J. Kissel, H-y. Huang
#BelatedaLOG
[1] WP LHO:11037
[2] D2000592 TIA Circuit Schematic
[3] S2100832 "SN02" Serial Number of TIA
[4] D2200215 Current Whitening Chassis Interface Assembly Drawing
[5] D2100630 Former Whitening Chassis Interface Assembly Drawing

In our haste to gather an updated transfer function of the OMC DCPD transimpedance amplifier (TIA) on 2023-03-06 (see LHO:67801), we neglected to measure the "measurement setup" that is typically taken to ensure there's not hidden frequency response in the data that might confuse the fitting routine that's used to generate the poles and zeros for the compensation of said frequency response (a la LHO:67809). On Friday 2023-03-10, we took advantage of the ETMX LVLN ESD Driver down time (LHO:67875) to actually gather that data. 

In this aLOG, I explore the different configurations one might imagine for this "measurement setup" with the new D2200215-style whitening chassis interface.

First, a reminder of the four analog measurements we have of these TIAs,
    20220215    D2100630    In-vac    No light on DCPDs, Pinout of Chassis Confusing, no good electrical ground reference, TIAs not thermalized
    20220419    D2100630    In-vac    No light on DCPDs, No good electrical ground reference, TIAs not thermalized  
    20230306    D2200215    In-vac    No light on DCPDs, New better measurement interface, Forgot to take "measurement setup"
    20230310    D2200215    In-vac    No light on DCPDs, Best one yet!

And we have had to fits of the data thus far, the 2022-02-15 (LHO:61823) and 2023-03-06 (LHO:67809) sets, and those informed two iterations of front-end compensation filters, "V2A" (LHO:62653) and "NewV2A" (LHO:68098).

We've recently confirmed that the "NewV2A" update to the compensation filters improves the residual systematic error in the electronics in LHO:68098.

So, there are are four remaining questions that we need to resolve:
    (1) Since we saw such a drastic change in frequency response between 2022-02-15, 2022-04-19, and the 2023-03-06 data sets -- is the frequency response of the TIA changing (irr)regularly, and thus will we need to have a live monitor of the frequency response? If it does, on what time-scale does it change?
    (2) Did *not* dividing out the frequency response of the "measurement setup" impact the fits that informed the compensation filters (or probably more directly -- did it impact the fit of the 10 kHz and 11 kHz poles that are compensated for the low-latency GDS pipeline)?
    (3) What *is* the right "measurement setup" to divide out of the primary data set, given the new buffered interface to the TIA that the D2200215-style whitening chassis provides?
    (4) Does the response of the TIA's change when there's 10 to 20 [mA] of current going through the circuit as it does in nominal low noise (as opposed to all the measurements thus far, which have been with no light on the DCPDs).

In this aLOG, we answer (1) and (3), in reverse order. Unfortunately, we've yet to really properly probe the sensing function / calibrated data stream's performance above 1 kHz, so we can't yet answer (2) confidently. And indeed, we've yet to see if we even *can* measure the DCPDs when we're in nominal low noise, so don't yet have any data showing that it changes with current through the circuit.

Answer to Question (3)
    (Check out the third attachment to orient yourself to the measurement scheme. These are collections of measurement diagrams that show

        pg1 -- The "primary" "Device Under Test (DUT)" measurement setup when the TIA is included in the signal chain.

The rest of the pages are various versions of the "measurement setup" which we want to divide *out* of the DUT configuration. We tried these various configurations, because there are buffer amplifiers between the *actual* TIA response path (that goes to further into the whitening then out to the ADC) and the monitor path that provides nice neat BNC spigots for SR785 interfacing.

        pg2 -- CONFIG 1, the first configuration we tried, using a DB25 breakout to connect the reference drive upstream of the buffer amplifiers, with the DB25 cable that interfaces with the TIA still plugged in and the TIAs powered on.
        pg3 -- CONFIG 2, the second we tried, same as CONFIG 1 except with the DB25 cable to the TIA's disconnected from the DB25 breakout and uninvolved with the data (and thus the TIAs are powered down.)
        pg4 -- CONFIG 3, the third and final we tried, the "traditional" measurement setup to takes of the SR785 accessory box alone.

Check out the second attachment, which compares the results of CONFIG 1 and CONFIG 2 against CONFIG 3.
We see that there is a significant response difference if one drives up-stream of the monitor path's buffer amplifiers (CONFIG 2), and even more response difference if one leaves the TIA included in the measurement.
That being said, in retrospect, it doesn't make sense to use CONFIG 1. In that configuration, the output impedance of the TIA system is unwantedly involved in the measurement, influencing the response.

From this, we conclude that we must use CONFIG 2 as out "measurement setup."
Indeed, though, as one can see in the right two panels of the second attachment, there is about a 1 deg phase loss by 10-20 kHz in this measurement setup, we may may a small bit of difference in where the fitter estimates the 10 and 11 kHz poles are in the full response measurement TIA. So, we do our due diligence and divide it out of the data.

Unfortunately, this means even though we can gather a primary "DUT" measurement of the TIA from the D2200215-style whitening chassis without powering down the TIAs and without "breaking in" to the DB25 signal chain (and thus we can gather a "fully-thermalized" measurement), we still need to "break out" the DB25 chain to gather the measurement setup transfer function, and thus there is a brief 15-20 minute moment where the DCPDs are powered down. Because we don't yet have a quantitative measure of how long these things take to thermalize (or de-thermalize), then there is *some* risk that -- in a really quite contrived scenario -- we take this measurement, and within ~30 minutes we're back to nominal low noise and observing, and thus there's some bit of systematic error. But I think that risk is very low. It's more just that it's an inconvenience for this measurement that we hadn't considered when we designed in these BNC monitor spigots.
     
Some Answer to Question (1)
    Now that we know which "measurement setup" configuration to use (CONFIG 2), we divide it out of the 2023-03-10 data and arrive a comparison between all primary "DUT" data sets taken thus far, in first attachment.

In the right plots, I divide each of the *other* data sets by the 2023-03-06 data set because
    (a) the 20230306 data set informed what's currently being used for "NewV2A" compensation, and
    (b) I want to see if the response has changed in the 4 days between the 2023-03-10 and 2023-03-06 data sets, and
    (c) I'm now much more confident that the measurement with the D2200215-style whitening chassis is much more accurate and reproducible.

Thankfully -- just like the remote, DAC driven measurements taken on 2023-03-21, seen in LHO:68098, it looks like there is negligible difference between the transfer functions between 2023-03-06, 2023-03-10, and 2023-03-21. This is very relieving. At least on the time scale of ~1 month, this "dark" measurements of the TIA response appears to no longer be changing. 

We'll continue to watch this response over the next few months (probably more with the remote DAC driven measurement than this more challenging analog measurement) to
    (a) continue monitoring whether the response is time-dependent over long time scales, and
    (b) to see if we can get definitive data on whether the response changes with current flowing through the circuit.

Stay tuned!