squares approximation to the differences zh (t) between the measured altitude hM (t), and the uncorrected inertial altitude hB (t) : ) =)-) (2) where the measured altitude hM (t) is obtained as the sum of the terrain altitude hT (t) and the radar-measured altitude above the terrain hr (t) as shown in the summation junction at the lower left part of Fig. 2: hM(t) = hR (t) + hT (t) The least squares estimate is determined by saving all of these past measurements up to the present time t, and completely re-solving for the least-squares estimate of the polynomial coefficients at a rate of 1 Hz. These polynomial coefficients in t are passed through switch 6 to computing unit 11 where they are used to correct the inertial altitude hB (t) to obtain the best estimate hc (t) of the true altitude at 10 Hz.

When Selective Availability (SA) is present, it is anticipated that the accuracy of the vertical position information generated will be 5 meters, 95%, assuming that only the C/A code is used in civilian applications. This accurate vertical measurement will result in improved horizontal accuracy of AIME. It is anticipated that the horizontal accuracy will be improved from 100 meters, 95%, to better than 75 meters, 95%. This corresponds to a horizontal integrity

protection level (four sigma) of less than 500 feet, which permits parallel approaches with runway spacing of 1000 feet.

If used with SA turned off, which was promised by the year 2006,3D AIME meets the accuracy requirements for CAT I. If used with the Wide Area Augmentation System (WAAS), or with the military P-Code, it has a vertical accuracy of 1.5 meters, 95%, which is the CAT II requirement.

The method of using tables of local terrain altitude above sea level is most easily visualized for the case of a straight approach lined up with the center line of the runway. For a given airport runway, a finite number of curved approaches will be handled in a similar way.

For a straight or curved approach, if the aircraft flew directly above a line exactly along the nominal approach path, it would only be necessary to store the local terrain altitude at points along this line.

Because of horizontal steering errors and navigation sensor errors, the aircraft will deviate laterally from this line. It is therefore necessary to store local terrain altitude at points adjacent to this line. These grid points are located at the vertices of square grids along an area of finite width along the nominal approach path.

For aircraft whose horizontal position is at points within a particular square in the region covered by the grids, the particular square will be referred to as the"reference square"for that measurement. Two dimensional linear interpolation is used within the reference square, using stored terrain altitude at the four corners of the square. The terrain altitude stored at each grid point will account for the highest point within the neighboring reference squares.

Assuming the data from the radar altimeter hR (t) in Figure 2 comes in at a rate of 10 HZ, and the aircraft speed during the approach is 150 knots, the measurements will be spaced at a horizontal distance of approximately 25 feet.

The radar altimeter measures the range to the nearest reflecting object within its beamwidth, which is as wide as 45 to 90 degrees. If the stored terrain tables are based on the terrain height without accounting for certain buildings or towers, the measured height above the local terrain will be in error for one or more seconds when near this terrain feature. This is true even though the aircraft is not directly over the feature, since the altimeter has a finite beamwidth.

The 3D AIME mechanization will delete these measurement by using a three or four sigma rejection criterion on the individual 10 Hz. measurements. In addition, it will use a three

or four sigma rejection criterion on the Kalman filter averaged measurement over the Kalman filter update interval, which may be 10 seconds in the preferred embodiment. If these terrain features are known in advance, an additional adjustment will be made to the terrain altitude stored at the grid points, so that this data rejection is unnecessary.

It is tentatively assumed that these grids cover an area 1000 feet wide, 500 feet on either side of the nominal approach path. The area covered by the grids extends up to 60,000 feet along the nominal approach. This 60,000 foot path extends from an initial point, when the aircraft initial altitude above the runway is approximately 3000 feet, to the projected touchdown point 1000 feet down the runway from the threshold. This assumes the slope of the approach path is 1: 20.

Since many radar altimeters have a maximum altitude range of 2500 feet, the grids may not extend all the way to 60,000 feet, as shown in Figure 1, depending on the local terrain altitude above sea level at this location. Only the measurements within the 2500 foot vertical range of the altimeter will be used. Similarly, whenever the aircraft is more than 500 feet laterally from the nominal approach path center, the measurements will not be used.

Since horizontal position coordinates are used as inputs, horizontal position errors will cause errors in the local terrain altitude obtained from the tables. However, the errors in horizontal position for 3D AIME are only 75 meters, 95%, rather than 100 meters, 95%, for GPS with SA. This improvement is a result of the effect of the accurate altitude measurement on the Kalman filter estimate of horizontal position.

To limit horizontal errors when using 3D AIME, an additional test is made. The residuals zh (t) in Figure 2 are saved, together with their horizontal position coordinates.

Adjustments to past residuals are made, based on horizontal position adjustments from the Kalman filter.

The root mean square, zhrms, of the adjusted altitude residuals Zh (t) in Figure 2 are compared with an alarm limit. If ZhRMS exceeds the alarm limit, this indicates an excessive horizontal position error or some other integrity problem with the data.

Additional tests would be made by shifting the stored position coordinates forward toward the runway, backward, left, or right, respectively, and re-computing the RMS residuals in each case. If one of these cases, or two cases in adjacent quadrants, provided a smaller RMS residual, this would indicate a horizontal position error.

This is similar to conventional techniques used in missile guidance terrain matching

systems to obtain horizontal position fixes. However, in this case, the terrain can be smooth, so that the integrity check is not always achievable. This does not cause a problem unless the terrain has a constant slope in one direction during the entire approach. For example, over water, the surface is smooth, but horizontal position errors do not cause errors in altitude.

In addition to testing the RMS residual at 10 Hz., the RMS of residuals averaged over several cycles would be tested. In particular, the RMS of the residuals of the Kalman filter, with averaging time of 10 seconds, would be tested.

When sources of terrain altitude near the airport are available, initial altitude entries near the airport will be generated by off-line software processing of this information. The sources of such information might be published contour maps or published altitude tables.

However, because of the beamwidth of the radar altimeter, it is difficult to predict the effects of buildings and rugged terrain features by off-line processing alone. Therefore, it is also necessary to validate the process by taking radar altimeter measurements while flying the actual approach. This would be done by using an accurate external reference, such as DGPS, to determine the true position and altitude at the time of each radar altimeter measurement.

Such flights are necessary in order to prove that the entries generated by the off-line software processing can be applied to the measurements when flying the actual approach. Such flights would also be required for FAA or ICAO certification at the airport.