Bottom line up front: how many distinct values of acceleration can the ADIS 16228 report in Realtime mode? Is it 2^16, or a subset?
Context: we began looking into an unusual periodic signal that was appearing in our data when we used the ADIS16228 in real-time acceleration measurement mode. We ultimately determined that the periodic signal was due to, in quiet periods, several continuous readings of the sensor all having identical values. It seemed unusual that this would happen, given that other readings around the quiet period would differ by 32 (e.g. when gravity was the only acceleration measured, the records would show that, +/- 0x20. We wondered why bits below the fifth were not changing as well. This seemed like a quantization problem.
We then tried to lower the dynamic range, using the bits in REC_CTRL2 to lower the measurement range to 1g (from 20g), and running a manually-triggered acceleration collect (since real-time collect ignores values in REC_CTRL2 per page 16, second pagaraph of the Dynamic Range Settings subsection). However, we ran into the same sort of quantization problem; in a "silent" period, we were seeing that acceleration would "wobble" around five or seven values, and these values were not bit-adjacent to each other.
We are trying to understand why we're not seeing more intermediate states between these gravity-with-noise figures. It seems like although the LSB of an acceleration register is 0.0006104g, the smallest change that the ADIS16228 is capable of reporting is larger than that. Is this correct? Should we be treating the 16-bit number that gets produced as a 10-bit number with a 32x sized scaling factor?
If not (i.e. in a normal case, we should see differences of plus or minus one bit), what are troubleshooting steps we can undertake to see this in our own data?