Though I'm not sure the exact time of increasing REFL power, PDA3_RF45_I frequently saturates during OBSERVATION_WITHOUT_LINES at ADC after that day. Saturation appeared since 12:30 on Apr. 10th.

Bottom panels of Fig.1 shows the overflow counts for RF45_I and RF45_Q. During OBSERVATION_WITH_OUT_LINES (= top left panel is ~10000), we can see non-zero values on bottom panels. Fig.2 shows same figures with more shorter duration.

In such case, it's difficult to use them as the witness sensors for the jitter subtraction. So I removed the 3rd stage of whitening filter and added 12dB gain to use ADC range effectively (Whitening settings was changed around 21:30). Because these channels are not used for any control, there should be no problem for the lock acquisition and stable operations.