Hello,
We don't usually see noticeable drift in our accelerometers. We do see post-board mount offsets. These can be corrected using the precision accelerometer trim function in the Version 7.00 sensor fusion library. I suggest you read section 2.8 (Inertial navigation - truth or fiction?) in the Version 7.00 User Guide. You can download this from nxp.com/sensorfusion.
Regards,
Mike
Thank you for response. It means that i must constanty correct unstable bias (subtraction right value of g)? Bias is random value. I tried use hipass filter (MATLAB), but my motion is in very low frequencies. Filter was useless. I must reducte drift likely with absolute sensors as magnetometer, but it is difficult calculation.
Regards, Juraj.
J.P.
You misunderstand. The one-time calibration of the accels does a nice job of subtracting out offsets. You should not have to constantly update the offset values. You DO have to update the magnetic calibration, but that function is already included in our fusion library.
Neither of the above addresses random noise. There's nothing you can do about that other than apply a low pass filter. If you are using our fusion library, you should NOT do that. We already include a low pass filter into the eCompass algorithm, and the Kalman filter already assumes random noise in the input (in fact, Kalman mathematics depend upon it). As noted elsewhere, I am not a believer in pure (unassisted) inertial navigation using commercial grade sensors. The calculations WILL blow up, it's just a question of when.
Mike
Mike,
Then, you do not recommend to make use of the internal filter functions in the sensors when using NXP fusion library? FXAS21002 has an Internal low-pass & high-pass filters with programmable cut-off frequency.
How are the sensor filters preset in the evaluation kit?
regards,
Javier,
Correct. Don't use the internal low-pass & high-pass filters. it's better if the Kalman filter deals with the raw signals. the FXAS21002_Initialization structure in sensorfusion/sources/driver_FXAS21002.c is documented with register settings that we used.
Regards,
Mike
Thank You very much for answers. I know already, that i need to use Kalman filter.