Check What a Cathode Capacitor Does to a TLV431A Reference
A voltmeter can show about 1.24 V while the cathode of a shunt reference repeatedly moves above and below that value. Build a low-current fixture with a Texas Instruments (TI) TLV431AIDBZR. Compare a probe-only cathode, a 100 nF capacitor directly across the device and the same capacitor with a series resistor. A second probe-only capture checks whether the fixture returned to its starting behavior.
A shunt reference regulates by sinking current from a node toward ground. Its internal amplifier adjusts that current as the voltage changes. Connecting the reference input to the cathode closes the feedback loop at unity gain, so the cathode sits near the reference voltage in steady regulation. A load capacitor changes how that loop responds to fast changes. This experiment uses the oscilloscope to look for recurring ripple on one assembly. The blank worksheet contains no bench results.
Choose the exact device and read the stability plot
Use TI TLV431AIDBZR, package DBZ, a three-pin small-outline transistor package (SOT-23). Use a passive adapter with confirmed pin numbering. The top view in TI data sheet SLVS139Z, revision Z from June 2024, places reference pin 1 and cathode pin 2 on the two-pin side. Anode pin 3 is on the opposite side. The revision updates pinout diagrams, so compare the package drawing and the pin-function table before wiring. The exact product listing and technical documentation were checked for separate electrical errata. The checked sources point to revision Z and list no separate electrical errata for this orderable.
The recommended cathode voltage in regulation is the reference voltage through 6 V, with 0.1–15 mA cathode current. The I grade is characterized from −40 to 85 °C. The A grade reference at 25 °C is 1.228–1.252 V under the stated 10 mA test condition. This room-temperature experiment uses lower bias, so that reference range is a useful comparison under its own test conditions. The 7 V and ±20 mA figures in the absolute maximum table are stress ratings. Keep this assembly far below them.
Figure 5-18 is specific to the TLV431A and shows typical stability boundary conditions at 25 °C. For the curve labelled VKA = VREF, TI uses a 150 Ω feed resistor and the reference pin tied to the cathode. Around 0.1 µF and a few milliamperes, that curve identifies conditions that can produce oscillation. Figure 5-19 gives typical phase-margin trends at approximately 1.25 V cathode bias. Phase margin describes how far a feedback loop is from reinforcing a disturbance at the frequency where its loop gain reaches one. These curves guide the experiment. Unit variation, the actual capacitance, the capacitor's losses and wiring can change what you see.
Build a fixed-bias fixture
| Node | Connections |
|---|---|
| VIN | Isolated supply channel 1 positive, one end of RF, oscilloscope channel 1 tip. |
| K | Other end of RF, reference pin 1, cathode pin 2, scope channel 2 tip, first J1 contact. |
| Capacitor branch | Second J1 contact to one lead of CL. The other lead of CL goes to G0. Leave J1 open, short it with a wire link, or insert 1.00 kΩ. |
| G0 | Supply negative, anode pin 3, the return lead of CL, and both scope probe ground springs. |
Use a SIGLENT SPD3303X adjustable output in independent mode, set to 1.80 V with a 30 mA current-limit setting. Use a 150 Ω, 1%, ¼ W metal-film feed resistor RF and a 1.00 kΩ, 1%, ¼ W series resistor for the capacitor branch. CL is a 100 nF, 63 V, ±10% nonpolar film capacitor, such as KEMET R82DC3100DQ50K. Mount the chip and both J1 contacts close together on a compact soldered board, with short return paths. Leave out other loads or output bypasses. The cathode still sees the probe and wiring capacitance when J1 is open.
At 1.80 V supply and 1.24 V cathode voltage, the feed current is (1.80 − 1.24) V ÷ 150 Ω = 3.73 mA. At steady direct current (DC) the capacitive branch draws negligible current apart from leakage. The same current mostly goes to the tied cathode/reference node. It includes reference-pin current and the measuring probe's small DC load. With a 1.85 V source and the minimum 148.5 Ω feed resistor, even a shorted K node would draw at most 12.46 mA from that steady ideal source through RF. That bound is about 23 mW in RF. It is a circuit estimate for the specified supply and resistor, separate from measuring a transient or the regulator's own peak current.
For a capacitor in series with a resistor, the resistor limits how strongly that branch loads the cathode at high frequency. A simple resistor-capacitor (RC) estimate for 1.00 kΩ and 100 nF gives a corner near 1 ÷ (2π × 1000 Ω × 100 nF) = 1.59 kHz, where the capacitor's reactance equals the resistor. That corner describes the added branch. The shunt's actual feedback loop includes other dynamics. The resistor is an experimental comparison, with its suitability for another circuit requiring a fresh stability check.
Take four comparable oscilloscope records
- With the supply output disabled, verify the adapter orientation, the short between pins 1 and 2, and the common return. Use a RIGOL DS1054Z oscilloscope with two compensated PVP3150 probes at 10× and matching menu settings. Use DC coupling, normal sampling and short ground springs. Put channel 1 at VIN and channel 2 at K. Both probe grounds must go to G0. An earth-referenced probe ground on K would short the reference node.
- Leave J1 open and enable the supply. Confirm approximately 1.80 V at VIN and 1.24 V at K. Set the vertical position and scale on each channel to resolve a few millivolts of variation without clipping either trace. The software uses an 8 mV peak-to-peak ceiling for quiet control captures. If you cannot establish a quiet control with both channels DC-coupled, improve the probing and capture resolution before interpreting the comparison. Keep the scale, channel offset, bandwidth limit and probe location the same for all four files. Inspect a longer view first to verify the capture is taken after power-up has settled.
- Save
probe-before.csv. Then disable the source and confirm the cathode and capacitor are discharged. In the open position, K and the free capacitor contact are separate. Verify both near zero relative to G0 before touching J1. If the capacitor retains charge with the output off, discharge its free contact to G0 through a 10 kΩ resistor and remove that resistor before the next capture. - Fit a wire link across J1 so CL is directly across K and G0. Restore power, wait at least one second, check the longer view, then save
direct-100nf.csv. Power off and verify discharge again. Replace the link with 1.00 kΩ, keep the same capacitor and probes, then repeat forseries-1k-100nf.csv. - Power off and verify discharge, then open J1 and record
probe-after.csvat the original bias. Use the blank worksheet for scale, time span, file and configuration notes. If the probe-only repeat drifts, check the return and any change in scope settings before running the comparison again.
Save both channels from one stopped acquisition each time. In waveform comma-separated values (CSV) storage choose Memory as the data source and select CH1 and CH2. A Screen export can contain too few samples. With two channels active, start at 100 µs per division. That covers about 1.20 ms across twelve divisions. Use an available native depth that provides an exported sample interval no greater than 0.20 µs, such as 6 kpts if the file verifies that spacing. Retain the original Memory export. The program accepts files up to 1.2 MB. If the cropped file is too large, try a suitable lower native memory depth and recheck spacing. Reducing saved samples offline requires a documented bandwidth limit or appropriate anti-alias filtering before the reduction. Save any filtered copy separately and record that the samples were modified. For the analyzer, crop a settled 1 ms interval and normalize its first timestamp to 0 µs. Keep the simultaneous samples in common rows. Convert seconds to microseconds if necessary and voltage values to volts. The exact header is time_us,supply_v,cathode_v. The capture template has these column names. Use an interval between adjacent timestamps no greater than 0.20 µs and check the export retains the sample timing. If clipping or an unexpectedly large oscillation forces a scale change, retake the four conditions at the new common setting. The chosen sample interval resolves the analyzer's frequency band. A scope with a wider time or frequency view can reveal changes outside that band.
Separate recurring ripple from a disturbed capture
The Python analyzer looks at 100–400 µs and 600–900 µs in each capture. It checks the 1.75–1.85 V input median and 1.15–1.33 V cathode median, requires no more than 10 mV peak-to-peak input ripple in either window, and uses the probe-only runs as noise and bias checks. Here peak-to-peak means the largest sample minus the smallest. Across all probe-only windows it requires the largest cathode span at most 8 mV and the spread of their medians at most 20 mV. These are experiment screens for the captured voltages.
For a qualified set, the cathode amplitude threshold is the larger of 20 mV peak-to-peak and four times the largest probe-only cathode span. In each window the detector uses the midpoint of its extreme cathode samples and looks for repeated low-to-high excursions between two levels one-quarter threshold below and above that midpoint. It requires at least four periods between rising crossings, a median period between 2 and 50 µs (20–500 kHz), and individual periods within 15% of that window's median. The two window median periods must agree within 15% of the shorter one.
A large excursion with inconsistent periods is reported as ripple with unresolved periodicity. A trace below the amplitude screen leaves open the possibility of smaller or out-of-band activity. Recurrence in both windows supports a repeatable feature in that acquisition. If the same frequency appears strongly on VIN or follows a changed ground connection, revisit the setup before attributing it to the shunt loop. Repeat the four-condition comparison to check whether the result persists. For a separate reference experiment focused on available DC current, see the LM4040 shunt current budget.
Run the complete analyzer
Download the experiment ZIP or the Python program. With Python 3.10 or newer, run python3 shunt_ripple.py probe-before.csv direct-100nf.csv series-1k-100nf.csv probe-after.csv. Exit status 2 means the capture format or range failed. Exit status 3 provides window summaries and identifies which fixture control failed. The included synthetic examples are constructed for code testing.
"""Screen synchronized TLV431A supply and cathode captures for recurring ripple."""
import argparse
import csv
import json
import math
from pathlib import Path
from statistics import median
import sys
HEADER = ("time_us", "supply_v", "cathode_v")
LABELS = ("probe_before", "direct_100_nf", "series_1k_100_nf", "probe_after")
WINDOWS = ((100.0, 400.0), (600.0, 900.0))
EPS = 1e-9
def read_capture(path):
if path.stat().st_size > 1_200_000:
raise ValueError("capture exceeds 1.2 MB")
with path.open(encoding="utf-8-sig", newline="") as stream:
rows = csv.reader(stream, strict=True)
if tuple(next(rows, ())) != HEADER:
raise ValueError("expected time_us,supply_v,cathode_v")
points = []
for row in rows:
if len(row) != 3:
raise ValueError(f"row {rows.line_num}: expected three fields")
try:
time, supply, cathode = (float(field) for field in row)
except ValueError:
raise ValueError(f"row {rows.line_num}: expected numeric readings") from None
if not all(math.isfinite(value) for value in (time, supply, cathode)):
raise ValueError(f"row {rows.line_num}: nonfinite reading")
if not (-EPS <= time <= 1050 + EPS and 1.5 <= supply <= 2.0 and 0.5 <= cathode <= 1.8):
raise ValueError(f"row {rows.line_num}: outside time or voltage window")
if points and not 0 < time - points[-1][0] <= 0.20 + EPS:
raise ValueError(f"row {rows.line_num}: require increasing times and gaps ≤0.20 µs")
points.append((time, supply, cathode))
if len(points) < 2 or points[0][0] > 0.20 + EPS or points[-1][0] < 999.80 - EPS:
raise ValueError("capture must cover approximately 0 to 1000 µs")
return points
def excursion(points, midpoint, threshold):
signal = [(time, cathode - midpoint) for time, _, cathode in points]
crossings = []
armed = False
level = threshold / 4
for time, value in signal:
if value <= -level + EPS:
armed = True
if armed and value >= level - EPS:
crossings.append(time)
armed = False
periods = [second - first for first, second in zip(crossings, crossings[1:])]
result = {"crossings": len(crossings), "period_us": None, "frequency_hz": None}
if len(periods) < 4:
return dict(result, status="fewer_than_four_periods")
period = median(periods)
if not 2.0 - EPS <= period <= 50.0 + EPS:
return dict(result, status="period_outside_2_to_50_us")
if any(abs(value - period) > 0.15 * period + EPS for value in periods):
return dict(result, status="periods_vary_over_fifteen_percent")
return dict(result, status="recurrent", period_us=period, frequency_hz=1_000_000 / period)
def summarize(points):
windows = []
for start, stop in WINDOWS:
part = [point for point in points if start <= point[0] <= stop]
cathode = [point[2] for point in part]
supply = [point[1] for point in part]
windows.append({
"points": part, "cathode_median_v": median(cathode),
"cathode_pp_v": max(cathode) - min(cathode),
"supply_median_v": median(supply), "supply_pp_v": max(supply) - min(supply),
})
return windows
def classify(windows, threshold):
observed = []
for window in windows:
points = window["points"]
measurement = {name: value for name, value in window.items() if name != "points"}
values = [point[2] for point in points]
midpoint = (min(values) + max(values)) / 2
measure = excursion(points, midpoint, threshold)
observed.append(dict(measurement, recurrence=measure))
signal = [row["cathode_pp_v"] + EPS >= threshold for row in observed]
recurrent = [row["recurrence"]["status"] == "recurrent" for row in observed]
if not any(signal):
status = "below_fixture_ripple_screen"
elif not all(signal) or not all(recurrent):
status = "ripple_seen_periodicity_unresolved"
else:
fast, slow = sorted(row["recurrence"]["period_us"] for row in observed)
status = ("recurring_in_both_windows" if slow <= 1.15 * fast + EPS
else "ripple_seen_periodicity_unresolved")
return dict(status=status, windows=observed)
def main():
parser = argparse.ArgumentParser(description=__doc__)
for name in LABELS:
parser.add_argument(name, type=Path)
args = parser.parse_args()
summaries = {}
try:
for name in LABELS:
try:
summaries[name] = summarize(read_capture(getattr(args, name)))
except (OSError, ValueError, csv.Error) as error:
raise ValueError(f"{name}: {error}") from None
except ValueError as error:
print(f"capture rejected: {error}", file=sys.stderr)
return 2
controls = summaries["probe_before"] + summaries["probe_after"]
windows = [row for capture in summaries.values() for row in capture]
quiet = max(row["cathode_pp_v"] for row in controls)
ctrl_medians = [row["cathode_median_v"] for row in controls]
checks = {
"supply_medians_1p75_to_1p85_v": all(1.75 - EPS <= row["supply_median_v"] <= 1.85 + EPS for row in windows),
"supply_ripple_at_most_10_mv": all(row["supply_pp_v"] <= 0.010 + EPS for row in windows),
"cathode_medians_1p15_to_1p33_v": all(1.15 - EPS <= row["cathode_median_v"] <= 1.33 + EPS for row in windows),
"probe_only_controls_at_most_8_mv_pp": quiet <= 0.008 + EPS,
"probe_only_medians_within_20_mv": max(ctrl_medians) - min(ctrl_medians) <= 0.020 + EPS,
}
threshold = max(0.020, 4 * quiet)
qualified = all(checks.values())
report = {
"scope": "Observed peak-to-peak ripple and recurrence in this fixture only.",
"controls_pass": qualified, "controls": checks, "ripple_screen_v_pp": threshold,
"windows_us": WINDOWS,
"conditions": {
name: classify(capture, threshold) if qualified else {
"status": "control_check_failed", "windows": [
{key: value for key, value in row.items() if key != "points"} for row in capture
],
}
for name, capture in summaries.items()
},
}
print(json.dumps(report, indent=2, allow_nan=False))
return 0 if qualified else 3
if __name__ == "__main__":
sys.exit(main())
Challenge
Your worst probe-only span is 7 mV. A direct capacitor capture has 24 mV peak-to-peak, and its frequency looks similar in both windows. Does it pass the amplitude screen?
The screen is the larger of 20 mV and 4 × 7 mV, so it is 28 mV. A 24 mV span falls below that screen. Record the observed 24 mV and improve or repeat the control if you want to resolve a feature of that size.
Manufacturer references
- TI TLV431AIDBZR exact orderable, TLV431A technical documents, and SLVS139Z revision Z. See §4 pin functions, §§5.3 and 5.6 operating limits, Figures 5-18 and 5-19, and §7.3.
- KEMET R82DC3100DQ50K exact part sheet for capacitance, tolerance, voltage and film dielectric.
- RIGOL DS1000Z data sheet and DS1000Z user guide for PVP3150 probes and CSV storage.
- SIGLENT SPD3303X documentation for isolated adjustable outputs.