by Ian Strachan
Chairman, GNSS Flight Recorder Approval Committee
of the FAI International Gliding Commission
Fédération Aéronautique Internationale
Dated 4 May 2000
Since midnight 1/2 May 2000, average GPS accuracy in lat/long using
12 channel receivers under the conditions given below, has improved from
36.2m to 13.2m, an improvement of 2.74 times over what has been recorded
since testing started in May 1996.
The deliberate Selective Availability (SA) error was removed from the
US GPS system from about 12:04AM 2 May US Eastern Daylight
Time. Therefore, additional accuracy tests have been carried out
using two types of GNSS Flight Recorders. These new tests following
a series of GPS accuracy tests that have been carried out since May 1995.
Each recorder tested has a different type of 12 channel GPS receiver. The tests were from a moving automobile which was driven over a number of accurately surveyed points at about 51N 001W. The GPS antennas were mounted on the top of the car, and satellites recorded varied from 6 to 12, averaging 8 or 9 at test points. The main test location is a crossroads in a shallow bowl with gently sloping terrain on three sides. This is the UK BGA Waypoint code LA4, at 51 10.179N 001 02.644W to the WGS84 Geodetic Datum.
Results up to end 4 May 2000:
Average overall lat/long accuracy: 13.17m from 202 samples, maximum
Average for recorder 1: 12.64m from 98 samples, maximum 32m
Average for recorder 2: 13.55m from 140 samples, maximum 28m
A tabular breakdown follows:
100% probability of being within 32m, average 13.2m
99% prob of being within 28.0m, average within 99% sample 13.0m
95% prob of being within 21m, average within 95% sample 12.5m
90% prob of being within 20m, average within 90% sample 12.1m
80% prob of being within 18m, average within 80% sample 11.3m
75% prob of being within 17m, average within 75% sample 10.9m
70% prob of being within 17m, average within 70% sample 10.45m
66.7% prob of being within 16m, average within 66.7% sample 10.1m
60% prob of being within 16m, average within 60% sample 9.5m
50% prob of being within 15m, average within 50% sample 8.3m
And for Sigmas:
There is a 1 Sigma (68.27%) probability of being within 17m,
with an average within the 68.27% sample of 10.3m,
There is a 2 Sigma (95.45%) probability of being within 22m,
with an average within the 95.45% sample of 12.6m.
There is a 3 Sigma (99.73%) probability of being within 29m,
with an average within the 99.73% sample of 13.1m.
The main error (as it always has been) will be in accurately recording
and plotting the ground coordinates to which the GPS lat/long refers. Many
maps worldwide have coordinate grids and topography based on surveys which
were last carried out many years ago. Sportflying sites and areas
are often not surveyed as accurately or as recently as major international
airports or areas of military interest. Finally, it should be borne
in mind that the SA error system could be turned on again by the US GPS
managers, probably on a regional basis as mentioned in the US President's
statement dated 1 May 2000.
Average accuracy figures which included the SA errors were:
SA ON, all tests, all receivers:
Average accuracy 43.9m from 2342 samples, starting 27 May 1995.
Maximum errors recorded since 1 Jan 97: 106m with good reception, 141m with poorer reception (before 1997 one sample was recorded on 23 Mar 96 with an error of 246m, this was a from a single channel receiver).
These tests have included 20 designs of GNSS Recorder from 9 manufacturers,
which used 13 types of GPS board from 5 manufacturers.
12 CHANNEL RECEIVERS
SA ON, 12 Channel receivers: Average 36.2m from 1157 samples.
Maximum errors recorded since 1 Jan 97: 95m with good reception, 141m
with poorer reception (before 1997 one sample was recorded on 23 May 96
with an error of 170m).
COPIES OF RESULTS
Details of these and other test results since this FAI/IGC/GFAC accuracy
testing program that started in May 1995, are available from IAN
STRACHAN in the form of Excel spreadsheets which can be emailed to
anyone who is interested.