So, good signal strength (good SNR) and all receive locations evenly distributed around the target and roughly the same distance from the target can yield very good results, skywave or not.
Yes, in fact back around the time that you wrote this, I realized that I had gravitated toward an understanding that one RX being much closer or farther away than the others would skew the results. I had more-or-less come to a belief of exactly what you say here - equidistant RXs are optimal - through trial and error.
Of course, this is more complicated if you don't know where the TX is (which is the point of the whole exercise). You use an iterative process to narrow it down. However, even with this knowledge, things can and will go wrong.
Because of my belief in "robust" solutions to things, I like to attempt to achieve basically the same location result with a mostly or completely different set of RXs. If it's a correct solution, I believe that you should achieve a corroborating result. If the RXs are roughly equidistant, I have been able to achieve a similar result this way, which is reassuring.
Sometimes I try to test it to be sure that it is leading me to the right solution by picking RXs that are not centered on the previous result, perhaps not surrounded by the RXs at all. The hope is that it at least tells me that the TX location is somewhere off in the direction of the previous result, i.e., corroborating. Unfortunately, this works maybe 50% of the time. The other 50% of the time, I get a completely different outcome that is also equidistant from those RXs in use for that second run, nowhere near the previous result and not a result that says “No, no. You are headed in the wrong direction.”
If you think about the implications of this, that is what makes the process of using TDoA on an unknown TX very difficult. It is very easy to be lead astray.
