Variance parameters

[antonellotartam]

Hi everyone,

I'm trying to understand how the "COMPUTE_6DOF_GY_KALMAN constants" have been set compared to the accelerometer and gyroscope datasheets. I'm referring to the file fusion.h (https://github.com/memsindustrygroup/Open-Source-Sensor-Fusion/blob/master/Sources/fusion.h)

Is there a matching between that constants and the relative noise/offset inside the datasheets?

Thanks in advance

[michaelestanley]

Yes,  fAccGL[3] is the linear acceleration in the global frame.  It is available in both the 6 and 9-axis Kalman filter structures.  Check section 3.4 in the user guide.

Regards,

Mike

[antonellotartam]

Ok I got it ... I have another question.

Is there a way to get the Kalman filtered accelerations ?

Thank you for your answers

[antonellotartam]

Are you saying these data can be computed empirically ?

Looking at the comments in fusion.h, I understood that FQVA_6DOF_GY_KALMAN and FQVG_6DOF_GY_KALMAN are

computed as RMS values of a stationary acquisition of time T.

Is this right ?

Is there a relation with the noise density value in the datasheet ?

What about FQWB_6DOF_GY_KALMAN and FQWA_6DOF_GY_KALMAN ?

[michaelestanley]

Conceptually, yes there is a correspondance.  But in practice, we adjust them to achieve best tradeoff between all use cases, including dealing with magnetic interference and sustained linear acceleration.  For instance, tightening up magnetic variances while leaving accelerometer variances alone will give you faster convergence relative to the earth magnetic field, but at the cost of degraded immunity to magnetic interference.

[michaelestanley]

We experiment with the numbers until we get desired result.  So yes, view them as empirical fudge factors that have some basis in a physical model.  The covariances have no datasheet equivalent.  Random walk can be viewed as the time integral of noise, so there is a relationship.  But again, there are other factors at play.