Status: Rev A schematic captured; no PCB yet. Tracks #56 (an extension of #21); the physical interface is #57; the board itself is #324 (the P1 stage) and #325 (panel, audio, power). Board register: hardware/README.md.

What it is

An ESP32-S2 running a Zimodem fork — Hayes AT command set plus telnet, and with two pieces of new work on top:

  • SSH as a first-class transport, with keys, host verification and session management from a web UI. An Osborne 1 dialling into a modern host with ssh.
  • Baud-rate buffering, decoupling the machine’s 6850 ACIA rate from the far-end link rate — which makes a turbo upgrade usable and quietly corrects the factory 600/2400 → 650/2600 quirk.

The S2 rather than an ESP8266 or a C3 is deliberate: its auto-baud pulse counter is 20-bit, and the C3/S3 are 12-bit, which puts 300/1200/2400 below their floor. The ESP8266 lacks the headroom for SSH entirely.

How it attaches

It sits in the left floppy pocket and hangs off the machine’s MODEM port (P1) by a short 10-conductor ribbon: a DE-9 plug into the machine, a 2×5 IDC header on the board. There is no case opening and no separate supply.

The port is TTL, not RS-232 — it feeds the 6850 ACIA directly (status 2A00H, data 2A01H). The levels are asymmetric, and both halves need work:

  • TXD and MCB arrive at 5 V and the ESP32 is not 5 V tolerant, so a dual-supply translator (74LVC4T245) sits in each direction.
  • MSB is open-collector and gets an open-drain driver (74LVC1G07).
  • RXD is bipolar (−0.5 … −10 V = 1), so a MAX3232 on 3.3 V supplies ±5.5 V from its own charge pump.

Power

DE-9 pin 7 is +12.6 V DC, and it is behind the machine’s 22 Ω — the Technical Manual §4.3.2 says so in words (“Connected to power supply through 22 ohms”), and sheet 7 shows R21 feeding both this pin and the keyboard branch (J6 → P4-19). So the port is a low-current supply: 22 Ω at 500 mA would drop 11 V, and at 100 mA it drops 2.2 V. The board’s average draw has to be tens of mA, and it shares the resistor with the keyboard adapter.

Two sources, diode-OR‘d so neither back-feeds the other: the port’s 12.6 V → 5 V → 3.3 V, and USB-C 5 V → 3.3 V for programming and bench work. The switch is ahead of the buck, so “off” is genuinely dark. The bulk capacitor at the buck input supplies the WiFi TX peaks, which R21 cannot.

The panel, and the noise

Eleven lamps on an AW9523B (16 constant-current channels, 256-step dimming, holds its outputs so a hung firmware leaves a readable frozen panel): radio, connected, HS, AA, CD, OH, RD, SD, TR, MR, RI. A second expander is a DNP option and carries the handshake lines.

Sound is an I²S class-D amplifier with its own DAC on the 5 V rail, driving a passive transducer — dial tone, DTMF and a ring cadence, not a one-note buzzer.

The case

A “COMM PAC-look” case fits the left pocket — period-authentic Osborne industrial design, so it reads as original kit rather than a modern dongle. The STL is already designed and fit-tested; left to right across the front: antenna, power switch, 2×5 IDC, LED row, USB-C.

Still to confirm

  • The 2×5 IDC orientation on a real machine (#324).
  • The RXD negative swing in hardware — ±5.5 V should be comfortable, but the receiving front end is a comparator with a 5.1 V clamp.
  • The audio transducer and where it points once the case is closed (#325).