In the simulation (JGWdoc), I found the sensor correction path was wrong and injecting noise at microseismic region.
After I fixed it and carefully compare the actual signals and simulated signals, I found the inconsistency in the FB signals in amplitude.
It can be solved if we have about 10 times less actuation efficiency than the design value.

The figure below is the simulated FB signals and actual FB signals. I assumed 10 times less actuation efficiency to calculate the FB signal.

The transfer function of TM and MN that can derive the consistent FB signals are below.

They has the actuation efficiency from LOCK_L_OUT to TM displacement of

If I looked up some klog,

Nakano-kun and Yamamoto-kun said the ETMX TM actuation efficiency is 5.4e-12 m/cnt @ SUMOUT @ 1Hz = 5.4e-14 m/cnt @ SUMOUT @ 10Hz.

In Ushiba-kun's measurements, it is reported that 1.6e-13 m/cnt @ 10Hz @ TM, 6.1e-14 m/cnt @10Hz @ MN. 
Also, the ration of actuation efficiency between TM and MN seems to be wrong as reported in the reply to the Ushiba-kun's klog.