S2 Veto Safety Studies

Peter Shawhan
August 13, 2003

I spent some time studying the safety of using the auxiliary interferometer sensing channels (AS_I, REFL_Q, REFL_I, POB_Q, POB_I) as veto channels, by seeing whether length excitations couple into these channels. As detailed below, I find that AS_I is very unsafe, while the others are very safe.

In-Band Study Using Burst Injections

We injected "sine-gaussian" burst waveforms with various frequencies and amplitudes several times during the S2 run. We used particularly large amplitudes during the big injection session on April 9-10, so I looked up the times and used DTT with filtering to pick out the injected signal in AS_Q and to look for it in the five auxiliary channels as well. When a signal was evident, I measured its peak height by eye, which should be accurate to roughly 5%. When no signal stood out from the noise, I guessed how large a signal could be and still be hidden; these values are fairly rough, maybe good to 25% or occasionally even more rough.

In all of the cases considered here, the burst waveform was injected into ETMY alone, i.e. it excited both differential- and common-arm degrees of freedom.

Burst injections at various frequencies

These waveforms all had Q=30. The "Scale" column is the overall scale factor applied by multiawgstream when injecting the signal into ETMY. (The complete list of injections is in an attachment to an elog entry.) The "Filter" column indicates the frequencies used in DTT to define a Butterworth filter, e.g. for 2000-Hz sine-gaussians, the filter used in DTT was butter("BandPass",4,1800,2200). I chose these filter frequencies to be 10% below and above the central frequency of the injection; this seemed to let through basically all of the power in these Q=30 waveforms. Each pdf link in the table leads to a three-page document, with filtered time series from AS_Q and AS_I on the first page, REFL_Q and REFL_I on the second, and POB_Q and POB_I on the third.

IfoTimeFreq (Hz)ScaleFilterPlots AS_QAS_IREFL_QREFL_IPOB_QPOB_I
L1 733997569 2000 61.12871800-2200 pdf 0.0100.30 < 0.015< 1.5 < 0.05< 0.04
733997589 1304 20.96971174-1434 pdf 0.0080.68 < 0.02< 0.5 < 0.045< 0.03
733997609 850 7.1935765-935 pdf 0.01250.28 ? < 0.07< 1.0 < 0.028< 0.02
733997629 554 3.7015499-609 pdf 0.0240.92 < 0.005< 0.8 < 0.01< 0.01
733997649 361 2.1163325-397 pdf 0.071.95 < 0.01< 1.3 < 0.025< 0.037
733997669 235 1.4520211-259 pdf 0.252.0 < 0.012< 0.5 < 0.01< 0.015
733997689 153 0.6973138-168 pdf 0.3114 < 0.01< 0.5 < 0.02< 0.025
733997709 100 0.341790-110 pdf 0.4817 < 0.02< 0.3 < 0.02< 0.01
H1 733997569 2000 52.35561800-2200 pdf 0.068< 0.005 < 0.05< 0.05 < 0.1< 0.05
733997589 1304 14.36811174-1434 pdf 0.0540.007 ? < 0.03< 0.2 < 0.08< 0.07
733997609 850 2.4644765-935 pdf 0.0320.004 ? < 0.04< 0.2 < 0.05< 0.03
733997629 554 0.8454499-609 pdf 0.0480.0037 < 0.06< 0.6 < 0.1< 0.1
733997649 361 0.2900325-397 pdf 0.0900.0078 < 0.01< 0.5 < 0.02< 0.02
733997669 235 0.1990211-259 pdf 0.220.013 < 0.01< 0.4 < 0.03< 0.04
733997689 153 0.1706138-168 pdf 0.680.071 < 0.005< 0.2 < 0.02< 0.02
733997709 100 0.093790-110 pdf 0.800.08 < 0.01< 0.3 < 0.02< 0.03
H2 733997569 2000 101.02211800-2200 pdf 0.130.04 < 0.09< 12 < 0.02< 0.13
733997589 1304 34.65481174-1434 pdf 0.1250.025 < 0.02< 5 < 0.05< 0.05
733997609 850 1.7832765-935 pdf 0.0230.004 ? < 0.05< 3 < 0.02< 0.02
733997629 554 0.8156499-609 pdf 0.0470.0105 < 0.1< 7 < 0.05< 0.05
733997649 361 0.3497325-397 pdf 0.070.015 < 0.1< 9 < 0.04< 0.04
733997669 235 0.1440211-259 pdf 0.0950.026 < 0.2< 10 < 0.05< 0.04
733997689 153 0.0988138-168 pdf 0.220.044 < 0.05< 5 < 0.02< 0.05
733997709 100 0.045290-110 pdf 0.320.085 < 0.7< 40 < 0.15< 0.03

Note that the signal is visible in AS_I in almost all cases, but it is never apparent above the noise in any of the auxiliary channels.

Large burst injections at 850 Hz

On April 10, we selected the 850-Hz sine-gaussian with Q=9 with a wide range of amplitudes, some of them very large. As in the study described above, I looked at them using DTT, with a somewhat broader filter due to the lower Q value. The table below gives the results for a fairly loud injection (not actually the loudest, since we had observed nonlinear effects for the one or two loudest signals we injected). This provides the most sensitive check for coupling to auxiliary sensing channels, at least at this frequency. Even for these very large signals, there was no sign of the signal in REFL_Q, REFL_I, POB_Q, or POB_I.

IfoTimeFreq (Hz)ScaleFilterPlots AS_QAS_IREFL_QREFL_IPOB_QPOB_I
L1 734002424 850 630.384700-1000 pdf 0.9012 < 0.01< 2 < 0.05< 0.05
H1 734002424 850 215.952700-1000 pdf 2.60.18 < 0.02< 0.65 < 0.05< 0.02
H2 734002424 850 156.288700-1000 pdf 3.00.33 < 0.13< 5 < 0.05< 0.05

Out-of-Band Study Using Swept Sine Injections

The studies above show that arm length displacements do not couple to REFL_Q, REFL_I, POB_Q, or POB_I at frequencies of 100 Hz and above. I wanted to check whether the situation is different at lower frequencies. To do this, I figured out when a swept-sine calibration was performed. The L1 autocalibration directory listing includes a calibrated spectrum taken on April 10 at 734066817. (The injection was apparently done into ETMX.) By trial and error, I determined that the swept-sine excitation was actually a bit after that, e.g. it passed through 100 Hz (stepping down in frequency) at about 734066965.

I used DTT to calculate power spectra for AS_Q and the five auxiliary channels for fairly short time intervals (Bandwidth=0.2 Hz, 4 averages, overlap=0%, so I believe I was integrating over 20 seconds of data) starting at 734066950. The plots of these, given in the table below, each include a few calibration lines that were part of the sweep. You can see that the injection shows up clearly in AS_I (as well as AS_Q, of course), but not in any of the other auxiliary channels. Thus, the lack of coupling at higher frequencies extends down to at least 40 Hz, which is the lowest frequency excited in this sweep.

Start timePower spectraInjected frequencies (Hz)
734066950pdf112, 103, 95
734066970pdf86, 80, 73
734066990pdf67, 62, 57
734067010pdf57, 52, 48
734067030pdf48, 43
734067050pdf40

FYI, another L1 sweep passes through 100 Hz at ~729314565, and yet another at ~729318065. But neither of these goes below 40 Hz.