I'm getting the acceleration data using the -[CMDeviceMotion userAcceleration]
I've noticed one interesting thing: I always get a small bias on the Z axis. It is about 0.0155 (with variance of 0.002). While on other axes the average values are near 0.
I'm testing this with iPod Touch 4G (and it is just laying on the table during testing). The question is: where this bias is from and is it device specific?
I noticed similar values although CoreMotion tries to eliminate bias. If you rotate your device so that x (or y) is parallele to gravity you will probably see the bias in x direction. Using raw sensor data showed the same tendency but with larger values and some more super-imposing effects like temperature dependency, time based shifting, ...
18 months ago I read a specification of the iPhone 3 devices' accelerometers and according to this the accuracy was about 1.8 % of g. (what a pity the bookmark to STM product page I set now leads to 404).
Basically this should not be a problem as long as you don't try to estimate exact positions (displacements) and this seems to be impossible with an acceptable accuracy - see the several discusion here on SO.
来源：https://stackoverflow.com/questions/7820447/cmdevicemotion-useracceleration-drift