Reports until 10:22, Friday 01 March 2019
H1 ISC
stefan.ballmer@LIGO.ORG - posted 10:22, Friday 01 March 2019 - last comment - 13:28, Friday 01 March 2019(47217)
Setting the OMC DCPD matrix - part II

This is a follow-up to alog 45734.

It answers the question of how to set the DCPD input matrix, with some more background.

Basics:

- We can measure two numbers (ratios):
     g = PD_A/PD_B coherent signal amplitude ratio (best done by looking at the PD_A/PD_B transfer function at low frequencies)
     h = PD_A/PD_B shot noise signal amplitude ratio (by matching the amplitude spectra - a transfer function does not work here)

- These are directly related to the OMC BS splitting ratio 'bA/bB' and the sensing chain gain ratio between the two photo diodes, gA/gB via
     g = gA/gB * bA/bB
     h = gA/gB * sqrt(bA/bB)
  in particular we have g/h = sqrt(bA/bB)

- I can write the photodiode signals as
     PDA = g * s  + h * nA
     PDB =     s  +     nB
  or, rescaling PDA to match the shot noise level of PDB, PDA_ = PDA/h
     PDA_ = g/h * s  + nA
     PDB_ =       s  + nB
  Here s is the signal as seen by PDB, and nA and nB are realizations of shot noise, scaled such that <nA^2>=<nB^2> (i.e. same power spectral density).
  They are independent, i.e. <nA*nB>=0.

- Since the two shot noise levels are equal and uncorrelated, we can now rotate the signal by any angle theta to maximize our readout SNR:

     / PDA~ \       / PDA_ \
    |        | = R |       | 
     \ PDB~ /       \ PDA_ /

  or in vector notation:
        x~     = R x_

  Note that the shot noise levels in PDA~ and PDB~ remain uncorrelated and equal during this rotation.
  Here R is the recular rotation matrix:

          / cos(theta)   sin(theta) \
    R  = |                           | 
          \-sin(theta)   cos(theta) /

- If we rotate all the signal into PDA~, PDA~ becomes DCPD_SUM, and PDB~ becomes DCPD_NULL. This is achieved for 

    theta = atan(h/g) = atan(sqrt(bB/bA))

- The matrix proposed in alog 45734 did exactly that, with
   i)  an additonal sign change for DCPD_NULL, i.e. DCPD_NULL = - PDA~
   ii) rescaling the overall matrix gain to some aribitrary, traditional value set by A+B=2 (see alog 45734 for details, also the attached MATLAB file).
   To be explicit, the alog 45734 matrix M is equal to

    M = gain * diag([1,-1]) * R * diag([1/h,-1]);

- The cross-power CP between the two photodiodes can be calculated for any rotation angle theta:
  With
          /   0     1/2 \
    C  = |               | 
          \ 1/2       0 /

  we get

    CP = < PDA~ * PDB~ > = < transpose(x~) C x~ = transpose(x_) transpose(R) C R x_  >
 
  or
                              / -sin(2*theta)   cos(2*theta) \
    CP = < 1/2 transpose(x_) |                                | x  >
                              \  cos(2*theta)   sin(2*theta) /

  or

    CP = cos(2*theta) < PDA_ * PDB_ >  + sin(2*theta)/2 * ( <  PDB_^2 > - <  PDA_^2 > )

  There are now two ways to choose the theta for this cross-power calculations:
    A) Stay in the original PDA_-PDB_ basis: theta=0deg
    B) Optimize the SNR ratio in the cross-power: theta = theta_SUMNUL - 45deg

    A) Guarantees that the cross-power measurement is independent of the measured gain-scaling - at least at high frequencies
       (Low frequencies require loop gain corrections anyway).
    B) This introduces a slight sensitivity to the measured shot noise ratio in the two diodes (sin(2*theta)/2 term).
       But since the loop gain correction needs to be done in the SUM_NULL basis anyway, that is a bit of a mute point,
       and should not matter for realistic OMC BS splitting ratios.


*****************************************************************
Bottom line:
If we follow the procedure from alog 45734, the cross-power estimator should work as advertised.

Practical suggestion:
The process will be less mysterious if we put the shot-noise-matching gain '1/h' up front in the PDA filter modules.
Then we can use a pure rotation matrix for the PDA_PDB to SUM_NULL matrix M.

Comments related to this report
stefan.ballmer@LIGO.ORG - 13:28, Friday 01 March 2019 (47221)

Attached are two hand-drawn illustrations.

1) Illustrations for the discussion above

2) What would change form 1 when we use squeezing. If we know the squeeze factor the cross-correlator can be made to work in principle. But precise knowledge of the squeeze factor is needed.

Also attached is the MATLAB file for calculating the input matrix.

Images attached to this comment
Non-image files attached to this comment