ESTB/EGNOS works. :-)

I have just done another ESTB experiment, using Garmin Perl code from
CPAN to log positions with an external antenna on the roof of a seaside
cabin with 360 degree visibility.

I got 10-11+ESTB sats all the time, the EPE numbers were in the 1.5 to
2+ m range, but the really important result is the actual deviations:

I had a total of 27631 sample points, collected with a custom perl
program which gathered mm-level data using Garmin protocol. This becomes
important when trying to determine a meter-level position!

Anyway, here is the output of my post-processing:

Average position (27631 points) = N 59.03928938 E 10.94126581
First half average position (13816) = N 59.03928886 E 10.94126925
Offset between them: 0.2053 m

I.e. the first half of the samples resulted in a position which was 20
cm away from the one I got using all the samples, so the second half
must be 40 cm away from the first half. The total sample time was nearly
8.5 hours.

Here is the distribution of offsets from the average position, given in
percentiles:

% offs(m)
0 0.01
5 0.25
10 0.36
15 0.44
20 0.51
25 0.57
30 0.63
35 0.69
40 0.75
45 0.80
50 0.86
55 0.91
60 0.98
65 1.05
70 1.13
75 1.23
80 1.33
85 1.46
90 1.62
95 1.87
96 1.94
97 2.06
98 2.25
99 2.55
100 3.46

This means that on average (including the ~20 cm uncertainty in the
average position) half of all samples fell within 1.0 meter, and the
95th percentile is still within 2.0 meter.

This is _very_ good indeed. :-)

Note also that even the 99th percentile (approx 3 std dev?) is within
2.5 m, and the worst single outlier is only 3.5 m away. If/when this
becomes repeatable/dependable, then it is _extremely_ good, and means
that GPS can be used for primary navigation in critical situations.

Terje
--
"almost all programming can be viewed as an exercise in caching"

On Fri, 15 Jul 2005 13:50:13 +0200, Terje Mathisen


Terje,

thanks for your good work! This is very nice to hear.

However, one day may not be enough for meaningful results, so
more cross-checking is needed.

Hans-Georg

--
No mail, please.

Hans-Georg Michna wrote:


I agree.

I am currently re-running the experiment. After a little less than 2
hours, the average position is within centimeters of the one I got last
night. The percentile values are very similar as well, while the worst
outlier is less than 3 m:

percnt.        prev        curr
67        1.08        1.02
95        1.87        1.85
99        2.55        2.45
100        3.46        2.72

The position last night was
N 59.03928938 E 10.94126581
while the current average is
N 59.03928885 E 10.94126736

The differences between those positions are about 5 cm NS and 8 cm NS,
which is less than the offsets I got between the first 4 and the last 4
hours last night.

I.e. averaged position data seem to be very repeatable on the sub-meter
level, and when I get more samples today I expect the maximum variations
to increase a bit, making them even closer to what I got previously.
...
After another hour, the data is still looking good, essentially the same
position (within 15-20 cm of last night) and the percentiles have
actually improved, to 0.99, 1.80, 2.41 and 2.89 as the single worst sample.

Terje
--
"almost all programming can be viewed as an exercise in caching"

On Fri, 15 Jul 2005 21:54:10 +0200, Terje Mathisen


Terje,

great! it's becoming more likely that EGNOS will be a really
good thing.

Keep measuring if it's easy for you to do this. Wait for a few
days with different weather as well.

Hans-Georg

--
No mail, please.

Hans-Georg Michna wrote:


I've been sailing for the last week, but now that I'm back at my wife's
family summer cabin (the location with such perfect view. :-) I'll make
another test tonight.

(The weather have changed from clear blue skies to overcast and some
drizzle.)

Terje

--
"almost all programming can be viewed as an exercise in caching"