Hi Erdin and thanks for your reply. For the benefits of all readers of this thread, let me give you a better description of the project.
No actually I just want to monitor precisely the rotations of a ball rolling freely on the ground. I am not primarily interested in accelerations. I am wondering if this is something feasible with a good level of precision.
It is true that it doesn't involve any feedback control or fancy servo, just data output. But so far I haven't found one single project dealing with this type of setup. It is quite different from what is achieved usually with this type of sensor (ie MPU-6050) where people try to stabilize attitude of an object and with it, the noise coming back through the sensor attached to this stabilized object (ie copter, robot...). In my case, it is the inverse that's happening. My object is not servo controlled, and its random movements feedback a lot of noise into the captor. I don't know if this type of 6 or 9 DOF sensor is even designed for such an application. I don't see why it wouldn't be, but I may be missing some technical limitations of these devices (cumulative drift, noise floor...).
So for my application :
1/ The rotation is indeed only 20 to 30 times a second maximum (not thousands), but considering centrifugal force, and vibrations it would be sufficient to badly scramble my readings, from what I understand.
2/ On the good side, as rotation is "relatively" slow, I may be able to use high quality Kalman filtering as I don't need a very high sampling rate and thus can use some computing power. From what I understand I need to extract filtered quaternions values from my sensor readings.
3/ I am interested primary in monitoring continuously the ATTITUDE of the ball for a few seconds only (for roughly 5 seconds which translates into 150 revolutions maximum). Mathematically, I believe this means that I need a series of 3 axis orientation data at various points in time.
Eventually I would integrate these data into rotation angles by period of time. This can be done in a second time asynchronously. First priority would be to record a good series of orientation data for 5 seconds.
4/ As the ball would be bumping around, the accelerometers would be shaken a bit, but not too much as the ball would be free rolling (probably not beyond 25g as the ball would not be bumping into walls or anything like that). I am not interested in these high values at all as I want rotational data only (meaning I can filter these acceleration peaks out).
5/ In a far far future, I would be interested in being able to track the overall speed and position of the ball, but this seems highly difficult to extract such data from a rotating 6 or 9 DOF sensor with any kind of meaningful precision.
6/ The captor would not be perfectly centered so there would some offset to consider in my calibration phase.
Hope this will help. Sorry if this is all a bit fuzzy, as it's a general concept, and there is no specific code to discuss yet, but I am really wondering if such an application is even possible using this type of captor, or if it must be put into a drawer until technology improves... And if it is possible, with which level of precision. I am not asking for precise or definite answers of course, just a general feeling...
Thanks for your help!
