Good triggered pulses on Detector 0 / Channel 0 (series 07221203_2025, dumps 2–5)

Note 5 · Author(s): Anthony Villano · 22 Sept 2026 · Status: In progress

Code Repository

07221203_2025_dump2_pulse.ipynb — the notebook behind this note. Pinned to c3d6754.
python/ — the library it uses (Note 1a). Pinned to c3d6754.

Goal

Find several clean, clearly triggered, comparably sized pulses in series 07221203_2025 and overlay them to see where the pulse onset falls inside the 4096-sample window. Dumps F0002–F0005 (8961, 8580, 8724 and 8661 channel groups) are pooled; restricted to Detector 0, Channel 0 this gives 11,642 traces.

The cut

Tightened relative to the single-dump version on every axis:

bstd below its own 10th percentile, pooled across the four dumps (bstd < 1.70): a much quieter pretrigger baseline, to keep out the contamination described in Note 2;
dev (maximum excursion) below 20 ADC counts, down from 100, to get small, comparably sized pulses;
ratio > 9, raised from 8 after looking at the plot: at 8 most examples showed no visible pulse, because some quiet traces cross a loose ratio threshold by extreme-value chance alone. Going to 9 cut the candidates from 29 to 10.

Result: 10 of 11,642 traces.

Ten selected traces from dumps F0002 to F0005, baseline-subtracted, with the end of the pretrigger window marked — The 10 selected traces, baseline-subtracted; dashed line at sample 1000 (end of the pretrigger window).

Check made while writing this note: most of the 10 are not pulses

Finding. Only 3 of the 10 selected traces contain a visible pulse. The other 7 pass the cut because their baseline drifts downward by roughly 10–17 counts across the post-pretrigger window: for those traces dev (14–17.4) is essentially the drift, and after removing a linear trend nothing larger than ~2.5 counts remains (noise level ~0.25 counts in a 40-sample average).

Trace	`dev`	`ratio`	Visible pulse?	Approximate onset
F0005:10881	20.0	12.1	yes	sample ~1650
F0005:11426	16.0	10.7	yes	sample ~3600
F0004:11416	14.7	10.2	yes	sample ~3900 (still rising at the end of the trace)
the other 7	14.4–17.4	9.1–10.7	no (drift)	—

Among the three real pulses the onset is different every time, all well past the nominal pretrigger/post-trigger boundary at sample 1000. That is consistent with there being no fixed “trigger sample” for this trigger configuration, which would matter for optimal filtering: Golwala’s thesis (Appendix B, section B.4) assumes a known template start time, so a time-offset search rather than a fixed-position fit would be needed. Three pulses is thin evidence for that, though.

Discrepancy with the notebook. The notebook’s closing “Observation” cell says pulse onsets land “anywhere from roughly sample 1150 to 3050”. That is not supported by the figure above or by the check in this note (the 1150–1860 “rise samples” that a half-maximum estimate gives for the drifting traces are drift crossings, not pulse onsets). The notebook text has not been changed yet.

Status / Next Steps

Fix the notebook’s observation cell to match the check above.
The cut needs a term that rejects slow drift; dev and ratio alone cannot tell a drifting baseline from a pulse. A drift or slope quantity, or requiring a localized excursion, would be candidates.
With that in place, redo the pulse selection to get more than three genuine pulses before drawing conclusions about onset time.
The ratio > 9 threshold was chosen by eye.
The cut now calls ratio_band (renamed from excursion_band on 2026-09-23; see Note 4) and still selects the same 10 traces.