Preliminary investigations suggest that it is not the magnetic field itself that is coupling the chiller signal to the gravitational wave channel. Instead, it may be that the coupling is through the UPS power supply. In this "power supply ripple" mechanism, slight fluctuations of the UPS voltage, caused by the varying current draw of the chillers, would couple to the gravitational wave channel by affecting sensitive electronics (an example might be A to D converters or coil amplifiers), that are also powered by the UPS.
Our preliminary conclusion, that magnetic fields from the chillers and UPS cables are not strong enough to produce the observed peaks in DARM, is based on the coupling function that was previously measured for magnetic injections. Based on these PEM injections, Irene estimated chiller peaks in DARM that were roughly 1/10th of the actual level in DARM, suggesting that the magnetic fields were too small. The accuracy of this estimate depends on how similar the ratio of the chiller magnetic field to the PEM injection magnetic field is at the magnetometer used in the calculation, to the ratio of the fields at the dominant magnetic field coupling site(s). Since the V1:ENV_CEB_MAG magnetometer is fairly far from the core optics, and there are UPS cables that run quite close to the presumed coupling sites, we set up a magnetometer between the beam splitter and the power recycling mirror and we repeated the magnetic field injection to compare the ratios of injected to chiller fields at the two magnetometer locations. Irene will analyze the data more carefully, but our preliminary check suggested that the chiller-peak to injection-peak ratios at the two magnetometers were similar, supporting the preliminary conclusion that the chiller fields were not strong enough to produce the observed peaks in DARM. Magnetic injections and magnetometers at more locations would further improve the accuracy of the coupling predictions and test the hypothesis that the magnetic fields are too small to produce the observed effect in DARM.
As mentioned above, our favored hypothesis is that the coupling is through power supply ripple. This hypothesis could be further investigated by, for example, calculating a coupling function for UPS voltage variation to DARM, based on the chiller peaks, and seeing how well it predicts whether other peaks in the UPS voltage spectrum appear in DARM. Further investigations, like the magnetic investigations above, would be needed to ensure that the coupling is not from magnetic fields in the UPS cables that pass near magnetic coupling sites.
A solution, if the power supply ripple hypothesis holds, is to power the two chillers on a different UPS. Personally, I think it would be a good idea for Virgo to have two UPS systems, one for digital and high-current devices like the chillers, and a second for other systems. At LIGO, we don't use UPS as extensively as Virgo (and we usually loose a day when we have power outages), but we do have a lot of problems with shared DC power supplies, and Virgo may run into similar problems with the UPS supply. We often cure coupling to DARM by placing offending sources on their own power supplies. At the same time, we have been pro-actively putting the most sensitive electronics, the electrostatic drives, on their own exclusive power supplies.
We repeated the test (moving the chiller's temperature by half degree), having a look at others signals: the effect (100Hz sidebands moving) is well visible in:
- CEB environmental magnetometers
- local magnetic probes (the one currently below SPRB)
- the electric ground noise probe (sensing the voltage drop between metallic towers body and racks in INJ EE_room)
- the UPS "sniffer" (a sample of the UPS voltage that supply the whole Virgo experimental area in CEB)
Worth noticing that:
- moving the temperature setpoint of both chillers affect the temperature in a different way (different time constant), allowing a better understand of who is who (which sidebands is related to which chiller)
- both chillers sidebands are visible in magnetic sensors, as well as in the ground electric noise probe
- only ONE of the two chillers sideband is visible in UPS sniffer
- nothing relevant is visible in IPS sniffer
Conclusions:
No definitive conclusion yet, we know where the problem begin (TCS chillers) and where it comes (DARM), but the path is unclear; could be a magnetic coupling, could be an "electrical" coupling (voltage drops un UPS line). For sure having this kind of signal (100Hz sidebands) well visible in a distant room (EE_room) as "voltage drop" between ground is not a good sign. We plan to repeat this test after all EE_room racks have been grounded (they are not, for the time being).
Many other unwanted signals seems present as sidebands around Mains harmonics, and these could have origin from other powerful devices (e.g. the beam-dump chiller) in EE_room, affecting DARM all togheter.
We identified the TCS chillers and some of the EER equipment (likely the laser beam dump chiller or the pumping diodes chiller) as responsible of the inner and outer sidebands of 100Hz and 200Hz measured by the CEB UPS voltage sensors: see 44151 and 44154.
Here some plots to correlate this noise to recent ITF data (this essentially confirms the plot by Robert Schofield, adding details):
- Figure 1 and Figure 2 (zoom-in): DARM peaks on the LEFT-hand side of 100Hz: this is a bunch of little peaks (maybe 3 peaks): two of them correspond to UPS sidebands ahd are moslty coherent with CEB UPS probes, the central one seems something else, mostly coherent with CEB magnetometers and ELECTRIC probe.
- Figure 3 and Figure 4 (zoom-in): DARM small bump on the LEFT-hand side of 200Hz. Here the peaks are not well resolved and appear as a small bump. The situation looks similar to the 100Hz described above: some coherence (small) with UPS left sidebands pair plus some coherence (as well small) with magnetometers and ELECTRIC probe.
