Uncle Magic's Liquid Turntable: Hardware Overview#

How the turntable is built: what the parts are, how they are connected, how power flows, and what protects what. For operating it, see the User Guide. For exact pin-by-pin detail and the history of design decisions, see the README and the changelog; the KiCad files in kicad_mainboard/ are the authority on connections.

1. System overview#

The turntable is built from two printed circuit boards, a stepper motor and the panel controls, fed by two external DC supplies.

Three computers share the mainboard, all running at 3.3 V:

All three are socketed: they plug into female headers, so each can be replaced or reflashed.

Power flow#

Signal flow#

2. The mainboard#

Schematic sheets#

Module orientation#

Each module fits its socket only one way up. The board is silkscreened to match, but a module plugged in backwards would put power on the wrong pins.

Key parts#

3. Power#

Rails#

The three rails do not share a protection path: a fault on one cannot ride through into another. The 5 V supply is used only for the LED ring; everything else runs from the 24 V supply through U2.

Design choices worth knowing#

• 3.3 V is generated on the board. It comes from a fixed-output converter rather than an adjustable module, because three logic chips with a 3.6 V maximum depend on it and a drifting trimpot in a sealed enclosure would be a real overvoltage risk. U2 needs no external capacitors.
• Both ESP32 chips run Wi-Fi. Peaks approach 1 A on the 3.3 V rail, which is why the 2 A converter is fitted and why the main board's Wi-Fi power is capped (15 dBm).
• The ESP32-S3-ETH's own 5 V regulator is disabled (its 3V3_EN pin is grounded through the socket) so it doesn't fight U2. Its USB-C port still works for programming but cannot power the board by itself.
• Switch-off behaviour. Without R27 the 24 V capacitors would linger at a few volts and leave the processors in brown-out. With it the rail is under 1 V in about 2 seconds.
• Switch-on order. 3.3 V can't rise before 24 V. The 5 V rail may arrive at any time; the worst case is a few milliseconds of 3.3 V on the data buffer's input before it has power, limited to about 10 mA by R18.

Power budget#

Mains side: two switch-mode supplies together draw about 100 to 120 W in steady state, with a 20 to 60 A inrush for a few milliseconds at switch-on. A slow-blow fuse and a switch rated at 10 A or more are recommended.

Power connector J-PWR-IN#

Molex Mini-Fit Jr, 8 positions, two rows of four. The 5 V is on two pins because of the ring current.

These pins connect ahead of the fuses. Never plug or unplug this connector with power on.

4. Signals and control#

Main controller pin use#

Every pin on the ESP32-S3-ETH socket was checked against Waveshare's published pinout, and the TMC2209 and DevKitC sockets against their makers' pinouts.

The motor driver#

• Mode: step and direction, configured over the UART at start-up. The driver's address pins and clock are tied to ground; the carrier's trimpot is ignored because the firmware sets the current.
• Settings: 800 mA RMS (the motor is rated about 1 A), 1/32 microstepping with the driver interpolating to 256, and the quiet StealthChop mode. At 32 microsteps a full turn is 6,400 steps.
• Health checks: every 5 seconds the firmware checks the serial link and the driver's over-temperature and short-circuit flags, and restores the settings if the driver has reset.
• Boot safety: three of the control pins (GPIO 0, 45 and 3) double as the processor's boot "strapping" pins. R23 (1 k to 3.3 V) holds STEP high, R25 (1 k to ground) holds DIR low, and R24 (10 k to 3.3 V) holds the driver disabled from power-on until the firmware is ready. This makes boot independent of whatever pull-ups the driver module has.

The lighting path#

1. The main controller sends the colour as a short text (JSON) message over the serial link at 115,200 baud.
2. The WLED DevKitC turns it into LED data on its GPIO 4.
3. The data passes through R18 (330 Ω) into U3, which lifts it from 3.3 V to 5 V (it must be the AHCT part, whose inputs accept 3.3 V as a "high"), and on to the ring.
4. R26 holds the data line low while WLED starts, so the ring stays dark rather than flashing.

Panel controls and the expander#

• The four encoders and the tap button go straight to processor pins, and the encoders are read by interrupts so fast turning isn't missed.
• The hue, saturation and value push buttons are on the MCP23017 expander, because the socket has no spare pins. They are read by polling every 20 ms.
• The expander also carries three spare encoder headers (J-ENC-EXP1 to 3; the third has no push-button channel). The current firmware does not read them. Two expander pins (GPA7, GPB7) are outputs only, as the chip's errata requires, and are not brought out.
• Static protection. Every panel signal pin goes through a 1 k resistor, with a small TVS diode to ground on the connector side, so a spark from a person touching the panel wiring is clamped before it reaches a processor pin. Each encoder's 3.3 V supply pin has a 100 Ω resistor in series, and the LED ring's 5 V supply has its own TVS (D32) beside J5.
• Encoder modules are the common KY-040 type. Their pull-ups are fed from 3.3 V, never 5 V, because the lines go straight into 3.3 V-only inputs.
• R28 (1 k) sits between GPIO 43 and the main encoder's push button. That pin is also the processor's boot-log output, so the resistor limits current if the encoder happens to be closed to ground during a reset.

Encoder header pin order#

All eight encoder headers use the same five pins, in this order:

(The board is silkscreened V, S, D, C, G.) The tap button header has two pins: GND and TAP. Suggested panel wire colours: main encoder blue, hue red, saturation green, value yellow, buttons orange, ground black.

Network#

Wired Ethernet comes from the W5500 chip on the ESP32-S3-ETH module. Wi-Fi is the module's own 2.4 GHz radio. Both are active at once and share the same OSC port and web pages.

5. Connector reference#

The motor connector follows the common stepper wiring, so a motor cable wired black, green, red, blue (A+, A−, B+, B−) plugs straight in. If the platter turns the wrong way, swap two wires of one coil, or tick Reverse motor direction on the settings page.

6. The LED ring board#

A straight-through six-wire cable joins J5 on the mainboard to J1 on the ring. Power and ground are spread across the board as copper planes (5 V on the top layer, ground on the bottom) instead of traces; an estimate of the voltage drop to the farthest LED came out under 12 mV. Both boards pass design-rule and electrical-rule checks with no errors.

7. Motor and mechanics#

• Recommended motor: StepperOnline 17HS08-1004S: NEMA 17, 1.8° per step, 1 A per phase, 3.7 Ω, 4.5 mH, 16 N·cm holding torque, 21.5 mm body, 400 mm leads (black, green, red, blue = A+, A−, B+, B−). It is a short ("pancake") motor, which is enough torque for a light platter. A heavy platter would justify a full-length NEMA 17 (40 to 60 N·cm); the electronics suit either.
• Coupling: a flexible shaft coupler joins motor and platter directly. There is no gearbox, so platter inertia is felt straight at the motor. This is why every speed change is ramped.
• Speed: one revolution per 12 beats; default top speed 17 RPM (adjustable). At 1/32 microstepping the platter needs 6,400 steps per turn.
• Layout constraints on the mainboard: the bulk capacitors C8 and C11 are 7.7 mm tall and fit between the two plug-in modules, so check clearance if you substitute taller parts.

8. Current, heat and protection notes#

• Ring current path. The 5 V route from the power connector through F3 and Q5 to J5 is carried by copper pours 6.5 mm wide or more at the narrowest point, with the return on a full bottom-layer ground plane. The power connector and ring connector each use two pins per supply, so each pin carries about 5.4 A at full white.
• Q5 heat. At full white it dissipates roughly 0.8 to 1.1 W. The firmware's WLED setup (a current limit of about 8 A) keeps this lower in practice.
• U2 heat. It runs only slightly warm at the loads above. No signal traces pass under it.
• TMC2209 heat. Fit a heatsink on the module; the maker calls for heat dissipation.
• Fuses. F1 (2.5 A) protects the 24 V side and F3 (15 A) the ring. Both clip types take the same standard-size fuse, so the board relies on its silkscreen labels to prevent mix-ups.
• Voltage spikes. D5 softens spikes on the 24 V rail, but a TVS cannot clamp as low as the motor driver's 28 V rating, so the design relies on a fixed, well-regulated 24 V supply and on never hot-plugging the power connector. An adjustable supply should not be set above about 26 V.
• No hot-plugging. Do not connect or disconnect the motor (J3), the ring cable (J5), the power connector or any of the plug-in modules with the power on. A motor unplugged live produces a voltage spike that can destroy the TMC2209 (the module's maker says the same about the module itself), and the ring cable carries up to about 11 A. The D5 clamp on the 24 V rail is not designed for this.
• USB back-feed. Plugging a USB cable into the WLED DevKitC while main power is off feeds the whole 3.3 V rail through the DevKitC's own regulator. It usually just causes a brown-out, but apply main power first. (A board revision could prevent it with a diode.)

9. Manufacturing and files#

Ordering from JLCPCB. Surface-mount only: 24 part types, 68 solder joints. Full assembly (through-hole parts too): 49 parts, 267 joints. Every part has an LCSC number (JLCPCB re-checks stock when you upload). Notable ones: regulator U2 C45406999, data buffer U3 C12495, TVS D5 C123820, bulk capacitors C8 and C11 C176683 (an "extended" part: confirm stock when ordering).

Before ordering the board, check:

• 2 oz copper is selected on the order form.
• In JLCPCB's placement preview, the orientation of Q5, C11 and U2 looks right.
• You have standard or low-profile ATO fuses (a mini fuse needs a different clip spacing).

10. Verification status and limits#

Checked on the computer

• Both boards pass design-rule checks (no violations, nothing unrouted) and the mainboard matches its schematic exactly. The schematic passes electrical-rule checks.
• Socket pin assignments match the manufacturers' published pinouts.
• The board meets JLCPCB's two-layer manufacturing limits (smallest hole 0.3 mm, narrowest track 0.2 mm).
• The speed-control logic has automated tests (tests/), and the firmware builds with no warnings.

Checked on a bench unit. A standalone ESP32-S3-ETH running the current firmware boots, joins Wi-Fi, announces liquidturntable.local and serves its web pages. The automatic renaming of a taken hostname was tested against a name owned by a PC on the network: the unit detected the clash and announced itself as -1.

Not yet verified. No fully assembled mainboard has been powered up yet. Still to confirm on real hardware: the motor driver link and motor behaviour, the boot straps with the driver module fitted, the WLED link and ring, encoder feel, and the fuse clip fit with real fuses. The project README lists these under "Open items".

Known limits

• The four spare encoder headers and the expansion header are wired but not used by the firmware.
• Wi-Fi shares the 3.3 V rail with the wired network and the WLED chip; its power is capped to keep the rail within budget.
• The panel and ring cabling are made by hand. Check the pin order on the silkscreen, not on the connectors' descriptions in text.