Displaying report 1-1 of 1.
Reports until 15:54, Tuesday 21 February 2023
H1 ISC (CDS, DAQ, DetChar, ISC)
jeffrey.kissel@LIGO.ORG - posted 15:54, Tuesday 21 February 2023 - last comment - 08:54, Wednesday 22 February 2023(67530)
First Look at OMC DCPD ADC Noise; 4x Copy only Benefits Regions outside of GW Band, Some Time-dependent, Channel-Cross-Coupling, related to IO Chassis Fans Present
J. Kissel, M. Pirello

I continue the campaign of budgeting the dark noise of the OMC DCPD sensing chain. The TIA was characterized in LHO:67272, the OMC Whitening was characterized in LHO:67119. 

In this aLOG, I *begin* the characterization of the data acquisition system, the pass-through and 4x copy "anti aliasing" chassis, and the new 524 kHz sampled, 18-bit, lower noise ADC. 

I say *begin*, my study happened to be coincidental with Marc's IO chassis fan change -- the *change* to reduce the IO chassis cooling fans from 2 down to 1 (see LHO:67519), and as such we were able to identify the following conclusions:

    (1) Averaging the ADC channels (discussed in LHO:67439 and LHO:66489) works with the expected ~sqrt(4) = 2 improvement, BUT
    
    (2) Because the whitened TIA self noise is well above the ADC noise floor, we only really see a benefit *outside* of the 10 - 10000 Hz region.
        (3) So, doing this more complex analog copy averaging using D2300115 may not be worth it.
    
    (4) It seems like there's a loss in gain between the sum and each channel.
    
    (5) Among other high-frequency lines, there seems to be a pretty ugly time-dependent line feature in the shorted channels around ~30 Hz (indeed, moving around between 20-40 Hz) that is coherent with the OMC DCPD channels at the level of ~0.5, even though the DARM noise of the DCPD TIA is an order of magnitude above the shorted channel. (Note, the statement is informed using 50 aveages of 8 second FFTs, so the sharp, quickly moving line, may be smeared out in ASD and coherence )
    
    (6) Aside from this, it seems like having 1 fan (blowing across the ADC card) is better than either 2 fans or 0 fans.
        (7) The high frequency content of the shorted channel is better with 1 fan.

    (8) There appears to be a pretty hefty thermal dependence in (4), witnessed during the time that we were fan testing.

    (9) We need to do these tests more carefully, and/or on the test stand.

I'll post the "0 fans vs 1 fan vs. 2 fans" data in another aLOG, when I get more maintenance day time, since the story is complicated by the apparent change in noise of the ADC card due to thermalization.
The data collection and display is also severely limited by the fact that one can only look at three 524 kHz test points at a time, and none of this data is stored in the frames.
Here's what I can prove using the attachments below.

All measurements were taken while the OMC DCPDs had no light on them (the IMC was offline and the shutter was closed), unlike in previous data (LHO:67297 and LHO:67328). As such, we *expect* the signal chain to be dominated by the self-noise of the Transimpedance Amplifier, propagated through its frequency response and the whitening chassis frequency response (both electronics' frequency response was delibrately *not* compensated for in the data to better understand what is being seen.). Of course we've also proven time-and-time again that ADCs and DACs are signal dependent, so we can't necessarily conclude that the ADC's noise should be the same between a shorted channel, a channel measuring ~0 V_{DC}, and a channel with plenty of V_{DC} and an active AC signal, and as such there's all the more reason to continue this study more carefully.

1st attachment: This shows the full frequency span ASD (binwidth 0.125 Hz) of the 524 kHz data channels -- comparing the following channels ~10 minutes apart ("shortly after" Marc unplugged the 2nd fan at 2023-02-21 20:13 UTC, and later at 2023-02-21 20:21 UTC)
    (a) H1:IOP-OMC0_MADC0_TP_CH0, marked in the legend as "ADC Raw CH0" -- one of the four, raw, analog copies of the DCPD A signal chain (again, with no light on it).
    (b) H1:OMC-DCPD_A0_IN1, marked in the legend as "ADC0 Sum(CH0, 4, 8, 12)* 0.25" -- the digital sum of the four analog copies, calibrated in the DTT template to then divide by 4 to complete the average.
    (c) H1:IOP-OMC0_MADC0_TP_CH16, marked in the legend as "ADC0 Raw CH16 [Differentially Shorted Channel]-- one of the four copies of the shorted channels (whose short was installed last Thursday; LHO:67465)
all calibrated into units of differential input voltage to the ADC with a DTT template gain of 40 [V_diff] / 2 ^18 [ADC ct] = 1.5259e-4 [V_diff / ADC ct].

Here, both Blue and Cyan traces -- which are the average of the 4 copies -- is below the Red and Magenta traces, demonstrating conclusions (1) and (2) above. 

Also, there is a clear an obvious feature in the data in the 20-40 Hz region that is quite different between the 20:13 UTC (Red and Magenta) and 20:21 UTC (Blue and Cyan) data that appears in both the the dark OMC DCPD A channel set and the shorted CH16 data set. At first, and had I only seen the 20:13 UTC data, I would have posited that this was the CH0, CH4, CH8, and CH12 copies cross-coupling into the shorted CH16 channel, but because it moves around, and it's seen in both channels, I instead confirm this is a feature that seems present on all channels, and highly-dependent on the thermal state of the ADC card / IO Chassis -- with the guess being that even though both data sets are with 1 fan, these were so close in time that we suspect thermalization. This demonstrates conclusions (5) and (8).

2nd attachment: This two panel plot shows the same ASD as the 1st attachment, but now adds the coherence between the shorted CH16 and the dark DCPD CH0 during the later, 20:21 UTC, measurement. This further validates point (5). Unfortunately, I didn't save the coherence during the 20:13 UTC data set, but subsequent (as of yet unpublished) data sets during the 0, 1, and 2 fan comparison convince me that the feature moves and is coherent when it moves.

3rd attachment: This 4-panel plot shows a transfer function between the calibrated single DCPD dark channel, CH0 (H1:IOP-OMC0_MADC0_TP_CH0) and the calibrated (and divided by 4) average DCPD dark channel (H1:OMC-DCPD_A0_IN1). In the region where the ADC channels are dominated by the coherent transimpedance amplifier and whitening chassis dark noise, I would have expected these to have a transfer function of identically unity, and they do not. This demonstrates the conclusion (4) from above, and gives me more motivation for conclusion (3).
Note that the phase, where coherent, is nicely 0 as expected. 



Images attached to this report
Comments related to this report
jeffrey.kissel@LIGO.ORG - 15:56, Tuesday 21 February 2023 (67534)
The DTT template for this data set lives in 
    /ligo/svncommon/CalSVN/aligocalibration/trunk/Common/Electronics/H1/SensingFunction/OMCA/Data/
        20230221_2021UTC_H1_TIAxWC_OMCA_S2100832_S2300003_DCPDA_vs_Short_524kHz_asd.xml
daniel.sigg@LIGO.ORG - 08:54, Wednesday 22 February 2023 (67544)

It looks like the dark noise of an individual channel is about a factor of 6 above its ADC noise. So, the sum channel has about a factor of 12 headroom between dark noise and ADC noise. This makes it possible to run without the whitening gain and only loose a factor of ~√2 in technical noise.

The ADC noise will make the coherence and transfer function less than 1 between the scaled sum and an individual channel.

Displaying report 1-1 of 1.