here we go again. I won't even go to audible differences between redbook, dsd64 and dsd128. because I don't know your system, and after all it is possible that you get a difference. if only some kind of IMD coming back into the audible range. so let's dismiss that part.
but plz try to find another justification than the past 20khz music being recorded. how many human adults can hear 20khz(not even trying to go above)? I'm 38 and have always had better results than even most youngsters when checking my ears for work. I've always been well protected, half luck(living for years in desolated places without much more that quiet old people), half care(I have used for as long as I can remember all kind of protections, I've never been a fan or nightclubs and when I went I used those protection. right now I have 3different models of foam, and some custom silicone protections. just to say I have always cared for my ears and listened to music quielty. yet over the last years I've slowly but surely lost 19khz, then 18khz, and I'm now perceiving slightly past 17khz but only a little boosted sounds at 17khz, if I use a sweep, then it's already lost and I can only grasp 16 something. and I believe that most adults are closer to me than they are to bats. that's one, age! but you could argue that perceiving consciously is different from our brain using it. that's true as long as we actually have enough working parts in our ears to perceive the sound. but lets keep at it.
to hypothetically have a handle on those ultrasounds, first we would have to listen really loud to flatten out fletcher & munson response. else we're pretty much sure that the midrange would cover the rest by being too loud. if something is much louder than the rest our brain will by himself discard most of the quiet message and focus on what's louder, so we need to listen to music real loud.
but at the same time, you don't really have a lot of margin to set the volume level, as the levels where flecther&munson gets pretty flat are close to the levels where sound starts to hurt and that's mostly true for high freqs and low freqs(and of course it's damaging for our ears if we listen a full album at those level, that put us back to age and hearing).
that's 2, loudness.
now we're talking about headphones for binaural, so we all agreed on this we need a special one, that at least wouldn't roll off like most headphones do. I really can't tell much about that, as all I know about headphones signature and distortions comes from FR graphs and they usually don't go high in freqs. so that's pretty much a mystery for me, but we can agree that it's usual for any driver to roll off at some point. also given that minute changes in placement can add or remove several DB in the trebles, I'm not super confident about how precise we can be in the ultrasound department.
also as most systems for measuring headphone don't care about what's past 20khz (and often less it seems), we would also need to be concerned about finding a headphone that doesn't distort too much the ultrasounds(because if as you say they are indeed important for spacial cues, then they have to be the right sounds, and not just distortion sounds). that's 3, headphones.
so if I had to make a wild guess, your horse wouldn't get my favors. because if I have no particular doubts that you perceive differences, even if again the odds aren't with you given past controlled trials. but you've seemed concerned enough about the subject to maybe get past just sighted bias and something probably really is audible. still I do have high doubts that the ultrasounds are responsible for those differences. and if they are, then I would still bet on IMD in the audible range caused by the ultrasounds instead of following you with ultrasonic cues helping for imaging positioning. maybe your 16bit version begins to rolls off around 16 or 17khz because of how the high freqs are filtered out, and that's what you're hearing? maybe your recording system induces some ringing on one or the other type of file? I really have no clue about the reason why you get a difference. but the chances of that reason being ultrasound aren't much for all the reasons I mentioned.
so your experience, I want to believe. you're conclusion as to why, I do not.
Like I wrote previously, it is NOT the mere audibility of the sinusoidal signal up to frequency of so and so much kHz. It is the timing of overtones, particularly those that do exceed 20 kHz. In nature, there is no mechanism ( besides air that does attenuate HF content over the distance ) to disrupt this timing among fundamental and harmonics. As noted before, these levels are LOW, LOW; LOW ; nowhere near anything like 0 dB. You can observe any piece of music ( in PCM ) in foobar2000's spectrum analyzer - there is seldom 20 kHz bar gets to see any action, and when it does, it is only ocasionally peaking at something like -20 or so dB.
ANY technical means to transduce from sound to an electrical signal, to record this signal and play it back by transducing it from electrical signal back to sound involves frequency response limitation(s) - and they add up in the end. Ideally, there should be NONE.
I am 53 and hear reliably to 14 kHz - I am no BatMan and can clearly recall having (much) better hearing when younger. Yet the difference I can hear between the limited and "unlimited" frequency response remains pretty much the same. The system can be pretty much deciphered from my profile - Stax Lambda Pro and AKG K1000
are around from mid eighties/early nineties and should not be anything unknown. I do tend to use phono cartridges with the extended frequency response - some go well past 50 kHz .
No, any of my (also) 16 bit devices are quite linear up to 20 kHz, none is rolled off before that. That "quite" is less than +- 1dB @20 kHz in all cases.
With ONE exception : the Pioneer PD-6J CD/SACD player has option to use one of the latest iterrations of Legato Link , which was pioneered by Wadia IIRC. http://www.head-fi.org/t/42951/pioneer-legato-link-appreciation-thread I will not go into detail here - suffice to say, when used "normally" without Legato Link in an "average" system, PD-D6J is a decent performer; kick in Legato Link and insert it into a FAST system, you get a slightly mellower, but much more 3 D presentation. It can become heated debate - which calms down the moment the difference is heard. Sorry, can not stash it into internet and expect it to be properly demoed on the other end.
Technically, Legato Link DOES limit the linearity in the audible range to 20 kHz somewhat - but bove 20 kHz does not filter as per usual brick filtering of the redbook CD. Its use IS system dependant - it is unlikely to be used in lower quality system(s) that lack extended bandwidth well beyond 20 kHz.
You raise one VERY important issue : headphones'/
EARSPEAKERS' capability to, firstly, offer response beyond 20 kHz, and even more important, allow for precise enough orientation of driver(s) relative to the ear in order so they can be "heard" by the pinna at least. One of the concepts that does allow for this and is with us for at least 40 years is Jecklin Float Electrostatic - it is simply large enough to cover the pinna no matter what and does have the frequency response extended "enough", limited in practice by amplifier/transformer combo or direct drive high voltage amp, not the transducer itself. It is also one of the three "headphones" that completely avoid touching or encircling the pinna - and is the pioneer of all three ( the second being AKG K1000, third Sony PFR-V1 Personal Field Speaker System , a lower priced "clone" of K1000 if you will ). I am familiar with the Jecklin and own K1000 - and they both excel on binaural, particularly the K1000. Never saw the Sony mentioned in flesh, so I can not comment. K 1000 can be made to sound pretty differently, depending how much one does open its "wings" - so this is too open to debate and can not be trully appreciated unless experienced. The response naturally does change a bit according to the openess of the "wings" - yet it can be safely said it is reasonably good/useable up to 25 kHz or so.
The reason why published measurements of headphones ( or any other piece of audio gear, with the exception of amplifiers ) only go to the 20 kHz is simple: any manufacturer would get godzillion questions WHY they publish such "irrelevant" information, why do they bother with it in the first place, etc - I think the thread(s) on head-fi are enough to get at least a vague idea. It also means investing in better, more costly measuring setup - only to get lambasted by 20.000,00 Hz limit crowd for trying to improve things ?!? But rest assured that above 20 kHz measurement have been and are being made by the better manufacturers - because, if nothing else, one day one of them will roll the first snowball off that "off the limits beyond 20 kHz" hill - and competition will force the others to follow. The idea is NOT to have the most extended response in absolute terms - but the most extended response WITHOUT any major abberations. That is to say, an audio device with response of say +2 - 5 dB up to 50 kHz is preferrable to one of +7 - 10 dB to 100 kHz. This usually automatically brings better sonics in 20 -20 K range as well.
If one goes to any musical instrument shop, tries some cymbals, violin etc - and records them with a microphone/preamp/recorder capable of response say to 50 kHz, one does get "more than zero" beyond 20 kHz. Nature does not filter it - and so should not we. It is OK to say "it is too expensive" - just as much is OK for me to say that a "Ferrari is too expensive for me" - yet it is not OK to ban the car of such performance, based on my incapability or unwillingness - as there are both people willing and roads/tracks capable of supporting this performance.
Regarding loudness - the only concession to "convinience" I see that definitely does have a place in audio is remotely controlled volume. A difference of 1 dB in overall loudness can be a make or break regarding naturalness of the reproduction - and recording levels that would support the natural sound pressure level of quieter acoustical sources (harpsichord, guitar, etc ) are not yet supported by the present technology; I mentioned harpsichord specifically, because if recorded at its proper - 20 or so dB, it would come off "dead" on the recording, particularly at redbook CD. It is better to have it recorded at higher level and reproduced at lower, in order to have as high resolution as possible. That is why a harpsichord track in a random/shuffle play can sound so LOUD - it can be louder than a rock band !
And that is why I try to match the monitor/playback level as much as possible to the real thing heard live; no other can be adjusted per ear as precisely as binaural when you are using your own head - so levels I use are normal and not loud(er) than live. Just for clarification: whenever recording, musicians/singers/conductors invariably require MORE volume than I am using for the K1000 - because they are closer to or are the source of sound and would like to match the level as heard by them. It is quite a difference if you are a conductor in front of a choir/or singer in the same choir - or a listener some distance further back in the hall. I always try to capture the sound as it would be heard by the listener in the best possible seat in any given venue.
I know these differences in perceived volume could be quantified by measurement(s) - I merely wanted to emphasize that I try to match the loudness between live sound and reproduction as closely as possible.
Furthermore, there is one problem not many people are aware of; we perceive loudness differently over various portions of the 24 hour day. For this reason, it is difficult to reproduce some quiet music most suitable for evening/night during most busy time in the morning/early afternoon , when our surroundings are usually much louder. And vice versa - I do not see likely anyone playing say Iron Maiden "in full cry" at realistic volume - at 04:00 AM.