A new mag calibration algorithm is needed for reducing data storage heap

[zarathustr]

The magnetometer calibration is based on a least square fitting of raw data. Currently, the algorithm requires a heap load of raw data storage arrays. This point significantly increases the online memory usage of the algorithm also increases instability of managing the RAM. A better issue is raised here to integrate my recent contributions into the PX4 that it no longer stores the raw data and uses a 10x10 matrix to compute the singular value decomposition results. Such method swaps the execution time with the space consumption, which is better for low-configuration platforms e.g. stm32.

[dagar]

Let's talk about doing this after v1.7.0 goes out shortly. Calibration code will be moving out of commander (#8200) and needs other improvements (#8265) and fixes (#8190).

[zarathustr]

@dagar So when will we receive the 1.7.0 release? Then what are the main advances included in this release?

[dagar]

We're in the process of testing and resolving a few remaining issues. https://github.com/PX4/Firmware/milestone/23

It's a huge list of things (everything that went into master over the last 4-5 months).

[mhkabir]

@zarathustr Could you open a PR with your changes to the calibration routine?

[dagar]

Related, I've been playing with mag calibration via UKF here. #15059