100% Natural Sounding Digital Reproduction: Is It Possible???

No. What you are missing is that the quantization error is randomized, so that it is not audibly correlated with the signal.

In an analog to digital converter, this error randomization happens pretty much automatically. In the digital domain, it can be done mathematically.

 

I've been following the discussion and find it quite interesting. Malc S, I find some of the things you mention quite enlightening, thanks.

Just one question: how does the use of 32-bit floating point aid in the amplitude rounding that goes on in the examples you have mentioned?

 

Although there is a lot of musical information above 20,000hz (see this link) nobody really talks much about this. Of course our present microphones might not pick this information up.

 

And of course for everybody that is not familiar with the Nyquist theory, this link has a lot of good information.

 

The discussion keeps getting back to "bits" and analog v digital and DSD v PCM. Yet the issue of realistic sound is quite different. The art of capturing a live performance (by whatever means), minipulating it (mixing and mastering it) and then reproducting it in the home has little to do with the underlying technology. There is something about live music that has yet to be successfully capture and reproduced (and understood so that it can be reproduced).

We all can name specific recordings with sonics we love and I bet that few have any specific feature in common we can identify. Want to give it a try?

Richard.

 

Hey Richard, This is the first post in this thread. I think that this is what the thread is all about, asking which digital resolution is needed to accurately reproduce an anolouge musical waveform. I think you lost track of the point of this thread.

 

There's really no mystery. 2 speakers (or 5 speakers for that matters) simply can't reproduce the original soundfield. It's impossible.

An alternative is binaural (dummy head) recording, but that has problems too. (1) Heads differ, and (2) When you move your head, the sound doesn't change like in real life.

 

Yes, I agree. I probably didn't say it right (not unusual). What I meant was with analog the waveform is continuous, so the "sampling rate" is virtually infinite as opposed to a digitized waveform created from a series of snapshot samples which need interpolation. Taking more samples of the audio signal helps, and by increasing the sampling frequency the aliased image you're referring to is moved further from the audible portion of the audio spectrum.

 

Of course everybody here is familiar with Ken Pohlman's excellent book THE PRINCIPLES OF DIGITAL AUDIO. This book is like the bible for everything digital. Well if you don't have time to read the entire book then click here for a very nice outline of Ken's great book.

 

Except that, once again, the "sampling rate" of analogue doesn't come into it. It doesn't have anything to do with the "sampling rate" of digital samples, which are not interpolated to create the audio (at least in a connect-the-dots sense).

 

Fact: Sampling does not involve interpolation.

 

Folks, FWIW, I will add my two cents here as I have done in other discussions like this. As some of you know from other discussions, I am a physicist and use to work in an area where I digitized signals that spanned 20 Hz to 20KHz, however, they were not audio signals. I want to try to help people understand that ultrasonic signals are important to what we hear even though we do not hear them.

None of us can hear a signal whose frequency is above 20 KHz and as SH has already pointed out we are lucky if we can hear a 15 KHz signal. The link below will take you to a .pdf file of Sony's white paper on SACD:

http://www.sel.sony.com/SEL/consumer/dsd/dsd.pdf

Open the link and go to page 7 in the presentation. Look at the graphs in the lower right hand corner. They show what you get when you digitize and the reconstruct the signal from the digital data for a 10 KHz square wave using a 44.1 KHz / 20 BIT PCM system and DSD which samples at a rate the is 64x faster than the PCM system. As you can see the best the PCM system can do is a sine wave approximation of the square wave, where as the DSD system comes very close to the square wave. Without getting into a lot of heavy duty mathematics the reason for the difference is that the PCM system has to filter out all of the signal content above 22.05 KHz to prevent aliasing (The Nyquist theorem). That ultrasonic content is what makes the square wave a square wave instead of a sine wave.

Now, here is the important part. If I had a signal generator that could produce both 10 KHz square waves and 10 Khz sine waves and hooked it up to an amp and then to speakers and all of you listened to the output, when I switched it from sine wave to square wave you would all hear a change. You might not be able to identify which is square or sine (unless you were trained to do so), but you would hear a difference. So now I hope you see how something that we cannot hear effects what we do hear.

The scientist who study "how we hear" have found that the leading edges of signals play a very important role in how we get our auditory clues as to what instrument is playing and where it is located and those leading edges are very effected by ultrasonic filters.

 

Wow! I would certainly expect a marketing paper from Sony to have a fair rundown of PCM vs. DSD...

I take much of Sony's propoganda with a grain of salt. I've seen enough "but what happens if THIS is done!" banter to lead me to the conclusion that such ephemera really isn't the point.

 

This is quite true Malc. I think you misunderstand me. My point was to blame the people who trot out the Theorem as the be all and end all of forgiveness of all of Red Book sins. The theorem is quite clever for what it is but explaining some bandwidth fundamentals does not explain all that exists within the audio phenomena. The fact is that science cannot measure all that happens in audio. Otherwise we would have perfect sound replication and yet we don't and while ever close are far from it.

Transient response, instrument tonality, soundstage, and many important other attributes of audio cannot be fully explained by the Nyquist application offered up by David. I don't disagree with the theorem in other words, just the incorrect interpretation (as you said) to explain 16/44 as sufficient for quality audio. It is, sadly, not even close. There is real value to be found in higher sampling rates.

When Jeremy Kipnis and I were more involved at Chesky Records he was among the first to record in 96k (remember the DAD era?). The difference in playback was simply astounding and this was one of the arguably top recording sessions of the day in terms of sound quality (Bob Katz was at the helm not Barry Wolifson isn't equally stellar) so we knew what quality Red Book sounded like. We noticed the impact particularly in soundstage, cymbals, tonality, and the drums transients attack sustain decay...

Even with greatly improved DACs since 1992-1995, sampling rates make a HUGE difference. That is one of many reasons I love 24/192 and DSD.

 

Wow. That's a lot. Explain to me how science cannot measure all that happens in audio? It seems science does a great job in areas that presumably need more precision than audio. Medical devices, astronomical devices...science measures all of these, and most of these require bandwidth allowances orders of magnitude above that of audio. What properties does audio have that prevent it from being accurately measured?

It is your opinion that science cannot measure all that happens in audio. I don't happen to think your opinion is very well supported by evidence. But opinions don't always have to be.

You're essentially creating a strawman here. Science can measure something, can know why something works the way it does, and yet not acheive "perfection." The problem with perfection is precisely that: it is unattainable. Nobody can get there. Science can measure all sorts of nifty visual resolution possibilities, yet we often settle with imperfection (in the form of 1200dpi, etc.).

What about it? Science knows what transients are. Science does, in fact, measure them. Rip any audio file to your computer. See the transients?

My faith in the verifiability of the transient experience has been failing as of late. I know that certain things "smear" them. Other than that, nobody had adequately pointed out exactly what they feel is present or missing.

What about it? The tonality of a group of instruments cannot possibly be reproduced correctly by a standard stereo set-up, but that has more to do with the two-speaker medium itself and less with any failing of the recording process.

Ahh, my favorite. Soundstage! The measurement of audio depth that somehow has nothing to do with discrete channels, which are always the same in any digital implementation. The above reason (the stereo set-up itself) is usually the factor limiting soundstage; surround implementations improve this by the virtue of having more discrete sources. This phenomenon is measure by science, incidentally.

That isn't entirely true. There's an explanation, right there in black and white. The issue isn't that there's no explanation, but that the explanation itself is problematic; it doesn't seem to fit the information, so is discarded out of hand instead of being considered as a very real possibility.

Ready for it?

The explanation, according to Nyquist, is largely that these things don't exist.

See? That's the explanation that's provided. It isn't a fun one, though. It's always the end-stage of these arguments, and it's never a pretty resolution.

Good, because disagreeing with an exceedingly-well demonstrated mathematical theorum is taking steps onto very perilous ground.

What real value? Smoothed transients? A wider soundstage? Crisper bass bite? More "presence" in the vocals? A less "thin" high-end?

See, here's a contradiction. You've said that you don't disagree with Nyquist, but also say that there's "real value to be found in higher sampling rates." Barring the admission of supersonic noise which may affect/color the sound (which I accept as a valid--albeit somewhat inspecific in application--possibility), Nyquist says the exact opposite: there is no real value to be found in higher sampling rates.

There it is!

 

Sony marketing techniques aside, 44.1kHz sampling can not reproduce a 10kHz square wave. I think I mentioned that earlier in this thread but oh well.

 

This one comes up over and over. Is your music comprised of all square waves?;-)

 

It's been proved that higher sampling rates are good for more than an extended frequency response. So what is meant by "real value"? Did you mean that those differences are negligible? I can see where a case can be made for that. Just curious because I like to know where everyone is coming from.

 

It isn't that the differences are negligable. Rather, it's that I don't know if they can be qualified in quite the way that they often are.

Which things (other than an extended frequency response) are you referring to, BTW?

 

As you probably know, a flute is only capable of odd harmonics. When a flute is played in a very energetic way, like Ian Anderson does, there are lots of odd harmonics created and these waves look somewhat square.

 

Somewhat square seems different from square.

Has there been an experiment done that shows that PCM can't reproduce this?

 

First, you explain to me what audio metrics explain soundstage capability? Unfortunately you can't. Another example: Think about MLSSA plots of speaker performance. Before these existed we were missing all sorts of information about loudspeaker performance. That is where we are with overall audio playback. And we are still missing information about loudspeaker performance. Science can do a reasonable job of describing certain aspects of it but not how those aspects interact in real time to create a natural sounding waveform.

Please note I have no problems with science-I own part of a software company that employs PhD mathematicians. I have just seen enough to know what we don't know.

Transient response is not easily measured. Seeing them on a ripped computer file does not equate to developing a metric for showing the difference between 44k and 96k capture of transient acoustic events.

The burden of proof is on you. How does the Nyquist show other audio events outside of bandwidth? How does it explain transients?

If that's true, then the experience of every top recording engineer I know (and I know a lot having engineered or produced two dozen albums) is wrong. That leads me to believe the Nyquist Theorem does not hold up well in practice.

Life is a reality show. Practice is what matters.

 

That wasn't what you were saying before, I don't think.

In any case, what scientific explanation do you want? I don't think "natural sounding" is necessarily within the lexicon of the sort of equation you describe above.

I have not a ****ing clue. Not one. I am not willing to be the avatar for Nyquist defense. I'm not an electrical engineer, nor do I pretend to be. My knowledge extends to a certain point, and no farther.
Which is why I've repeatedly asked that this be taken over to a place where people know the answers.

I really don't understand your reasoning here, but I guess we'll agree to disagree. I don't see how a theorem which describes how a system works can simultaneously not hold up in the practice of said system.

And, thankfully, reality's measurable.

 

Hydrogenaudio folks don't appear to have much sympathy towards audiophiles. I won't mention anything about the War.

 

It still remains the best place to engage in Tech Speak about this sort of subject. If science is the topic, hydrogenaudio is frequently the answer.