Unexpected level change in Rx

We are testing a custom board with AD9375. In reception the level change for unknown reason. In the picture below I had the chance to capture it. Raw I and Q data of the same channel are plot. The input signal is a kind of AM, the carrier is at the same frequency as the Rx OL, the modulation is a sine. In this situation, usually the component (I or Q) with the least level is changing, the other stays at a constant level.

What I usually do is to compute the RMS value of a set of received samples, again and again, and the result is changing in less than a second. On the other side I rarely see it happening an a small set of data plotted (which last less than a microsecond).

I have checked the gain, it is manual, and it does not change (anyway it should change both I and Q). I have checked the gain compensation and slicer, they are disabled. I have tried to disable all tracking calibration, the result is the same.

Thank you

PS. In another situation where the input signal is a CW, at a frequency slightly different from the OL frequency (in such a way that I and Q have the same level), I do not see the same think (I have other problems, by the way).

Add "another situation"
[edited by: tcachat at 9:28 AM (GMT 0) on 18 Apr 2019]