Sheila, Evan
Summary
By increasing the gains on the common-mode and IMC boards, we pushed the CARM UGF up to 35 kHz and saw a reduction in the high-frequency noise floor of DARM.
Some further loop tuning is required to make this viable as a long-term change.
Details
Based on Sheila's earlier measurement of the IMC OLTF, we felt it was safe to increase the gain at the IMC error point (both the error signal and the AO) by 2 dB. Then we increased the CARM gain by 6 dB (using the IN2 slider on the CM board). These changes gave a UGF of 35 kHz with 25° of phase. [See plot and zip file.]
This gave a noticeable improvement in the frequency noise as seen by REFL_A 9I, which is currently the out-of-loop CARM sensor. [See plot.]
Consequently, there was a small but noticeable improvement in the high-frequency noise floor of the DARM spectrum. [See plot with red and gold, taken while still feeding DARM back to ETMX.] In the full locking configuration, the noise floor now touches the GWINC curve from 300 Hz to 3 kHz [See plot with purple.]
Obviously this CARM phase margin is quite thin, and we don't want to run like this as a matter of course. In order to win more phase, perhaps we need to look at the IMC loop and the FSS. Peter K last measured the FSS UGF to be 200 kHz with 30° of phase (on the low end of the phase bubble). In comparison, at LLO the FSS UGF is 500 kHz with 60° of phase.
More details
Nominal gains are 0 dB for CM IN2 [the CARM error signal], 5 dB for MC IN1 [the IMC error signal], and 0 dB for MC IN2 [the AO signal].
The gains used here are 6 dB for CM IN2, 7 dB for MC IN1, and 2 dB for MC IN2.
At the start of the evening, REFL_A 9I had no whitening filters engaged and 0 dB of whitening gain. It now has 1 stage of whitening, 21 dB of gain and a −21 dB filter engaged in FM4 (I and Q). There is some saturation during the lock acquisition, but in full lock the I and Q inputs are now at least a factor of 5 in ASD above the ADC noise floor everywhere. I also changed the digital phase rotation from 77° to 90°, as was hinted at earlier.
The above work was done as usual with CARM controlled by REFLAIR.
However, in-vac REFL will also work for controlling CARM. The following screenshot shows the settings required and an OLTF of the CARM loop.
This has the unfortunate effect of making the DARM spectrum worse at high frequencies; a broad lump appears between 2 and 5 kHz. More investigation required.
A noise budget of this morning's lock is attached, and includes intensity and frequency noise. For the coupling TFs I used what Koji and I measured previously.
This isn't our most sensitive lock in terms of inspiral range, in part because of the bump around 100 Hz. This is new as of a few days ago, and we are hoping it's just the HAM6 cleanroom.
The intensity noise is taken from IM4 trans, and the frequency noise from REFL_A 9I.
From the spectra shown in the parent entry, the noise in REFL_A 9I from 40 Hz to 2 kHz is mostly flat, and does not scale with CARM loop gain. It could be that the CARM loop is limited by sensor noise (e.g., from REFLAIR_A), or that it is just masked by noise in REFL_A. So the frequency noise trace in this plot should be taken as an upper limit between 40 Hz and 2 kHz.
In the future, when measuring intensity noise we should perhaps do what Alexa suggested and use the ISS second loop PD array as an out-of-loop sensor, since it is acquired faster than IM4 trans.