M8P FIXED vs FLOAT Accuracy

0 votes
We have a pair of the M8P RTK eval boards.  Both are stationary, although the rover is configured with the portable dynamic model.

At one point in time, RELPOSNED reports a FLOAT RTK solution with 0.01m (1 cm) relative accuracy in all three directions and a given relative position.

This relative position can vary by up to a meter or more from that measured with a tape measure.  This is much larger than 1cm error, it is also much larger than the"typical"  20cm errors quoted for a FIXED solution in another post.

Some time later, RELPOSNED reports a FIXED RTK solution, also with 0.01m (1cm) stated relative accuracy, but a relative position that differs from the FLOAT solution by a meter or more, and is closer too the measured position difference, but  but still not within 1cm (or even 10cm) of the measured position difference.

I assumed the 1cm "relative accuracy" would have been an 1-sigma position error in each of the three directions, meaning anything more than 3cm error would be highly unlikely (in agreement with a different post quoting 2-3cm typical total error in FIXED mode) .

Is this not the case?  Obviously neither of the above FLOAT or FIXED relative solutions actually has a 1cm 1-sigma accuracy, and neither seem to be withing the 3cm to 20cm accuracy quoted elsewhere...

asked Jan 20 by dlisoski
I used my own antennas, which had clear sight lines and known survey locations. It is important to understand that specification number are arrived at under optimal conditions. I suspect u-Blox used geodetic grade antennas, which likely cost $1K a piece, not ones costing $10

If using the antennas in the kit you'd want to put them on a large metal ground plane (say 12" diameter), and keep them away from the human body. The roof of a car would work well, and pair of cars in the middle of a large parking lot would make for a good test case.

A tape measure would be a good way to establish a baseline length, less so for position in 3D space. I prefer to work with ECEF coordinates so I can automate the computation of the rover with respect to the truth location.

It is also important to understand that using "Survey In" (ie averaging GPS location from stand-alone receiver) is likely to get numbers that are ~1m from a survey grade position using post-processing against a governmental reference (CORS)
I guess my concern/question was more about the definition of the "Relative Accuracy" field in the RELPOSNED message.  Both FIXED and FLOAT solutions were telling me "Relative Accuracy" of 1cm in this message.

Obviously (by measurement and by comparison of the two solution types), neither one was that accurate.

The Float solution was REALLY not that accurate.

I used the antennas and ground planes that came with the boards. As you guys have noted, there are ways to get better accuracy. Given my rather low $ setup I wasnt really expecting the solutions to be 1cm accurate, but I was wondering why RELPOSNED is telling me they are.

Is there a better thing to look at to determine how accurate the current solution might be?

3 Answers

0 votes
Seems a bit strange.  I think users on this forum have been able to get much better relative (not absolute) position accuracy.  Fixed should provide better accuracy compared with Float.  You should also have antennas with good ground planes to minimize reflections, etc.  If you are not running latest u-blox firmware then I would strongly recommend upgrading both base station and rover and repeating the testing.
answered Jan 20 by grumpy Senior Principal Expert
0 votes

To add some information, here are a couple of screen grabs of the RELPOSNED message at different times (antennas were stationary)

And no, this example with a 82m baseline was not the one I used a tape measure on :-)

One float solution, with 1cm accuracy noted:

A while later, another Float solution, still with 1cm accuracy claimed but more than a meter different relative position:


A while later, a FIXED solution, also 1cm accuracy noted:

They cant all be right...


answered Jan 21 by dlisoski
Yeah, I don't think I buy into the self expressed accuracy numbers, in this case it is really just an expression about how noisy it thinks the measurements are.

I've torn down my original test range, need to set one up in my new location...
0 votes
I've been doing similar tests, mainly to test the repeatability of the positions under a FLOAT or FIXED solution.

The results are not great when only in a FLOAT solution. With a vertical Accuracy between 0.01m and 0.03m and a horizontal Accuracy of 0.01m I can calculate up to 2m variances in GPS positions.

What I have learnt thus far is that only the RTK FIXED solution provides repeatable positions. The GPS position variances in a FIXED solution tends to be in the cm range.

It would be great if they gave a confidence interval of each of the accuracy measurements.
answered 3 days ago by LukeSkyWire