Measure the First Low and the Handoff on a TCA9517 Repeater
A low on an Inter-Integrated Circuit (I²C) wire is a voltage range. On the Texas Instruments (TI) TCA9517DGKR repeater, the B-side input sees a falling voltage from idle under different conditions from a low that another device must take over. Build a two-rail fixture and compare the same weak B-side pull-down in both sequences. A stronger pull-down provides a control for the handoff.
Use three switches and a digital multimeter (DMM) to examine the settled voltages. The worksheet leaves the readings blank. The accompanying Python program checks the controls and keeps ambiguous observations visible.
Why the two sides have different lows
The I²C data wire is called SDA and its clock wire is SCL. A device with an open-drain output can pull a wire toward ground or release it. A pull-up resistor raises the released wire toward its supply. The TCA9517 buffers SDA and SCL across two such pairs of wires. Each side has its own pull-up rail. In this experiment the two clock pins stay grounded as specified for unused pins. The data pins serve as test nodes and no I²C transactions are needed.
When an external switch pulls SDAA low, the repeater drives SDAB to a low voltage above ground. TI calls this a buffered low with a static offset, the voltage by which the low sits above ground. The A-side output can pull much closer to ground when an external B-side low starts the event. The offset helps keep the chip's own B-side output from feeding back as a fresh input low.
At the initial falling level, the recommended B-side low-input limit is 0.3 × VCCB. At 3.30 V that is 0.99 V. The data-sheet footnote distinguishes that first low from subsequent B-side lows. When the repeater is already driving its B-side low, a device taking over needs to pull it farther down. TI specifies at least 70 mV of separation below the buffered output low for that internal recognition mechanism and gives a 0.4 V typical contention parameter. Contention here is the interval when an external pull-down acts while the repeater still holds the B-side low. Section 10.2.2.2 instructs system designers to get below 0.4 V. Treat that value as the application criterion and leave margin. It is separate from the 0.3 × VCCB first-low limit.
Section 7.5 lists the B output low from 0.45 to 0.60 V, with 0.52 V typical, under its stated 100 µA or 6 mA output-current and zero-volt input test conditions. The A output low table shows 0.20 V maximum at 6 mA. The fixture uses its own smaller pull-up currents and measures the actual voltages. A data sheet low-input range classifies an observed voltage. A verified transition sequence is still needed to show which driver held the line low.
Wire one TCA9517DGKR
Use the exact catalog part TI TCA9517DGKR, package DGK, an eight-lead very thin shrink small-outline package (VSSOP). Use a passive VSSOP-8 adapter and identify pin 1 from the top-view package drawing before soldering. TI's SCPS242D data sheet was revised July 2017. The exact orderable page and product technical documents identify revision D. The checked TI product sources yielded no separate electrical errata for this device.
The recommended A-side supply range is 0.9–5.5 V, and the B-side range is 2.7–5.5 V. TI's product information specifies VCCA ≤ VCCB. This circuit uses 1.80 V and 3.30 V, with a common return G0. The logic and bus input/output circuitry are powered from VCCB. VCCA establishes the A-side input reference and participates in the power-good check. Keep the levels and current within the recommended operating conditions. The absolute maximum voltage and current ratings describe stress limits.
| Pin | Connection |
|---|---|
| 1 · VCCA | VA from isolated supply channel 1, set to 1.80 V. Put 100 nF ceramic from pin 1 to G0 close to the package. |
| 2 · SCLA | Direct to G0. This clock channel is unused. |
| 3 · SDAA | RA, 4.70 kΩ, to VA. Separate switch S_A and a 100 Ω resistor in series to G0. Red meter probe point A. |
| 4 · GND | Direct to G0. |
| 5 · EN | Direct to VB for active-high enable. The experiment keeps this connection in place. |
| 6 · SDAB | RB, 4.70 kΩ, to VB. S_BS with 100 Ω in series to G0 and S_BW with 1.50 kΩ in series to G0. Red meter probe point B. |
| 7 · SCLB | Direct to G0. This clock channel is unused. |
| 8 · VCCB | VB from isolated supply channel 2, set to 3.30 V. Put 100 nF ceramic from pin 8 to G0 close to the package. |
Use five ±1%, ¼ W resistors, two 100 nF X7R ceramic capacitors rated at least 16 V, and three latching single-pole single-throw switches. X7R is the designation of the ceramic dielectric inside the capacitor. Use the two adjustable outputs of a SIGLENT SPD3303X in independent mode, with a 50 mA current-limit setting on each. Join both negative terminals at G0 with both outputs off. Set one positive terminal to VA and the other to VB. The supply positives stay on separate nets. Leave all three switches open while setting up and check pin continuity, ground and the two rails before taking readings. This low-voltage fixture has no attached controller, target or bus cable.
Choose the pull-downs from the voltage targets
When the internal driver on a node is released, a pull-up and a closed resistor-to-ground branch act as a voltage divider. For a first estimate divide the supply voltage by the sum of the two resistors to obtain the branch current. Multiply that current by the pull-down resistance to obtain the node voltage. Once the repeater itself sinks current on the same node, the divider alone cannot give the voltage.
Vnode = Vrail × Rdown ÷ (Rpull-up + Rdown)
| Branch | Predicted node voltage | Branch current |
|---|---|---|
| S_A · 100 Ω on 1.80 V | 0.0375 V | 0.375 mA |
| S_BS · 100 Ω on 3.30 V | 0.0688 V | 0.688 mA |
| S_BW · 1.50 kΩ on 3.30 V | 0.7984 V | 0.532 mA |
The A-side strong branch easily clears the 0.54 V first-low limit. Both B branches are below the nominal 0.99 V first-low limit, but only the strong branch has margin to the 0.4 V criterion in the simple divider model. With a measured B-side buffered low of 0.52 V, the B pull-up supplies (3.30 − 0.52) V ÷ 4.70 kΩ, or about 0.591 mA. Across the worksheet rail windows and ±1% resistors, the released-driver model puts the strong B node below 0.072 V and weak B at roughly 0.774–0.824 V. While A holds the buffered low, the actual B voltage with a switch closed depends on the local B driver as well as the two resistors. Measure that overlap before releasing A.
Measure the order of events
Use one Fluke 115 in the 6.000 V direct-current range, with the red lead in its volts jack and the black lead in the common jack. Leave the black lead on G0. For each stable switch position, measure VA, VB, SDAA and SDAB in that order and enter the four indicated voltages in the corresponding row of the blank observation worksheet. Open the switches and verify both data lines return high between independent trials. Use latching switches so you can measure the four nodes without releasing the contacts.
- With all switches open, turn on the rails and wait for the readings to settle. Record
idle. With the multimeter, high at A should sit near 1.80 V and high at B near 3.30 V. If either rail is wrong, turn off the outputs and check the independent-channel settings. - Close S_A alone. Record
a_only, then open S_A and verify both lines return high. The contrast between the local A low and the buffered B low gives a baseline for the next steps. - From that idle state, close S_BS alone. Record
b_strong_first. Open it and verify both lines return high. Repeat from idle with S_BW alone forb_weak_first. Open it and verify idle again. - Close S_A first and check that B is at its buffered low. With S_A still closed, close S_BW. Record
weak_overlap. Open only S_A and leave S_BW closed. Record the settledweak_after_a_releasevoltages. Then open S_BW and verify both lines return high. - From idle, close S_A, then close S_BS while A is still low. Record
strong_overlap. Open only S_A, keep S_BS closed and recordstrong_after_a_release. Open S_BS, recordidle_afterand repeat the entire sequence on a fresh worksheet if a level varies.
Compare the two weak sequences only after the strong controls pass. If the same weak branch leaves A low from idle and A high after release, you have observed a difference tied to the order of switching. A different stable outcome is also a reading to record. Mechanical contacts can bounce and a handheld meter reports settled voltage. It cannot resolve the transient order around release. If the displayed readings drift or toggle, repeat from a known idle state and use a properly grounded two-channel oscilloscope on SDAA and SDAB to study the transition separately. Keep the two scope ground clips at G0.
A full B-side handoff matters when two devices drive the same line at different moments, such as when a target acknowledges a byte. TI disallows a B-side-to-B-side connection between these static-offset ports. The resistor fixture uses one repeater to reveal the voltage condition behind that rule. For a different repeater problem, see the TCA4307 stuck-bus recovery experiment.
Check the readings with Python
Download the complete experiment ZIP or the Python program. Save a completed sheet as my-readings.csv, keep the nine row labels in order and use volts with decimal points. Run python3 repeater_levels.py my-readings.csv with Python 3.10 or newer. Exit status 2 identifies a malformed or out-of-window sheet. Exit status 3 returns the measured classifications but withholds the weak comparison because a control failed.
The program uses Fluke's ±(0.5% of reading + 2 counts) specification on the 6.000 V range. One count is 0.001 V. Use a meter within its specified accuracy period of one year after calibration, at 18–28 °C and 0–90% relative humidity. The interval varies both the node reading and the rail reading. The low classification requires the entire node interval to be at or below 0.3 times the bottom of the supply interval. The high classification requires the entire node interval to be at or above 0.7 times the top of the supply interval. Values between those bounds remain indeterminate. These calculations apply the published meter specification and DC data-sheet ranges to the entered measurements. They cover neither contact bounce nor a transient threshold test.
For B contention, the program compares the full indicated voltage interval with 0.4 V and reports if it straddles the criterion. For the weak pair it requires the indicated B voltages in the three weak rows to remain at or above 0.4 V and inside the first-low range. This rejects a disconnected weak switch that leaves B near 3.30 V. A failed weak qualification or indeterminate A level leaves the A comparison undecided. The included constructed readings exercise the code and do not represent a built fixture.
"""Compare measured DC low levels in the TCA9517DGKR bench fixture."""
import argparse
import csv
from dataclasses import dataclass
from decimal import Decimal, InvalidOperation
import json
from pathlib import Path
import sys
D = Decimal
HEADER = ("condition", "vcca_v", "vccb_v", "sdaa_v", "sdab_v")
ORDER = (
"idle", "a_only", "b_strong_first", "b_weak_first",
"weak_overlap", "weak_after_a_release",
"strong_overlap", "strong_after_a_release", "idle_after",
)
@dataclass(frozen=True)
class Reading:
vcca: Decimal
vccb: Decimal
sdaa: Decimal
sdab: Decimal
def interval(volts):
"""Fluke 115 6.000 V DC range, specified conditions, volts."""
error = abs(volts) * D("0.005") + D("0.002")
return volts - error, volts + error
def classify_bus(volts, rail):
low, high = interval(volts)
rail_low, rail_high = interval(rail)
if high <= D("0.3") * rail_low:
return "within_low_input_range"
if low >= D("0.7") * rail_high:
return "within_high_input_range"
return "indeterminate_between_limits"
def b_contention_margin(volts):
"""Compare the indicated SDAB interval with TI's 0.4 V design criterion."""
low, high = interval(volts)
if high < D("0.4"):
return "below_0p4_v"
if low >= D("0.4"):
return "at_or_above_0p4_v"
return "straddles_0p4_v"
def read_rows(path):
if path.stat().st_size > 8_192:
raise ValueError("use one nine-row fixture worksheet at a time")
with path.open(newline="", encoding="utf-8-sig") as stream:
rows = csv.reader(stream, strict=True)
if tuple(next(rows, ())) != HEADER:
raise ValueError("use the exact observation worksheet header")
readings = {}
for expected in ORDER:
row = next(rows, ())
if len(row) != len(HEADER) or row[0] != expected:
raise ValueError(f"expected exactly one row named {expected} at line {rows.line_num}")
try:
values = tuple(D(field) for field in row[1:])
except InvalidOperation:
raise ValueError(f"{expected}: enter four meter readings in volts") from None
if not all(value.is_finite() for value in values):
raise ValueError(f"{expected}: readings must be finite")
if not all(D("-0.020") <= value <= D("4.000") for value in values):
raise ValueError(f"{expected}: reading is outside the fixture voltage window")
if any(value.as_tuple().exponent < -3 for value in values):
raise ValueError(f"{expected}: enter the display resolution of 0.001 V or fewer decimals")
reading = Reading(*values)
if not (D("1.750") <= reading.vcca <= D("1.850")
and D("3.250") <= reading.vccb <= D("3.350")):
raise ValueError(f"{expected}: verify the 1.80 V and 3.30 V rails")
if reading.sdaa > reading.vcca + D("0.030") or reading.sdab > reading.vccb + D("0.030"):
raise ValueError(f"{expected}: measured bus voltage exceeds its rail by more than 0.030 V")
readings[expected] = reading
if next(rows, None) is not None:
raise ValueError("unexpected extra worksheet row")
return readings
def analyze(readings):
table = {}
for name, row in readings.items():
b_low, b_high = interval(row.sdab)
table[name] = {
"vcca_v": str(row.vcca), "vccb_v": str(row.vccb),
"sdaa_v": str(row.sdaa), "sdab_v": str(row.sdab),
"sdaa_input_range": classify_bus(row.sdaa, row.vcca),
"sdab_first_input_range": classify_bus(row.sdab, row.vccb),
"sdab_interval_v": [str(b_low), str(b_high)],
"sdab_vs_0p4_v": b_contention_margin(row.sdab),
}
low = "within_low_input_range"
high = "within_high_input_range"
strong = "below_0p4_v"
checks = {
"idle_before_and_after": all(
table[phase]["sdaa_input_range"] == table[phase]["sdab_first_input_range"] == high
for phase in ("idle", "idle_after")
),
"a_drive_propagated_to_b": (
table["a_only"]["sdaa_input_range"] == table["a_only"]["sdab_first_input_range"] == low
and table["a_only"]["sdab_vs_0p4_v"] == "at_or_above_0p4_v"
),
"strong_b_first_reached_a": (
table["b_strong_first"]["sdaa_input_range"] == low
and table["b_strong_first"]["sdab_vs_0p4_v"] == strong
),
"a_held_during_b_overlap": all(
table[phase]["sdaa_input_range"] == low for phase in ("weak_overlap", "strong_overlap")
),
"strong_b_handoff_reached_a": all(
table[phase]["sdab_vs_0p4_v"] == strong
for phase in ("strong_overlap", "strong_after_a_release")
) and table["strong_after_a_release"]["sdaa_input_range"] == low,
}
comparison = None
if all(checks.values()):
from_idle = table["b_weak_first"]
after_a = table["weak_after_a_release"]
weak_qualified = all(
table[phase]["sdab_vs_0p4_v"] == "at_or_above_0p4_v"
and table[phase]["sdab_first_input_range"] == low
for phase in ("b_weak_first", "weak_overlap", "weak_after_a_release")
)
comparison = {
"a_when_weak_b_was_first": from_idle["sdaa_input_range"],
"a_when_weak_b_followed_a": after_a["sdaa_input_range"],
"same_observed_a_level": (
from_idle["sdaa_input_range"] == after_a["sdaa_input_range"]
if weak_qualified and from_idle["sdaa_input_range"] in (low, high)
and after_a["sdaa_input_range"] in (low, high)
else None
),
"weak_conditions_observed_at_b": weak_qualified,
}
return {"controls_pass": all(checks.values()), "controls": checks,
"weak_comparison": comparison, "readings": table}
def main():
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("worksheet", type=Path)
args = parser.parse_args()
try:
result = analyze(read_rows(args.worksheet))
except (OSError, csv.Error, ValueError) as error:
print(f"invalid worksheet: {error}", file=sys.stderr)
return 2
print(json.dumps(result, indent=2, allow_nan=False))
return 0 if result["controls_pass"] else 3
if __name__ == "__main__":
sys.exit(main())
Challenge
With VB at 3.30 V, what ideal ground-side resistor would put B at 0.40 V if only the 4.70 kΩ pull-up drives it? Would that resistor have a comfortable margin for handoff?
Solving the divider gives Rdown = 4.70 kΩ × 0.40 V ÷ (3.30 V − 0.40 V), or about 648 Ω. That value lands on the application criterion in the ideal model. Resistor tolerance, supply variation, the meter interval and the already-active B driver all matter near this value. Use the 100 Ω branch for the clear control and verify the actual voltage while A is still held low.
Manufacturer references
- TI TCA9517DGKR exact orderable part and TCA9517 product and technical documents.
- TI SCPS242D Rev. D data sheet. See the pin table, §§7.3 and 7.5, and the offset, pull-up sizing and power-up descriptions in §§9.3.3, 10.2.2.2 and 11.
- Fluke 115 user manual for DC voltage range and accuracy.
- SIGLENT SPD3303X product documentation for isolated outputs and independent channel operation.