Does somebody can tell how compass offset calibration made?

[and-sh]

I see that it is a system of linear equations, but can not understand which.

[sensei-hacker]

I can tell you at a high level what it's doing. You may already understand the high-level view, or maybe not.

I haven't pored through the actual Gaussian elimination equations.

At a high level, it's very similar to calibrating the current sensor. On each axis, the sensor has a maximum value read based on the sensitivity of the sensor and the strength of the local magnetic field. We want to scale that to get consistent units. It also has a a minimum value which correlates to a constant offset from zero - the sensor would read a nonzero value even if the craft were out in space, away from Earth's magnetic field.

At a high level, the calibration computes the offset and the scale.

During run time, the scaled value is computed something like:

CurrentX.scaled = (x.raw - x.offset) * x.scale

That gives the value of the axis in uniform units, adjusted for the two variables that come from the sensor and the environment.

If you're wanting to understand the actual Gaussian elimination that calculates the offset, I can't help you there.
For most practical purposes, one can pretend that it takes the minimum reading as the offset. That's not exactly accurate because at no point in the calibration is there actually zero magnetic field. The Gaussian elimination accounts for that to compute what the offset should be.

I don't know if that was helpful, or perhaps you already understood that much.

[and-sh]

Thank you
I guess this is something like this
http://juddzone.com/ALGORITHMS/algorithms.html

[sensei-hacker]

Thank you I guess this is something like this http://juddzone.com/ALGORITHMS/algorithms.html

Yeah I think that gives a good explanation of a) what the basic idea is and b) why it's not quite as simple as "use the minimum value as the offset", though that's a good enough simplified view for most pilots to think of it that way.

[and-sh]

It should be noted that algorithm give gains too. But it is not used. May be there are some well hidden error

[and-sh]

good review
https://thecavepearlproject.org/2015/05/22/calibrating-any-compass-or-accelerometer-for-arduino/
It is interesting that accelerometer can be calibrated by the same way

[sensei-hacker]

It should be noted that algorithm give gains too. But it is not used. May be there are some well hidden error

I think you will find magGain used on line #461 of src/main/sensors/compass.c and in the various compass drivers under src/main/drivers/compass/

[and-sh]

I mean that zero calculated from elipsoid
https://github.com/iNavFlight/inav/blob/master/src/main/sensors/compass.c#L440
But gain from max/min
https://github.com/iNavFlight/inav/blob/master/src/main/sensors/compass.c#L452

[and-sh]

There is even function for gain
https://github.com/iNavFlight/inav/blob/master/src/main/common/maths.c#L516
But it not used