Repository navigation
Conversation
…carry it through the random-trigger merge [rtj] In real data the FDC cathode strips are read out by fADC125s, and the reconstruction uses the pulse peak amplitude (DFDCHit::pulse_height, with q = a_scale * pulse_height). mcsmear wrote only the integrated charge q, so for MC the reconstruction fell back on q where the data use the peak. FDCSmearer now adds an fdcDigihit to every smeared strip hit, with peakAmp = q / FDC_ADC_ASCALE (/FDC/digi_scales), i.e. on the data scale. If FDC_ADC_ASCALE cannot be read from CCDB no fdcDigihit is written and the behaviour is as before. The FDC strip merge with random triggers (hddm_s_merger, the earliest-hit rule of PR #420) now applies the same rule to the peak amplitude as to q: the earliest random hit replaces the MC hit (time, q and peakAmp) if it is earlier, both are summed if they fall in the same 8 ns sample, and a later random hit is ignored. A hit without an fdcDigihit contributes q, which is what the reconstruction would use for it. The merge also returns immediately if the random strip has no hits, instead of dereferencing src.begin() of an empty list. Validated in the GlueX rho0 data/MC study of 2026-10/11: with this mcsmear (and the hdgeant4 FDC fixes of rjones30/HDGeant4 PR #37) the change of the rho0 yield between recon-2018_08 and recon-2019_11 on the same data is reproduced by the MC to 0.3% (70 chunks of 10k events, runs 51048-51052). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…MA and FDC:EFFVSDOCA_PAR0..3; defaults leave the output unchanged [rtj] - FDC:DRIFT_SMEAR_NS (ns, default 0 = off): a Gaussian drift-time shift drawn once per track per chamber and applied to the wire hit and to the strip hits of that track together. The strips see the charge induced by the wire avalanche, so a fluctuation of the drift delay moves both by the same amount; the existing independent smearing (FDC_TDRIFT_SIGMA) is kept. No random numbers are drawn when the option is 0. - FDC:CATHODE_SIGMA (microns, default < 0 = CCDB): overrides FDC_CATHODE_SIGMA from /FDC/fdc_parms, which sets the strip pedestal noise, so the strip resolution can be tuned without a CCDB change. - FDC:EFFVSDOCA_PAR0..3: the parameters of the doca-dependent wire efficiency, previously hard-coded; the defaults are the former values. The settings in use are printed at startup. In the GlueX rho0 data/MC study of 2026-10/11 the tuned values were FDC:DRIFT_SMEAR_NS=3, FDC:CATHODE_SIGMA=85 and FDC:USE_EFFVSDOCA=0. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…o the wire hits [rtj] FDC_THRESH_KEV was read from /FDC/fdc_parms but never used by mcsmear, so the only wire threshold in the simulation chain was the one applied by hdgeant4 to each track segment. mcsmear now drops an anode truth hit whose deposited energy is below FDC_THRESH_KEV. The cut acts on the energy that hdgeant4 summed over the segments it combined into the hit, which is closer to what the discriminator sees, and it makes the effective wire threshold adjustable without rerunning hdgeant4 (by giving mcsmear a CCDB in which FDC_THRESH_KEV differs from the one used for hdgeant4). The value in use is printed at startup. No random numbers are drawn by the cut. With the same CCDB for both programs the output is unchanged: every hit that hdgeant4 kept already has dE above the threshold. (In the drift-cluster mode of hdgeant4, DRIFTCLUSTERS != 0, dE holds the integrated pulse instead of a deposited energy, so FDC_THRESH_KEV should be set accordingly there.) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Build status for this pull request: FAILURE Build log: make_fdc_strip_peakamp_rtj.log |
|
I assume it fails the test because of the nightly-build. I'll restart it tomorrow |
|
El mar, 3 de nov de 2026, 4:26 a.m., Alexander Austregesilo <
***@***.***> escribió:
… *aaust* left a comment (JeffersonLab/halld_sim#422)
<#422 (comment)>
I assume it fails the test because of the nightly-build. I'll restart it
tomorrow
—
Reply to this email directly, view it on GitHub
<#422?email_source=notifications&email_token=CIFHIAJ63Z7T7XMRFCNYXGT5XFF6BA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTMNJZGQ3DKNRWHEY2M4TFMFZW63VKON2WE43DOJUWEZLEUVSXMZLOOSWGM33PORSXEX3DNRUWG2Y#issuecomment-6594656691>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/CIFHIAOKZWGV2UM6TIVBVWD5XFF6BAVCNFSNUABFKJSXA33TNF2G64TZHMYTIMRZGI4DGMZRHNEXG43VMU5TMMJXGU3DAMBRGMY2C5QC>
.
Triage notifications, keep track of coding agent tasks and review pull
requests on the go with GitHub Mobile for iOS
<https://github.com/notifications/mobile/ios/CIFHIALUHVZJQBEFUOBNYYD5XFF6BA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTMNJZGQ3DKNRWHEY2M4TFMFZW63VKON2WE43DOJUWEZLEUVSXMZLOOSVGM33PORSXEX3JN5ZQ>
and Android
<https://github.com/notifications/mobile/android/CIFHIAOB2GYNELONEMUPG2D5XFF6DA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTMNJZGQ3DKNRWHEY2M4TFMFZW63VKON2WE43DOJUWEZLEUVSXMZLOOSXGM33PORSXEX3BNZSHE33JMQ>.
Download it today!
You are receiving this because you are subscribed to this thread.Message
ID: ***@***.***>
|
|
Test status for this pull request: SUCCESS Summary: summary.txt Build log: make_fdc_strip_peakamp_rtj.log |
nsjarvis
left a comment
nsjarvis
left a comment
There was a problem hiding this comment.
The new merge code with the peak amp in it looks good.
I don't like what you have done in the smearer with the peak amp and a-scale.
What I think you should do to create the peak amp is this:
Set it equal to q.
Make a separate raw_amp = q/a_scale
Compare raw_amp against the threshold and discard the hit it if is too small.
This will model the effect of changing pressure using the gain factors that have already been provided in a_scale by calibrating the later runs. I can easily calibrate the earlier run periods similarly. We do not need a different a_scale for simulations.
In the CDC smearer we even have code to model the saturation of raw_amp, that might not be necessary for the FDC as its readout has 3 more bits for pulse height than the CDC's readout does.
|
Thanks, Naomi. I agree with modelling the gain through the per-run a_scale, and with cutting on raw_amp = q/a_scale. I would like to keep peakAmp = q/a_scale, though, and I'll explain why below. a_scale already follows the gain. /FDC/digi_scales FDC_ADC_ASCALE has per-run values for 2019-11, entered on 2022-03-22:
The ratio is 0.60. That is exactly the drop we measure in the FDC strip signal between those runs, which is behind the apparent beam-current dependence of our 2019-11 rho(0) cross section. Run 71350 and the September 2020 runs 73223-73227 have no entry and use the default 0.2. From the measured signal, the September runs should be near 0.36. It would be great if you could calibrate those. Why peakAmp should be q/a_scale. For MC, DEventSourceHDDM takes the fdcDigihit peakAmp as DFDCHit::pulse_height. In data, DFDCHit_factory computes q = a_scale × pulse_height, so the data pulse_height is q/a_scale. DFDCPseudo_factory uses the charge built from pulse_height in absolute terms, in three places:
So the MC pulse height has to be on the data scale. These are median pseudopoint charges (DFDCPseudo q) for pion tracks within 0.2 cm of the wire, in the first January 2020 runs:
With peakAmp = q/a_scale the MC pseudopoint charge also follows the January step: relative to the first runs it is 0.62 in MC and 0.60 in data. With peakAmp = q it would stay flat. Proposal:
If you agree, I'll update the PR accordingly. |
|
The MC data is supposed to mimic calibrated data, so its peakAmp should be constant with time. I have no idea what happened to the 2019-11 calibrations. We can see in the Demon plot that dE/dx is not flat. I have new values in /work/halld/njarvis/gluex20/gains/rest1/newgains - I think these should not go into the main ccdb yet, maybe they need a new variation. TBD. No problem using them in sqlite for tests. We should take the thresholds from the experimental data, they can be extracted from the BOR records. This will take a while. Is there anywhere in ccdb to put them? There should be one threshold per channel, I suppose we could cheat by using one threshold for all. |
|
there's not currently a table for per-channel thresholds, but adding one is easy enough |
|
"The MC data is supposed to mimic calibrated data, so its peakAmp should be constant with time." Should the MC response be run-independent and the corrections always applied to real data when running the reconstruction? In principle, but in practice that can get prohibitively expensive. That would mean that we have to rerun all of the reconstruction on 2020 data before we can get a cross section with forward charged tracks that we can believe. Is standing by that principle worth the cost? Or can we move into mcsmear some of the corrections that one would prefer to have had already when the data were reconstructed? The reconstruction is what it is, and the job of the simulation (mcsmear) from this time forward is to produce sims that look like the data did ("calibrated" is time-dependent) at the time the reconstruction was run. I would agree with your statement if it were applied to the output from hdgeant4. Applied to the output from mcsmear, I would question the wisdom of it. |
|
We do plan to re-reconstruct most of the older run periods with the correct geometry anyway. |
|
Actually, I don't think it's necessary to redo the 2019-11 reconstruction just because we changed the FDC gain calibration. As nobody cuts on FDC dE/dx, the only thing that uses it is the PIDFOM. If it was necessary to include new gains that could be done with a scaling factor per run in the analysis trees. While we are improving the FDC smearing, we should do it properly. Using my new calibrations, this should remove simulated hits when the gas gain is too low. Apart from losing hits at small dE/dx, the sim dE/dx should be approximately unchanging throughout the run period, which will match the real properly calibrated dE/dx (although nobody is using this anyway). |
|
For this gain and threshold, it is not the PID that is relevant, it is the tracking itself. We get the tracking efficiency wrong in MC without having this run-dependent inefficiency right. |
New command-line options for the FDC smearing. Every default leaves the
output unchanged.
FDC:GAIN_SCALE gas-gain factor on the strip charge q and on the
anode wire dE, applied before the thresholds (1)
FDC:GAIN_PKG1..4 extra gain factor per FDC package (modules 1-2,
3-4, 5-6, 7-8), times FDC:GAIN_SCALE (1)
FDC:GAIN_SIGMA s avalanche gain fluctuation: one log-normal factor
exp(s G - s^2/2) per chamber and track, shared by
the wire hit and the strips it induces (0)
FDC:DOCA_ATTEN a wire dE and the track's strip q in the chamber
scaled by max(0.05, 1 - a doca^2), doca in cm (0)
FDC:STRIP_THRESHOLD strip charge threshold in pC; < 0 keeps
FDC_THRESHOLD_FACTOR * FDC_PED_NOISE from CCDB (-1)
FDC:WIRE_THRESHOLD_KEV anode threshold on the gain-scaled wire dE in keV;
< 0 keeps FDC_THRESH_KEV from CCDB (-1)
Motivation: the FDC gas gain of the real detector changes with time and
differs between packages (e.g. a 40% drop of the strip signal between
2019-11 runs 71362 and 71483), which changes the FDC hit efficiency and
with it the tracking efficiency. With per-period gains fitted only to the
data's FDC plane-used fractions (not to any cross section), the rho(0)
photoproduction cross section of ten run segments (2018-01 to 2023-01)
agrees to about 2% (RMS, beyond statistics).
Builds on the FDC options and peakAmp of PR #422; independent of #423.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…rtj] New option FDC:RAW_AMP_THRESHOLD (ADC counts; < 0 = off, the default): a smeared FDC strip hit is dropped if raw_amp = q / FDC_ADC_ASCALE is below the threshold. With the per-run a_scale of /FDC/digi_scales, raw_amp is the fADC125 amplitude at that run's gain, so a gain drop that is calibrated into a_scale removes small strip signals in MC as the readout threshold does in the data (as suggested in the review of this PR). The fdcDigihit peakAmp is unchanged (q / FDC_ADC_ASCALE). The default leaves the output unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
I pushed two commits to this PR. The new options all default to off, and with the defaults the output is byte-identical to before.
I kept peakAmp = q / a_scale. The reconstruction reads the MC fdcDigihit peakAmp as DFDCHit::pulse_height, and q as DFDCHit::q, and it uses the two for different things:
In data, q = a_scale × pulse_height. Carrying both lets the MC reproduce that relation without any change to halld_recon. With these options, ten run segments from 2018-01 to 2023-01 give a rho0 cross section that agrees to about 2% (RMS beyond statistics). The gains were fitted only to the FDC plane-used fractions of the data, not to the cross section. These are the command-line values used:
All segments: Run-dependent constants belong in CCDB rather than on the command line. I am open to a friendly amendment that puts them there before this PR is merged. |
|
Test status for this pull request: SUCCESS Summary: summary.txt Build log: make_fdc_strip_peakamp_rtj.log |
No description provided.