Help MPU9250 find North

[flidsky]

Hello Kris

I'm making project that include a compass with tilt compensation. I bought a few MPU9250 and 9255 from different manufactures and connected it to ESP32 running on 240Mhz. Gyro and accel are awesome, but magnetometer never get north and its raw data is noisy. I use your code for this chip where I removed all stuff for air pressure sensor. I spent a week reducing noise, avoiding EMI, waving a figure eight and playing with settings and I never got north for more than a turn around. All of my modules have the same behavior: slowly drifting to a random position or waving between two values. Often if I let them lie down for a while it shows north correctly. I checked all what I could but still not sure about correct timings of filtering or integration. Also now i'm not sure that using 9250 is a proper solution.

As the last attempt can I ask you help me set it up?

[flidsky]

https://youtu.be/FcbioW3APCw

[kriswiner]

How did you calibrate your sensors? How did you verify proper calibration?

On Wed, Dec 23, 2020 at 10:24 PM flidsky notifications@github.com wrote:

https://youtu.be/FcbioW3APCw

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/kriswiner/MPU9250/issues/448#issuecomment-750765788, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABTDLKWROUNSPVD53WWNWCDSWLNCDANCNFSM4VH2ZCOQ .

[flidsky]

Thank you for answering. I did calibration in many different ways trying to guess the best, slow rotation in three axes, full revolution around roll pitch, yaw, and the same a bit faster, slower and more faster, more amplitude, less, just randomly, etc. I check calibration by comparing values with three compasses: digital, analog and one more in my phone. To lessen interference I did it outside also. It can't track precisely a single rotation, my guess is that the compass works only using accel and gyro and they can't track yaw. Cumulative error was more than 50° after one minute use.

[flidsky]

Waiting you answer we research around calibration and made our implementation but it is still drifting as shown here

[kriswiner]

Mag calibration doesn't look too good. You also have to calibrate the accel and gyro.

When calibrated and with the board at read flat on a table for example the accel should read 0.0, 0.0. 1000 milligs to within ~10 milligs on all three axes, the gyro should read 0, 0, 0 dps to within 0.01 dps. The mag should read whatever your local magnetic H and V values are, which you can look up via USGS site or equivalent. Also, if the board is flat on a table and reads Mz = X, it should read Mz = -X when the board is flipped upside down.

The sensor order fed into the Madgwick function also matters, and should conform to NED convention. Does it?

On Thu, Dec 24, 2020 at 6:21 AM flidsky notifications@github.com wrote:

Waiting you answer we research around calibration and made our own algorithm but is stll drifting as here https://youtu.be/K-P8V6ZddM0 [image: Compass_calibration-01] https://user-images.githubusercontent.com/33526480/103093790-6d2ab780-460c-11eb-963a-c5a1e6e75848.png

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/kriswiner/MPU9250/issues/448#issuecomment-750896105, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABTDLKXQX6RMVYZMSQES6ADSWNE67ANCNFSM4VH2ZCOQ .

[kriswiner]

Send me a link to the sketch you are using (don't send me code!).

On Thu, Dec 24, 2020 at 9:24 AM Tlera Corporation tleracorp@gmail.com wrote:

Mag calibration doesn't look too good. You also have to calibrate the accel and gyro.

When calibrated and with the board at read flat on a table for example the accel should read 0.0, 0.0. 1000 milligs to within ~10 milligs on all three axes, the gyro should read 0, 0, 0 dps to within 0.01 dps. The mag should read whatever your local magnetic H and V values are, which you can look up via USGS site or equivalent. Also, if the board is flat on a table and reads Mz = X, it should read Mz = -X when the board is flipped upside down.

The sensor order fed into the Madgwick function also matters, and should conform to NED convention. Does it?

On Thu, Dec 24, 2020 at 6:21 AM flidsky notifications@github.com wrote:

Waiting you answer we research around calibration and made our own algorithm but is stll drifting as here https://youtu.be/K-P8V6ZddM0 [image: Compass_calibration-01] https://user-images.githubusercontent.com/33526480/103093790-6d2ab780-460c-11eb-963a-c5a1e6e75848.png

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/kriswiner/MPU9250/issues/448#issuecomment-750896105, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABTDLKXQX6RMVYZMSQES6ADSWNE67ANCNFSM4VH2ZCOQ .