OAT DS18B20 Temperature Node

One wire, many probes, every reading landing somewhere you own.


Status	Live — flash from the browser, no IDE needed
Interface	1-Wire · DS18B20 (64-bit address per probe, up to ten on one bus)
Chips	ESP32 · S3 · C3 · C6
Delivers	Webhook (HTTP POST) · MQTT — oat-ods/0.3
License	Apache-2.0

A self-configuring ESP32 node that reads one or many DS18B20 temperature probes on a single 1-Wire data line and pushes the readings as oat-ods to an endpoint the grower owns, by webhook or MQTT.

Everything except the sensor read is lifted from the SHT-30 Node, which took it from oat-ble-listener v1.4.8: config + NVS, the shared field registry (one applyField/readField behind both the web form and the serial Console), WiFi/AP/captive portal with connect-then-confirm, mDNS, the push engine (https-preferred with signed-http fallback, HMAC signing, oat-ods batch, monotonic seq), the 60 s health heartbeat, and the two-way USB Console. Set up one OAT node and you have set up all of them.

A grower flashes it from the browser at openagriculturetechnology.com/build/sketches/ds18b20-node/ (ESP Web Tools / Web Serial), then configures it on the device's own page. No IDE, no command line, no code editing.

Wiring

DS18B20	ESP32	wire colour on a sealed probe
VDD	3V3	red
GND	GND	black
DQ (data)	the data pin (default below)	yellow, sometimes white or blue

Plus one 4.7 kΩ resistor between DQ and 3V3 — see below. Every probe on the run shares all three wires.

The default data pin, per chip

Board	Default	Why not the same number everywhere
Classic ESP32	GPIO 4	free; not flash, not strapping, not input-only
ESP32-S3	GPIO 4	free
ESP32-C3	GPIO 4	free
ESP32-C6	GPIO 6	GPIO 4 is a strapping pin on the C6

A 1-Wire bus idles HIGH through its pull-up, which is exactly the nudge that flips a strapping pin into the wrong boot mode — so the C6 gets a different default rather than a shared one that would sometimes break a boot. pinOkForOneWire() refuses flash, strapping, USB-JTAG, UART0 and input-only pins per chip, with the reason, because a bad pin is persisted in NVS and re-applied at every boot: it is not a mistake you get to notice and correct.

Pin lists are from Espressif's per-chip GPIO documentation, not folklore.

The resistor is not optional

1-Wire is an open-drain bus: devices only ever pull the line DOWN, and a pull-up resistor is the only thing that pulls it back UP. With no resistor there is no bus at all, just a floating pin reading noise — which looks exactly like a broken sensor.

Fit one 4.7 kΩ resistor between the data line and 3V3, once for the whole bus, not once per probe. Anywhere on the run works; near the board is easiest.
Many sealed probe modules and breakout boards have it fitted already. A bare TO-92 chip or a plain 3-wire cable does not.
Long run, or several probes, and reads go intermittent? 2.2 kΩ is the next thing to try: a stiffer pull-up recovers the edges that cable capacitance rounds off. Beyond that: linear topology rather than a star, and eventually an active pull-up (Maxim application note 148 covers real 1-Wire networks).

How many probes on one wire

The protocol sets no limit. Every DS18B20 carries a unique 64-bit address lasered in at the factory (family code 28h + 48-bit serial + CRC), and the node enumerates whatever is present with the ROM SEARCH command (F0h, the algorithm from Maxim application note 187). Addressing is not the constraint.

The electrical limit is real: every device and every metre of cable adds capacitance, which rounds off the edges of a 60 µs bit until the bus stops being readable. On a short run with the standard 4.7 kΩ pull-up, eight to ten probes is comfortable. This firmware caps its list at MAX_DS (10) for that reason — an honest bound of ours, not a protocol one, and it says so on the console when the list fills.

What it reports

measurement	unit	fold over the push window
`temperature`	`Cel`	mean

One stream per probe. source.physical_id carries the probe's own 64-bit address (ds18b20:28ff641f8c1a3b02), so a reading is traceable to a chip, not to a position in a list: swap a failed probe and the history of that spot carries on with the change of hardware visible in the record.

Probes are listed in address order, so the numbering on the setup page is stable across reboots even when one is added.

The parts that matter

The bus is bit-banged here, not a library. Reset/presence, read and write slots, the CRC and the ROM search are all in this file, about two hundred readable lines, with the datasheet's timings quoted where they are used. One less pinned dependency between a grower and a working node, and the CRC is genuinely checked rather than assumed.
Only the edges hold interrupts off. A bit slot is protected by a critical section because a task switch mid-slot is a corrupted bit. The 480 µs reset LOW is not: it only has to be at least 480 µs, so blocking WiFi's interrupts through it would cost packets for nothing.
CRC-checked, and a failed CRC is discarded, not reported. A long or damp wire produces plausible-looking garbage; a wrong number is worse than a missing one. "No answer" and "bad CRC" are counted separately, because a pulled wire and a marginal bus are different problems with different fixes.
85.00 °C is discarded. That is the datasheet's power-on value of the temperature register: seeing it means the probe reset or the conversion never ran, so the register was never written. Publishing it would be publishing a default as an observation. (If you genuinely need to measure 85 °C, this is the wrong node — better to say so than to quietly make it up.)
Parasite power is detected and called out. A DS18B20 can run on two wires, stealing power from the data line, but conversions then need a strong pull-up this node does not drive. READ POWER SUPPLY (B4h) reports it and the status page says "wire VDD to 3V3" rather than pretending the readings are fine.
Reading happens in two halves. A 12-bit conversion takes up to 750 ms. Blocking the node for that every cycle would stall the setup page and the push engine, so the sample timer issues the conversion for every probe at once (SKIP ROM + CONVERT T) and the loop collects the results once the window has passed. A person pressing "read now" gets the blocking version, because a person is waiting.
A re-scan keeps each probe's history. Counters are carried across by address, because resetting "reads failed" to zero in the middle of the fault the re-scan exists to recover from would hide the one thing worth seeing.
A probe that stops answering is dropped from the live view, rather than left showing its last reading in live-value type. The failure detail replaces it.
agg.window_s is the window actually covered, not the configured interval. Sampling continues while the network is down, so the first push after an outage can carry hours of readings, and it says so.
Remote diagnosis. When a probe dies, the 60 s heartbeat keeps arriving while the readings stop. That contrast — alive but quiet — is what the endpoint watches, and it is why the heartbeat is sent before the sensor push.

Resolution

9 to 12 bits, set on the setup page and written to every probe on the wire.

bits	step	conversion
9	0.5 °C	93.75 ms
10	0.25 °C	187.5 ms
11	0.125 °C	375 ms
12	0.0625 °C	750 ms

Default is 12. The unused low bits at lower resolutions are masked rather than reported as decimal places they do not carry.

Specs above are from the DS18B20 datasheet (Analog Devices / Maxim): ±0.5 °C from −10 °C to +85 °C, ±2 °C over the full −55 °C to +125 °C range.

Files

File	What it is
`oat_ds18b20_node.ino`	The firmware source (Arduino / ESP32).
`platformio.ini`	Build config. Four envs: classic ESP32, S3, C3, C6. Pinned libs.
`merge_bin.py`	Post-build hook → one merged factory image per chip at `out/firmware-<mcu>.bin`, flashable at offset 0.
`make_manifest.py`	Assembles `manifest.json` + `manifest-<mcu>.json` from whatever bins exist.
`build.sh`	Builds, then copies bins + manifests + source downloads into the site assets.
`docker-build.sh`	The same compile with no host toolchain (Docker + PlatformIO).

Build

pipx install platformio        # or: pip install --user platformio
./build.sh

No host toolchain?

./docker-build.sh              # compile all four chips in Docker
SKIP_BUILD=1 ./build.sh        # then the asset steps

build.sh produces one merged, flash-at-offset-0 image per chip in out/, plus the manifest and download bundle the OAT site's flash-from-browser page serves.

oat-sht30-node/ — the wired sibling: air temperature and humidity, on I²C.
oat-ble-listener/ — the reference node this core comes from.
../lib/oat_ods/ — the shared oat-ods encoder + measurand vocabulary.
../lib/oat_sign/ — the shared HMAC push signer.
The oat-ods standard — the wire format.

Code: Apache-2.0 · Docs: CC BY 4.0 · An OAT sketch — an OpenCDC initiative.