Re: Deternining when GPS data is good

T Driver wrote:

On Jun 25, 5:45 am, Marc Brett <use...@xxxxxxxxxxxxxxxxxxx> wrote:
On Wed, 20 Jun 2007 20:20:42 -0500, Ted <nos...@xxxxxx> wrote:
pwel...@xxxxxxxxx wrote:

You could have a look at the DOP (dilution of precision) values in the
NMEA strings. This is an estimate of the inaccuracy of the position.

Peter

HDOP is very poorly correlated to the actual error. I've taken 100's of
thousands of data points, and done a regression and r**2 of the error in
position vs. HDOP, and the correlation is not very good.

Me too!

In surveying a fixed position with a non-DGPS receiver over 24 hours,
I've tried to do a weighted average with DOP values, and the result is
always /worse/ than a straight arithmetic mean. Go figure!

You'll do better if you compute a weighted average with the
number of satellites used in the solution. Another good trick
is to watch the individual satellite SNR values and throw out
(or weaken) the measurements which include satellites which
have appeared recently, or are about to disappear.

The same applies to low elevation satellites, which are more
likely to cause multipath effects.


I made a plot of the typical situation - here:
http://www.teddriver.net/PositionAccvsPDOP.jpg
You're right that it's poorly correlated. I did a quick linear
regression, but the residuals from the regression are very large!
Ted

The slope of you linear regression is roughly correct. DOP is just
a multiplier for your existing range errors. However, range errors
are completely uncorrelated with the DOP and show significantly
larger excursions than the DOP does.

Unless you can estimate the range errors, DOP is a poor indicator
of accuracy.

Kind regards,

Iwo

.

Quantcast