Satellite chain

The chain is the reason a Rodeo panel can grow without the motherboard changing. Each satellite carries a cent-scale microcontroller, reads its own controls, and passes frames along a ring. A satellite’s position in the chain is its address, so there is nothing to configure and no jumpers to set.

This page is the short version. The contract in full — discovery, commands, fault codes, programming — is hardware/rodeo-motherboard/docs/RODEO_SATELLITE_CHAIN_CONTRACT.md.

Connector

Every satellite has an IN and an OUT, both JST SH 5-pin. The motherboard has one chain connector wired like a satellite’s OUT.

Pin	Signal	On a satellite
1	GND	—
2	3V3	Supply in, passed straight through to OUT
3	DATA_FWD	MCU RX on IN, MCU TX on OUT
4	DATA_RET	Straight through, with the MCU’s return tap on it
5	PRESENT	Tied to GND on IN; an input with a pull-up on OUT

DATA_FWD is point to point — one transmitter and one receiver per cable. DATA_RET is a single wire from the end of the chain back to the Seed, passing through every satellite.

Line

Format	UART, 8N1, LSB first, idle high
Rate	1 Mbaud
Levels	3.3 V CMOS; satellites run directly from the chain supply
Seed side	USART3 — TX on pin 3, RX on pin 2

The ring closes itself

At boot and every millisecond after, each satellite reads the PRESENT pin on its OUT connector.

• Low — something is plugged in downstream, so transmit on DATA_FWD.
• High — this is the last satellite, so move the UART’s transmit onto the DATA_RET tap. Frames then travel back along the return wire to the Seed.

No terminator plug, and nothing to configure when a panel grows. A satellite drives the return tap only while it is actually transmitting; the rest of the time the tap is high-impedance and the motherboard’s pull-up holds the line idle, so two satellites can never fight over the return wire.

Frames

A 4-byte header followed by N slots of 4 bytes, at most 260 bytes — 2.6 ms at 1 Mbaud.

Header

Byte	Field
0	SOF, 0xA5 for protocol v1
1	TYPE — 0x01 scan, 0x02 discover, 0x03 command
2	N, slots that follow, 0–64
3	HOP, satellites passed through so far

Slot — KIND, D0, D1, then a CRC-8 over the first three bytes (polynomial 0x07, init 0x00, CRC-8/SMBUS).

Slot kinds

Satellites report state, not events. If a frame is lost nothing is missed: counters wrap, and the Seed works out what changed from the difference with the last good value.

KIND	Name	Carries
0x01	POT	16-bit position, scaled 0–65535 whatever the ADC’s resolution
0x02	BUTTON	Pressed-now bit, plus a press counter that wraps at 256
0x03	ENCODER	Detent counter wrapping at 256, plus switch state and its counter
0x04	SWITCH	Position index
0x7F	FAULT	Fault code and detail
0x10	LED	Seed to satellite: brightness, and steady or blinking

Board kinds

A resistor to ground on an ADC pin tells the one satellite firmware image which board it is running on. Position in the chain is still the address; the kind only says what the board has on it.

Kind	Board	Slots
1	potentiometer (rodeo-pot-1)	1 × POT
2	button (rodeo-button-1, specified)	1 × BUTTON
3	encoder with push switch (rodeo-encoder-1)	1 × ENCODER
4	rotary switch, up to 8 positions (rodeo-rotary-switch-1: 4 positions, specified)	1 × SWITCH
5	pot bank ×4 (rodeo-pot-bank-4)	4 × POT
6	encoder bank ×4 (rodeo-encoder-bank-4)	4 × ENCODER
7	button bank ×4 (rodeo-button-bank-4)	4 × BUTTON
14	custom board on its own firmware (the satellite dev board)	1–4, as it declares
15	no resistor fitted — reports EMPTY	1

Kinds 8–13 are kept for future Rodeo satellites. A new board type needs a resistor value and a firmware update, not a firmware of its own. Kind 14 is the exception: a board Rodeo doesn’t make runs its own firmware and declares its slots when the Seed discovers it, so the Seed needs nothing new to read it.

Kind 7 is written but not yet on a board

The contract gives rodeo-button-bank-4 kind 7 and an RGB output slot — a colour as RGB565, which fits the two data bytes every slot already has. The satellite firmware and the Seed’s driver implement both, along with the WS2812 driver. None of that has run, because no button bank has been made.

Timing

One frame is in the ring at a time. The Seed sends the next after the previous returns, or after a timeout.

• Ring time ≈ (4 + 4N) × 10 µs + 40 µs per satellite. For thirty one-slot satellites that is about 2.5 ms.
• Default scan period 4 ms, so 250 scans a second. A control change reaches the Seed within roughly one period plus one ring time — about 6.5 ms worst case across thirty satellites.
• Scans are paced from a timer, never from the audio callback.

What has actually run

A ring of a motherboard, a pot bank and an encoder bank, on 2 November 2026: about 24,600 scans in two minutes, none bad, every reply the full 36 bytes. All four pots swept their range, all four encoders turned both ways and pressed, and the Seed drove each board’s LED. Pulling a satellite mid-scan cost seven bad scans, all at the moment of the unplug, and the chain recovered.

It did not work first time. Both boards were discovered, but most scans came back short until the satellite firmware moved to passing frames through in its UART interrupt rather than in its main loop.

Untried: breaks within the ring, three or more satellites, and every board that has not been made — the button bank and the single-control boards.

Power

What limits the chain is voltage drop, not current. It is fed from one end at 3.3 V, every satellite has to work down to 2.7 V, and each link — 150 mm of 28 AWG plus two connector matings, there and back — is about 0.14 Ω carrying the current of everything past it.

The baseline is ≈6 mA per satellite including its indicator LED, for thirty on one feed through a 0.75 A resettable fuse. That puts the last of thirty roughly 0.4 V down, leaving about 0.2 V of headroom.

You can spend that headroom. A satellite drawing ΔI more than the baseline at position p lowers the far end by ≈0.14 Ω × p × ΔI, and the rule is to keep Σ (p × ΔI) under about 1.4 A across the whole chain. Four single-colour lit buttons cost 85 mV even at position thirty, and under 30 mV ten boards in.

LEDs that want more than 3.3 V or tens of milliamps — addressable RGB, meters — take their LED power from a separate 5 V feed and keep their microcontroller on the chain, so logic levels do not move. Rev B does not bring 5 V to a connector yet; that is on the motherboard’s list for its next revision.