Hi Tanja. Sounds so familiar. I think I agree, swap OFFSETA and OFFSETB. I struggled with this a lot in 2009 when I did my first Viewsheds, I tried it back then and reported/complained to Esri Finland that the guidance is very, very confusing. Had to teach them what the things mean. But think it this way at least if you have some towers the visibility of which you wish to find out from whichever location: While OFFSETB is added to each pixel, it should be about 2 m, if a human being is observing from each cell, therefore OFFSETB should actually be said to be at the observation point in such cases at least. One observer cannot be in many towers (or pixels) the same time... OFFSETA "to be added to the z-value of the observation point" is to be added only to those few vector points in your feature class, not to each DEM cell. If they are the towers, they could be 100 m high or so.
Again yesterday out of office and without my old examples at hand, I run Viewshed by just looking at the on-line help, and did it the wrong way. The help is very counterintuitive for the case mentioned at least. Back then I also pointed out several confusing things about the image they use.