Using a pair of accelerometers instead of a single analog flex sensor ?

Hello,

In a dataglove project I am trying to replace the usual flex sensor with a pair of 3D accelerometers (ADXL345). One reference sensor is put on the back of the hand while a second one is attached to a finger. Currently only one axis of each accelerometer is used (the x axis, set parallel to the length of the finger). The reason to have two accelerometers is to compare the inclination of the two sensors and use the delta, which makes the measure independant from the absolute inclination. Actually the hand remains nearly horizontal all the time and do not turn.

The circuit returns consistent data when the hand doesn’t move or moves slowly, however when the hand moves fast I have high variations in the measurements even if the finger inclination doesn’t change.

My questions:

  1. Is my basic strategy, that is comparing data from two accelerometers to keep only the difference of inclination on one axis, a good one?
  2. If the answer to the first question is “yes”, then how can I avoid the variations caused by the fast movements? Could I benefit from using also the measurements of acceleration on y and z axes in some way? Should I change the acceleration range (currently it’s set to 4G)? Up to which point is the temporal synchronization of the measurement on the two sensors important ? Should I smooth the data on time using median/lowpass filter before comparing ???

Thank you in advance.

  1. Yes.

Could I benefit from using also the measurements of acceleration on y and z axes in some way? Should I change the acceleration range (currently it's set to 4G)? Up to which point is the temporal synchronization of the measurement on the two sensors important ? Should I smooth the data on time using median/lowpass filter before comparing ?????

a. I don't think so.
b. Range has to be correlated to max acceleration, there must not be a "clipping". Can you dump a data to buffer or PC, to verify clipping issue?
c. Obviously data has to be synchronous.
d. All depends on temporal scale : how fast do you need update rate ? 1 / 100 / 1000 per second?
How you sample a data, can you be sure freq. rate is exact? Filtering doesn't make sense w/o strictly defined two conditions.