It seems to be because of the oscillation of NBDAMP_R (fig1).
I found that NBDAMP_R was not tuned from 300K, so I tuned it.
I didn't measure OLTF but it seems to wrk well according to the time series of the feedback signals (fig2).