Using the calibrated DRMI channels created and described by Kiwamu in entry 18742, I grabbed data from the lock of August 3, 2015, starting at 04:20:00 UTC. The attached 4 page PDF shows spectra of the open loop and residual displacement noise for SRCL, MICH and PRCL. The 4th page shows the coherence of SRCL with the other 2 degrees of freedom.
Degree of freedom | Residual rms | Shot noise level |
---|---|---|
SRCL | 8 pm | 1.3 x 10-15 m/rtHz |
MICH | 3 pm | 1.5 x 10-16 m/rtHz |
PRCL | 0.8 pm | 4 x 10-17 m/rtHz |
The SRCL spectra has a curious shape: it comes down quickly with frequency to 10 Hz, then is fairly flat from 10 Hz to 50 Hz, then falls by a factor of 5 or so to the (presumed) shot noise level that is reached above 100 Hz. What is this noise shelf between 10 Hz and 80 Hz? That is exactly the region where the SRCL noise coupling to DARM is troublesome. Our usual approach is to send a SRCL correction path to DARM to reduce the coupling, but this spectrum shows that there should also be some gain to be had by reducing this noise shelf.
The last page of the PDF shows the coherence between SRCL and MICH & PRCL, and it indicates that the SRCL noise shelf could be coupling from PRCL noise -- the coherence is fairly high in this band, though not unity. This suggests that the DRMI signals could use some of the demodulator phase and input matrix tuning that Rana has recently done on L1, reported in LLO log entry 19540.
To complete this log entry, it would be useful if someone at LHO could add the open loop transfer functions for each loop (models), and other pertinent info such as DC photocurrents for these detectors and the input matrix coefficients.
The shelf appears to be gain peaking in SRCL. We have an 80 LPF to get rid of SRCL control noise in the bucket, but it makes the control noise from 30 to 60 Hz a bit worse.
I had the filter off between 2015-08-23 00:36:00 Z and 00:39:00 Z. The attachment shows the error and control signals with filter off (dashed) versus filter on (solid).
Evan, is the shelf in Peter's open loop spectrum there because of OLTF model without LPF? Otherwise, we still need to investigate.