B5Z Magnetic Particle Brake Controller¶

Moved here from kicad-libs/docs/ on 2026-10-07. Paths into that repository are written kicad-libs/…; the datasheets it cited from its tmp/ cache are in the private store kicad-libs-nonfree/datasheets/.

Status: topology selection and the brake purchase list in this note are superseded as of 2026-08-31 by kicad-libs/docs/brake-current-sink-design.md, which owns the circuit, its values, and its bill of materials. What remains current here: the brake requirements, the previous-board prior art, the brake/torque noise-isolation rules, and the validation plan.

This note records the background for controlling a Placid Industries B5Z-24-1R magnetic particle brake from the mouse-joystick-pcb nominal 36 V rail. The older spelling B5Z-24-IR in this repository appears to confuse the digit 1 with the letter I; verify the purchased brake nameplate before freezing fields or silkscreen text.

Recommendation Summary¶

The brake and the low-level torque bridge will operate simultaneously. Treat brake-driver noise as a measurement-system requirement, not merely an EMC cleanup item.

Settled 2026-08-31. The DAC-programmed linear current sink is the design, not one of three candidates. The evaluation ladder below is retained as a record of what was rejected and why:

Chosen: a smooth, DAC-programmed linear current sink using an external pass device. At only 49 mA the worst pass-element dissipation is 0.71 W, so the switching converter is avoided with no meaningful thermal penalty. The pass device is now a stocked MOSFET, not a high-voltage NPN — see kicad-libs/docs/brake-current-sink-design.md for the base-current argument. Full values, BOM, and open items live in that document.
Rejected as the final design, retained as prior art: the stocked BCR421UE6327 with its 30 ohm set resistor. It programs average torque by PWM, has little voltage margin on a 36 V rail, and its regulated current drifts with self-heating. It remains a valid corrected bench prototype if one is ever wanted, under the conditions listed below.
Dropped, not deferred: LM3404HV. Total brake load is 1.76 W. Placing a 400 kHz switch node beside a 15 mV bridge to recover 0.7 W of linear dissipation is a bad trade, and no thermal argument forces it. The purchases listed for it in Switching Fallback Purchases are cancelled.

Two requirements consequences were missed on the first pass and are now recorded in kicad-libs/docs/brake-current-sink-design.md: coil I²R (1.11 W at rated current) is not slip heat and is the term that sets case temperature, which makes the NTC case probe optional rather than Priority A; and the datasheet's approach-direction torque hysteresis is percent-level, which caps the useful precision of the whole current path at about 1% and makes the torque sensor the calibration reference.

Do not use the torque sensor as a substitute for a hardware current limit. The electrical current measurement and torque measurement answer different questions:

current confirms the driver, wiring, and coil are behaving electrically;
torque includes brake hysteresis, friction, mechanics, and the applied load;
a hardware current ceiling remains necessary even if firmware closes a loop around measured torque.

Initially use torque feedback for calibration and supervisory observation, not as a fast closed loop. A fast loop around the animal, mechanism, brake hysteresis, and sensor could create dynamics that are harder to interpret than the open-loop brake.

Brake Requirements¶

Source datasheet: kicad-libs-nonfree/datasheets/B5Z-metric-data-sheet-1.pdf

Winding: 24 V nominal
Rated coil resistance: 467 ohm
Rated current: 49 mA at 100% input current
Rated torque: 3.5 N-cm at 100% input current
Torque is proportional to current, with different curves when approaching a current from 0% and from 100%
Recommended supply type: adjustable constant-current DC supply
Response: 6 ms unforced, 3 ms forced
Maximum brake case temperature: 75 C
Continuous slip heat limit: 2 W
Continuous slip heat formula: W = RPM * torque_N_cm / 960
Always electrically ground the brake body

At 49 mA the cold 467 ohm winding needs about 22.9 V. A roughly 560-570 ohm hot winding needs about 27.4-27.9 V. A 36 V supply therefore provides useful current-regulator compliance. A 24 V rail cannot maintain rated current through the hot winding.

The brake must not be driven directly from 36 V. A cold 467 ohm winding would draw about 77 mA, well over its 49 mA rating. Supply stability does not make fixed-voltage drive safe; the winding resistance changes with temperature.

Previous-Board BCR421 Circuit¶

The fabricated previous-generation board used:

U801: BCR421UE6327
R801: 30 ohm
C801: 100 uF, 50 V
J801: 2-position 3.5 mm terminal

That sheet has no current monitor, external PWM generator, or visible flyback diode. It is useful prior art but should not be copied unchanged.

The BCR421 is stocked as OW-PWR-BCR421UE6327, JLCPCB C112923, quantity 495 in the current inventory export. Its relevant characteristics are:

low-side constant-current regulator with logic PWM input;
30 ohm gives approximately 49-50 mA from the typical curves;
40 V maximum output voltage;
up to 1 W package dissipation only with an appropriate thermal layout;
typical current temperature coefficient of -0.2%/K;
PWM input rated to 10 kHz.

At 36 V and 49 mA, approximate BCR421 dissipation is:

cold 467 ohm coil: (36 - 22.9) * 0.049 = 0.64 W;
hot 560 ohm coil: (36 - 27.4) * 0.049 = 0.42 W.

Those values are thermally plausible, but self-heating causes the regulated current to fall. The 40 V output rating is the larger concern. Whenever a connected brake is off, its resistance pulls the low-side output toward the 36 V rail. A flyback diode can add a forward drop above that rail during current decay. There is too little margin for an unspecified 36 V supply maximum, hot-plug ringing, and cable inductance.

If the BCR421 is prototyped:

measure the actual 36 V rail maximum and transient overshoot first;
place a Schottky flyback diode directly across the two coil terminals, cathode to the 36 V side and anode to the BCR421 side;
use substantial copper under and around all three OUT pins;
add a pull-down at EN so reset and disconnection mean brake off;
prohibit live connector removal unless the harness/load-side inductive energy has its own safe path;
measure peak current, average current, BCR421 temperature, and torque versus duty after thermal equilibrium;
do not infer production accuracy from the typical 30 ohm graph alone.

Preferred Low-Noise Linear Current Sink¶

The concept below is the chosen design. Its values are now settled in kicad-libs/docs/brake-current-sink-design.md; use that document for anything you intend to draw. The remaining text here is the reasoning, and two of its specifics have been revised: R_SENSE is 10 ohm as suggested, but the pass device is a stocked MOSFET rather than an MJD340-class NPN, and the independent ceiling is a third stocked OPA197 rather than a TLV7041.

The preferred concept replaces PWM amplitude control with a continuously adjustable analog current loop:

+36V_SW ----+---- brake coil ----+---- QPASS collector/drain
            |                    |       QPASS emitter/source ---- RSENSE ---- BRAKE_PGND
            +-----|<|------------+                    ^
               flyback diode                          |
                                           op amp current loop
                                                    ^
                                                    |
                         DAC -> scaling + low-pass -+-- default-off clamp
                                                        / BRAKE_ENABLE

RSENSE -> buffer/gain -> BRAKE_ISENSE
RSENSE -> independent overcurrent comparator -> clamp + ~BRAKE_FAULT

The control law is simply:

I_BRAKE = V_SET / R_SENSE

A reasonable design start is R_SENSE = 10 ohm, which requires V_SET = 0-490 mV and dissipates only about 24 mW at 49 mA. Scale and filter the DAC output so its maximum possible output, not merely its programmed normal range, cannot command more than the rated current. Use 0.1% or better and low TCR for R_SENSE and the scaling network.

With a cold coil and a roughly 0.49 V sense drop, QPASS dissipates about:

(36 - 22.9 - 0.49) * 0.049 = 0.62 W

That is modest for a DPAK, SOT-223, or exposed-pad power package with a real copper heat spreader. The transistor still needs a verified DC safe operating area at approximately 10-15 V and 50 mA. Low RDS(on) and pulsed-current ratings do not establish linear-mode suitability.

A high-voltage NPN BJT is attractive at this small current because its DC safe operating area is explicit and it avoids the thermal-instability questions of many switching-optimized MOSFETs. An MJD340-class NPN is a useful search starting point: 300 V VCEO, 0.5 A continuous current, DPAK, minimum current gain 30 at 50 mA/10 V, and published active-region SOA. A suitable rail-to-rail op amp can supply the roughly 1.7 mA worst-case base current. A MOSFET remains acceptable only after its DC SOA is checked at the actual board temperature.

Reversed 2026-08-31. The BJT is no longer the preference, and the reason is measurement rather than thermal. A ground-referenced emitter shunt measures I_C + I_B, so it reads high by 1/(hFE+1) — 0.7% to 3.3% across the guaranteed gain spread, varying with current and temperature. That is a systematic error in a torque-calibrated system, and a MOSFET removes it rather than requiring it to be characterized. The DC-SOA objection to a MOSFET does not survive the numbers at 0.71 W and 49 mA: see the SOA and hot-spot analysis in kicad-libs/docs/brake-current-sink-design.md. The stocked BUK7D36-60EX is the pass device, with the 80 V BSC072N08NS5ATMA1 placed as a DNP alternate against the documented +36V_SW ringing. No MJD340T4G, MJD122T4G, or LinearL2 MOSFET purchase is needed.

Safe State And Independent Current Ceiling¶

The DAC and main op amp must not be the only current limit.

Select a zero-scale power-on-reset DAC variant.
Pull the pass device off in hardware.
Add BRAKE_ENABLE as an active-high arm input with a passive default-off state. This signal enables the analog stage; it does not modulate torque.
Add a separate comparator at roughly 52-55 mA that overrides the op amp and removes pass-device drive. Give it hysteresis or a latch appropriate to the fault policy.
Export the comparator state as ~BRAKE_FAULT if that board interface is kept.
Size the analog scaling so a full-scale DAC command is no greater than 49 mA.
Do not rely on firmware register initialization, the torque sensor, or the upstream 36 V smart switch as the only protection.

Corrected 2026-08-31 — the scaling bullet above is not a ceiling on its own. A DAC's absolute-maximum output is its VDD, which for the DACx0501 family in BUFF-GAIN = 1 is twice its programmed full scale. A divider that maps 2.5 V to 49 mA therefore still permits 98 mA at VDD = 5 V, or 62 mA at VDD = 3.3 V, under a firmware or DAC fault. Two changes close this, and both are in kicad-libs/docs/brake-current-sink-design.md:

Power the brake DAC from 3.3 V. REF-DIV = 0 is the reset default and is out of spec at VDD = 3.3 V, and the datasheet states that improper reference-divider configuration shuts down the reference buffer and drives all outputs to 0 V. The 0-2.5 V span becomes a hardware property and the wrong-GAIN failure becomes fail-safe rather than fail-hot.
Use a third stocked OPA197 as the independent ceiling, not a TLV7041. This note's own audit flagged the TLV7041's ±8 mV offset and 3-25 mV hysteresis as something that must be budgeted into the threshold — ±3.3% of a 550 mV threshold. The OPA197's ±100 µV offset is ±0.02%, so the ceiling can sit at 56 mA with a comfortable 13% margin over the command maximum. It is a better comparator here, not merely a cheaper one.

The TPS3431 hardware watchdog is dropped for rev 2, reversing the guidance in the purchase plan. "Brake stuck at 49 mA" is the brake's rated continuous operating condition rather than a damage mode, and the board already gives the MCU an unconditional hardware kill through HV_EN on the TPS1HTC100-Q1, which deliberately bypasses the GPIO isolation bus switch. Reversible by purchase if bench experience disagrees.

The current-loop bandwidth need not be high. The brake response is 6 ms unforced, and trial-level torque commands are likely slower. Start with a well-damped loop and a DAC-output RC filter; raise bandwidth only after measuring step response and stability with the actual coil and cable.

Current Feedback¶

The low-side sense voltage is already ground referenced and represents the regulated current. A second channel of a dual op amp can buffer and scale 0-490 mV to a convenient 0-2.5 V or 0-3.0 V BRAKE_ISENSE output.

Route BRAKE_ISENSE to the MCU ADC if closed-loop diagnostics are wanted.
It is also a good candidate for InstrumentPort AIN1 after buffering.
Do not put raw high-side shunt nodes on InstrumentPort.
Keep a local Kelvin test pair across R_SENSE, even if the amplifier is populated.
Compare commanded and measured current in firmware to detect an open coil or a pass-stage fault.

A low-side shunt below QPASS does not measure the brief recirculating current that flows through a diode placed directly across the coil after QPASS turns off. Measure that decay at the coil during validation if it matters to torque step response; do not mistake the low-side trace for all transient coil current.

The stocked INA230 is not a comfortable high-side alternative: its 36 V common-mode limit is exactly the nominal rail and therefore provides no useful supply or transient margin. The analog sink does not require it.

DAC Interface Instead Of A PWM Generator¶

A serial DAC provides the same MCU portability sought from a dedicated PWM chip while avoiding chopped brake current. A 12-bit DAC gives approximately 12 uA per code over a 49 mA range, already much finer than the expected brake accuracy and torque hysteresis. More resolution can help calibration but is not required for basic command granularity.

Prefer a DAC with:

I2C or SPI selected by a hardware pin;
an internal precision reference or a well-defined external-reference input;
zero-scale power-on reset;
monotonic output;
a package that can be assembled and inspected reliably;
output range and supply rails that permit a passive, hardware-safe maximum current scaling network.

The PCA9685 remains a useful PWM option, but it does not create a safety boundary. Its last duty setting persists if firmware hangs. If it is used:

pull OE high for a disabled reset state;
pull every brake DIM input low locally;
lower OE only after all channel registers are written to off;
arrange for reset or a hardware watchdog to return OE high;
observe the actual output waveform, not only the requested duty;
verify what happens when the 36 V switched domain is power-cycled while the always-on PCA9685 retains its previous state.

LM3404HV Switching Fallback¶

Dropped 2026-08-31, not deferred. Total brake load is 1.76 W; a 400 kHz switch node beside a 15 mV bridge is not worth 0.7 W of linear dissipation, and the headroom analysis in kicad-libs/docs/brake-current-sink-design.md shows the linear stage has 4.2 V of margin at the worst-case hot winding and low rail. The section is retained only so that a future revision does not have to re-derive it, and the purchases it calls for are cancelled.

Use the HV version for nominal 36 V. The plain LM3404 has too little practical operating margin for supply tolerance and ringing.

Starting values from the earlier analysis remain useful for a prototype:

Ref	Starting value	Notes
U1	`LM3404HV`	Prefer PowerPAD package when practical
RON	430 kohm, 1%	Approximately 400 kHz near cold full-current operation
RSNS	4.22 ohm, 1%	First-pass 49 mA setting; tune with real hardware
L1	2.2 mH	Also evaluate 4.7 mH for lower ripple
D1	>=60 V Schottky, >=1 A	Stocked `DSK36` is a starting candidate
CB	10 nF X7R, >=25 V	Bootstrap capacitor from `BOOT` to `SW`
CF	100 nF X7R, >=25 V	`VCC` bypass
CIN	2.2-4.7 uF X7R, 100 V	Close to `VIN` and power ground
Bulk	47-150 uF, >=63 V	Local to the power entry or brake zone
CS resistor	1 kohm	Datasheet-recommended protection

Approximate predicted current with 36 V, 430 kohm, 2.2 mH, and 4.22 ohm was 49.5 mA cold and 47.8 mA hot. Buy 4.02, 4.12, 4.22, 4.32, and 4.42 ohm values for tuning. If a 2.2 or 4.7 mH inductor cannot tolerate the controller's fault current without saturating, document and test that fault instead of relying on the 49 mA normal load rating.

For PWM dimming, start around 250-500 Hz only as a test point. Measure current decay and torque ripple; the brake inductance is not specified, so average current cannot safely be assumed to equal duty times 49 mA. If PWM remains in the final design, consider coherent torque sampling and PWM-synchronous averaging rather than allowing an unrelated PWM carrier to alias into the sensor data.

Brake And Torque-Sensor Noise Isolation¶

The torque sensor is a millivolt-level bridge. Tens of milliohms of shared ground impedance carrying 49 mA can create errors comparable to its signal. The most important isolation is control of current return paths, not an arbitrary split in the ground plane.

Placement And Routing¶

Put the brake power stage at the board edge near the 36 V entry and brake connector. Put the bridge connector, excitation, PGA/ADC, reference, and anti-alias filter at the opposite side of the board.
Route brake supply and return as a close pair. Return brake current directly to the power-entry/power-stage ground junction; never let it traverse torque AGND, bridge return, reference ground, or ADC ground copper.
Give bridge excitation and bridge return dedicated Kelvin routes to the torque front end. Do not share a connector return pin or trace with the brake.
Keep a continuous reference plane where practical, but partition placement and current flow. Do not create a split-plane gap that digital or analog signals must cross.
Keep the brake connector and torque connector physically separated. Use a twisted pair for the brake coil. Use a separate shielded/twisted harness for the bridge signals and excitation.
Ground the brake body through a deliberate chassis/protective path. Do not route case-ground current through torque-sensor analog ground.
Do not route brake command, pass-device drive, flyback, PWM, or switch-node copper under the torque connector, bridge traces, reference, or ADC.
Add grounded guard copper and a via fence between zones when it does not increase capacitance on a switching node.

Supplies And Digital Coupling¶

Feed the torque front end from its own low-noise LDO and local RC/ferrite filtering. Do not share the brake driver's local decoupling path.
Filter the DAC/reference supply as an analog rail. A ferrite bead plus local bulk and ceramic bypassing may be sufficient for the op amp; verify rather than assuming the existing 5 V switcher is quiet.
Slow serial clock edges with source series resistors where timing allows and keep I2C/SPI clocks away from bridge traces.
Schedule DAC updates outside critical ADC conversion intervals if measured digital feedthrough warrants it.
Galvanic digital isolation is not the first choice: it requires isolated power, often adding another switching converter. Add it only if the physical zoning and return-path design fail measured noise requirements.

If A Switching Brake Driver Is Required¶

Use a shielded inductor and keep the VIN capacitor, controller, catch diode, sense resistor, and switching loop in the smallest possible area.
Keep SW copper minimal and entirely inside the brake zone; never use a large switch-node copper pour as a heat spreader.
Give the brake zone a local input filter so switching current does not flow through the torque analog supply path.
Place the torque anti-alias filter at the ADC/PGA input, not at the remote connector.
Measure conducted noise on bridge excitation, bridge return, analog supply, and ground as well as radiated coupling from SW.
Prefer a fixed carrier with known spectral lines. Synchronize acquisition or reject those lines digitally only after confirming the analog front end is not saturating.
Do not use spread spectrum merely to make a spectrum look smoother; it can move interference into more torque-measurement bins.

Stocked Supporting Parts¶

Based on kicad-libs/db/source/jlc_inventory_export.csv during this review:

Role	Stocked part
Linear PWM prototype	`BCR421UE6327`, `C112923`, qty 495
BCR set resistor	30 ohm 1% 0402, `C3036820`, qty 2990
PWM generator	`PCA9685BS,118`, `C5191634`, qty 95
Flyback/catch diode	`DSK36`, `C106913`, 60 V 3 A, qty 3000
Higher-voltage Schottky option	`S115Q`, `C5362393`, 150 V 1 A, qty 3000
Input ceramic	2.2 uF 100 V, `C2981733`, qty 105
Small HV bypass	100 nF 100 V, `C77058`, qty 4000
Input bulk	150 uF 63 V, `C51018328`, qty 400
Brake connector	`1844210`, `C3584780`, 2-position 3.5 mm, qty 485
Pass device (final, 2026-08-31)	`BUK7D36-60EX`, `C3288253`, qty 1360; 60 V, 2.3 W at `Ta`. Sole pass device — no alternate is placed. Linear-mode operation at 0.71 W / 49 mA analysed in `kicad-libs/docs/brake-current-sink-design.md`; confirm `V_GS(th)` and the SOA curve against NXP's datasheet and a measured case temperature
~~Pass device DNP alternate~~	~~`BSC072N08NS5ATMA1`~~ Rejected 2026-08-31. Its `V_GS(th)` is 3.8 V max at 36 µA, so at 49 mA the gate node lands near 4.6 V against an `OPA197` ceiling of about 4.85 V on a 5 V rail — 0.25 V of margin, consumed by a cold start. And a 60 V part cannot be the weakest thing on the rail: the TMC5130A's inductive-load absolute maximum is 49 V, which is 80% of 60 V, so ringing that threatens the brake FET has already destroyed the driver. Fix that case with a TVS at the 36 V entry, not a bigger brake FET
Precision op amp on hand	`OPA2210IDRGR`, `C2866278`, qty 50; input common-mode does not make it the default low-side loop choice

The stocked TVSs remain unsuitable as a general 36 V clamp: SMBJ28A has too low a standoff voltage, while SMF48A and SMBJ58A clamp too high for the weaker 40-60 V devices in the system.

Parts To Search For And Purchase On JLCPCB¶

Cancelled 2026-08-31. The whole brake column resolved to stock. Every part the brake driver needs is already in JLCPCB inventory and already in kicad-libs/db/source/internal_parts.csv with a symbol and footprint — see the bill of materials in kicad-libs/docs/brake-current-sink-design.md. Specifically cancelled from the tables below: DAC60501ZDGSR, OPA2197IDR, MJD340T4G, TLV7041DBVR, the precision current shunt, the precision DAC scaling network, the default-off clamp transistor, and the brake case-temperature probe. TPS3431SDRBR, MJD122T4G, and IXTA6N50L2 are cancelled from the companion purchase plan for the same reason.

The only two SKUs still worth ordering for the brake are replenishment, not new parts: DAC80501ZDGSR (50 in stock, 2 per board with the angle block) and OPA197IDBVR (100 in stock, about 6 per board across brake, angle, and torque).

The tables below are retained as a record of the search criteria, which remain correct if any of these positions is ever reopened.

Preferred Analog Linear Prototype¶

Function	Search target	Selection guidance
Zero-scale serial DAC	`DAC60501ZDGSR` (12-bit) or `DAC80501ZDGSR` (16-bit)	I2C/SPI selectable, internal reference, zero-scale POR. Prefer the `Z` zero-scale variant, not the `M` midscale variant. A cheaper monotonic 12-bit I2C DAC is adequate if its POR and full-scale behavior are explicit.
Dual loop/sense op amp	`OPA2197IDR` or single `OPA197IDBVR` plus a second buffer	True rail-to-rail input/output, 4.5-36 V supply, low offset, enough output current for a BJT base, and usable at ground on a 5 V supply. Verify loop stability with the pass device and coil.
High-voltage NPN pass device	`MJD340T4G`	300 V, 0.5 A DPAK with published DC SOA and minimum gain at 50 mA. Search alternates by DC SOA at >=15 V/50 mA, not only voltage/current headline ratings.
Precision current shunt	10 ohm, 0.1%, <=25 ppm/C, 0805 or 1206, >=0.125 W	Produces 490 mV at 49 mA and only 24 mW. Buy nearby values for calibration or use a parallel trim position.
Precision DAC scaling resistors	0.1%, <=25 ppm/C matched network or discrete pair	Scale the DAC's worst-case possible output to <=490 mV. Matching and tracking matter more than absolute package size.
Independent overcurrent comparator	`TLV7041DBVR` or similar open-drain comparator	The audited TLV703x/704x datasheet identifies `TLV7041` as open-drain and `TLV7031` as push-pull. Its input range includes ground; include maximum offset and internal hysteresis in the current-ceiling calculation and give the output enough authority to clamp pass-device drive.
Low-noise DAC/reference LDO	`TPS7A2033PDBVR` or similar 3.3 V low-noise LDO	1.6-6 V input, low output noise, high PSRR, enable pin. Feed from the board 5 V rail through appropriate local filtering.
Default-off clamp transistor	`2N7002` or small NPN such as `MMBT3904`	Used to force `V_SET` or pass-device drive low when `BRAKE_ENABLE` is absent or a fault occurs. Choose the polarity so an open control wire means off.
Brake case-temperature sensor	10 kohm NTC probe or board-compatible NTC with harness provision	The datasheet limits case temperature to 75 C. A PCB sensor does not measure a remote brake case; plan a mechanically attached probe if temperature is safety relevant.

Buy enough of the DAC, op amp, pass transistor, comparator, shunt, and scaling network to build at least three independent channels or prototypes. Matching two joysticks and retaining one instrumented spare is more useful than buying only the exact board quantity.

Switching Fallback Purchases¶

LM3404HV, not plain LM3404;
4.02, 4.12, 4.22, 4.32, and 4.42 ohm 1% or better low-TCR sense resistors;
2.2 mH and 4.7 mH shielded inductors, with saturation/fault behavior checked;
36 V-rail TVS candidate in the SMBJ40A or SMBJ43A class, selected only after the supply maximum and TVS clamp curves are known;
output/open-load protection parts selected from the final topology rather than copied from a linear sink.

Proposed Design-Block Boundary¶

Settled 2026-08-31 as brake_current_sink_0_49ma_dac80501, with the port list and the boundary decisions in kicad-libs/docs/brake-current-sink-design.md. Three things changed from the proposal below: the connector is inside the block (the flyback diode has to sit across its pins, and separating the two invites a board that forgets it, so this follows the hv_input_* blocks rather than analog_output_0_3v3_dac80501); the case-temperature connector is dropped because coil heat rather than slip heat sets case temperature at this duty; and brake-body grounding is mechanical, because no 3-position 3.5 mm terminal block is in stock. The proposal is otherwise intact.

For the low-noise implementation, a first private block can expose:

+36V_SW, +5V_SW or the selected analog supply, and BRAKE_PGND;
SCL and SDA, or the chosen serial DAC interface;
active-high BRAKE_ENABLE, default off;
BRAKE_ISENSE, ground-referenced analog output;
~BRAKE_FAULT, if the independent comparator is populated;
the two brake-coil connector pins;
a separately documented brake-body/chassis-ground connection;
local Kelvin current test points and optional case-temperature connector.

Join BRAKE_PGND to board ground once at the defined power-stage return point, using a net tie if separate net names are needed to enforce routing. If one continuous GND net is retained instead, use placement and routing constraints that guarantee the brake return still reaches that point without crossing the torque analog zone.

The existing mouse-board plan assigned InstrumentPort DIG11 to BRAKE_DIM. If the analog design is selected, revisit that name rather than manufacturing a fake PWM signal. BRAKE_ENABLE, ~BRAKE_FAULT, or a DAC update marker may be a more truthful digital observation. AIN1 remains a natural home for buffered BRAKE_ISENSE.

Resolved 2026-08-31: DIG11 becomes BRAKE_CS, the DAC frame, which lets a capture timestamp every brake command — the DAC-update marker option above. DIG12 keeps ~{BRAKE_FAULT}, and AIN1 takes BRAKE_ISENSE. Propagated to mouse-joystick-pcb/docs/interface-map.md.

Keep the first block private and project-specific until bench results validate it. Promote a reusable abstraction only after the mouse board verifies it and a second use establishes which boundary is actually reusable.

Validation Plan¶

Validate the brake and torque paths together, with the motor driver operating, because quiet standalone brake measurements do not establish simultaneous-use performance.

Electrical and thermal:

actual 36 V minimum, maximum, ripple, startup, shutdown, and hot-plug ringing;
brake current accuracy at several commands, cold and after warm-up;
current-loop step response, overshoot, stability, and decay to zero;
pass-device, BCR421, shunt, diode, and PCB temperatures at 100% command;
independent overcurrent threshold and latch/recovery behavior;
boot, MCU reset, DAC communication loss, watchdog loss, and power-domain cycling all produce zero brake current;
open coil, shorted connector, and intermittent harness behavior;
brake case temperature and calculated slip power under expected operation.

Torque and noise:

torque noise floor with brake unpowered, powered at zero, and at several DC currents;
simultaneous motor motion plus brake current at worst expected operating points;
spectra of raw bridge/PGA data, not only filtered torque output;
conducted coupling on bridge excitation, bridge return, analog supply, and reference ground;
increasing-current and decreasing-current torque calibration curves;
unit-to-unit brake matching and temperature drift;
if PWM is tested, carrier harmonics, aliasing, acoustic behavior, mechanical torque ripple, and synchronous-averaging effectiveness;
verify the analog front end never saturates before applying digital filters.

Preserve current, torque, temperature, supply, command, and timing data from the same run. That evidence is what decides whether the linear stage is sufficiently quiet and stable or whether the LM3404HV fallback is necessary.