Files
hexapod-robot/2026-current/dongle/README.md
T

205 lines
8.4 KiB
Markdown
Raw Normal View History

# USB nRF24 dongle
A cheap "USB NRF24L01+ 2.4G" stick (AliExpress, board marked **USB_24RF
Ver 3.1**, Rainsun chip antenna) reflashed with our own firmware so it can be
the PC-side radio for the hexapod: same nRF24 air settings as the robot and the
remote, talking to Unity over USB serial.
It shipped with a closed "AT command" bridge firmware. We replaced that with the
two-stage firmware in `firmware/`. **The factory firmware could not be backed
up** (read-protected, see below), so this is a one-way change — done on the
owner's say-so.
## Hardware (reverse-engineered from the board)
| Part | Chip | Notes |
|------|------|-------|
| MCU | **STM32F030F4P6** | Cortex-M0, 16 KB flash, 4 KB RAM, TSSOP-20 |
| USB-serial | **CH330N** | shows up as "USB-SERIAL CH340" on Windows (COM port) |
| Radio | **Si24R1** | nRF24L01+ clone, driven over SPI; 16 MHz crystal |
| PA/LNA | **AT2401C** | 2.4 GHz power amp + low-noise amp (range booster) |
| antenna | Rainsun chip antenna | |
### Pin map (MCU)
Found by probing the board over SWD and confirmed at runtime by our firmware:
| Function | STM32 pin |
|----------|-----------|
| USB serial (USART1) TX / RX | PA9 / PA10 → CH330N |
| SPI1 SCK / MISO / MOSI | PA5 / PA6 / PA7 → Si24R1 |
| nRF CSN / CE / IRQ | PA3 / PA4 / PA2 |
| status LED (active-high) | **PA0** — confirmed by eye |
| AT2401C TX/RX enable + unused | PA1, PB1, PF0, PF1 — **not yet distinguished** |
PA0 drives the LED (HIGH = on). The AT2401C control lines are among PA1, PB1,
PF0, PF1 (the TSSOP-20 has no other GPIO). Listening for ambient 2.4 GHz with
the radio's carrier-detect bit was too noisy to tell them apart, so they need a
real link test (dongle ↔ robot/remote, trying each pin while watching whether
packets get through). The radio still works without driving the amplifier, just
at reduced range. The app firmware can drive any pin (`0`-`4`), sweep them all
(`a`), and run a carrier-detect finder (`f`/`n`).
### Programming header (5 holes next to the CH330N)
Square pad = hole 1.
| Hole | Signal |
|------|--------|
| 1 | BOOT0 |
| 2 | 3V3 |
| 3 | GND |
| 4 | SWDIO (PA13) |
| 5 | SWCLK (PA14) |
The factory firmware repurposed PA13/PA14, so SWD only answered with **BOOT0
tied high** (hole 1 → hole 2) at power-up. With our firmware this header is not
needed at all — see flashing below.
### Flash readout protection
The STM32 was at **RDP level 1**: reads of flash are refused over both SWD and
the ROM bootloader, so the factory firmware cannot be dumped. Level 1 (not the
permanent level 2) still allows reflashing — unlocking does a mass-erase, which
is what wiped the factory firmware. The only way past RDP1 to read it would be
the F0 cold-boot stepping exploit (thousands of power cycles); not worth it.
## Firmware: two stages, so wire-free flashing can never break
The whole point of the design: **once installed, the dongle is reflashed over
USB forever, with no BOOT0 wire and no ST-Link — and no firmware update can take
that ability away.**
```
flash (16 KB):
0x08000000 +--------------------+ page 0
| safeboot (4 KB) | stage 1: permanent recovery bootloader
0x08001000 +--------------------+ page 4
| app (12 KB) | stage 2: the radio firmware
0x08004000 +--------------------+ page 16
RAM (4 KB): 0x20000000 = magic word (app -> safeboot signal); .data/.bss from +4
```
- **`safeboot`** (`firmware/safeboot.c`) runs first on every reset. It speaks a
tiny, fixed-baud (115200 8N1) flash protocol over USB and can always erase and
rewrite the app. It has no dependencies and is flashed once.
- **`app`** (`firmware/app.c`) is the actual radio firmware. It is the only part
future updates rewrite.
- **The app updater only erases pages 4-15 — never safeboot.** So a buggy app
can't remove the recovery path.
Recovery is guaranteed three ways, none needing the wire:
1. normal update — the running app resets into safeboot on the `b` command (the
CH330N keeps its USB power across an MCU reset, so the COM port doesn't drop);
2. a corrupt/missing app — safeboot sees the app's reset vector is invalid and
stays in flash mode;
3. a booting-but-wedged app — safeboot opens a 1.5 s knock window on every cold
boot; the updater spams `S` after a replug to catch it.
The only thing that could break it is corrupting safeboot itself, which the
updater never writes.
### safeboot flash protocol (115200 8N1)
Each command replies one byte, `K` = ok / `E` = error, unless noted.
| Cmd | Bytes | Action |
|-----|-------|--------|
| `P` | | ping |
| `V` | | prints `SBV1` (identify) |
| `E` | | erase app region (pages 4-15) |
| `W` | addr(LE32) n data[n] | program n bytes (n even, ≤128) |
| `R` | addr(LE32) n | read n bytes (returns n bytes, no `K`) |
| `G` | | `K`, then boot the app |
Banners: `SBF` = entered flash mode via the app; `SB1` = cold boot (knock
window open); `SBR` = no valid app, staying in flash mode.
## Building
Needs the ARM toolchain PlatformIO ships:
```
C:\Users\you\.platformio\penv\Scripts\pio.exe pkg install -g -p ststm32
```
Then, from `firmware/`:
```
.\build.ps1
```
Produces `build/safeboot.bin`, `build/app.bin`, and `build/dongle_combined.bin`
(safeboot padded to 4 KB + app), plus the sizes.
## Flashing
### One-time install (the only time the BOOT0 wire is used)
Puts safeboot + app on a stick that still has the factory (or any) firmware.
1. Wire **hole 1 → hole 2** (BOOT0 high) and plug the stick into USB.
2. `python tools/dongle_flash.py firmware/build/dongle_combined.bin`
- unlocks (mass-erase, the step that wipes the factory firmware),
- writes and verifies the combined image,
- starts it.
3. Remove the wire. Never needed again.
`dongle_flash.py` drives the STM32 **ROM** UART bootloader (one-shot auto-baud
per power-up, so power-cycle between attempts).
### App updates (normal, wire-free, forever after)
```
python tools/safeboot_flash.py firmware/build/app.bin
```
This resets the running app into safeboot and rewrites only the app region. If
the app is dead or doesn't answer, start it with `--knock` and then replug the
stick: it keeps knocking (reopening the port as it comes back) until it lands in
safeboot's 1.5 s cold-boot window.
## Air settings
Matches `shared/HexapodProtocol/HexapodProtocol.h`: 2.4 GHz channel 96
(2.496 GHz), 250 kbps, 5-byte address `HXPD1`, 16-bit CRC. The app sets these on
the Si24R1 over SPI (same register layout as the nRF24L01+).
## Status / next
- [x] identify the hardware, map the radio + serial pins
- [x] permanent two-stage firmware, wire-free USB flashing
- [ ] identify the LED and AT2401C TX/RX-enable pins (app toggles the 5 free pins)
- [x] bridge mode: the dongle stands in for the robot's radio, Unity runs the
robot firmware (sim) behind it — see "Bridge mode" below
- [ ] the other direction: Unity as the remote, driving the real robot
- [ ] AT2401C pins: at desk range and even a few rooms away with the remote at
minimum power, every on/off combination of PA1/PB1/PF0/PF1 received
every packet, so they couldn't be told apart; the link works without
driving them
## Bridge mode (Unity ↔ real remote)
The app listens on the hexapod's air settings as if it were the robot
(ch 96, 250 kbps, `HXPD1`, auto-ack, dynamic payload lengths, ack payloads)
and passes everything to the PC. Framing both ways over the USB serial
(115200 8N1):
```
A5 type len payload[len] sum sum = type + len + payload bytes (mod 256)
```
| Direction | Type | Payload |
|-----------|------|---------|
| PC → dongle | `L` | — : start bridging (LED on) |
| PC → dongle | `X` | — : stop |
| PC → dongle | `A` | ≤ 32 bytes: the robot's `Telemetry`, sent back on the auto-ack of the remote's next packet (reloaded after every packet, like `robot/Comms.cpp`). Dropped if not refreshed for ~0.5 s, so the remote never shows stale telemetry as live |
| dongle → PC | `R` | a packet the remote sent (its `CmdPacket`, any length; the firmware checks it) |
| dongle → PC | `S` | once a "second": packets received (u16 LE), ack payloads loaded (u16 LE). Its second runs ~30% long (blocking UART sends miss SysTick ticks); it's only for display |
Plain text never contains `A5`, so the PC skips anything between frames;
while bridging the app prints no text and only takes frames (and `b` for a
reflash). USB serial input goes through DMA into a ring buffer (no
interrupts in this app), and the USART's overrun stall is disabled, so a
burst can't deafen it. Unity's side is
`Hexapod_Simulation_Unity/Assets/HexapodSim/DongleLink.cs`.