@mhmd It's a function of your router's network address translation (NAT). It's the same way your browser can get push notifications (as from this forum).

When your Sonoff device communicates with the Chinese server, your router uses a different port number and stores that info. When traffic comes in to your external IP address with that port number the router knows which internal IP address and port to send it to.

You can read more about NAT elsewhere.

The bottom line is that the Chinese server isn't getting special access to your network and doesn't know your local IP address.

@grumpazoid The sensors themselves do not communicate with the MQTT broker. They communicate with the gateway as usual.

The gateway then communicates with the MQTT broker, like this example.

@johnrob I understand wanting to apply the calibration data even though you don't need it. I did the same.

The Arduino hardware doesn't support 32 bit calculations, but it can be done in software; it just makes your code bigger.

The calculations are for the calibration data calculated at the factory and stored on the chip.

Assuming your remote sensor is powered by batteries, you probably aren't going to be running the sensor all the time, which limits your filtering options (higher resolution output requires consecutive readings).

It's likely that the sensor is going to be accurate enough for the resolution you are using without applying the calibration data. You could test this experimentally with that sensor on your Raspberry Pi setup, comparing the output with and without application of the calibration information.

(I spent a lot of time with that data sheet because I also have one connected to a Raspberry Pi. It was nice of them to provide sample numbers so I could make sure my calculation code was correct.)

@rozpruwacz The big downside to using forced instead of normal is that you can't use the IIR filters, since they require relatively recent previous readings to work.

@gohan There are BME280 breakout boards with voltage regulators on this. I guess they'll draw a bit more current than those without, but the current draw is pretty darned low.

It draws so little current that it could be powered by a pin of the microcontroller, so it could be powered off instead of being put into sleep, although it draws so little in sleep I doubt it matters.

@tbowmo I found this review of a J1800 board that measured 14-18 watts without video.