An Analog Force-Sensing Knee Controller

A design study for a microcontroller-free phase control, isolated from the mains, driven by a force-sensing resistor. A companion to A Force-Sensing Pedal

Claude (Anthropic), at the direction of Robert Drzyzgula

Development Build v82, November 2026

1 Summary

About this paper. The text, analysis, figures and simulations were written by Claude, an AI model made by Anthropic, at the direction of Robert Drzyzgula, who set its scope and questions. It is a companion to A Force-Sensing Pedal (“the pedal paper”), Inside a Switching Circuit (“the switching paper”) and Speed Control of Universal Motors in Sewing Machines (“the speed-control paper”). It is a design study: nothing in it has been built or measured yet. The sources are listed in the last section; please report errors as issues on the project’s GitHub repository.

The question. The pedal paper designed a force-sensing pedal for a stock cordset, where everything has to live at mains potential and run from two wires. The knee controller in the VSM Base has none of those constraints. It has an earthed IEC inlet, room for a certified isolated 5 V supply, and the MCU controller’s optocouplers to carry signals across an isolation barrier. What does a force-sensing knee controller look like when it is designed for that setting from the start, with no microcontroller and no firmware, and owing nothing to the KB3151C?

The answer is the circuit at the heart of industrial phase-control chips, built from a handful of parts on the low-voltage side:

The mains side is the MCU controller’s, unchanged: inlet, fuse, MOV, triac, opto-triac, zero-cross detector and supply. The two designs can share it.

What the simulation shows.

A foot pedal too. §6 adds a passive low-voltage pedal on a two-conductor cable, in parallel with the knee’s sensor, with the MCU controller’s relay restored as the motor’s cutoff and a check for a shorted cable. A printable draft of the pedal is in standalone/fsr-pedal/.

What it is not. It has no speed regulation, needle stops or calibration; the MCU controller (speed-control paper, §9) provides those. This is the “no code” alternative, built to the same safety standard.

2 The starting point

2.1 What the base provides

The pedal paper’s constraints came from the cordset: two wires, no earth, unknown polarity (polarized cordset plugs are wired to the pedal at random, so even those give no help), and everything live. In the VSM Base:

So the knee, its sensor, its cable and all the control circuitry can be SELV, touchable and wired like any low-voltage gadget. Only the triac, its driver’s output, the zero-cross detector’s input, the supply’s input and the protection parts are live, and they sit in one place, on one side of one barrier.

2.2 The principle: ramp and compare

The KB3151C, and the pedal paper’s circuit, fire the triac when a capacitor charging from the line reaches the diac’s breakover. That ties the timing to the line’s voltage, to the diac’s tolerance and to the capacitor’s leftover charge from the last half-cycle, the source of the snap-on (switching paper, §4).

Industrial phase controllers do it differently, and the classic phase- control ICs, Siemens’ TCA785 among them, were built around the difference. They generate a timing ramp that restarts at every zero crossing, and compare it against a control voltage. The triac fires when the ramp crosses the control voltage. The firing angle is then proportional to the control voltage and nothing else: not the line voltage, not the last half-cycle.

The MCU controller does the same thing in firmware: its zero-cross interrupt starts a timer, and the timer’s alarm fires the triac. This design does it with a capacitor and a comparator.

3 The circuit

The low-voltage side. U2’s output transistor, pulled up by RZ, goes high for a few degrees around each zero crossing (Z); Q3 then empties the ramp capacitor CR. The LM334 charges CR at a constant 45 µA the rest of the time. RM and the FSR divide the 5 V supply into the command D, filtered by CF. When the ramp rises past D, U4’s output goes low and lights U3’s LED, which fires the triac. Everything to the left of the barrier is SELV; the mains side is the MCU controller’s.

3.1 The mains side

It is the MCU controller’s line side, unchanged (hardware/mcu-controller/, and the speed-control paper’s §11): the earthed IEC inlet, the T2A fuse, the 150 VAC MOV, the snubberless triac (the BTA16-600BW the project has bought) fired by the MOC3023, the H11AA1 zero-cross detector fed from the line through R7 and R8 (22 kΩ each), and the IRM-05-5 on its own T250mA fuse. The pads for an RC snubber stay, unfitted.

Two things differ:

3.2 The low-voltage side

The zero-cross window (U2, RZ). U2’s two LEDs carry the line current through R7 and R8, so its output transistor is on, holding Z low, for most of each half-cycle. Near a zero crossing the LED current falls too low, the transistor turns off, and RZ pulls Z high. How wide that window is depends on the pull-up: with 100 kΩ, the transistor needs only 50 µA, and with the H11AA1’s guaranteed current-transfer ratio of 20% that takes 0.25 mA in the LEDs, about 11 V on the line, within 4° of each zero crossing. The H11AA1’s transfer ratio is specified at 10 mA and falls at small currents, so a real window may be wider; §4 tests ±8°, which barely matters. (With the MCU controller’s 10 kΩ the window would be ten times wider, which is fine for timing an interrupt but would eat into the firing range here.)

The ramp reset (Q3). Z drives a 2N7000’s gate. While Z is high, Q3 shorts CR, so the ramp starts every half-cycle from zero.

The ramp (LM334, RSET, CR). The LM334 is a three-terminal adjustable current source: its current is 67.7 mV divided by RSET, at room temperature. With RSET at 1.5 kΩ it gives 45 µA, which charges CR (0.1 µF, film) in a straight line to about 3.6 V by the end of the half-cycle. A straight line matters: it makes the firing angle simply proportional to the command. The LM334 needs about a volt of headroom, which 5 V leaves.

The command (RM, FSR, CF). RM (22 kΩ) runs from 5 V to the command node D, and the FSR from D to ground. With the knee at rest the FSR is open, D sits at 5 V, above the ramp’s peak, and nothing fires: no current, no hum. Pressing lowers the FSR’s resistance and pulls D down. CF (0.1 µF) filters D, with a time constant of at most 2 ms, too short for a knee to notice. The FSR never sees more than 5 V or carries more than about 0.23 mA, well within Interlink’s typical drive and current limit.

The comparator (U4). An MCP6562, a dual comparator with rail-to-rail inputs and a push-pull output that runs from 5 V. The ramp goes to its inverting input and D to its non-inverting one. While the ramp is below D the output is high and U3’s LED is off. When the ramp rises past D the output goes low, and the LED lights through RL (330 Ω, about 11 mA, against the MOC3023’s 5 mA trigger current). Rail-to-rail inputs matter because D sits at 5 V at rest; a comparator like the LM393, whose inputs work only up to 1.5 V below its supply, would need the divider scaled down. The second comparator in the package is spare.

Gating to the end of the half-cycle. Once the ramp passes D it stays above it until the next reset, so the LED stays lit from the firing angle to the end of the half-cycle. That is deliberate. A universal motor’s current lags the voltage, and a triac given only a short gate pulse may let go before the current has built up past its latching level. Held gate drive keeps the triac on until the current flows. And because Z resets the ramp just before each zero crossing, the drive ends before the line reverses, so the triac can’t be fired at the very start of the next half-cycle by a leftover drive.

3.3 Why it starts smoothly

Each half-cycle starts from a reset ramp, so nothing carries over from the last one. As the knee presses past the start, D falls just below the ramp’s peak, the triac fires at the very end of the half-cycle, and the motor gets a sliver of each half-cycle: it starts from zero. There is no snap-on, the RC pedal’s habit of jumping to a quarter of line voltage (pedal paper, §4.3; switching paper, §4).

3.4 Parts

Part Value Notes
U4 MCP6562 (Microchip), DIP-8 Dual comparator, rail-to-rail inputs, push-pull
LM334 LM334Z (TI), TO-92 Adjustable current source
RSET 1.5 kΩ, 1% Sets 45 µA
CR 0.1 µF film, 5% The ramp’s timing; film for stability
Q3 2N7000 (TO-92) or 2N7002 Ramp reset; on order
RZ 100 kΩ Zero-cross pull-up
RM 22 kΩ, 1% Sets the force range (§4)
CF 0.1 µF ceramic Command filter
RL 330 Ω U3’s LED current, about 11 mA
FSR Interlink UX 406 or 402 At 5 V either is within its current limit
Rest switch Mains-rated snap switch In series with the motor; the knee design’s existing switch
Mains side As the MCU controller Triac, MOC3023, H11AA1, R7/R8, IRM-05-5, fuses, MOV, inlet

Apart from the MCP6562 and the LM334, everything is already on the project’s order or in the drawer.

4 The circuit simulated

4.1 The model

The figures come from tools/make_analog_knee_figures.py. Unlike the pedal’s RC network, this circuit has no state carried between half-cycles, so its behavior can be computed directly:

4.2 One operating point

At 6 N on the sensor, the FSR is about 14 kΩ and the command 1.92 V. The ramp (blue) crosses it at 96°; the MOC3023’s LED (purple) is on from there to the reset window (green) before the next zero crossing, and the motor gets the shaded part of each half-cycle.

4.3 Force against speed

Motor voltage against force on the sensor, for three values of RM.

With RM = 22 kΩ:

Force on the sensor Motor voltage
1.0 N Starts, from zero
2.3 N 25%
4.1 N 50%
7.9 N 75%
15 N 90%
50 N About 99%

RM sets where the range falls: 10 kΩ moves the start to 2.8 N and 90% to 40 N; 47 kΩ moves them to 0.4 N and 5.8 N. The motor can reach nearly full voltage, because the earliest firing is just after the reset window ends. The curve is roughly logarithmic in force through its middle, as in the pedal, from the same cause: the FSR’s power law.

Line voltage doesn’t matter. The firing angle depends only on D and the ramp. A low line gives the motor less voltage at a given angle, as any phase control does, but doesn’t move the angle.

4.4 What moves the curve

The nominal curve, and the curve with each source of variation at its likely extreme.
Variation Starts at 90% at
Nominal 1.0 N 15 N
Ramp 10% slower 1.5 N 17.5 N
Ramp 10% faster 0.6 N 12.8 N
FSR 25% higher resistance 1.4 N 19.8 N
FSR 20% lower resistance 0.8 N 11.3 N
Reset window ±8° 1.3 N 16.9 N

None of these approaches the factor of two the diac’s tolerance causes in the pedal circuit (pedal paper, §4.4). RM as a trimmer, or a fixed RM chosen after measuring the sensor, covers all of them.

5 The knee

5.1 Force, or position?

An FSR measures force. The knee controller as it stands is a position device: the knee pushes a bar through about an inch of speed range against a light spring, and a pot measures where the bar is (the User’s Guide’s knee-controller chapter). The original Singer knee controller worked the same way. A force sensor can still serve a moving lever, because a spring turns position into force. But the shape of that conversion matters a great deal.

Speed along the knee’s travel, from the rest switch closing at 1 inch to full travel at 2 inches, when the sensor is loaded by a linear spring (solid) and by a progressive stack (dashed), for 20 N and 50 N on the sensor at full travel.

A linear spring crowds the range. If the knee compresses an ordinary spring onto the sensor, force rises in proportion to travel. But the FSR’s resistance falls as a power of force, so the first few newtons do most of the work. With 50 N at full travel, the simulation puts 50% of line voltage 1.6 mm past the switch point and 90% at 7.2 mm, out of 25 mm of speed range. The rest of the inch does almost nothing. A softer spring (20 N at full travel) helps a little but not much.

There are two ways out.

1. A progressive stack. If the force on the sensor rises geometrically with travel, each millimeter multiplying it by the same factor, then the FSR’s power law and the log-shaped speed curve cancel, and the speed spreads over the whole range: 25% at 5.4 mm, 50% at 9.1 mm, 75% at 13.3 mm and 90% at 17.5 mm past the switch point, with 50 N at full travel. Practical ways to get a progressive force:

2. A near-isometric knee. Alternatively, make the knee lever nearly rigid: a few millimeters of travel against a stiff spring, and speed set by how hard the knee presses, not how far. That is the FSR’s natural use, and the carbon pile’s way of working (speed-control paper, §4): force in, speed out. It would feel different from the Singer knee controller, firmer and more like pressing a button with the knee, and whether that is better or worse is a question for the bench, not the simulation.

The FSRs on order make this an easy experiment. A 402 under a printed lever, read by this circuit or simply by a meter, answers how each arrangement feels before any of them is built into the base.

5.2 Fitting the existing knee

The knee controller’s hub, crank, spring housing, rest and stop screws and snap switch all stay. What changes:

The sensor’s tail and its two wires are SELV, so they run to the board like any low-voltage lead.

A printable draft (standalone/fsr-knee/fsr_knee_housing.scad, with its README) puts the sensor beside the return spring rather than under it: the housing’s bottom is already less than 5 mm above the table, and the 402’s pad is wider than the spring’s bore. A sensor tower on the shaft side, under the crank, holds a 402, a puck, a soft and a stiff spring nested together, and a plunger that the crank meets just before the switch point. At 38 mm from the shaft the plunger moves 3.2 mm by full speed and 4.0 mm by the stop, and the stack reaches 54 N at full speed, sized for §6’s 10 kΩ divider. Since the sector gear goes, the rest switch’s cam needs a new home on the shaft.

6 Adding a foot pedal

The same circuit can take a foot pedal as well as the knee lever, or instead of it. At low voltage the pedal can be passive: a force sensor and a resistor, on a light two-conductor cable to a jack on the base. This section covers the changes to the base’s circuit; the pedal itself is a printable draft in standalone/fsr-pedal/ (§6.4).

The additions. The pedal’s FSR, with 470 Ω in series, reaches the command node D through jack J7, in parallel with the knee’s sensor (also with 470 Ω in series); RM drops to 10 kΩ. Comparators C2 and C3 close the relay K1 whenever a sensor is pressed and the command isn’t shorted. C1 is §3’s firing comparator, unchanged.

6.1 One command node, two sensors

The pedal’s sensor and the knee’s are wired in parallel at D. With neither pressed, both are open and D sits at 5 V. Press either and its resistance pulls D down; press both and the lower resistance, the harder press, wins. No switch is needed to choose between them.

Two things change on the base’s board:

Motor voltage against force on either sensor, with the shared command.

Simulated with the same assumed UX curve, the motor starts at about 2.8 N on the sensor, reaches 25% at 6.6 N, 50% at 12 N, 75% at 24 N and 90% at about 50 N, and tops out at 98%. Those are forces beyond whatever the pedal’s springs carry: the springs hold up a resting foot (§6.4), and the sensor feels only the push beyond it. A firm press is roughly 50 to 100 N, so this suits a foot pressing the sensor directly, with no force divider. The knee’s sensor stack should then be sized to reach about 50 N at full travel, the “progressive, 50 N” curve of §5.

6.2 The relay comes back

With a pedal, the knee’s mains rest switch can’t be the motor’s cutoff: it would be open, at rest, whenever the pedal is in use. So the MCU controller’s relay K1 returns, with its contact in series with the triac on the mains side, and the low-voltage side closes it only when there is a reason to:

C1, C2 and C3 fit one MCP6564, the quad version of the MCP6562, with one comparator spare.

6.3 When things go wrong

Fault D Result
Pedal unplugged 5 V (open) Off; the knee still works
Cable cut, or a sensor’s lead broken 5 V (open) Off
Cable or jack shorted 0 V C3 opens K1 at once: off
A plug inserted while running (a jack shorts briefly as the plug goes in) 0 V for an instant K1 drops out; press again
Triac shorted — Runs only while a sensor is pressed (K1); off at rest
Sensor shorted inside 0.22 V C3 opens K1: off

The faults the circuit can’t catch are the slow ones: a sensor drifting so that a light press reads as a harder one. That shows up as a pedal that feels touchy, and a check with a meter answers it: unloaded, the sensor should read above 1 MΩ.

6.4 The pedal and its cable

The connector. A two-conductor cable, shielded, with the shield as 0 V and the center conductor as D. An RCA (phono) plug works; a 6.35 mm (1/4”) mono phone plug, the kind on a keyboard’s sustain pedal, is sturdier and harder to pull out by accident. Either way, use a connector that no mains plug can mate with, and expect the brief short as it goes in; C3 handles it. Strain-relieve the cable at both ends.

The pedal’s board. The pedal needs almost no circuit: the 470 Ω resistor, the sensor’s tail soldered to two pads, and the cable’s two conductors. On perfboard that is about 25 × 20 mm; the draft allows 35 × 25 mm, for room to solder.

The draft (standalone/fsr-pedal/fsr_pedal.scad, with its README) is a heel-hinged tread over a rigid pedestal. The UX 406 lies flat on the pedestal, with the board beside it on the same plane, so the sensor’s tail never bends. A 22 mm puck under the tread, faced with a 1.5 mm rubber pad, presses the sensor’s center. Two springs near the toe hold the tread up, carry a resting foot and return it; two posts beside the pedestal stop the tread after 2 mm of travel at the puck, so a stamp can’t crush the sensor. It is low voltage throughout, so it needs no insulation beyond keeping the cable’s strain relief honest.

7 Safety and construction

7.1 The barrier

The design’s safety rests on keeping the mains side and the SELV side apart, with only three parts bridging them: the IRM-05-5, the H11AA1 and the MOC3023, each certified for the isolation. On the board:

7.2 Earth

The inlet’s earth should go to anything metal a user can touch that a fault could make live. In the base, that is chiefly the machine itself: an earth lead to the machine’s bed, through a ring terminal under one of its screws, earths the cast iron, which a two-wire cordset never did.

The SELV side can float, or be tied to earth at one point to keep noise down. Either is safe: the barrier, not the earth connection, protects the user.

7.3 The rest switch

The snap switch at the knee’s rest position is in series with the motor, on the mains side. It must be rated for the motor’s current at 125 VAC, and mounted so its terminals are covered. Because it opens whenever the knee is at rest, a shorted triac, a stuck comparator or a damaged sensor can at worst run the motor while the knee is pressed, never when it is released.

8 Testing it, and the open questions

8.1 Testing order

  1. The low-voltage side alone, on the bench. Power it from a 5 V bench supply, with a signal generator’s 60 Hz square wave standing in for Z. Watch the ramp, D and the comparator output on the scope; press the FSR and see the comparator’s edge move along the ramp. No mains involved.
  2. The sensors. Measure each FSR’s resistance against force (the pedal paper’s §6 describes a kitchen-scale method), and rerun tools/make_analog_knee_figures.py with the fitted curve. Pick RM.
  3. The whole controller, with a lamp. On the isolation transformer, with a 60–100 W incandescent bulb as the load: check the zero-cross window’s width, the firing against force, and that the LED drive ends before each zero crossing.
  4. With a motor on the rig: check that the held gate drive fires the triac reliably at every angle, and how the start feels.
  5. The knee mechanics: a printed test lever with a 402, trying a linear spring, a nested pair and an elastomer pad, before committing a design to the base.

8.2 Open questions

9 Sources

  1. Interlink Electronics, FSR 400 Series Integration Guide and FSR X & UX Series Integration Guide: resistance against force, drive voltage, current limit, repeatability and drift.
  2. MOC3023M random-phase triac driver optocoupler, onsemi data sheet.
  3. H11AA1 AC-input optocoupler, Vishay data sheet: current-transfer ratio (20% minimum at 10 mA) and isolation.
  4. LM334 three-terminal adjustable current source, Texas Instruments data sheet: set current, temperature coefficient, headroom.
  5. MCP6561/2/4 push-pull output comparators, Microchip data sheet. G5LE relay, Omron (now Aratas) data sheet: coil current.
  6. BTA16, BTB16 and T16 series 16 A triacs, STMicroelectronics data sheet.
  7. Siemens (later Infineon), TCA785 phase control IC data sheet: the synchronized ramp and comparator this design follows.
  8. A Force-Sensing Pedal, in this repository (docs/fsr-pedal/).
  9. Inside a Switching Circuit, in this repository (docs/switching/).
  10. Speed Control of Universal Motors in Sewing Machines, in this repository (docs/motor-control/), §9 and §11; and the MCU controller’s schematic and README in hardware/mcu-controller/.
  11. The User’s Guide’s knee-controller chapter, for the knee’s mechanism and feel.

The simulation and the figures are generated by tools/make_analog_knee_figures.py in the project repository. The schematic is drawn by the same script; there is no KiCad source yet.