Inter Sample Overs vary with digital process ?

[Correction: Subsequent data has called most of the theoretical speculation below into question.  It turns out that SACD's do not have consistently low ISOs, they can possibly have the highest ISOs of all, as I detected with the the 'Poulenc' SACD released by Linn.  This is an orchestra with a large pipe organ, and if I had associated low ISOs with higher quality, let alone previously believing all SACD's to have low ISO, I would have believed this to require a minimal 2.0dB headroom for ISOs.  But in fact it set my current record...it required a a mind boggling 7.5dB headroom.  And it was somewhat inconsistent.  It might work with as little as 6dB headroom....or it might not.  It apparently produces a peak around the same size as the 96kHz sampling interval, and how high it registers depends on how much of the peak occurs in any one sampling interval.  So there is a large random element to how much headroom is required.  I've also noticed that on some DVD-Audio discs with 96kHz sampling rate.

It does still seem that Reference Recordings and 'Discipline' (the brand behind the 40th Anniversary King Crimson DVD-Audios) produce consistently low ISOs.  I would venture they seem to know what they are doing.  But the peak ISO level seems not to depend on DVD-Audio vs SACD.  Nor does it even have a fixed relation to music genre (rock vs classical) or does it necessarily have anything to do with the recording "Dynamic Range" or how much compression was obviously used---though it's possible some compression was slipped into the Poulenc at key moments and that's what gave it such high ISOs.]

I have begun fairly systematically copying all my SACD's and DVD-Audios into 24/96 copies for my hard drive.  And I'm noticing a few weird things.

Remember that I set my gain structure so that with an 880 Hz tone recorded at precisely 0dB (generated by Audacity) a +4 level would be the highest possible before clipping.

But for actual recordings, I have to set the level between -2.5dB and +2.5dB, giving them at least 1.5dB extra headroom and as much as 6.5dB appears to be needed in some cases.  This headroom is required for inter sample overs (ISOs) where in between the samples the signal peaks higher than 0dB.

Benchmark claimed that ISOs could be as high as 3.01dB and that many digital decoders failed to allow sufficient headroom.

BUT I am finding 3dB headroom is way insufficient.

EXCEPT, in some cases it isn't.  Some discs seem to cluster around 2dB headroom required, up to 2.5dB in some cases.  Which discs are these?

1) Reference Recordings HRx (174kHz/24bit)  seem to only require 2-2.5dB headroom.  These are some of the best sounding recordings ever.

2) SACD's generally only seem to require 2dB headroom  (I would have expected SACD to be the 'wildest,' but in actuality, it's the 'tamest.')

3) The DVD-Audios in the King Crimson 40th anniversary DVD-Audio boxes, which seem only to require 2.5dB headroom.  These are truly spectacular in audio quality (or in the case of Court of the Crimson King, merely way better than ever before).

On the other side?

1) Many hot sounding recordings on DVD-Audio, including Elton John, and especially Steely Dan.  (I've long noticed that.  I figured Steely Dan cranked up the compression so high it's bleeding out the ISOs). 

I wrote this off for awhile, but now I'm getting a pretty clear feeling that the recordings which require the least ISO headroom are the ones that sound the most natural, laid back, and 3D.  The recordings that require the most headroom sound highly processed.

You may be astonished to see some of the pictures when I post them.  Having the huge ISO's means the rest must be scaled back in a recording, losing dynamic range in the midrange.

I think the excessive ISO's occur when the the anti-alias filtering is insufficient, and full scale high frequency garbage gets into the digital encoding.  THAT's what's causing so much overshoot.

DSD has such a high rate it captures and controls all the HF crap by design.  So SACDs do not suffer from excessive ISO's.

Likewise, the PMI analog to digital encoders used by Reference Recordings, which must have superb filtering.

And whatever King Crimson was using in the 40th anniversary DVD-Audio set.

Now, it would be interesting to know what Steely Dan used in such things as the Everything Must Go DVD-Audio, that produces such high ISOs.

Perhaps it's not the digital process, but the amount of processing (including compression) used before digital encoding that is the culprit.  But the biggest ISOs look too big for just that, IMO.  Still, the sonic variation might just be coincidental, the least processed recording just happening to use the digital processes which produce the least ISOs.

This is also a function of the Oppo BDP-205, which is apparently not headroom constrained itself.  But I've found the height of the ISOs not to change with different reconstruction filters.  I think they are inherent in the data and and the degree of oversampling used, with higher amounts of oversampling 'revealing' the true ISO height (because you are filling in more of the points in between the points).

******

Update:

Now it appears that that the ISOs in DVD-Audio discs are all over the map.  They can be < 2.0dB, for example, in Queen's A Night At The Opera.  They can be < 2.5dB such as in every King Crimson 40th Anniversary DVD-Audio box.  Or, they can be as high as 6.5dB, as in Steely Dan's Everything Must Go, Elton John's Goodbye Yellow Brick Road, or Frank Zappa's QuAUDIOPHILIc.

SACD's still appear to be consistent at around 2.0dB, and everything from Reference Recordings is there too.

The Audio Hobby

The audio hobby means different things to different audiophiles, and isn't that wonderful?

Not always, perhaps.   It can be made awful in many ways.  People can fall victims to unfounded beliefs that cause them to waste time and/or money.  It can lead to the virtual inability to listen to music anymore, in extreme forms of audiophilia nervosa.

It can lead to endless bullying.  I think generally the best attitude is non-judgemental: live and let live.  However all the same I believe audio is filled with frauds of many kinds, both on the big and small levels.  You probably already know I side mostly with the audio objectivists as to what kinds of audio beliefs are founded and which are not.  All the same, I don't take it as my mission to change anyone's mind.  I suppose, in some cases, even I could be wrong too.

Anyway, nowadays I think generally it's a waste of time to compare good amplifiers, DACs, cables, power conditioners, or anything of that ilk.

All have been tested endlessly by audio objectivists getting only null results.  With my lack of patience, I'm unlikely to do better, if I follow all the proper procedures.  If I don't, then the result may be meaningless anyway.

My normal result with good amplifiers is this:  I start the auditory level matching process by matching the apparent loudness with both amplifiers to make it identical in fast A/B testing.  Once I have matched the level, I attempt to match the quality of the sound.  If one seems to have more bass, or highs, I assume that's because it's actually louder.  In this second phase I make only the smallest adjustments, 0.25dB at at time (I'm fortunate to have a first generation Emotiva Stealth DC-1 with 0.25dB adjustment.  That means, on average, the best case is within 0.125 dB of being exactly correct, which is close enough to pass the 0.1dB minimum in my experience.)  Ultimately, on every amplifier I've tested, I can make them sound identical simply by matching the level that closely.

Now I did not do such procedures when I thought for several years my then go-to amplifier, the Aragon 8008 BB, was sounding a bit harsh.  Back in those days, I found myself avoiding listening to the FM for very long.  It drove me insane.

I later found the distortion had risen to 0.7% because of low bias.  After bias adjustment, I got it back to 0.07% and sounding fine.

But I know that from years of experience and not A/B tests.  And measurements which make that experience believable.

And that's another thing.  One should often believe measurements, when they are meaningful and honest.  Not necessarily specifications.

Anyway, distortion can be a factor down to 0.1%, so products having higher than 0.1% should generally be avoided.  (In electronics, anyway, where it's easy to do better.  There's hardly any speakers that can do as good as that.)

In audiophile land, there are often electronic products with higher than 0.1% distortion.  And sometimes they sound better.  I think what's happening is that in some cases products with predominantly 2nd order distortion may fix recordings that were made with high amounts of 3rd order distortion.

Furthermore, boosting the amount of 2nd order distortion, which tends to occur with zero feedback designs (feedback works best at suppressing 2nd order distortion) can add additional "spaciousness" and air to recordings lacking those things because of poor production.

Things like this may work on some recordings and not others.  It may work best on recordings that are fairly simple, like a few instrumentalists.  Not on works of great complexity, like a full symphony orchestra.

Euphonic adjustments are like that.  Generally it's best to stick with low distortion, wide response, low noise, because that's the combination that works overall best on everything.  Basically what the objectivists say.

Other claimed magic requirements in design, however, are sold on the basis of faulty comparisons, typically failure to match levels very well.

Anyway, if feedback free amplifiers, and electrically charged cables, or whatever makes sense to you, go for it.

The best is when we're not bullying people over such things, one way or the other.

Even being forced to make a decision is a kind of bullying.

We are generally not designed to discriminate among audio reproduction systems.

We don't have a 'memory' that works very well for making such comparisons.  We don't store 'experience A' in anything like the raw form that would make for a good comparison with 'experience B.'

To be reliable at all requires, as the objectivists always say, instantaneous A/B switching.  Other than that, perhaps exhaustive training.

Furthermore, always being assigned to make the equipment comparison detracts from the process of having the most enjoyable and enlightening experience from the music, appreciating the music itself rather than arcana of possible sonic differences caused by different audio equipment.

Fine, the uber subjectivists say, just see which piece of gear gives you that most transcendent audio experience.

That's basically impossible, because each time you listen to the same piece of music you get a very different experience.  Symphony orchestras often like to prove this by playing a Premier (first ever) performance of some work twice, sometimes even without warning.  Few guess it was an identical repeat.  The identical music doesn't provoke an identical response.  And for a very important reason.

The you listening to any work the second time is now older and wiser, having already heard the music before.  The brain has already stored memories and made new connections.   Usually that opens up entirely new realms of experiences.  While closing down others.

(Curiously a work with more 'movements' can be heard more times without seeming repetitious.)

As many have often opined, that is what audio should be mostly about, the music, and quite often isn't.

(However in line with non-judgmentalism, I prefer to say that any mode of enjoying an obsession with audio reproduction is fine.  If your thing is building amplifiers that test conventional theories -- fine.  If your thing is arguing about whether such things can or do make any difference -- fine.  As long as I'm not to much detained by your obsessions.)

But along those line, I fear the excessive denigration of standard audio engineering practices, as often occurs in the writing of audio subjectivists (including the one I always loved to read anyway, Harry Pearson) tends to promote rather than discourage audiophilia nervosa and therefore lack of being able to enjoy music.

But some people swim in one thing or the other, so whatever works for you.  Doing 'comparisons' is also a way of just hearing things twice, which may itself be beneficial.

I have dubbed such demonstrations magic shows, and typically enjoy them even when (that is, all the time) my core beliefs are unshaken.  

I myself only drifted into near audio objectivism in my late 20's, after doing carefully constructed experiments I felt would 'confirm' my beliefs in tweakdom.  That's the path of many noted audio objectivists.  First they were fully taken in, then they decided to do some tests.

After the recording, DAC, and amplifier problems are 'solved,' is there anything left?  Two things, the loudspeaker/room (or headphone) interface, and the selection of the music itself.  Neither is a problem that will ever be 'solved.'

R128

Here is the EBU R128 standard for measuring "dynamic range" that Roon uses:

https://tech.ebu.ch/docs/tech/tech3342.pdf

Dynamic range can mean different things.  In the context of technical measurements of an amplifier or transmission system, dynamic range is pretty similar to "Signal to Noise."  What is range from the lowest signal that can barely be resolved to the maximum.  So a dynamic range of around 120dB is where many of the best units are (chips can be as good as 130dB).

But this has nothing to do with they "Dynamic Range" of program material.  But what has long bugged me is, what does this mean anyway, because every signal, no matter how high the peak, ultimately has to go back through zero again.  And how close it gets to zero depends on how finely you can measure it, assuming it's a continuous waveform.  So this is back to to the "dynamic range" of amplifiers again.

But this is not what audio/music people mean by the Dynamic Range of program material.  Their window of analysis is not 1 uS, say, the resolution of a decent digital scope.  Their resolution is the loudness in a 3 second time window which must be overlapping.  The spec is unclear to me, but in specifying "dBFS" it is clear to me they don't mean instantaneous levels but something like average or probably RMS levels...that is to say levels related to the equivalent sine wave measured with RMS.

And then the Dynamic Range is specified as the difference between the 10th percentile and the 95th percentile of these 3 second time windows.

Now it's still not clear to me how Roon uses this in level normalization.  There must be other parameters such as the maximum level, etc, because the R128 I've just described only relates to relative and not maximum levels.