Any progress on this? I had two more ideas that might be influencing your observations.
- Are you able to turn the device 180 degrees, while collecting data and validate the 180 turn through an integration of the recorded gyroscope data?
- How did you validate that you were processing the 14-bit, twos complement number correctly? I can see that as a potential area that could create a false impression of a lot of noise.
Hope that helps. Look forward to hearing of any updates on this.
Best,
NevadaMark