Calibrating error signal to beat frequency;
  I injected 0 dBm RF sine wave into the beatbox and sweeped the frequency(just like we did in elog #6815).
  This time, we have different whitening filters. I sweeped the frequency from 0 to 100 MHz in 200 sec.
  The length of the delay line is ~1.5 m for COARSE.


Y arm length;
  Here, I think we need some assumption. Let's assume wavelength of IR lamb_IR = 1064 nm and Y end green frequency is nu_g = 2*nu_IR.
  There is a relation
    dnu_g / nu_g = dL / L
  So,
    dnu_g / (dL/lamb_IR) = 2*nu_IR * lamb_IR / L = 2c/L
  We know that dL/lamb_IR = 1/2 for difference in beat frequency between TEM00s. Therefore, slope of the dnu_g vs dL/lamb_IR plot gives us the arm length L(figure below, middle plot).

  Error estimation is not done yet, but I think the COARSE_I_IN1 error signal to the beat frequency calibration has the largest error because it seems like the amplitude of sine wave changes ~10% day by day.

Calibrating beat frequency to Y arm length change;
  I used L = 32.36 m (figure above, bottom plot).
    dnu_g / dL = c / lamb_g / L = 1.74 MHz/m