How does Genuino Arduino 101 Intel Curie Autocalibration (Ax Ay Az) process works?

Question

I am reading the source code file CurieIMU.cpp#L409, made by Intel for Arduino/Genuino 101. I have some questions:

• How does auto-calibration work in this specific case? This is not very clear to me.
• Why the offset can be assumed to be linear?
• Is there any reference stating that it is safe to assume a linear offset for 1 axis calibration?

The CurieIMUgetAccelerometerOffset reference says that the sensitivity is 3,9 mg.

Given a sensitivity of +-2g, I have to use this formula

3,9*19,6/32767

in order to convert the value in m/s^2.

Here is the library code, made by Intel, that computes the correct offset:

{
    int bmiOffset = offset / 3.9;

    if (bmiOffset < -128) {
        bmiOffset = -128;
    } else if (bmiOffset > 127) {
        bmiOffset = 127;
    }

    if (axis == X_AXIS) {
        setXAccelOffset(bmiOffset);
    } else if (axis == Y_AXIS) {
        setYAccelOffset(bmiOffset);
    } else if (axis == Z_AXIS) {
        setZAccelOffset(bmiOffset);
    }

    setAccelOffsetEnabled(true);
}