The LoRa Field Node.

What it does.

Wire what you have to it. It finds what is there and reports it: DS18B20 temperature probes on one wire (up to sixteen), an SHT-30 air temperature and humidity sensor (one or two), a photoresistor for light level, and up to eight capacitive soil-moisture probes. It also reports its own battery if one is fitted, and it carries anything plugged into its pod port: a Radio Sensor Listener, an Apogee meter, another OAT sketch. Nothing is configured per bus sensor: a probe added today shows up at your endpoint on the next cycle under its own factory serial number. Analog inputs are the one exception, and the wiring section says why.

On its cadence, three minutes by default, it reads everything, packs the readings into a frame of a few dozen bytes, listens for a moment to make sure nobody else is talking, and transmits. Then it waits. There is no Wi-Fi radio running, no internet, no account, no pairing, and no acknowledgement. If a frame is lost, the next one three minutes later carries fresh readings, and the gateway reports the gap.

It is the Collect layer for the places Wi-Fi cannot reach: the back field, the pump house, the greenhouse at the end of the row. The LoRa page explains why this radio carries a reading a kilometre on almost no power.

Flash it.

Plug your board in, find it below, and press its button. The LoRa radio is soldered onto these boards, so the image is built for the exact board, and the installer checks the chip matches before it writes anything.

Attach the antenna before you power the board. A LoRa radio transmitting into nothing can damage its own output stage. The little stub antenna in the box is enough for the bench; a real antenna is where range comes from.

Prefer the source? Browse it on GitHub: the full project, openly licensed, ready to build yourself.

What you need.

How it works.

1. On its cadence the node reads every sensor it found.
2. It packs the readings into one small frame: for each sensor, a one-byte tag, a measurement code, and a scaled value. Ten sensors fit in about sixty bytes.
3. It listens for a moment, and if the channel is quiet, transmits. A random pause of up to three seconds keeps a dozen nodes from ever falling into lock-step.
4. The gateway hears the frame, files each reading under the sensor's own serial number, averages the window, and pushes oat-ods to your endpoint.

Every fifth cycle, and at boot, the node also sends a roster: the map from each one-byte tag to the sensor's full hardware address. That is how a DS18B20's 64-bit factory serial rides the air as a single byte, and why a probe moved from a Wi-Fi node to a LoRa node keeps its history at the endpoint.

It is a private radio LAN with the manners of UDP. No handshake, no acknowledgement, no routing, no network server. The gateway is the only thing that needs an internet connection, and it needs it for the same reason every OAT gateway does: to push your readings to a place you keep.

Wiring.

The defaults below are for the pins each board leaves free once its radio and screen are spoken for. Every pin is a setting, so if your board or your sensor wants a different one, change it on the console rather than on the bench.

Buses are discovered; analog is declared. A DS18B20 announces its serial and an SHT-30 answers at its address, so the node finds them by itself. An analog pin cannot be discovered: a pin with nothing on it floats at a few hundred millivolts and reads exactly like a wet probe. So analog pins ship off, and you declare the ones you wired with one console line. A declared pin reading under 150 mV is reported as raw millivolts only, never as a percentage, because that is a wire off rather than saturated soil.

Power the sensors from the board's Vext pin, which the firmware switches on at boot, or from 3V3. A photoresistor needs a fixed resistor of about 10k to ground to make a divider. A capacitive soil probe outputs a voltage on its own; feed it 3V3, not 5 V, so its output stays inside what the chip can read. The DS18B20 and SHT-30 pages carry the pin-by-pin hookup and the reasons behind each choice.

Set it up.

This node has no Wi-Fi, so it has no setup web page. Setup is the USB console: plug it in, open a serial terminal at 115200 (the browser console on this site works), and type help.

1. Flash the board and leave it plugged in. It boots, finds its sensors, and prints what it found.
2. Type status. Every probe is listed with its serial number and its first reading. If a sensor is missing, check its wiring, then type scan.
3. Leave the radio plan alone unless you changed it on the gateway. The default plan on both sketches is the same: 915.0 MHz, 125 kHz, spreading factor 7, sync word 12.
4. Set the cadence if three minutes is not what you want: set cadence 120. Sixty seconds is the floor.
5. Type tx to send a frame right now, then look at the gateway's setup page. Your node appears under Field nodes heard with its signal strength.

The screen. The board’s small display shows what a person standing next to the node needs: its id, the radio plan and whether the radio started, how many probes of each kind it found, the last transmission with its sequence number and how long ago, when the next one is due, and the latest reading with the battery voltage. A node cannot know it was heard, since nothing answers it, so the screen says sent, honestly, and the gateway’s screen says heard. A blank screen on a powered board means the firmware did not boot, which is a flashing problem rather than a radio problem.

Settings survive reboots. A fresh flash wipes them, on purpose, so a re-used board never carries the last owner's radio plan. Calibrating the soil probes is a bench step before the node goes out, one probe at a time, over the same console: hold the probe in dry air and type set cal 3 dry, stand it in a glass of water to the line and type set cal 3 wet, using the probe's own GPIO. Each probe keeps its own pair, because two probes from one bag read hundreds of millivolts apart in the same glass. A probe with no calibration sends its raw millivolts only and no percentage; the raw value is always sent, so you can recalibrate at the endpoint later without visiting the node. show prints the table, and set cal 3 2876 1232 types a pair back in after a reflash. The Soil-Moisture Node shares this exact code and explains the calibration at length.

Open the printable setup card (PDF)

A three-page field card: what to have on hand, the flash and setup steps, blanks for this node’s settings and ids, the console commands, and the fixes for the usual snags. Print one per node.

What it sends.

Over the air, not oat-ods. A LoRa frame carries about 200 bytes and one oat-ods reading is 330 bytes of JSON, so the node sends a compact binary frame instead and the gateway expands it. Your endpoint sees ordinary oat-ods, tagged oat-lora-gateway, with each field sensor as its own stream and the node's signal strength, noise margin and battery as gauges beside it.

The whole over-air contract is one header file, oat_lora_frame.h, shared by both sketches. Adding a measurement is one line in its code table, on both sides. Codes are never renumbered.

The pod port.

Some sensors are not a probe on a wire. A weather station is a radio broadcast that needs decoding. An Apogee meter speaks serial. A neighbour has an ESP32 running something clever. The node cannot know about all of them, and it does not need to: it has a pod port, a second serial port at 115200 baud, and anything that prints readings on it in one small text grammar becomes a cluster of sensors on this node. The thing on the other end of the cable is a pod. It does one job on its own board, and the node carries what it says.

Every OAT sketch prints its readings in this grammar by default, so the Radio Sensor Listener cabled to this port carries an AcuRite’s wind and rain to the house over LoRa under the station’s own id, with no Wi-Fi anywhere in the field. Three wires: the pod’s TX to this node’s pod RX, ground to ground, and optionally the pod’s RX to this node’s pod TX.

The grammar: oat-line/1.

Plain text, one record per line, space-separated, newline-terminated, under 120 characters. Four record types, one letter each. Everything else on the line is ignored and counted, so a pod can print debug text without breaking the reader.

M <stream-id> <measurement> <value> [unit|-] [c|g|k|s|e]    a reading
S <stream-id> brand=<..> model=<..>                        who made it, sticky
L <stream-id> [rssi=<dBm>] [battery=<pct>]                  link health, latest wins
H <pod-id> <firmware> [uptime=<s>]                         the pod's own heartbeat

An AcuRite station, as a Radio Sensor Listener prints it:

S acurite-5n1:2716 brand=AcuRite model=Acurite-5n1
M acurite-5n1:2716 wind_speed 3.4 m/s c
M acurite-5n1:2716 wind_direction 225 deg g
M acurite-5n1:2716 rain_total 128.5 mm k
M acurite-5n1:2716 temperature 21.4 Cel c
M acurite-5n1:2716 humidity 63 %RH c
M acurite-5n1:2716 battery_low 0 - s

This node prints its own readings the same way on its USB port, so a field node plugged into a gateway by USB is itself a pod. The listener page shows the wiring from its side; the developer reference says what a pod’s streams look like to an endpoint, which is: exactly like every other stream.

Writing a pod. Read your sensor however it wants to be read. Print one M line per reading with the vocabulary name and unit, an S line once for brand and model, and an H line now and then. That is the entire contract, and it works from any board in any language.

Range, airtime and power.

The default plan, spreading factor 7 at 125 kHz, is the campus setting: a 60-byte frame costs about a tenth of a second of airtime, which is why a hundred nodes on a three-minute cadence share one channel with room to spare. Line of sight with the gateway antenna up high, it reaches hundreds of metres to a kilometre or two. Metal-framed greenhouses eat signal; a node whose antenna pokes outside the frame hears and is heard far better than one on the bench inside.

For a node genuinely far away, step the spreading factor up on both sides: 9 reaches a few kilometres at four times the airtime. The US 400 ms dwell rule makes 10 the practical ceiling at 125 kHz. The LoRa page walks the trade-off.

Power. This node stays awake. It does not sleep between readings, on purpose: an awake node can listen before it talks, take a reading on demand, and keep its console live. The cost is about half a watt. On mains, that is nothing. Off grid, size for December rather than June: a 20 W panel and a 12 V 10 Ah LiFePO4 battery carry it through a week of overcast. The solar page has the sizing. Oversize the panel and skip the firmware complexity; that is the appropriate trade here.

For makers & trainers.