I did some more research on the subject. It seems that the PPD42 simply is the way to go. The GPY2Y1010 doesn't seem very useful for this application. It's does not necessarily always give wrong readings, but it's considered unpredictable since the results are between reasonably good to almost random.
I also saw tests where the same was said about the PPD42. This is probably caused by the fact that it has two signal pins while an often used wiring only uses one of them. For setups using both channels very cool results were achieved.
What is particularly cool is that these also contain scientific references, dragging away the subject from pure 'hey you guys with resistors and soldering irons, you're just playing around and we don't believe your results'. For this reason I would also not recommend the copied DSM501A. Even if it would perform similar it might raise questions you prefer to avoid.
Waag society also used this for their new experiment. The first smart citizens project primarily taught them how you shouldn't do this and based on this experience they made a new attempt. It's based on WiFi but they consider porting it to LoRa (anyone knows more about this?)
http://waag.org/en/blog/making-sense-making-sensor. Also check the github link for more detailed docs. It also contains temperature/humidity and nitrogen sensors.
For that nitrogen sensor they used a more expensive sensor. I couldn't find actual prices, but think about 50-100 euro because of calibration costs. Reasons I found so far to measure nitrogen is that it's an easy proxy for dust. Interpreting this nitrogen sensor is also the reason you need the temperature/humidity because changes effect its effectiveness. But given that the nitrogen measurement makes it much more expensive than this cheap dust sensor I think when you're on a budget it makes sense to drop both nitrogen and temp/humidity.
They base their concept on the dustduino.
.. and this is what I'm currently porting to LoRa
I'll add a lab post in the near future. An indoor version seems to work pretty nice.
Summary from all links mentioned above:
- calibration is a subject that has not been addressed yet. But if you calibrate them yourself using a few identical sensors and compare them to each other your get reasonable results
- do not try to make a low power device except if your find a way for 'free' air flow. If you don't, expect somewhere in the range of 100-150mA power consumption
- take the average of 1 hour readings as starting point. Everything shorter is too unreliable. With 1 hour it reaches acceptable accuracy.
- I can't recall where I read it, but while browsing I saw multiple reports that a relatively high number of sensors (perhaps 10% or so) of the individual sensors is simply unreliable. So order at least 3 to be able to compare them.