I corrected the mistakes to the PDH box. The LISO estimate of the noise from the previous post is no longer quite right, however, as I had to reduce the gain at different points than I had originally intended (originally I was going to do at least part of the reduction at the output stage, but this proves difficult because doing so without being very careful results in a different gain for the inverted and non-inverted modes the way the output stage is designed). The result is that the high-frequency noise is higher than anticipated. Taking the increase in optical gain into consideration, though, the high-frequency contribution to the gyro noise is still lower than before.
Here is a transfer function, showing an overall gain decrease of about 35 dB from the previous case (this post, about halfway down):
Here is the input-referred noise spectrum, along with the LISO estimate for this circuit. The excess low-frequency noise is now absent, but the noise at high frequencies is higher than estimated. It looks as though this might actually be the noise of the spectrum analyzer (the output noise level here is on the order of 20 nV/rHz, and though I had it auto-ranging it could have gotten hung up). I will check this in the AM. Either way, there is a big improvement in broadband and the servo's contribution to the gyro noise is below requirement in a big chunk of our operational band (see previous post).
As I outlined in this post, the primary PDH box needed some modifying if we are to see any significant improvement from the new setup. I made these changes:
The input-referred noise is now at the level explained by the LISO model---unlike it was before---though this calls for a funny story:
When I plotted the noise and divided by the transfer function I measured, I got a noise level that was exactly 10x higher than predicted by LISO. After quite a while of scratching my head and checking my code, I realized that I mistakenly changed the gain of the wrong stage (despite writing down the right one in my notes), causing the noise from the nasty LF356 switchable stage to have 10x the influence. I will correct it and re-measure tomorrow, but the important thing is that we have removed the noisy 8336 from near the input.
Here is a picture of the modified portion of the board and a LISO prediction of the noise when I've corrected the mistake. You can compare with the plot in the link above, but this corresponds to a true low-frequency voltage noise betterment of 20 dB at low frequency, which adds to the 40 dB improvement in the servo's contribution from the expected optical gain increase. This actually puts the servo noise below the aLIGO requirement between about 100-350 mHz.