Calibration function is wrong

[MArimont3]

Hi, I think your calibration function is very wrong.
You initialize the calibration values to zero. Ok.
But then in a loop, you

• read values -> this method is using the calibration values !
• set the calibration values to the values read

So you are getting some random values out...

Also: the example GY521_performance_calibrate.ino has a method test with an argument, but this is never used. It is hardcoded the test should run 100 times.

[RobTillaart]

@MArimont3
Thanks for the issue.

Hi, I think your calibration function is very wrong.

Can you explain or show why you think it is wrong?
Do you have investigated the cause and do you have a proposal how to solve the problem?
(just asking as time is my most scarce)

The idea of the calibration is to have the sensor in a stable, horizontal, non moving position.
Then you make multiple measurements and these should in theory (ideal sensor) read zero however in practice you read noise.
The calibration routine sums this noise per axis and averages them with a factor float factor = 1.0 / times;.
Yes I know this is only a first order correction at best.

The underlying assumption is that this average error at the "no movement" level is the same as when the sensor is undergoing movements and accelerations in XYZ planes. I am no physicist, and I know the correction needed will probably be acceleration dependent and the three axis might even affect each other. However investigating a more elaborated (2nd or more order) calibration is outside my scope.

---

Also: the example GY521_performance_calibrate.ino has a method test with an argument, but this is never used. It is hardcoded the test should run 100 times.

I will have a look in that one asap.

[RobTillaart]

Also: the example GY521_performance_calibrate.ino has a method test with an argument, but this is never used. It is hardcoded the test should run 100 times.

Error confirmed, I will update this example right away (will not publish a new release for it)

[RobTillaart]

@MArimont3
Pushed a new version of the example in the master branch. I do not have a hardware setup nearby to test it on at the moment.
If you have time please post the output of the "fixed" example.

[jens-kuerten]

@RobTillaart

The issue with the calibration function is that you are modifying the errors while you are performing the calibration.

That breaks the averaging, because you are only measuring the "full" error on the first iteration.

[RobTillaart]

@jens-kuerten

OK, that gives some information where to look, I'll dive into it asap.

Thanks!

[RobTillaart]

@jens-kuerten @MArimont3

Aha, stupid me, I see the problem, will create a develop branch for a fix.
and need to find some hardware to verify :).

[RobTillaart]

Created develop branch + PR - #55