[FOSS-GPS] RTKLib - Constant offset between UBlox NEO-M8T receivers

Tue Apr 14 02:57:55 PDT 2015

Thanks a lot David for your detailed answer!

First, let me explain a bit more about my setup: the 3 rovers were aligned on the front of the car's roof, separated from each other by 45cm. Rover #1 was on the right side of the car, rover #2 was in the middle, and rover #3 was on the left side. Each rover were equipped with a magnetic GPS-GLONASS antenna (this model http://www.taoglas.com/images/product_images/original_images/AA.161.301111.pdf , the ones that comes with UBlox EVKs)

Like you, I've first suspected that the errors could be due to some kind of projection-error in the car's coordinates, but if I look at the trajectories on a 2D map, it seems to confirm the results I get: rover 1 and 2 are always correctly aligned (i.e. perpendicular to the trajectory), but rover 3 is ahead of the 2 other rovers in one half of the track and behind them in the other half (which led me to think the error came from an almost-constant offset in the Earth coordinates).

> I presume that during the ten loops, that RTKLIB was more of less in a fixed mode all the time?  Your line about removing the initial data concerns me that you are not fixed or understanding how to check for that.   Some folks come to rest on the test track close to a known point for a few minute to "rest and re-lock" between laps.   If these things check out, I suspect you are using low cost mag puck antenna and that every time you place it the effective phase center changes.   We have a circle track we do much the same runs with, and there one tactical grade type of antenna fails at one part of  the loop for such reason, so its not just a cost issue.  If we presume your Novatel choke ring ant is ideal (reasonable) what are you using for the three rovers - that normally where the compromising issues come in. 

Actually, the positions I obtained were almost never fixed. I got <1% fixed positions on all the rovers and the fix mode never seemed to hold for long. All the other solutions were floating. I didn't think it mattered a lot as I'm not really interested in centimeter-level accuracy, +-20 centimeters is good enough for my needs. In other tests I made, floating results seem to be good enough to achieve this kind of accuracy but maybe I misunderstood the way it works. Is there any obvious reason why after a 15-minute run, I can't hold a fix for more than 3 seconds?

Thanks again for your help,
Simon

> Le 13 avr. 2015 à 18:40, David Kelley <DavidKelley at ITSware.net> a écrit :
> 
> 
> Simon 
> Several sanity check questions for you.  Frankly it hard to say because once you have translated to the vehicle frame of reference it is hard to tell if there was a projection error there, or if  an angle of arrival error (antenna or environment caused) in one or more of the devices.  Also, if your kinematic path involves enough lateral stress to  tilt the vehicle, or the ground is tilted, you will need to account for that as well (but as you are only looking at one vehicle and presumably the roof did not deflect you can ignore that for now).   
> 
> I find a better way to do this comparison is to just plot the overall 3D or 2D distance between each set of rovers point streams first (before any frame of reference translation is made), and ignoring time issues.  If you see odd repeating patterns there, look for antenna nulls and the phases shift that come with them.  [And if you are going crazy, try the whole test with one shared antenna to confirm that your three rovers do in fact track the same to some limit with the same input signal]  When you plot the rover paths do you see (per each loop on the track) concentric circles?  Do these circles get corrupted when you see the sign change in you other plot?  What was the antenna separation distance (one hopes >>19 cm), and is this observable from these circles?  Always a good idea to snap a photo of your roof setups.
> 
> I presume that during the ten loops, that RTKLIB was more of less in a fixed mode all the time?  Your line about removing the initial data concerns me that you are not fixed or understanding how to check for that.   Some folks come to rest on the test track close to a known point for a few minute to "rest and re-lock" between laps.   If these things check out, I suspect you are using low cost mag puck antenna and that every time you place it the effective phase center changes.   We have a circle track we do much the same runs with, and there one tactical grade type of antenna fails at one part of  the loop for such reason, so its not just a cost issue.  If we presume your Novatel choke ring ant is ideal (reasonable) what are you using for the three rovers - that normally where the compromising issues come in.  
> 
> Regards,  DC Kelley
> 
> On 4/13/2015 8:49 AM, Simon Trény wrote:
>> Hello all,
>> 
>> I'm using RTKLib in Kinematic mode to obtain the relative positions between several rovers (they are all UBlox NEO-M8T at 1Hz with GPS+GLONASS). I use an additional NEO-M8T at 1Hz with a Novatel GPS-701-GG antenna as the base station.
>> 
>> To evaluate whether or not the relative positions are good or not, I've fixed 3 rovers on the roof of a car and I drove for 15 minutes on a looping race-track. Then, I take the positions outputted by RTKLib, I project them in the car's local coordinate-system (X and Y) and I plot the X (normal to the trajectory) and Y (tangential) differences between each rover. As the rovers are fixed in the car's local coordinate-system, the differences should be ideally constant over time.
>> 
>> Here are the results I get (yellow curve is the tangential difference, and red curve is the normal difference) :
>> - Differences between rovers 1 and 2 : 
>> http://postimg.org/image/ploa4ttpt/ <http://postimg.org/image/ploa4ttpt/>
>> 
>> - Differences between rovers 1 and 3 : 
>> http://postimg.org/image/hu7k69pkh/ <http://postimg.org/image/hu7k69pkh/>
>> 
>> 
>> As you can see, rovers 1 and 2 are quite well positioned relatively to each other: the yellow plot varies between -20cm and +20cm (54% below 10cm).
>> But with rovers 1 and 3, the tangential difference varies between -50cm and +50cm (only 12% below 10cm). I get approximately the same plot between rovers 2 and 3.
>> 
>> The sinusoidal aspect of these plots seems to indicate that the error comes from a constant absolute offset of rover 3. Each time a half lap is completed, the sign of the difference changes.
>> 
>> Is there any reason that could explain why I get better results between rover 1 and rover 2, rather than between rover 1 and 3? If I redo the same run, results may vary, sometimes all the results seems good between all rovers, sometimes it may be another rover than #3 that performs badly.
>> I've also noticed that sometimes, if I regenerate the positions by removing the first minutes of the run, the results may improve greatly .
>> 
>> Also, is there any settings that may be more appropriate than the default ones if I'm more interested by the relative positions of the rovers than by their absolute positions?
>> 
>> Thanks by advance for your help, and also thanks for all the hard work put into RTKLib!
>> 
>> Best regards,
>> Simon
>> 
>> 
>> 
>> _______________________________________________
>> This message is sent to you from 
>> FOSS-GPS at lists.osgeo.org
>>  mailing list.
>> Visit 
>> http://lists.osgeo.org/mailman/listinfo/foss-gps
>>  to manage your subscription
>> For more information, check 
>> http://wiki.osgeo.org/wiki/FOSS-GPS
> 
> _______________________________________________
> This message is sent to you from FOSS-GPS at lists.osgeo.org mailing list.
> Visit http://lists.osgeo.org/mailman/listinfo/foss-gps to manage your subscription
> For more information, check http://wiki.osgeo.org/wiki/FOSS-GPS