Four wires can keep the voltage drop in the current leads out of a low-resistance reading. A small series voltage in the sense circuit can still change the answer. Here you will keep all four contacts at the same positions on a 0.100 Ω resistor, intentionally add about +1.50 mV to the high sense lead, and see whether the resistance recovered from opposite currents stays the same.

The fixture illustrates the kind of additive voltage produced when junctions of dissimilar metals sit at different temperatures, an effect called thermoelectric electromotive force (thermal EMF). The battery divider supplies a deliberately much larger test offset. The examples below are calculated and the program has been exercised with simulated captures.

Connect the force and sense leads

Four fixed contacts at a 0.100 ohm resistor. In the offset run SENSE HI goes to the midpoint of an AA divider with a 10 ohm connection to the high resistor pad. A calculated time plot shows two positive-current observations bracketing two negative-current observations.
In the offset run, use the divider in the high sense path. Use the separate jumper shown below the circuit for the clean and restore runs. The calculated plot illustrates the first observation block. Open circles show interpolated voltages, and filled circles show sample positions. Full-size PNG · Editable SVG

At 10 mA, the nominal resistor drop is 1.00 mV and dissipation is 10 µW. The E91's 1.5 V nominal value and the 10 kΩ plus 10 Ω divider give about 150 µA in the battery loop and +1.4985 mV at S relative to A. The divider loop closes locally on pad A. R1 carries only the forced current in the ideal circuit. The meter input is specified above 10 GΩ in this voltage range, in parallel with the measured voltage between SENSE HI and SENSE LO. For these millivolt signals the loading through the stated input resistance is negligible against the 10 Ω leg. Real bias, leakage and dirty connections remain possible.

Use an electrically insulated fixture, shrouded leads, and a cover rated for the instrument as described in the 2450 reference manual. Preserve protective earth. Turn output off, power the instrument down and unplug it before touching any connections or changing the network. The 2450 can generate hazardous voltage. The 2 V overvoltage protection (OVP) setting limits the force terminals in this procedure but does not make them touch-safe. An open sense lead can defeat the separate current-mode voltage limit. Verify all four connections with the meter disconnected, close the fixture, verify OVP before each run and stop if the 2450 reports a limit, overvoltage, overrange or unstable reading.

What two polarities reveal

With unchanged contacts, write the measured voltage as V = I × R + E, where E is the additive voltage. At +10.00 mA and E = +1.50 mV, the ideal display reads +2.50 mV. At −10.00 mA it still reads +0.50 mV. Dividing +2.50 mV by +10.00 mA gives an apparent 0.250 Ω. Subtracting the two measured voltages and dividing by the signed current separation instead gives (2.50 − 0.50) mV ÷ 20.00 mA = 0.100 Ω. This is a voltage-versus-current slope. No exact match between the two measured current magnitudes is needed.

Offset can move between readings. Take four readings with signs +, −, −, +. For each inner time t, interpolate the same-sign outer current and voltage separately with weight w = (t − t₀) ÷ (t₃ − t₀). The interpolated voltage is V̂ = V₀ + w(V₃ − V₀), and the interpolated current is Î = I₀ + w(I₃ − I₀). Then compute R = (V̂ − V) ÷ (Î − I) and E = V − I × R. If the resistance stays constant and the additive voltage varies linearly within the block, the interpolated additive voltage equals the inner additive voltage, so the subtraction removes it. Repeat as −, +, +, −. A nonlinear drift, intermittent contact or temperature-dependent resistance can produce disagreement. Similar slopes only establish agreement under the conditions tested.

Ideal steady readings for the two circuit states
State+10 mA−10 mA
Clean+1.00 mV−1.00 mV
Offset+2.50 mV+0.50 mV
Restore+1.00 mV−1.00 mV

Record actual current and voltage

  1. With the output off, choose FUNCTION → Source Current and Measure Voltage. Select the front terminals, 4-Wire Sense, a fixed 10 mA source range and a fixed 20 mV measure range. Set the voltage Limit to 20.0 mV and recheck it on the home screen after setting the range. The fixed 20 mV range supports at most 21 mV of magnitude for this limit. Under MENU → Source → Settings, set Overvoltage Protection Limit to 2 V and Source Readback to On. Use the measured source value shown as MEAS. PROG identifies the programmed setpoint.
  2. Keep the numeric display at full resolution, autozero On and any moving filter or math offset Off. Set the voltage aperture to 1 power-line cycle (1 PLC, shown as NPLC in settings), so the example comparison uses a consistent voltage integration time. The manufacturer specifies at least an hour of warm-up for rated accuracy. After each power cycle recheck all settings, keep the fixture closed and use comparable warm-up and ambient conditions. Start in the low-current setup described above. Never connect externally powered hardware to these terminals.
  3. Set +10.00 mA in the clean condition. Select Manual Trigger Mode by holding TRIGGER and selecting it. Once the output is enabled and the reading is settled, trigger one reading. Edit the source with its numeric-entry dialog and commit the full value at once. Do not use the live digit-by-digit editing mode. For each update, wait around 1–2 s, press TRIGGER, and keep successive triggers roughly 2–10 s apart. Take + − − + followed by − + + −, even when two consecutive signs are the same. Check for a yellow Limit, OVP, an event, or an invalid reading on each step. Switch output off after the eight readings.
  4. Before turning the instrument off, use MENU → Views → Reading Table to retrieve the stored time, source readback and voltage from the matching reading rows. Open a row's details for timestamp or limit information if needed. Record the elapsed seconds from the start of the experiment, keeping the three phases on an increasing time axis. Note each compliance/limit flag as 1 if active and 0 otherwise. An invalid/overrange reading without a finite measured value calls for a new run. The buffer may be cleared at power-off, so save or transcribe it first. Use actual timestamps and signed readback, rather than assumed equal sampling intervals or the programmed current.
  5. Power down as described above, fit the floating divider and repeat under offset. Shut down, remove it, restore the direct SENSE HI connection and repeat under restore. Save just the 24 chosen readings in order, with amperes converted to milliamperes and volts to millivolts. Use the original contact sites throughout.

The 20 mV range lists 10 nV measurement resolution but an absolute 1-year accuracy term of 150 µV plus 0.100% of reading under the specification conditions. Resolution alone cannot justify a precision claim. For scale, two voltage errors of about 150 µV each with opposite signs would contribute roughly 0.015 Ω when those readings are subtracted across 20 mA. Shared offsets can cancel and errors can be correlated. The software's checks are experiment limits, separate from manufacturer specifications or a calibration uncertainty budget.

Analyze the three runs in Python

Use Python 3 and the standard library. Download the blank worksheet, synthetic example, complete program, or files and tests. The program reads a comma-separated values (CSV) file on the host. It issues no instrument commands.

python3 current_reversal.py observations-blank.csv
python3 current_reversal.py synthetic-observations.csv

Fill the blank worksheet first. The program requires exactly 8 readings for each of clean, offset and restore, with times increasing within one day of elapsed time. Inside each eight-row phase successive readings must be 0.5–15 s apart. Readback must have the expected sign, be between 9.8 and 10.2 mA in magnitude, and be paired with an in-range reading within ±5 mV and a zero limit flag.

The program takes the median of the four interpolated slopes and offsets in each phase. It also reports the median of the four positive-current quotients. To qualify the experiment, individual slopes within a phase and the three phase medians must each span at most 0.012 Ω. All three medians must lie in 0.075–0.125 Ω. The median clean and restored offsets must be within ±0.30 mV. Insertion must increase the offset by 1.15–1.85 mV against the average of clean and restore, and raise the positive-current quotient by at least 0.080 Ω.

Exit 0 means the selected observations support cancellation in this fixture. Exit 1 means the file could not be read as valid input. Exit 2 means the data was readable but the controls failed. Check the original photographs, recorded polarity, readback and limit flags first. If the corrected results spread, investigate a warm joint, slow nonlinear drift or an intermittent connection before repeating. Agreement does not measure the cause of an unknown offset or prove correction of arbitrary drift.

#!/usr/bin/env python3
"""Analyze manual four-wire voltage/current captures with a deliberate offset."""

import argparse
import csv
import io
import math
import statistics
from pathlib import Path

HEADER = ["condition", "time_s", "current_ma", "voltage_mv", "limited"]
MODES = ("clean", "offset", "restore")
SIGNS = (1, -1, -1, 1, -1, 1, 1, -1)
ROUNDING = 1e-10


class Unresolved(ValueError):
    """The file can be read, but the experiment failed a qualification gate."""


def within(value, minimum, maximum):
    return minimum - ROUNDING <= value <= maximum + ROUNDING


def read_capture(path):
    with path.open("rb") as stream:
        data = stream.read(8193)
    if len(data) > 8192:
        raise ValueError("CSV exceeds 8192 bytes")
    rows = csv.reader(io.StringIO(data.decode("utf-8-sig"), newline=""), strict=True)
    if next(rows, None) != HEADER:
        raise ValueError("expected header " + ",".join(HEADER))
    result = []
    for line, row in enumerate(rows, 2):
        if len(row) != 5 or row[0] not in MODES or row[4] not in ("0", "1"):
            raise ValueError(f"line {line}: expected condition, time, current, voltage and 0/1 flag")
        numbers = tuple(map(float, row[1:4]))
        if not all(math.isfinite(value) for value in numbers):
            raise ValueError(f"line {line}: all numeric fields must be finite")
        result.append((row[0], *numbers, row[4] == "1"))
    if len(result) != len(MODES) * len(SIGNS):
        raise ValueError("expected 24 records (eight per condition)")
    return result


def block_estimates(block):
    _, first_time, first_current, first_voltage, _ = block[0]
    _, last_time, last_current, last_voltage, _ = block[-1]
    for _, time, current, voltage, _ in block[1:3]:
        weight = (time - first_time) / (last_time - first_time)
        outer_current = first_current + weight * (last_current - first_current)
        outer_voltage = first_voltage + weight * (last_voltage - first_voltage)
        resistance = (outer_voltage - voltage) / (outer_current - current)
        offset = voltage - current * resistance
        yield resistance, offset


def condition_results(capture):
    previous_time = None
    reports = {}
    for index, mode in enumerate(MODES):
        start = index * len(SIGNS)
        group = capture[start:start + len(SIGNS)]
        for position, (row, sign) in enumerate(zip(group, SIGNS)):
            label, time, current, voltage, limited = row
            if label != mode:
                raise Unresolved(f"expected condition {mode} at capture row {start + position + 1}")
            if not within(time, 0, 86400) or (previous_time is not None and time <= previous_time):
                raise Unresolved(f"{mode}: timestamps must be increasing within 0–86400 s")
            if position and not within(time - previous_time, 0.5, 15):
                raise Unresolved(f"{mode}: successive readings must be 0.5–15 seconds apart")
            if limited or not within(sign * current, 9.8, 10.2) or not within(voltage, -5, 5):
                raise Unresolved(f"{mode}: sign/readback, ±5 mV, or compliance screen failed")
            previous_time = time

        estimates = [*block_estimates(group[:4]), *block_estimates(group[4:])]
        resistances, offsets = zip(*estimates)
        spread = max(resistances) - min(resistances)
        if spread > 0.012 + ROUNDING:
            raise Unresolved(f"{mode}: bracketed resistance estimates span more than 0.012 Ω")
        positive = statistics.median(row[3] / row[2] for row in group if row[2] > 0)
        reports[mode] = (positive, statistics.median(resistances), statistics.median(offsets), spread)
    return reports


def qualify(reports):
    clean, inserted, restored = (reports[mode] for mode in MODES)
    corrected = [result[1] for result in reports.values()]
    if not all(within(value, 0.075, 0.125) for value in corrected):
        raise Unresolved("a corrected resistance is outside 0.075–0.125 Ω")
    if max(corrected) - min(corrected) > 0.012 + ROUNDING:
        raise Unresolved("the three median corrected resistances differ by more than 0.012 Ω")
    if not all(within(result[2], -0.30, 0.30) for result in (clean, restored)):
        raise Unresolved("clean or restored offset exceeds ±0.30 mV")
    contrast = inserted[2] - (clean[2] + restored[2]) / 2
    if not within(contrast, 1.15, 1.85):
        raise Unresolved("the injected offset contrast is outside +1.15–1.85 mV")
    if inserted[0] - clean[0] < 0.080 - ROUNDING:
        raise Unresolved("the positive-current apparent resistance rose by less than 0.080 Ω")


def main():
    parser = argparse.ArgumentParser(description=__doc__)
    parser.add_argument("csv", type=Path)
    arguments = parser.parse_args()
    try:
        capture = read_capture(arguments.csv)
    except (ValueError, OSError, csv.Error) as error:
        print(f"INPUT ERROR: {error}")
        return 1
    try:
        reports = condition_results(capture)
        print("condition  positive_only_ohm  reversed_ohm  additive_mv  bracket_span_ohm")
        for mode, values in reports.items():
            print(f"{mode:<10}" + "  ".join(f"{value:.6f}" for value in values))
        qualify(reports)
    except Unresolved as error:
        print(f"UNRESOLVED: {error}")
        return 2
    print("QUALIFIED: the offset changed the one-way estimate; reversal recovered a consistent slope.")
    return 0


if __name__ == "__main__":
    raise SystemExit(main())

The synthetic file uses uneven timestamps, slightly unequal measured current magnitudes and deliberately linear offset drift. The three corrected resistance medians are 0.100000 Ω computationally. The synthetic offset phase still reports an apparent positive-current resistance above 0.250 Ω. These values describe the supplied generated file.

Challenge

Both displayed voltages are positive. Can the reading for negative test current be valid?

Yes. At −10.00 mA a 0.100 Ω resistor contributes −1.00 mV. Adding +1.50 mV in the high sense lead gives +0.50 mV. The readback sign gives the current direction. The difference between the two measured voltages removes the shared additive term.

Related reading

Primary references