While updating the noise models, I thought of a few additional parameters that may be useful to simulate real IMUs (and potentially other sensors).
Sensor Properties
Dynamic Range - The maximum range that a sensor can measure before the readings are clipped. It’s possible to saturate sensors that have higher sensitivity over a smaller dynamic range.
Sensitivity - This may already be captured by precision value, but we may need to rework how precision is mapped over a given dynamic range.
Sensor Bandwidth - Typically sensors will have either mechanically or electrically-imposed bandwidth limits such that higher frequency components are filtered out. For instance the ADIS16448 has a front-end lowpass filter at approximately 330 Hz. Other sensors like a magnetometer are even lower (25Hz).
Other Error Sources
Sensor timing - we could potentially model things like clock jitter or skew.
Scale Factor - A multiplier between the physical input and the measurement, typically parts-per-million
Nonlinearity - A non-linear function that maps the physical input to the measurement.
Misalignment - Could probably already be captured by slight perturbations in the pose of the sensor.
Non-orthogonality - If the sensor isn’t manufactured perfectly, then some physical inputs on the body axes will “leak” into other channels
Acceleration effect on Bias - In MEMS gyroscopes, high accelerations can lead to additional biases in the measurements.
Tempertature effect on Bias - In MEMS gyroscopes, varying temperatures can lead to additional biases in the measurements.
Original report (archived issue) by Michael Carroll (Bitbucket: Michael Carroll, GitHub: mjcarroll).
While updating the noise models, I thought of a few additional parameters that may be useful to simulate real IMUs (and potentially other sensors).
Sensor Properties
Other Error Sources