[TriEmbed] magnetometer libraries for the rest of us for soft iron and pitch/yaw correction

[Pete Soper]

I struggle (translation: I am a doofus) with linear algebra, let alone 
the matrix theory involved with "soft iron calibration" of magnetometer 
chips when they are in arbitrary orientations close to ferrous metals. 
I'm looking for OSS solutions and suspect they're plentiful in some 
problem spaces such as drone software. But when searching the repos I 
run into code that is either trivial (only smart enough to pull X/Y/Z 
samples out via polled I/O) or requiring a return to 'State's math 
department for a few months before I have any hope of recognizing the 
specific code I need. My application involves a six axis 
accelerometer/magnetometer and a Particle board with bags of compute and 
memory resources.

I understand the principles of hard iron correction (i.e. a competing 
magnetic field "pulls" the origin of the magenetometer and it's a simply 
a matter of using X, Y, and Z corrections to "recenter" the response to 
be linear and describing a sphere). Some chips even fully or partially 
automate this process. But soft iron correction (nearby metal distorting 
the sensor response to Earth's field lines and changing it from a sphere 
to an elipsoid that has to be corrected back to a sphere) is currently 
mystical.

The scheme for restoring linearity of response with soft iron distortion 
is to get bunches of samples (the ubiquitous "figure eight" manoevers we 
do with our smart phones). It's the "plugging in of these samples and 
getting and applying the resulting correction" that I don't understand 
properly to get the sensor back to an effectively spherical pattern of 
response in the presence of soft iron.

Anybody come across a library that one can simply poke a lot of samples 
into and end up with the necessary corrections to handle this case? Any 
sources of OSS solutions for this you've come across? Nirvana would be 
something like the old Usenet groups like 
comp.sensor.magnetometer.calibration-for-the-math-impaired that is 
likely to have a solution for an appliance operator like me in this 
particular space.

Also, I'm looking for simple pitch/yaw correction of magnetometer 
measures to have a "tilt insensitive" response such that an 
accelerometer can provide the axis measures needed to correct the 
magnetometer's measures for a given pitch and yaw (and roll). The best 
discourse I've found so far is this one 
<https://stackoverflow.com/questions/24639529/tilted-magnetometer-output-pitch-and-roll-tilt-compensation-i-am-lost> 
in which the petitioner shows some example code but goes on to explain 
that he had to invert the roll radian angle parameter and swap X and Y 
to compensate and get the proper results after the revelation that his 
Bosch datasheet had bad mistakes. My hope is I can mutate his example 
code until my test fixture matches the behavior of a smart phone running 
a compass application after it has been calibrated, but this is tangled 
up with the calibration process. I assume I can keep that simple and get 
joy (i.e. see the above function work properly with my sensor and the 
specified mods to it) by keeping any ferrous metal or magnetized 
material well away from the mag test fixture during my tests. Does that 
seem sensible?

Best Regards,

Pete

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.triembed.org/pipermail/triembed_triembed.org/attachments/20210623/ecd30e07/attachment.htm>