I do not know if there are generally agreed "recommended guidelines". SMART have a method. This is a cut-down version of the standard SMART method. My concern with this approach is that the use of PPS sampling needs up-to-date lists of settlements and their populations. In some settings we may lose some settlements and also have some new settlements. We will almost always not have accurate and up-to-date data on populations when displacement has occurred or is underway.

There are a few ways round this ...

You can make some effort to identify new settlements. You can do this using interviews with key sources. I have also used aerial observation (i.e. from light aircraft / helicopter ... I suppose drones could be used now but I have no experience with using them) or NEW satellite images (Google Earth imagery may be several years old even if Google claims otherwise and imagery may be downgraded in conflict settings to avoid its use by belligerents).

PPS will work with relative population sizes. You may be able to get some idea of the classes (small, medium, large) of settlements and use these as "populations"". It is a "rough and ready" solution but better, I think, than using old population data. An alternative is to use the RAM (rapid assessment method). This uses a spatial sample and weights by population during data analysis. In this case population data is ascertained by data collection as roof or door counts multiplied by mean population under each roof / being each door (which we collect as part of the survey). RAM was developed with UNICEF for use in conflict-affected parts of Sudan.

I hope this is of some use.