Frequently Asked Question
What are the major error sources for inertial sensors?
Most inertial sensors have several error sources. Some are easy to deal with, others less so. Here are a few:
Null bias error (or zero g bias error in the case of accelerometers)
Null bias error is simply the deviation from zero when the inertial sensor is experiencing no stimulus - zero g for an accelerometer or no rotation for a gyro. In general, all sensors exhibit at least two forms of null bias error: initial null bias error and null bias error due to temperature. Initial bias error is very easy to correct for - simply measure the sensor's output with no stimulus present and subtract it from what it should be ideally. Store that value and add it to all subsequent measurements. Null bias error due to temperature is harder to deal with since the correction coefficient generally varies from sensor to sensor and each unit must be individually temperature compensated. Data sheets should provide specifications for both null bias errors.
Scale factor error
Scale factor error is the deviation from ideal of the sensor's sensitivity. There may be an initial as well as an over temperature component. Just as with null bias errors, both can be corrected by calibration. Scale factor calibration is more difficult as it requires the user to apply known stimulus to the sensor. Analog Devices inertial sensors generally have little initial scale factor error and even less scale factor error due to temperature.
All inertial sensors suffer from noise. In general, noise is proportional to bandwidth so the user must calculate the significance of noise based on the bandwidth they will be using. The noise of all Analog Devices inertial sensors is Gaussian in nature (the noise energy is the same at all frequencies),so it is rather simple. With a single pole output filter, the RMS noise of a sensor is: RMS NOISE = Noise Density * √(Bandwidth * π/2). Noise density should be specified in the sensor's data sheet. There are several other error sources that are much less significant such as nonlinearity, ratiometricity,etc. that are beyond the scope of this FAQ.