regancooper:
Ideas I have rulled out:
-double integral from accelerometer due to inaccuracy
Why? I'd have thought this is a pretty good solution.
The inaccuracy of doubly integrating an accelerometer is drift. Drift that increases with time.
Here you know when the item is stationary, and can clamp position and velocity values during
this time, and calibrate the acceleration due to gravity.
During an impact the readings will be clearly distiguishable due to the large magnitude, so
turn on position estimation only during the impact and rebound. Should be able
to calibrate this from the other sensors.
Issues might be the high accelerations involved, and the need for rapid sampling.