It's completely true that any immersive audio system may not always be as easy to "hear" as listeners expect. This gives rise to hyped demos with pushed height and surround channels, and specific demo material with in-your-face immersive effects. But with real program material played on a calibrated system, particularly when it comes to actual Atmos tracks where what is placed in those additional speakers is the result of a deliberate artistic decision, they may spend quite a bit of time at lower levels or not lit up at all, while other times they contain highly obvious fly-overs and other effects. And the degree to which an effect is heard is also deliberate. The comparison has to be made by turning off Atmos processing reverting to standard 5.1, unfortunately not always exactly an "instant" change. In the early days of 4.0 surround we used to run into low surround audibility because of directorial preference (they didn't understand the format), which, in turn, resulted in people pushing S as much as 10dB above cal so they could hear their new surround speakers.
The same bypass comparison should be made for channel extraction systems, Atmos, DSX, or anything else that upmixes. Switching off all immersive processing, extraction or real, is the only actual way to hear the otherwise often subtle differences.
Upmixing is always an abstraction of sorts. Something that extracts additional channel information from a standard 5.1 mix like DSX may often be more easily heard as active. But the reason for that is the same as the reason it's not exactly "correct". Extraction always makes assumptions as to what should or should not be placed in those channels. The assumptions are all technically wrong because the track wasn't mixed in that channel plan, but if the extraction algorithm is well done, and DSX was one of the better ones, it ends up as pleasing. But where Atmos playing an Atmos track takes an object with directional vector and performs a predictable downmix to the speakers you have, DSX and Atmos extraction from 5.1 can never do the equivalent with an upmix, it must always "guess". The guesses are very general and often result more activity in those speakers, which may be more pleasing to some. DSX left a door open for discrete tracks in that channel plan, but it never happened.
A historical side note, the original Dolby Stereo 4.0 tracks decoded with the theater surround processor using their Cat. 150 card involved pre-ProLogic decoders that were sloppy. They produced several undesired decoding effects, one which was called "magic surround", the random decoding of material into the S channel that wasn't intentional, technically a form of wild upmix. For that and several other reasons, those tracks were mixed while monitoring through a reference decoder in the dubbing stage so the "magic surround" effect would at least be known if not minimized. Early home Dolby Surround processors didn't use any steering logic (there were no chips in quantity for that market, the cinema processors used repurposed Tate Audio quad steering chips), so they produced even higher levels of "magic surround", but those were the days when the whole surround idea was still new and novel. Today's ProLogic is actually far superior to the old Cat. 150 cards in many respects. Reducing unexpected decode artifacts is one of those aspects, but then they can throw some back in with algorithms like the Music setting of ProLogic.
Personally, I've always wanted some form of user-programmable metadata system with upmixing systems so the user could pick not only the algorithm but the specific surround effect level for any given bit of media for upmix, and have the system auto-recall that preference every time that specific media is played. Wouldn't it be great to throw on a favorite music track and have your system pick Neural Surround at a reduced level, exactly as you like it just for that track, then revert to pure stereo for the next track? Just another on my list of ideas that will probably never happen.
