40m QIL Cryo_Lab CTN SUS_Lab TCS_Lab OMC_Lab CRIME_Lab FEA ENG_Labs OptContFac Mariner WBEEShop
  40m Log  Not logged in ELOG logo
Entry  Tue Oct 13 11:13:54 2020, gautam, Update, BHD, Investigation into RF44 sensing phaseNoisePSD.pdfloopTF.pdf
    Reply  Thu Oct 15 10:42:39 2020, gautam, Update, BHD, More investigation into RF44 sensing analogZetaLock.pngrelPhaseNoise.pdfsigMagPhase.pdf
Message ID: 15623     Entry time: Tue Oct 13 11:13:54 2020     Reply to this: 15628
Author: gautam 
Type: Update 
Category: BHD 
Subject: Investigation into RF44 sensing 

Attachment #1: spectra of the phase noise between LO and IFO output fields sensed using the RF44 signal.

  • Measurement setup:
    • LO an IFO fields are combined on a beamsplitter, with ~60% mode-matching efficiency.
    • One port of the BS goes to a DCPD.
    • The other port goes to an RF sensing photodiode, PDA10CF. The spec-ed dark noise NEP is ~12 pW/rtHz at 1.6 um, (so let's say 25 pW/rtHz) and transimpedance is 5kohms into a 50 ohm load. We can convert this to an equivalent sensing noise at the error point of this loop, though it's more likely that the electronics (demod, ADC etc) noise downstream dictate the sensing limit, which I measure by blocking light on the photodiode.
  • The demodulation is done on one of the newly received D0902745 boards - this was just a more compact setup than many cascaded minicircuit components. We don't have the hardware to package this into a chassis to shield against electronics noise pickup yet, so I'm using a bench supply to power this for now (via a voltage regulation board, D1000217.
  • "Dark Noise" = ASD with no light incident on the photodiode. "LO field only" = ASD with only the LO field incident on the photodiode.
  • The "Dark noise" trace and "LO field only" traces are converted from cts/rtHz to rad/rtHz by noting that when the Michelson is locked on a dark fringe, the demodulated RF44 quadratures have a pk-pk amplitude of ~160 cts (corresponding to pi radians of phase shift). Since in these conditions the demodulated quadratures do not undergo any fringe wrapping, I converted the spectra by simple multiplication.
  • For the "RF44 open loop" trace:
    • The DC offset in the demodulated signal (due to the RF44 signal from the LO field only) is digitally compensated, so that the fringing has (roughly) zero offset.
    • The Michelson was locked on a dark fringe, and the demodulated RF44 quadratures were monitored for ~5 mins. Then arctangent (specifically, arctan2 to get the correct quadrant in the IQ plane) of the two signals was taken to convert the fringing signals to phase noise.

Closing a feedback loop:

  • Since it seems like we are sensing a signal (below ~1kHz at least), I tried to close a feedback loop (modelled loop shape shown in Attachment #2, it's just a model because I have to guess what the sensing and actuation gains are, and they're both assumed to be flat, digital delays etc aren't accounted for). I've also added the inferred loop gain by taking the ratio of the in loop and unsuppressed ASDs (though of course I don't account for the flat sensing noise at higher frequencies). At least qualitatively, things line up...
  • While I can get the light level on the DCPD to stabilitze somewhat, the loop is not at all stable, and the suppression isn't very good at all.
  • Not sure how meaningful any of the spectra with the loop closed are, but FWIW, I've put in the spectra of the demodulated RF44 signals with the loop engaged (RF44 Q is used as the error signal). A clear problem is evident at ~120 Hz, and the forest of lines isn't helping for sure. Also unclear to me why the I and Q signals don't have the same profile at low frequencies.

Conclusions/Questions:

  1. What is the reason for the huge forests of lines in the "RF44 open loop" ASD, that are absent in the other two traces? If this were electrical pickup, it should be there in all three traces?
  2. Is the shape of the spectrum reasonable? The roll-off above ~5 Hz doesn't seem quite steep enough to be seismic noise from the suspensions. Can it really be that the Michelson dark field has such high phase noise?
  3. How can we get this scheme to give us cleaner sensing?
  4. The actuation chain was verified to work fine with the single bounce beam from an ITM interfered with the LO field, and using the DC light level as an error signal and locking to the half-fringe point. So the problem is not due to insufficient actuation range. Seems like the error signal is so polluted with these forests of lines that even though there is some suppression of the error signal at low frequencies, the unsuppressed noise is still significant. I can't solve the problem by simply increasing the loop gain...
  5. It is not shown here, but with only the LO field incident on the RFPD, I see a drift of the demodulated signals on the ~5 minute timescale - is this just due to fiber length change? If so, this is potentially problematic, as on long time scales, the true zero of the error point of the servo would be changing on the ~5 minute timescale. This would be true even for the final suspended scheme - if the path length between PR2 and the homodyne BS changes by some microns, we would have to correct this at DC?
Attachment 1: phaseNoisePSD.pdf  215 kB  Uploaded Tue Oct 13 13:09:47 2020  | Hide | Hide all
phaseNoisePSD.pdf
Attachment 2: loopTF.pdf  116 kB  Uploaded Tue Oct 13 23:17:46 2020  | Hide | Hide all
loopTF.pdf
ELOG V3.1.3-