pico-serial — hardware spec (draft 0.3)¶

A USB-C to DB25 RS-232 adapter built on a bare RP2040 (the chip on the Raspberry Pi Pico). There is no Pico module: the chip, flash, crystal, regulator and every passive part are on this board.

It works in both serial roles through one female DB25:

A gender changer is wired straight through, so it only changes the plug shape. The board switches roles electronically: it has one transceiver for each role, both on the same DB25 pins, and only one is powered at a time. The role is auto-detected from whatever is plugged in, or can be fixed in firmware (§5.2).

The main DCE use is a phone running the D-Modem app plugged into a vintage computer: to the computer it looks like an external modem. DTE mode turns the same board into an ordinary USB-to-serial adapter for talking to real modems.

Sourcing: one LCSC order. - JLC assembles every surface-mount part from LCSC stock. - The DB25 connector and hood also come from LCSC (quantity 1 is fine) and are soldered and fitted by hand. - DigiKey options are listed as backups. - Stock was checked on 2026-10-17. The parts picker page compares every option.

1. Requirements¶

2. Block diagram¶

 USB-C ──┬─ CC1/CC2: 5.1k Rd each ─ GND
 (UFP)   ├─ VBUS ─ PTC 0.5A ─ TVS ─┬─ LDO 3.3V ─── 3V3 rail
         │                         └─ VBUS sense (÷2) → GP29
         └─ D+/D- ─ USBLC6 ─ 27Ω ─ RP2040 USB

                     ┌── UART0 pins set A (GP0–3) + GP4–7 ──┐
 RP2040 ─────────────┤                                      ├── DB25 female
   │                 └── UART0 pins set B (GP12–15) + GP8–11┘     (same 8 pins)
   │                      │                        │
   │               SP3238E (5 TX / 3 RX)    MAX3243E (3 TX / 5 RX)
   │                 "chip A": DCE            "chip B": DTE
   │               SHUTDOWN/R1OUT/STATUS    FORCEOFF/INVALID → GPIO
   ├─ QSPI ── W25Q128JV flash
   ├─ 12 MHz crystal
   ├─ BOOTSEL + RESET buttons, SWD pads
   └─ optional LEDs

3. Bill of materials (main parts)¶

Rough parts cost per board, before the PCB and JLC's assembly fees: about $12–14 including passives.

Parts avoided and why: - ADI MAX3238E: tiny stock at $8–14 each (LCSC and DigiKey). - TI MAX3238E: 0 stock or obsolete. TI MAX3238IPWR (not "E"): 49 on LCSC. - Amphenol L77SDB25SOL2: shows up in DB25 searches but is a vertical through-hole part, not solder cup. - MH MHDTPPK25-DB25P-K: the hood kit comes with a male connector. - ST ST3243EBTR: none until April 2027. - Plain SP3243E (no "B"): only 120 kbps. - W25Q16JVSSIQ: marked obsolete. - Bourns MF-MSMF050-2: not recommended for new designs. - True straddle-mount female DB25s: none stocked at qty 1 (§7.1).

LCSC asks for 5–10 of some cheap passives and protection parts; that adds only cents. The connector and hoods have a minimum of 1.

Passives are 0402 (0603 where noted). X7R or better for anything on the RS-232 charge pumps.

4. Circuit detail¶

4.1 Why RP2040 and not RP2350¶

• The RP2040's core regulator is a linear LDO that needs only two capacitors. The RP2350's is a switcher that needs a 3.3 µH inductor and careful layout, which is hard to fit in a DB25 hood.
• The job needs one UART, under 30 GPIO and full-speed USB.
• The RP2040 is $0.76 with 100k+ in stock.

4.2 RP2040 support parts (Raspberry Pi's minimal design)¶

Firmware: PICO_XOSC_STARTUP_DELAY_MULTIPLIER=64 for crystal start-up margin.

4.3 Power¶

• VBUS → 0.5 A PTC → SMF5.0A → AP2112K-3.3 → 3V3.
• VBUS → 100 kΩ / 100 kΩ → GP29 (ADC3), as on the Pico.
• Budget: RP2040 ~25 mA, the active transceiver ~1 mA plus 10–20 mA into its loads, the idle one ~1 µA in shutdown, LEDs (if fitted) ~10 mA. Under 80 mA.
• Keep both transceivers' VCC on all the time; only the shutdown pin puts one to sleep. Not every part's datasheet specifies off-state leakage with VCC unpowered.

4.4 USB-C¶

• CC1 and CC2 each get their own 5.1 kΩ to GND.
• Both D+ pads tied together, both D− pads tied together.
• Shell → GND through 1 MΩ ∥ 4.7 nF (0 Ω option to bond directly).
• SBU1/SBU2 not connected.
• D+/D−: short and matched. 90 Ω differential if the stack-up allows.

4.5 The two transceivers on one connector¶

Why two chips (decided): an RS-232 chip's pins have fixed directions. In DCE mode pins 3/5/6/8/22 are outputs and 2/4/20 are inputs. DTE mode is the exact opposite. A gender changer is straight-through and can't swap directions, so every signal pin needs both a driver and a receiver available: a 5T/3R chip plus a 3T/5R chip. Alternatives ruled out: - one chip + a mechanical crossover switch: still the wrong number of drivers and receivers, needs 8 poles, and is unreachable in a closed hood; - one 8T/8R chip: none found in common stock; - ±12 V analog switches: cost and space; - one DCE chip + a null-modem adapter for DTE use: a standard null modem merges DCD and DSR and usually drops RI, so DTE mode would lose separate carrier and ring sensing.

Each of the 8 signal pins has one driver and one receiver: the driver from one chip and the receiver from the other.

In DCE mode chip A is on and chip B is held in shutdown; in DTE mode it's the other way round. Chip A's SHUTDOWN pin and chip B's FORCEOFF pin do the same job. Both are pulled low at power-up, so both chips start off. Why this works:

• Idle drivers are high-impedance in shutdown.
• MAX3243E: ±25 µA max at ±12 V, even with VCC = 0.
• SP3238E: its datasheet says only "High-Z" and gives no figure at ±12 V. Check the leakage on the bench, or fit the TI TRS3238E (±25 µA at ±12 V) from DigiKey.
• Idle receivers' outputs are high-impedance. The exceptions are the always-active outputs: SP3238E's STATUS and non-inverting R1OUT (pin 16), and MAX3243E's INVALID and ROUT2B. Those go to their own GPIOs.
• Even with both chips on (a firmware bug), no two of our drivers share a pin. The worst case is fighting the far end's driver, which RS-232 drivers are short-circuit protected against.

Known trade-offs:

1. Extra load. The idle chip's receiver (3–7 kΩ, 5 kΩ typical, to GND) stays on every pin the active chip drives. A driver therefore sees ~2.5 kΩ typical and 1.5 kΩ worst case instead of 3–7 kΩ.
2. The datasheets guarantee ±5 V only into 3 kΩ. Into 2.5 kΩ, expect about ±5 V.
3. Receivers switch at 2.7 V or less, so there is still margin, but check the swing on the bench at 230 400 baud with all five DCE drivers loaded.
4. Larger charge-pump capacitors (below) help.
5. Overvoltage on idle drivers. Their outputs are rated to ±13.2 V absolute maximum. Normal ±12 V-rail RS-232 kit stays below that under load. A rare far-end driver running from ±15 V could exceed it. If that matters, fit bidirectional TVS diodes on pins 2–6, 8, 20, 22; there are no footprints for them by default.

Capacitors, per chip: - C1 (flying cap): 0.22 µF. Both SP3238E and TRS3238E are specified with 0.22 µF across a 3.3 V ±10 % supply. - C2–C4 (V+/V− reservoirs): 1 µF. - VCC: 1 µF. TI asks for ≥1 µF here to meet the IEC ESD rating. - All X7R, ≥16 V on V+/V−. Keep each chip's pump loop short.

ONLINE (SP3238E) and FORCEON (MAX3243E) share one GPIO and go high together. That turns automatic power-down off, so whichever chip its shutdown pin enables stays on.

4.6 GPIO map¶

UART0 can be routed to GP0–3 or GP12–15, so firmware gives the UART to the active chip's pins and leaves the other set as inputs. No analog switches are needed.

4.7 Buttons¶

BOOTSEL and RESET are on the board but can't be reached with the hood closed. Normal firmware updates don't need them: the USB reset interface (§5.1) reboots the board into the bootloader. The buttons are there for bring-up and recovery with the hood off.

4.8 LEDs (optional)¶

There are six 0402 LED + 1 kΩ footprints on GP22–27, not fitted by default. They're only useful with a clear hood or with the hood off. They are the first thing to drop if layout space runs out. With no LEDs fitted, GP22–27 become spare pins.

5. Firmware (outline)¶

Implemented in firmware/ (v0.1, wired board). Details settled there: - detection samples every 50 ms and needs 3 matching samples; - the vendor protocol is in firmware/PROTOCOL.md; - no 1200-baud reboot, since 1200 baud is a real modem speed; - TinyUSB 0.21 is used for SERIAL_STATE; - the baud rate is learned from the computer's typing (§5.7).

5.1 USB¶

• Pico SDK + TinyUSB composite device: CDC-ACM + the SDK's reset interface (so picotool or the app can reboot it into BOOTSEL).
• SET_LINE_CODING → UART format. RP2040 doesn't do mark/space parity or 1.5 stop bits; reject those.
• SEND_BREAK → UART break. Errors and breaks received → SERIAL_STATE.
• USB IDs: a free PID from Raspberry Pi's PID program (VID 0x2E8A) or from pid.codes.

5.2 Choosing the role¶

The mode setting is auto (default), DCE or DTE. It is stored in flash and set with a vendor request (§5.5).

In auto mode, both chips start in shutdown. Each chip's status output keeps working in shutdown and goes high when any of its own receiver pins carries a valid RS-232 level. The board checks about every 250 ms: - GP20 (A STATUS) high → something drives pins 2/4/20, i.e. a computer is attached → turn on chip A (DCE). - otherwise GP21 (B INVALID) high → something drives pins 3/5/6/8/22, i.e. a modem is attached → turn on chip B (DTE). - otherwise nothing is attached → stay off.

GP19 (chip A's always-on, non-inverting R1OUT) shows activity on the computer's TXD while chip A is off. Idle mark reads low on this pin, the opposite of a normal inverting receiver output.

While a role is active, if the active chip's status output stays low for about 1 s (the cable was pulled or the far end switched off), both chips go back to shutdown and detection starts again. When the mode changes, the board reports it in GET_INFO and sends a SERIAL_STATE with all lines off.

5.3 DCE mode: line mapping (matches a null-modem cable)¶

5.4 DTE mode: line mapping (an ordinary USB serial adapter)¶

5.5 Flow control¶

An RS-232 input that isn't connected reads as "off". If the UART always obeyed CTS, a 3-wire cable would never let any data out. So the UART's CTS input is ignored until the far end has asserted that line at least once since it was connected, and obeyed from then on. A vendor request can force flow control on or off.

5.6 Vendor requests (v1)¶

• SET_MODE(auto|dce|dte), saved to flash
• GET_INFO: firmware version, board revision, current mode, setting
• SET_MODEM_OUTPUTS(mask, value): drive each output line directly
• GET_MODEM_INPUTS: read every input line plus both status outputs, for hosts whose CDC driver ignores SERIAL_STATE
• SET_FLOW(auto|on|off)
• SET_LEDS (if LEDs are fitted)

5.7 Baud rate¶

As DCE, the board learns the computer's baud rate from what it types, as a Hayes modem does, and the host's SET_LINE_CODING speed is ignored. So the phone app doesn't need to know what speed the terminal is set to.

• How: a PIO state machine times each low and high stretch on the RX pin (GP1 as DCE, GP13 as DTE) while the UART keeps the pin. The firmware tries every standard rate from 110 to 230 400 baud and picks the one the timings fit (firmware/README.md, "Automatic baud rate").
• Nothing lost: the characters it learned from (for example "AT") are decoded from the timings and passed on. Two characters are enough, so pressing Enter (CR LF) works, or Enter twice on a CR-only terminal.
• Re-learning: 3 framing errors within 2 s (the computer changed speed), a vendor request, or a new peer on the DB25.
• Data format (data bits, parity, stop bits) still comes from the host.
• As DTE, the host's speed is used by default: a modem follows the speed it's spoken to at. Learning can be turned on for DTE too (a device that talks first, such as a GPS), or off for both, with a vendor request.

6. D-Modem app compatibility¶

Works today with no app change (DCE mode): - The app finds any CDC-ACM device (its USB filter matches the CDC interface class, and usb-serial-for-android 3.9.0 detects CDC-ACM by its interfaces). - The app's default "its DTR is the modem's carrier" setting becomes the computer's DCD, the same as a null-modem cable on an ordinary adapter. - Data and format pass through. The speed is learned from the computer (§5.7), so the app's speed setting doesn't have to match the terminal's.

Needs app work: - Hanging up when the computer drops DTR. usb-serial-for-android's CDC driver only sets RTS/DTR and never reads input lines, so the app's DSR read fails and it skips the hang-up. Fix in the app: read the CDC notification endpoint (SERIAL_STATE). That helps every CDC adapter, so it's better than relying on GET_MODEM_INPUTS. - Ring (RI), separate DSR/DCD, and choosing the mode: a "pico-serial board" option in the app's serial settings, using the vendor requests.

DTE mode is for computers using the board to reach a real modem; The D-Modem app doesn't need it.

7. Mechanical¶

7.1 DB25 connector: solder cups straddling the board¶

No true straddle-mount female DB25 is stocked at qty 1. Amphenol's G17M straddle series isn't stocked anywhere I checked, and DigiKey's "straddle mount" category contains mislabelled vertical parts (e.g. Assmann A-DF 25 PP/Z).

Instead, use the standard industry practice: slide the 1.6 mm board between the two rows of solder cups and solder the cups to pads on both sides. The 13-pin row goes on top and the 12-pin row underneath. NorComp's own material says its solder-cup D-subs take a 0.063" board between the rows. KiCad has a matching footprint: DSUB-25_Female_EdgeMount_P2.77mm.

The cup gap isn't standardised across brands, so check a sample with a 1.6 mm scrap board before ordering PCBs.

Pick a connector with plain, unthreaded Ø3.05 flange holes. The MHDB25SS has Ø3.1 holes, confirmed from its drawing, and a 2.84 mm row pitch. The hood's 4-40 thumbscrews pass through those holes into the computer's or gender changer's standoffs. Clinch nuts or board locks would get in the way.

7.2 Hood and board outline¶

The hood is the MH Connectors MHDPPK25-K (LCSC C22372846): - PA66 plastic, outside 53.5 × 47.2 mm (40 mm across the body), 15 mm thick; - comes with 4-40 thumbscrews; - Ø11.5 mm cable exit, which leaves room for USB-C (8.9 × 3.2 mm); - the kit includes only two strain-relief screws at the cable end, so the inside looks clear of screws. Not yet confirmed whether the hood snaps or hinges shut; - no datasheet gives inside dimensions. With a 15 mm outside thickness, expect about 11–12 mm of inside height.

Backups: - Amphenol FCI 86303639BLF: snaps together with nothing crossing the inside, best drawing, ~5 mm shorter. 15 on LCSC, 3k+ on DigiKey. - Adam Tech DB25-HD-PB-TS (DigiKey): two body screws cross the inside, so the board needs ~Ø5 mm notches at x = ±15 mm.

Target outline. Inferred, so confirm on a sample: - About 36–38 mm wide at the connector end, narrowing toward the back. - About 28–30 mm usable length behind the connector's insulator. The front ~3 mm of board is inside the cup field. - A USB-C tongue about 10–11 mm wide reaching into the cable exit. The receptacle's mouth sits flush with the opening. Leave out the strain relief. - Height: about 5 mm free on each side of the board, so components go on both sides. JLC can assemble both sides, for an extra fee.

7.3 PCB¶

Laid out in hardware/kicad/ (rev A, made by build_pcb.py): about 37 × 38 mm with the USB-C tongue, 4 layers, routed, DRC clean and matching the schematic. Differences from the notes below: - Decoupling sits on the bottom about 2 mm from the RP2040's pins, joined through the planes, not within 1 mm; close placement left the chip's pins no room to escape. Each power pin has its own via under the package. - Layers: In1 is a solid GND plane; In2 carries signals as well as the +3V3 pour. - Test pads are on the bottom, on a 2.6 mm grid. - No traces under the crystal on the top layer (keep-outs between its pads); below that, the In1 GND plane shields it.

• 4 layers, 1.6 mm (GND / power on the inner layers). The board is about 36 × 29 mm with two 28-pin transceivers, a QFN-56 and around 50 passives on both sides. Two layers might fit, but four makes the RP2040 decoupling and USB routing much easier.
• Suggested placement:
• top: RP2040, crystal, flash, USB-C, LDO, USB ESD;
• bottom: both transceivers and their charge-pump capacitors, close to the DB25 cups they drive.
• Crystal next to XIN/XOUT, with ground around it and no traces underneath.
• Every IOVDD/DVDD capacitor within ~1 mm of its pin.
• Test pads: VBUS, 3V3, 1V1, GND, SWCLK, SWDIO, GP28, chip A and chip B TX/RX (logic side).
• Silkscreen: name, revision, "DCE / DTE auto", DB25 pin 1, "male–male gender changer for DTE".
• Silkscreen must also carry the licence notice: CERN-OHL-S-2.0 and github.com/D-Modem/hardware. This is the Source Location required by the licence (LICENSING.md); the W board's hood also has it debossed underneath.

7.4 Wireless variant¶

A Bluetooth version (BLE for iPhone, classic Bluetooth serial for Android) is specified in SPEC-W.md. Its antenna needs more room than a DB25 hood has, so it is a separate, slightly larger board.

8. Open questions¶

1. Hood: confirm the MHDPPK25-K's inside space and how it closes. If it doesn't fit, use the FCI 86303639BLF.
2. SP3238E off-state leakage: measure it with ±12 V on an idle driver. If it's poor, switch the DCE chip to the TI TRS3238E from DigiKey. That chip is TSSOP-28, so decide before layout, or lay out a TSSOP footprint and use the MaxLinear SP3238EEY (also TSSOP).
3. RS-232 TVS footprints for ±15 V-era kit: they cost board space; leave them out unless a target machine needs them.
4. Galvanic isolation and phone charging through the board: still out of scope.
5. DE-9 variant: same circuit, different connector, pinout and hood.

9. Next steps¶

1. Buy samples from LCSC (~$7): 2 × MHDB25SS, 2 × MHDPPK25-K, 1 × 86303639BLF. Check that a 1.6 mm offcut fits between the cups. Measure each hood's inside length and width and any screw bosses. Then lock the outline.
2. Bench prototype on a Pico dev board with SP3238E and MAX3243E on SSOP-28 adapter boards, wired as §4.5. Prove the role detection, the driver swing with both chips on the bus, and both line mappings with the D-Modem app (DCE) and a real modem (DTE).
3. Netlist, schematic and PCB: done, hardware/ (generated and checked). Next: a person reviews the layout, and the outline is confirmed against the sample hood (step 1).
4. Firmware v1 (§5).
5. App: read CDC SERIAL_STATE; add the pico-serial options.
6. Order 5 boards. Bring-up: 3V3, 1V1, crystal, enumeration, UF2, loopback in both roles at each speed, every DB25 line in both roles.