I'm starting to form the hypothesis, that group delay (as well as frequency response; particularly around 8 kHz and possibly around 2 kHz - aka, the Blauert bands) is responsible for much of the out-of-head soundstage (or is it instrument separation? I dare not answer that question yet) phenomenon. To wit, compare the Focal Celestee with the Utopia, Ananda and HD800S:
Celestee: "soundstage is mostly whacked in the middle of your head"
Utopia: "You get a halo of sound about 25 to 30% outside of your head with very nice instrument separation and excellent clarity."
Ananda: 'You do get nice sensation of "speakers away from your ear"'
HD800S: "What was remarkable and uncanny was separation of instruments."
I've played around with increasing and decreasing the 8 kHz peak^1 of the HD800 (not S), and have noticed, that the soundstage collapses when the peak is decreased; and increases when the peak is increased. I've also tried putting this peak on other headphones, with mixed results: though the soundstage might increase, an audible resonance peak is then detected. Clearly, frequency response alone isn't responsible for soundstage. I'm now suspecting, that the cause is reflections, as evidenced by group delay over specific frequency ranges. This also explains why fiddling with frequency response alters soundstage: as response changes, so does reflection intensity.
I'd love to further test this hypothesis with iems: I suspect, that they have no group delay at all, indicating no reflections, indicating a lack of soundstage. I'd also like to test whether or not simulating reflections and group delay, say by increasing reverb over specific ranges, increases soundstage. I tried this once, but the plugin I used was insufficiently precise. I wonder if the famed Smyth device does something like this....
Thoughts and opinions? I wish Axel Grell or other headphone designers were members of this forum: surely they have the knowledge we seek!
^1: Or is it a peak? If the frequency response is flat, but there are a tonne of reflections, would the frequency response curve show a peak where none exists? In such a case, the frequency response curve is not just measuring frequency response - but also reflections. Perhaps a free-field measurement could answer this question.