The Halifax GPS Precision Approach Trial: A Report on the In-Depth Data Analysis
Attila Komjathy and Richard B. Langley
Both at: Geodetic Research Laboratory, Department of Geodesy and Geomatics Engineering
University of New Brunswick, Fredericton, N.B. E3B 5A3 Canada
Transport Canada Aviation, in conjunction with Cougar Helicopters Inc. (specializing in oil rig, search and rescue, and emergency medical services), carried out an in-service trial of helicopter precision approaches using local differential GPS (LDGPS) at Halifax International Airport between the 2nd of February and the 3rd of March 1994. The trial was co-sponsored by the National Energy Board and the U.S. Federal Aviation Administration. The immediate objective of this trial was to demonstrate helicopter precision approaches in an operational setting. The long term objective was to obtain operational approvals of GPS for precision approaches. The trial used a Trimble LDGPS ground station to provide differential corrections to a Trimble avionics receiver in the helicopter (Sikorsky SN76-SCH) via a VHF radio data link. Simultaneously, both at the base station and the remote station, Ashtech LM-XII single-frequency geodetic receivers collected data, using the same antennas supplying the Trimble equipment, to provide "ground truth". A modification to the software of the Trimble airborne receiver was performed approximately mid-way through the trial to increase the position update rate.
We investigated the validity of using the position solutions from the post-processed Ashtech carrier-phase data as "truth" after which we used the Ashtech solutions as a benchmark against which we assessed the accuracy of the Trimble LDGPS results.
We used three different processing techniques to validate the C/A-code/carrier-phase solution that we subsequently used as the "ground truth" solution for the Ashtech-Trimble LDGPS comparison. The estimated positions and position errors provided by the C/A-code/carrier-phase solution agreed (2 sigma) with the accuracy levels that we would expect from the other two solutions. However, the estimated position accuracies provided by the Ashtech "ground truth" were lower than anticipated (at the 3 meter level) mainly because carrier phase ambiguities could not be resolved presumably due to observations being biased by multipath. Also, since single-frequency GPS receivers were used during the trial with short observation periods (10 minutes on average), the data sets might not have been long enough for the Kalman filter in the data processing software to converge on a solution (overcoming the "settling-down" period) which would have resulted in higher estimated position accuracy.
After analyzing all 73 approaches (four approaches were eliminated due to extended periods of VHF link loss and inadequate quality of "ground truth" data), it was found that the LDGPS solutions met the horizontal accuracy requirements for Special CAT 1 approaches in 97 percent of the cases, whereas the vertical accuracy requirements were only met in 19 percent of the cases. The primary reason why the vertical accuracy requirements were not met is that there are jumps in the Trimble solutions mainly due to fact that differential corrections for one or more satellites were unavailable because of temporary loss of the data link or because the base receiver and the airborne receiver were tracking different sets of satellites. Analysis of cross-track and vertical error plots revealed that these jumps affect the calculated positions both in cross-track and vertical components. As well, one has to realize that the vertical accuracy requirements are more stringent than the horizontal ones which also contributes to the fact that the vertical requirements were only met in 19 percent of the cases.
We computed the mean and the standard deviation of the cross-track errors and the vertical errors for each individual runway approach. We generated histograms of these results with a bin size of 0.5 metres. This analysis shows that in the case of the means of the cross-track errors, the largest number (17, 23%) of approaches falls into the bin size of 0 to 0.5 metres whereas 42 approaches (58%) can be categorized with a bin size ranging from -0.5 metres to 1 meter. The histogram of standard deviations of the cross-track errors shows that the largest number (22, 30%) of the approaches fall into a bin size of 0.5 to 1 meter. In the case of the means of the vertical errors, the analysis shows that the largest number (11, 15%) of approaches falls into the bin size of 1 to 1.5 metres whereas 29 approaches (40%) can be categorized with a bin size ranging from 0 to 1.5 metres. Also, the histogram of standard deviations of the vertical errors shows that the largest number (15, 21%) of the approaches fall into a bin size of 1 to 1.5 metres.
After looking at the entire data set, we broke the data set down into categories that were looked at individually to develop further statistics. A primary objective of this part of our investigation was to find out what is the worst error (worst case scenario) that we should expect for cross-track and vertical errors. Based on the means and the standard deviations of the cross-track errors and the vertical errors, we computed the range of mean variation, the range of standard deviation variation, and the range of absolute value of the mean + 2 sigma variation for each individual category we created (3 degree, 6 degree glidepath approaches; 100 foot, 200 foot decision height approaches; 50 knot, 70 knot approaches; missed approaches). In the case of cross-track errors, the range of the absolute value of the mean + 2 sigma error varied between 1.5 metres (minimum) and 18.0 metres (maximum). In the case of vertical errors, the range of the absolute value of the mean + 2 sigma error varied between 2.7 metres (minimum) and 89.8 metres (maximum).
We also looked at the approaches falling into the same categories at 0.1 NM increments away from the threshold beginning at 4 NM from the threshold. Taking these readings at every 0.1 NM, we computed the means and the standard deviations of the cross-track and vertical errors. Based on the average mean, the average standard deviation, and the maximum standard deviation of the cross-track and vertical errors of corresponding approaches (3 degree vs. 6 degree; 200 foot decision height vs. 100 foot decision height; 50 knot vs. 70 knot approaches) falling into the same category, we arrived at the following conclusions:
In the case of cross-track errors:
In the case of vertical errors:
Furthermore, there appeared to be a statistically significant improvement after the receiver software modification both in the horizontal and vertical sense when comparing approaches prior to and after the modifications.
In addition, the trial provided a great deal of operational information which will be useful in designing procedures to allow helicopters to take full advantage of GPS. This was not the first time that LDGPS was used for helicopter precision approaches but it was the first such trial run in Canada.
Based on our analysis of the data from the trial, we recommend the use of a dual-frequency GPS receiver such as the Ashtech Z-12 as a "ground truth" system in any future trial. Such a receiver should be able to provide a more reliable comparison between the LDGPS and the "ground truth" solutions. The estimated position accuracy that an Ashtech Z-12 receiver, for example, would provide under similar trials is at the 5 cm level. Furthermore, when developing future helicopter precision approaches it is recommended to include a statistically significant number of approaches in each individual category.
Return to Attila's Home Page
This page was compiled by Attila Komjathy and last updated
15 September 1997. If you have any questions, comments please
feel free to get in touch with me.
You can reach me at: firstname.lastname@example.org