I performed post-run calibration measurements on 2016/4/28.
The data and the Jupyter Notebook to analyze them are in the follwoing URL:
https://www.dropbox.com/sh/hmpfhiqb9r2bvku/AACt1mQzKvERwNQW6kM7THMma?dl=0
You can see the notebook here:
The strange LPF is now 2nd order !!
Actually, the LPF does not appear in the TFs from the individual ETM. So analog origin is excluded.
Sorry for the very late comment on this issue.
I noticed that the apparent LPF in the transfer function came from my stupid mistake of saturating the excitation signal during the transfer function measurements.
The excitation signal was injected in front of the feedback filter, which has a HPF shape up to 2kHz. Since the injected excitation signal amplitude was too high, the DAC saturated above a few hundred Hz. The saturation makes the loop gain look smaller. This effect gets worse at high frequencies because of the HPF of the feedback filter. Therefore, a fake LPF appears in the measured open loop gain.
I edited the Jupyter notebook to include the effect of the saturation. Now the model and the measurements agree very well !