September 15 plasma run, plasma current and channel noise

The evening produced one real discharge, 18:54:14 to 19:10:10, driven by hand in millivolts with the Kikusui in current mode. This page compares the Hall sensor against the supply, measures the noise on every analogue channel, follows the ignition staircase, and records what each instrument reads once it is switched off. All times are JST from the rig's own clock.

Files analysed

Read-only copies in the repository's git-ignored data/rig/2026-09-15/. Nothing on the rig was touched.

File	Rows	Span	SHA-256
cu_20260915_184419.csv	24,708	18:44:19.339218–19:25:31.373103	252d1c1181e77f00fc1a9356cb313c90895a9d019aa984f2202cb5d45549ad26
kikusui_20260915_184419.csv	3,472	18:44:19.222562–19:25:28.223277	635b8b8471b2be285d4bc77ddaaeeb4114dedf5a1a1faf67f2937ed06c649e3b
cu_20260915_170626.csv	47,075	17:06:26.296854–18:24:56.336935	424309599a0b36c25127cfb7cef3db57a3fa65312dc64647fe19c9830f51dfdb
kikusui_20260915_170626.csv	9,400	17:06:26–18:24:56	bf7f4e881855d2687263683a74c493c9d704911272ef37cbc047b64acdfb880a
cu_20260915_161413.csv	1,828	16:14:13.901120–16:17:16.703378	3b6b9a2a88398d87c5a85fecd7ac81316cd4c03030e44eabac885051c4e87815
cu_20260915_192533.csv	214	19:25:34.162261–19:37:18.647529	48a5b439d2d3bd1dfaeac4e2b7bbaf427939ca630005a152e242ae396944f673
kikusui_20260915_192533.csv	142	19:25:35–19:37:15	e42bb0d271235b4a28cb9da4dd94879625e8774964c187690ae6a64c040092b7
controlunit.log	11,662	2023-06-02–2026-09-15	373d06b4d4742dc4a37d89482c3431c4e47ad2669b807434be2ee46856b677b3
controlunit.stderr.log	627	—	4be238de27068d4d348d7140b6a0251d3810993475497109ae586187067bd261

cu_20260915_192533.csv was still being written when it was copied.

Plasma current against the supply

Ip_c is not zeroed in the file. The minute before the cathode drive rose, 18:51:18–18:52:18, gives median -0.330102 A over 600 rows (sd 0.024100 A). That zero is stable: 18:44:20–18:52:00 gives -0.329719 A and 19:12:00–19:25:00, long after everything was off, gives -0.334313 A. Every "zeroed" number below subtracts -0.330102 A.

The 19:05 web reading reproduces from the files:

Web UI, 19:05	From the files	Source
Ip 0.778 A zeroed	0.7710 A (Ip_c 0.440906 less the baseline)	ADC, 19:05:00–19:05:10 medians
Bu 3.39e-3 Torr	3.403e-3 Torr (Bu_c 3.6675e-3 less the 18:48:38 baseline 2.6415e-4)	ADC
Cathode V 7.792 V	7.6057–7.8185 V over that 10 s	kikusui_…184419.csv
Cathode I 16.537 A	16.2392–16.6089 A over that 10 s	kikusui_…184419.csv

So the web's "Cathode V/I" is the Kikusui, and at 19:05 the supply read 16.5 A, not 1 A. The Kikusui current is the filament heating current; Ip is the Hall sensor on the discharge path. They are different currents, and the plot below compares them only to show that they are not coupled.

Nothing in these files holds a 1 A reading. The 2026-09-15 event log contains no plasma current N A request, and plasma_target_a is 0.0 in all 3,472 sidecar rows, so it is not a PID target either. The most plausible reading of the owner's "1 A by the PSU" is the front panel of the separate discharge supply, which ControlUnit does not log at all. That cannot be confirmed from the data; it can only be said that the 1 A is in no file, and that the comparison the owner has in mind is between two instruments on the discharge path of which only the Hall sensor is recorded.

A straight fit over every output-on row (n = 11,647) gives Ip = 0.04805·Ic - 0.04922, r = 0.864. That correlation is entirely the contrast between "supply off, no plasma" and "supply on, plasma running"; it is not a transfer function. Inside the settled discharge, 18:54:20–19:10:05 (n = 9,546), the filament current moves over 16.189–16.727 A (sd 0.161 A) and the fit collapses to Ip = 0.00129·Ic + 0.76005, r = 0.006. Over the same window Ip zeroed is 0.7814 A mean, 0.7898 A median, sd 0.0349 A.

The relation is a threshold, not a scale or an offset. At 18:53:30 the filament already carried 15.45 A at 6.28 V and Ip zeroed was 0.081 A. The per-minute ratio Ip/Ic sits between 0.0473 and 0.0482 from 18:54 to 19:09, constant to 1 %, but only because both quantities are separately constant.

Noise

Windows: no plasma 18:45:00–18:52:00 (4,200 samples) and plasma 18:55:00–19:10:00 (9,000 samples), both from cu_20260915_184419.csv at 10 Hz, resampled onto an exact 0.1 s grid, detrended and Hann-windowed for the spectra. cu_20260915_161413.csv (16:14:13–16:17:16, 1,828 rows) served as an independent quiet check.

Channel	Window	Mean	sd	sd / mean	sd after removing a 2 s running mean
Ip_c	no plasma	-0.33676 A	0.01818 A	5.40 %	0.01498 A
Ip_c	plasma	0.45324 A	0.02093 A	4.62 %	0.01427 A
Bu_c	no plasma	5.997e-4 Torr	9.553e-4 Torr	(gas rising)	1.218e-5 Torr
Bu_c	plasma	3.7207e-3 Torr	6.714e-5 Torr	1.80 %	6.592e-6 Torr
Bd_c	no plasma	-1.0815e-3 Torr	4.348e-4 Torr	40.2 %	1.795e-5 Torr
Bd_c	plasma	-8.7228e-4 Torr	1.706e-4 Torr	19.6 %	1.574e-5 Torr
Pu_c	no plasma	1.0139e-4 Torr	2.720e-4 Torr	(gas rising)	3.197e-6 Torr
Pu_c	plasma	1.0875e-3 Torr	2.086e-5 Torr	1.92 %	2.317e-6 Torr

The plasma adds almost nothing to the noise on Ip: 0.0182 A with no plasma, 0.0209 A with 0.79 A flowing, 0.0196 A after everything was switched off (19:12:30–19:25:00), 0.0224 A in the unrelated 16:14 file. Roughly 2 % of the discharge current, and it is instrument noise, not discharge fluctuation.

Autocorrelation of the detrended series, lags 1 to 10:

Series	1	2	3	4	5	6	7	8	9	10
Ip_c, no plasma	0.221	0.151	0.716	0.110	0.067	0.604	0.022	-0.011	0.516	-0.036
Ip_c, plasma	0.684	0.572	0.638	0.460	0.399	0.449	0.284	0.236	0.284	0.136
Ip_c, 16:14 file	-0.002	-0.027	0.729	-0.058	-0.082	0.647	-0.110	-0.125	0.602	-0.150
Bu_c, plasma	0.961	0.945	0.946	0.945	0.943	0.943	0.940	0.940	0.939	0.938
Bd_c, plasma	0.998	0.996	0.995	0.993	0.990	0.987	0.985	0.983	0.980	0.980
Pu_c, plasma	0.992	0.986	0.985	0.984	0.984	0.983	0.983	0.983	0.982	0.982

Two lines stand out on Ip_c, and only on Ip_c.

• 0.1333 Hz, a 7.5 s period, amplitude 0.0127 A with no plasma and 0.0176 A during the discharge, with harmonics at 0.267 Hz (0.0101 A) and 0.400 Hz (0.0090 A). It is present before ignition, during the discharge, after the cathode drive was off, and in the 16:14 file recorded before the Kikusui was even configured. It is not the plasma and it is not the telemetry poll.
• 3.3333 Hz, exactly one third of the 10 Hz sampling rate, amplitude 0.0076 A with no plasma, 0.0027 A during the discharge, and 0.0198 A in the 16:14 file. The lag-3, lag-6 and lag-9 autocorrelations above are its signature, with lags 1 and 2 near zero. Landing on fs/3 makes an acquisition artefact more likely than a physical 3.33 Hz oscillation, but the files cannot settle that; it needs a run at a different sampling rate.

No mains-related line was found. At 10 Hz both 50 and 60 Hz alias to DC and would appear only as a shifted baseline, which is consistent with the stable -0.330 A zero.

Bd_c carries a periodic wander at 0.00917 Hz (109 s) with a harmonic at 0.0179 Hz (56 s), amplitude 1.47e-4 Torr, peak-to-peak 1.198e-3 Torr over 18:45–19:25. It is on the channel with the plasma on and off. A heated Baratron head's temperature controller would produce something like this; that is a hypothesis, not a finding. Bu_c and Pu_c are quiet above 0.5 Hz, with median amplitudes of 1.1e-7 and 3.3e-8 Torr respectively during the discharge.

Filters

Measured on the same two windows. Group delay of a causal boxcar of length N is (N-1)/2 samples.

Filter on Ip_c	Delay	sd, no plasma	sd, plasma	0.133 Hz line left	3.333 Hz line left
none	0	0.0182 A	0.0209 A	0.0176 A	0.0027 A
boxcar N=3	0.10 s	0.0124 A	0.0183 A	0.0176 A	0.00000 A
boxcar N=5	0.20 s	0.0120 A	0.0175 A	0.0175 A	0.0005 A
boxcar N=11	0.50 s	0.0105 A	0.0156 A	0.0170 A	0.0003 A
boxcar N=21	1.00 s	0.0087 A	0.0127 A	0.0154 A	0.0000 A
boxcar N=75	3.70 s	0.0019 A	0.0037 A	0.0000 A	0.0000 A
1-pole, tau 0.5 s	—	0.0100 A	0.0148 A	0.0163 A	0.0009 A
1-pole, tau 1.0 s	—	0.0079 A	0.0117 A	0.0136 A	0.0004 A
median N=3	0.10 s	0.0146 A	0.0187 A	—	0.0005 A

Per channel, at least cost in delay:

• Ip — a 3-sample boxcar costs 0.1 s and puts an exact zero on the fs/3 line; a 3-sample median does not (0.0005 A left) and smooths worse (0.0187 A against 0.0183 A). For a displayed or control value, cascade it with a 21-sample boxcar: 1.0 s total delay for 0.0127 A, 1.6 % of a 0.79 A discharge. Where a settled number matters more than latency, a 75-sample boxcar is exactly one period of the 7.5 s wander and nulls it outright: 0.0037 A at 3.7 s delay, a factor 5.7 better than raw. Nothing between N=21 and N=75 helps much, because what remains is that one line.
• Bu, Pu — already quiet. Removing a 2 s running mean leaves 6.6e-6 Torr on Bu and 2.3e-6 Torr on Pu, so a 1 s boxcar (N=11, 0.5 s delay) is enough and further smoothing only buys drift. Their whole-window sd is real pressure change, not noise: lag-1 autocorrelation 0.96 and 0.99.
• Bd — filtering is the wrong tool. Its sd is the 109 s wander plus a standing offset of about -0.8 mTorr; no filter short enough to be useful touches a 109 s period. Null the offset and find the source first.
• Pd, Pu2 — see below; on this evening neither was a live measurement for most of the run, so no filter recommendation is meaningful.

Cathode drive, gas and the staircase

PresetV_cathode is 0 in all 24,708 rows, as on 2026-09-14. The manual drive is carried only by commanded_cathode_mv in the Kikusui sidecar, which agrees with the event log exactly: 1000 mV at 18:52:18, 1800 mV at 18:52:42, off at 18:52:59, 1800 mV at 18:53:26, 1850 mV at 18:53:42, 1900 mV at 18:53:58, off at 19:10:10.

PresetV_mfc1 and PresetV_mfc2 are 0.0 in all 24,708 rows too. The gas was opened outside ControlUnit: Bu_c rose from 2.641e-4 to 3.798e-3 Torr and Pu_c from 1.229e-5 to 1.141e-3 Torr between 18:50 and 18:52, with no MFC command in the log. It was closed at 19:10:16–19:10:20, six seconds after the cathode drive went off: Bu_c 3.744e-3 to 7.255e-4 Torr and Pu_c 1.184e-3 to 1.476e-5 Torr inside one 5 s bin.

5 s medians. Ip is zeroed; V, I and P are the Kikusui.

Time	drive	Ip	Ic	V	V·I	Bu
18:52:20	1000 mV	-0.006 A	0.011 A	-0.005 V	0.0 W	3.798e-3 Torr
18:52:25	1000 mV	0.007 A	8.647 A	1.210 V	10.5 W	3.805e-3 Torr
18:52:35	1000 mV	-0.016 A	8.669 A	2.311 V	20.1 W	3.817e-3 Torr
18:52:40	1800 mV	0.017 A	15.744 A	4.507 V	71.1 W	3.821e-3 Torr
18:52:50	1800 mV	0.101 A	15.788 A	6.128 V	96.8 W	3.790e-3 Torr
18:52:55	1800 mV	0.122 A	15.777 A	6.125 V	96.9 W	3.786e-3 Torr
18:53:00	off	0.000 A	0.019 A	-0.001 V	0.0 W	3.817e-3 Torr
18:53:25	1800 mV	0.004 A	15.470 A	3.379 V	52.7 W	3.828e-3 Torr
18:53:35	1800 mV	0.104 A	15.401 A	6.264 V	96.7 W	3.815e-3 Torr
18:53:40	1850 mV	0.149 A	15.813 A	6.508 V	103.2 W	3.817e-3 Torr
18:53:55	1850 mV	0.260 A	16.136 A	6.634 V	106.8 W	3.817e-3 Torr
18:54:00	1900 mV	0.377 A	16.577 A	6.982 V	115.6 W	3.819e-3 Torr
18:54:10	1900 mV	0.415 A	16.653 A	6.929 V	115.5 W	3.817e-3 Torr
18:54:15	1900 mV	0.791 A	16.670 A	6.944 V	115.7 W	3.828e-3 Torr
18:55:00	1900 mV	0.789 A	16.416 A	7.199 V	118.2 W	3.828e-3 Torr

Ip followed filament power, not the DAC command and not the gas. From 97 W to 116 W, a 20 % rise in power, took Ip from 0.10 A to 0.79 A, a factor of eight. Bu moved by 0.7 % across the whole staircase, from 3.798e-3 to 3.828e-3 Torr, so the gas explains none of it.

Ignition itself took two samples. The last pre-ignition row is 18:54:14.252300 at 0.4039 A; 18:54:14.355946 reads 0.7293 A and 18:54:14.455239 reads 0.7997 A. The cathode drive had been at 1900 mV for 16 s by then and did not change.

After ignition the discharge was insensitive to further filament heating. With the drive held at 1900 mV, the Kikusui voltage drifted from 6.93 V to 7.79 V and the power from 115.5 W to 128.6 W, +11 %, as V/I rose from 0.4254 Ω at 18:54:20 to 0.4656 Ω at 19:09 — the filament warming through. Ip over the same fifteen minutes went from 0.790 A to 0.786 A.

For a cold baseline: during 17:25–17:37 in the earlier run the drive sat at 599 mV, giving 1.0699 V, 5.0862 A, 0.2100 Ω and 5.385 W. Hot-to-cold resistance ratio 2.2. That stretch had no gas (Bu_c 2.79e-4 Torr) and Ip_c never left its baseline, so 17:06–18:24 was a filament conditioning, not a plasma attempt.

What "off" looks like

Kikusui. The drive went to 0 mV at 19:10:10 and the supply's own output stayed enabled for another 98 s: from 19:10:11 to 19:11:47 output_on is 1 with voltage_v -0.0053 to -0.0060 V and current_a 0.0083 to 0.0090 A. At 19:11:48.076632 output_on becomes 0 with the same V and I. At 19:11:58.086334 the supply loses power and every remaining row reads status=unavailable, error=timed out, query_ms about 1001 ms, and empty voltage_v, current_a, output_on and identity — 163 such rows to 19:25:28.223277, retried every 5 s, matching Kikusui telemetry LOST in the log at 19:11:58. All 142 rows of kikusui_20260915_192533.csv are the same. While powered the link was excellent: 3,309 ok rows, query_ms median 4.80, p99 25.85, max 48.51.

That is the one instrument that self-reports. The analogue channels do not.

Ion gauges. Pu2's controller was switched off at about 18:49:55: its raw voltage fell from 3.212 V to 0.0041 V within five seconds and settled at 0.00031–0.00092 V, which at range -6 converts to Pu2_c = 3.06e-10 Torr — a perfectly plausible-looking ultra-high vacuum. It was switched back on at 19:12:06.984792 (raw 1.93 V), with the range moved -6 to -5 to -4 inside 0.2 s and back to -5 at 19:15:39.809981.

Pd is the harder case. Its raw voltage ran from 1.236 V at 18:44 down to 1.0385 V at 19:25 and 1.0565 V at 19:37, drifting monotonically and never responding to the gas that moved Pu by 1.17 V and Bu by 35.6 mV. At range -8 that converts to 1.04e-8 to 1.24e-8 Torr for the whole evening. It looks like a clean base-pressure reading and it is not a measurement of a chamber that was at 3.8e-3 Torr.

Correction, the same night. queezz, shown these readings as off-state numbers: "Imagining things. IGs are on." Both ion gauge controllers were on all evening. So the two paragraphs above over-read the file: Pd's steady 1e-8 Torr is a live downstream reading — downstream of the pumping stage a gauge holding 10⁻⁸ Torr while the upstream side sits at 10⁻³ is the apparatus working, not a dead channel — and the Pu2 raw-voltage drop at 18:49:55 is real but is not a controller switched off; what the controller was doing then (emission, range, degas) is not in these files. The point of the section stands the other way round: the file alone cannot tell an instrument that is off from one that is on, in either direction, so the flag has to come from the operator or the controller.

Baratrons. Both controllers stayed on. Bu tracked the gas correctly. Bd sat at -8.7e-4 Torr throughout, a negative pressure, that is, a standing zero offset of about -0.8 mTorr on a 0.1 Torr head, and it did not respond to the gas either.

The 19:25 vacuum-only file, 19:26:00–19:37:18 medians: Ip_c -0.32230 A, Bu_c 3.1832e-4 Torr, Bd_c -7.4774e-4 Torr, Pu_c 1.2323e-5 Torr, Pd_c 1.0374e-8 Torr, Pu2_c 2.2194e-5 Torr.

The consequence for a future "instrument off" flag: no converted value from an off instrument looks invalid, and the off signatures differ between instruments — Pu2 off is under 0.01 V, Pd off is 1.0 to 1.2 V and drifting, Bd off (if it is off) is -0.08 V. A single voltage threshold cannot carry the flag. It has to be operator-set in the UI or read from the controller, and the file schema needs somewhere to record it, exactly as status does for the Kikusui.

Timing and other findings

• The 18:44 run kept cadence. 24,708 rows, median interval 0.099950 s, mean 0.100050 s, p99 0.1213 s. Exactly one interval above 0.3 s: 0.347 s at 19:05:09.043266. Nothing in the log at that time.
• The 17:06 run had one real fault. 47,075 rows, one interval above 0.3 s: 0.712 s at 17:57:01.831070, matching the logged TimeoutError('ADS1115 at 0x49 never finished a conversion') at 17:57:01, recovered after one failed read. Different from the two OSError(121, Remote I/O error) faults of 2026-09-14.
• The 19:25 file's 69 gaps above 0.3 s are deliberate: sampling was set to 10 s at 19:25:48, so the file runs at 0.1 Hz from 19:25:58 onward.
• No tracebacks. controlunit.stderr.log contains zero occurrences of Traceback; its only 2026-09-15 content is six process-start markers, at 16:14:01, 17:04:37, 18:29:50, 18:32:02, 18:35:49 and 18:42:54.
• Bu went electrically noisy during shutdown, 19:20:50–19:23:50: sd rose from 7.5e-6 to as much as 1.36e-4 Torr, with excursions to -2.87e-4 and +8.27e-4 Torr, that is, negative pressures. Nothing in the event log at those times; Ip and Pu were unaffected. Worth watching the next time instruments are switched off with acquisition running.
• Nine ADC files for one day. Four short files between 18:30 and 18:44 came from repeated app restarts while the gauge ranges were being set. Each restart starts a new CSV and a new Kikusui sidecar.

What the data cannot answer

The discharge supply is not logged, so the owner's 1 A cannot be checked against the 0.78 A the Hall sensor recorded; that needs the second supply on the record. Whether the 3.333 Hz line on Ip is an artefact of the read schedule needs one quiet run at a different sampling rate. Whether Pd was switched off or merely unresponsive, and what drives the 109 s cycle on Bd, both need a deliberate check at the instruments.