Reports until 16:36, Tuesday 21 March 2023
H1 ISC (CAL, DetChar)
jeffrey.kissel@LIGO.ORG - posted 16:36, Tuesday 21 March 2023 - last comment - 17:18, Tuesday 21 March 2023(68098)
Compensation for OMC DCPD Tranimpedance Amplifier Response Updated; OMC-DCPD_A0 and OMC-DCPD_B0 Banks Reorganized
J. Kissel, H-y. Huang, D. Sigg

Daniel is pushing forward with starting to design a giant collection of notch filters that kill many of the lines that are aliased from the new OMC DCPD 524 kHz system after down sampling to the nominal 16 kHz. 

In doing so, he began clearing out some of the old, now unused filters in the OMC-DCPD_A0 and OMC-DCPD_B0 filter banks to make room ("antiWh" and "antiB" that were compensating the now-removed D2100630-style whitening chassis, formerly in FM3 and FM6). 

This drove Hsiang-yu and I to finally push forward with updating the digital compensation for the trans-impedance amplifiers (TIAs) -- creating new FM1 "NewV2A" 2023-03-21 filters -- based on the poles and zero Louis' fits to our 2023-03-06 measurements (from LHO:67809). The former FM6 "V2A" 2022-02-15 compensation filters where informed by very early measurements taken via not-so-great whitening chassis interface (LHO:61810).

Further, because we then didn't have enough filter modules with Daniel's 3 new modules worth of notch filters (now FM3 "Notches1", FM5 "Notches 2", and FM8 "Notches3") to do a quickly-revertable "before" and "after" test with the FM6 "V2A" and "NewV2A", so I've moved the "4chn" gain of 0.25 into an additional gain in the "cts2V" module, and overwrote FM1 with "NewV2A."

In order to test the success of the updated compensation, we used the digital DAC-driven excitation, through the TEST input of the TIA that we've been using to explore the electronics signal chain (e.g. recently re-described in LHO:67760 and LHO:67730). 

The new novelty in this is that with the D2200215-style whitening chassis, we can turn OFF the whitening entirely, and thus, for the first time, can measure the trans-impedance amplifier (TIA) response directly without being confused by any whitening response.

Attached are the results showing success in four panels worth of transfer functions by comparing three different configurations:
    (1) With the analog whitening OFF (H1:OMC-DCPD_[A,B]_GAINSET = 1 = "Low"), compensating the TIA with the old FM6 "V2A" 20220215-informed filter. Here, we see the large residual in the Analog TIA * 20220215 V2A Comp product that's been there for months.
    (2) With the analog whitening OFF (H1:OMC-DCPD_[A,B]_GAINSET = 1 = "Low"), compensating the TIA with the new FM1 "NewV2A" 20230321-informed filter. Here, the residual in the Analog TIA * 20230321 NewV2A Comp product flattens out nicely.
    (3) Taking configuration (2), but then turning the analog whitening ON  (H1:OMC-DCPD_[A,B]_GAINSET = 0 = "High"), and *its* digital FM2 "NewAW" compensation ON,  we see the residual in the Analog TIA, Analog WH, and both of their compensation filters product remains quite frequency-independent.

Now, below ~200 Hz, the maximum deviation from frequency independent mean of 3.04e-6 [mA/DAC ct] is +0.2% / -0.1%. 
Very good!

We're still unsure why the data is not flat above 200 Hz, but is now confirmed to be unrelated to the TIA, the Whitening Chassis, or either of its compensation. The investigation continues elsewhere, by modeling all the anti-imaging and anti-aliasing that's going on -- our primary suspect at this point (see, for example, LHO:67766).

So, now the new OMC-DCPD_A0 and OMC-DCPD_B0 configuration, when the analog whitening is ON is,
    FM1   "NewV2A"  -- compensation for the analog TIA, updated on 2023-03-21
    FM2   "NewAW"   -- compensation for the analog Whitening, updated on 2023-02-01
    FM4   "cts2V"   -- calibration of the 18-bit ADC *and* the divide-by-four gain that used be in the "4chn" module, updated on 2023-03-21
    FM7   "A2mA"    -- gain of 1000 to convert amps into milliamps, for convenience -- no recent change.
    FM9   "Dec65k"  -- 8x digital down sampling filter for decimation from 524 kHz to 65 kHz
    FM10  "Dec16k"  -- 4x digital down sampling filter for decimation from 65 kHz to 16 kHz
as shown in the last medm screenshot attached.
Note -- Daniel's new notch filters in FM3, FM5, and FM8 are NOT YET in use.

THIS WILL REQUIRE AN UPDATE TO THE pyDARM MODEL PARAMETER SET because we've re-ordered filter modules are in use and rearranged the gain allocation. We *might* be able to get away with it, given that we've only re-arranged and updated the gains and frequency dependence that are nominally not included in the pyDARM model, but it will be worth it in preserving our sanity to do so.
Images attached to this report
Comments related to this report
jeffrey.kissel@LIGO.ORG - 17:08, Tuesday 21 March 2023 (68102)CAL
Details on the "NewV2A" design:
    DCPDA    OMC-DCPD_A0     FM1    "NewV2A"     zpk([5.766+i*22.222;5.766-i*22.222;32.77],[2.613;2.195;6.556],1e-5,"n")gain(459.31)
    DCPDB    OMC-DCPD_B0     FM1    "NewV2A"     zpk([4.808+i*19.818;4.808-i*19.818;32.59],[2.525;1.766;6.52],1e-5,"n")gain(466.105)
 
    (1) The poles and zeros are straight-forward: I take the poles from LHO:67809, and put them in the zeros. I take the zeros from LHO:67809 and put them in the poles. 
    (2) I set the gain of the zpk filter to be 1e-5 -- which is the *inverse* of the expected overall transimpedance gain in the gravitational wave band, 100e3 [V/A].
    (3) I then let foton to do the standard math one needs to do to normalize the gain for a given set of poles and zeros, by asking it to set the gain at 1000 Hz to be 1e-5. That's the source of the gain terms at the end.

That latter bit of gain finagling is a little bit different that Daniel chose to do in the former "V2A" filters (see LHO:62653) -- he put the standard zero-pole normalization in the gain field of the zpk filter along with the 1e-5 gain, and then "nudged" the rest to set the gain at 1000 Hz to be *exactly* 1e-5. I think my choice of gain makes the design choices a bit more clear.

As one can see by the attached bode plots of the filters for both channels, there's absolute no change at high frequency (i.e. the two methods of normalization result in the same answer), and really only a little bit of change (on this plot scale) around and below the complex pole pair.
Images attached to this comment
jeffrey.kissel@LIGO.ORG - 17:12, Tuesday 21 March 2023 (68103)CAL, ISC, SYS
In addition to updating the primary DCPD paths, I also updated the TIA compensation in the OMC-PI path as well -- namely, in the H1:OMC-PI_DOWNCONV_SIG_OUTPUT filter bank. 
There's now a "NewV2A" in that bank too, also in FM1. It's a copy of the DCPDA filter, since the sensor matrix that feeds that bank has been set to use only DCPDA.

Attached is a screenshot of that filter bank (during a time when the OMC whitening is OFF, so FM2, "NewAW" is not engaged.)
Images attached to this comment
jeffrey.kissel@LIGO.ORG - 17:18, Tuesday 21 March 2023 (68104)CAL, CDS, ISC
I have SDF accepted these changes, and committed the changes to the OMC-DCPD_A0, OMC-DCPD_B0, and OMC-PI_DOWNCONV_SIG filter banks -- all of which live in the H1IOPOMC0.txt filter file -- to the userapps repo under 
    /opt/rtcds/userapps/release/cds/h1/filterfiles/H1IOPOMC0.txt