Polarity does not matter functionally. The + / − marking on the terminal records which pin is which electrically, so a cable can be wired consistently across a panel.

The signal chain is the same on every input: a 10 kΩ resistor from 3.3 V to the terminal, the NTC to ground, then a 33 kΩ + 1 µF low-pass into the ADC. That filter has a time constant of 33 ms - a corner around 4.8 Hz. It is what makes the readings quiet next to switched mains, and it is also why a sensor reading lags a real temperature step by about a tenth of a second. For an HVAC loop that is free; for anything fast it is a hard floor.

Used as a contact input the same filter applies, so a contact change takes roughly a tenth of a second to appear. That is irrelevant for anything a person, a float or a thermostat does, and it debounces the contact for free.

A useful consequence of the pull-up: a disconnected sensor reads full scale. Nothing else on the input can produce that value, so firmware can tell a broken cable from a real temperature and say so, rather than reporting an implausible reading.

Model 02 only:
The 1 µF sits on the sensor side of the 33 kΩ, so a channel change does not have to charge it - but the 33 kΩ then works into the multiplexer's own 100 nF, and that is what the reading has to wait for. Allow about 35 ms after selecting a channel. See Input multiplexer.
Note that because of the multiplexer, the pin is an ADC input, never a logic input. Firmware compares a number against a threshold; there is no digital high or low to read.

Two 100 kΩ 2512 thick-film resistors in series limit the mains side - 200 kΩ total, each part rated 500 V and 1 W. Two in series rather than one is a creepage and single-point-of-failure decision: if one resistor fails short, the surviving one still holds the current to about 2.5 mA and dissipates well inside its rating. The isolator is a Vishayopen_in_new SFH6286-3Topen_in_new, an AC-input optocoupler - its input is a pair of anti-parallel LEDs, so it conducts on both half cycles and does not care about polarity.

On the board side the 1 µF capacitor bridges the zero crossings. Without it the output would be a 100 Hz square wave; with it the output is a steady level while mains is present.

Sensitivity - what voltage actually turns it on. This is not on the datasheet, it is derived, and the spread is wide because the part is sold in CTR bins. The -3 bin is 100-320 % at 1 mA, and 50 % minimum / 160 % typical at 0.5 mA, which is the region that matters here. Pulling the 10 kΩ down to saturation needs about 0.29 mA of collector current:

So on a 230 V system it is on with a very large margin, and it stays on down to somewhere between a fifth and half of nominal. It is a presence detector, not a voltage measurement, and those numbers are a calculation - measurements on both models land at or below the typical row, and the figures are on the 01 and 02 version pages.

The 200 kΩ is high enough that stray capacitive coupling on a long unloaded cable will not usually trip the input: a phantom voltage would have to push a few hundred microamps.

And DC? Electrically yes - anti-parallel LEDs conduct either way, and DC is actually the easier case, since there is no zero crossing for the 1 µF to bridge. Thresholds are in the table above. DC is not a rated mode, and the optocoupler's insulation rating is quoted per working voltage rather than per waveform.

The coil is driven by a small MOSFET with a flyback diode across it, from the 5 V rail, and the drive pin is ordinary PWM-capable GPIO - see Relay coil PWM hold.

A local 10 µF reservoir sits on the 5 V rail beside each relay, so that PWM drive does not modulate the supply the rest of the board is running from.

Under load it is the PCB copper under the relay that heats, not the relay body. That is what the thermal measurements on the version pages show.

This is a firmware technique that the hardware is built for, not a feature you switch on. Drive the pin statically high and the relay works - it just draws about seven times more.

The hardware side of it is the reservoir capacitor beside each relay, which keeps the switching current off the shared 5 V rail.

The coil as designed picks up at 3.75 V and drops out below 500 mV, over about 70 Ω. Those two numbers are the whole design space: energise at full duty to get above pick-up, then hold anywhere comfortably above drop-out.

The triac is driven through a digital transistor and a series resistor (from the 5 V rail), so the mains side is isolated from the logic with no extra parts and no isolation for you to provide.

It is a zero-crossing type. It only ever turns on as the mains waveform passes through zero, which is why it generates almost no switching noise and why the inrush into a resistive load is gentle. The same property is a limit: phase-angle control and dimming are not possible.

AC only. A triac cannot switch DC, for two independent reasons, and both are absolute. A triac latches: once triggered it conducts until the current through it falls below its holding current, which on AC happens automatically at every zero crossing and on DC never happens at all - so there is no way to command it off. And because this is a zero-crossing type, the trigger is gated on the load voltage being near zero, a condition a steady DC supply never reaches, so it will usually refuse to turn on in the first place. The dangerous combination is that it can latch during a supply ramp and then never switch off again. Use a relay for DC.

Off-state leakage is a specified 100 µA maximum, at the device's full 600 V blocking rating and therefore lower in service - roughly twenty milliwatts into a 230 V load. That is far too little to keep even a small motor turning.

Each output is protected three ways:

• a resettable PPTC fuse in the 5 V feed - 100 mA hold, 300 mA trip, 24 V;
• a flyback diode across the terminal, for an external coil;
• an 8 V ESD and TVS clamp on the terminal.

The flyback diode is the one worth knowing about. Most low-voltage outputs on comparable boards leave that to you, and leaving it out is how an inductive load destroys a driver.

The same 10 µF reservoir as the relay outputs sits on the rail beside each driver, so PWM on this output does not disturb the supply either.

That sharing is what the ? means in RELAY4 [IO4?] and SIG2 [IO5?] on the silkscreen: the board declines to state a pin's function when a jumper decides it.

One asymmetry worth knowing: the output side of each pair runs through the output buffer, the TIO side does not. A TIO line is a direct MCU pin. That is what makes it bidirectional and unaffected by the global output enable, and it means nothing sits between the pin and your signal to distort it - but equally, nothing sits there to protect it.

Allow about 35 ms on a channel before trusting the reading. That is the number that matters for firmware, and it is much longer than the multiplexer itself needs - the multiplexer switches in well under a microsecond.

The wait is the filtering. Each input carries a 1 µF capacitor on the sensor side and reaches the multiplexer through a 33 kΩ isolation resistor, and the multiplexer's common output carries a further 100 nF. Selecting a channel means charging that shared 100 nF through the 33 kΩ, from whatever voltage the previous channel left on it. The time constant is around 4 to 5 ms once the multiplexer and the divider are included, so a fully settled reading needs seven or eight of them.

Read sooner and the value is pulled towards the channel read before it - a warm sensor after a cold one reads too cold, and an input left unconnected drags the channel after it upwards. It is a smooth exponential, not a glitch: half the dwell does not give you half the error, it gives you a reading that is confidently wrong.

The reference firmware waits 80 ms per channel, which sweeps all eight inside a second. Nothing in an HVAC system moves faster than that.

The board prints the whole procedure on its own bottom silkscreen, next to the full GPIO map. Here is the complete channel selection table:

Configure IO11, IO10 and IO1 as general purpose outputs and set levels to low or high to get the corresponding input on the IO0 pin.

It exists so the board comes up in a known safe state. During reset and boot the MCU's pin states are undefined, and on a board wired to mains that is exactly when you do not want an output to twitch.

The enable line is held in the disabled state by a 100 kΩ pull-up, so the safe state is the hardware default rather than something firmware has to establish. A board that never runs its application never switches anything.

That the eBUS transmit line is gated too is deliberate - the TX can be physically disconnected - and it is why the board will not talk on eBUS before the application is ready either.

Note that the global output can be physically enabled, too. In an emergency, solder the J1 holes together and you will get a permanently enabled global output.

Doing signal isolation and power isolation in one part is what makes this fit on the board at all, and it is why the adapter is completely passive as far as the eBUS system's power budget is concerned. The barrier is a reinforced digital isolator; the converter it carries is short-circuit and overload protected.

The receive side is a resistive divider into a comparator with hysteresis rather than a bare logic input, so the bus's wide and rather forgiving level definitions resolve cleanly. The transmit side is a single transistor pulling the bus down, with a zener clamp.

The EBUS ON jumper feeds 3.3 V to the adapter. Remove it and the adapter is completely off and draws nothing - from the board or from the bus.

Both the transmit and the receive line have their own enable jumper, and each jumper does two jobs at once: two of its pins connect the line between the MCU and the port, the other two enable that line's indicator LED. So you can disconnect TX or RX from the MCU entirely - useful when a third-party device is driving the same line - and silence an LED, from the same header.

Model 01 only:
The port is a 5-pin header in PICkit pin order, so the same cable that programs the PIC can be used as a serial link. That is why one pin is marked NC and the port is silkscreened PICKIT PORT additionally.

Bus pull-ups: footprints, not parts. The board carries through-hole positions for the I2C pull-ups, silkscreened I2C PU, and they are deliberately left unfitted. Every OLED module sold for this socket already brings its own, and doubling them up is one of the easier ways to make an I2C bus misbehave.

If you attach a device that has none, solder a pair of resistors - around 3.3 kΩ - into the positions provided. The positions exist precisely so that you have somewhere sensible to put them.

Debouncing differs between the models, and anyone porting an application between them should know it. The 02 filters each button in hardware with an RC of about a millisecond. The 01 does not - it has a pull-up and a series isolation resistor only, and its firmware has to debounce.

Model 02 only:
The left button doubles as the MCU's BOOT pin, and there is a third button, RESET. Holding the left button while pressing RESET puts the MCU into download mode - the standard way to flash it over USB-C.

Both CC pins carry the standard 5.1 kΩ pull-downs that identify the board as a sink. Those resistors are what makes a USB charger turn its output on at all.

There is no USB Power Delivery. PD needs a controller communicating on the CC line, and there is none. The board asks for 5 V at default current and nothing more, which is all it needs.

Model 01 only:
The connector is a 6-pin power-only part with no data pins wired. Do not expect to flash a PIC through it - the PIC is programmed through the PICkit ICSP header.

The front end is a bridge rectifier of four schottky diodes - which is where "either polarity, AC or DC" comes from - into a wide-input buck converter, with a 600 W transient-voltage suppressor across the input rated for the printed range.

Polarity insensitivity is a property of the topology, not a protection device - nothing is harmed and nothing has to be reset if you wire it backwards.

The thermal limit moves with input voltage, and the two ends of the range load different parts. At low input the bridge rectifier is the hot component; at high input the buck converter is. Neither exceeded 53 °C in air at room temperature at roughly twice the worst-case load either board draws on its own - the numbers are on the version pages.

Every LED runs through at least 10 kΩ, so the current per LED is negligible against everything else on the board - it does not appear as a line in any consumption figure.

A single RISC-V core at up to 160 MHz with 8 MB of flash, and an on-board PCB antenna. The complete GPIO map is printed on the board's bottom silkscreen. The radio stack and the control loop share one core.

Comes with the example application pre-flashed for demonstration and testing only.

The board is designed around the PIC18F16Q40, the largest-memory device in the 20-pin PIC18-Q40 family, with 64 KB flash, 4 KB RAM and 512 bytes of EEPROM. Other 20-pin DIP PICs may work if their peripheral pin mapping matches, but that is your verification to do. All 18 of its I/O pins are used - 14 for terminals and buttons, 4 for the UART and I2C. Every socket pin is silkscreened by name.

Comes with the example application pre-programmed on the optional PIC for demonstration and testing only.

In practice the board survives being fed up to 5 V from a PICkit. The PIC itself is 5 V tolerant, so is the eBUS adapter, and bench testing found that even the OLEDs come through it unharmed. That may not hold for a third-party device on the QWIIC connector or the headers. Treat it as an observation, not as a rating.

A blocking diode stops current flowing back into the 5 V rail, so powering the board from the programmer leaves the 5 V rail dead - which is worth knowing before it surprises you.

The PICkit senses the voltage on the power line, so programming also works with the board powered normally, from USB-C or the alternative power supply.

Two pins on this header do double duty, and both were designed around deliberately:

• VPP/MCLR# is also the mains signal input. A diode between the two keeps the programming voltage out of the optocoupler circuit, so the input and in-circuit programming coexist.
• ICSPDAT/PGD and ICSPCLK/PGC are also the two buttons. Each button carries a pair of resistors chosen so that the programmer cannot be disturbed by the button, or the button by the programmer. Do not press a button while the PIC is being programmed.

Each board carries a name, its version, a copyright, a WEEE mark, a lead-free mark, a QR code with a link to the specifications, a serial number, and havaxer.com.

The full board layout is published as a PDF for each model, on the version pages. Printed at 100 % it gives exact dimensions and hole positions - what you need to cut a panel or plan an enclosure.

The board documents itself. Every terminal carries its functional name and the MCU resource behind it - TEMP1 [Ch6], TRIAC3 [IO22], RELAY2 [RC6], PBL [RA0] - and every rating is printed next to the terminal it belongs to: RELAY OUT 250VAC 10A, TRIAC OUT 220VAC 1.2A, ANALOG TEMP SENSOR IN 3.3VDC 10kΩ, DRIVER OUT 5VDC 100mA, ALT. POWER IN 7-33VDC 7-28VAC, and each rail's total budget.

The 02 goes further: its bottom silkscreen carries the complete GPIO map and a worked example of selecting a multiplexer channel. The 01 silkscreens the DIP socket pin by pin.

Standing at a panel with the board in front of you, you do not need this page, a datasheet or a laptop to know what a terminal is or what it is rated for. The board is the reference.