I made a mistake in entry672. The noise is injected in the position of CAVITY(G3). So we need to use open loop TF and G1,G2,G3 to calibrate this signal.

So I measured the open loop transfer function, it agrees with the simulation of Eleonora's thesis. For the measurement, the magnitude for 20Hz is about 10^4.(See picture 1)

Then using CH2*(open loop/(G1*G2*G3))=CH1. From this formula, we can do the calibration like entey672 again. This time, the result of calibration approximatly equals to the measurment.

By the way, I also tried to insert noise of 500Hz. It aggres with the prediction, the suspension can filter this high frequency noise. After injecting this noise, we cannot see any peak on the spectrum of PZT or error signal.(See comparation of picture 2 and 3)