I was using AOM scanning speed as 4000Hz/1.7s in the calibration. However, since the scanning speed for IR is 1/2 of the value for GR, the figure in the old elog was wrong.
Calibration for the measured spectrum should be: calibration = 2000/1.66666*11.5/11.2 #Hz/V (PDH: 11.2mV/11.5ms) (AOM: 4000Hz/1.66666s)
There was also problem for the calibration for off-center on-resonance, I modified the plot by using a more reasonable calibration. It comes from the center on-resonance. The new plot is shown in the attached figure.
We can see the new stable optical axis makes especially the low frequency length noise reduced. However, the high frequency noise is increased a bit.