SACD fundamentally flawed?

By this logic "Fortunate Son" that stays below 15kHz would be perfectly captured in 32kHz/16-bits, right?

 

It seems that every time the "CDs Are Good Enough For Human Hearing" thread comes along, someone will invoke Nyquist. While it is true that 44.1k PCM encoding can meet or exceed the requirements of the ear, that is a far cry from saying that real-world 1Fs ADCs and DACs can do the same. A proven mathematical theorem is one thing, but when you're talking equipment, issues like clocking and filtering have a snarky tendency to get in the way. Especially the anti-aliasing (adc) and anti-imaging (dac)filtering required by the very same Nyquist theorem. These filters have a well known deleterious effect on the accuracy of the recorded/reproduced sound. Over-sampling at 2Fs (and possibly higher) is just one of the ways these issues can be side-stepped. With all due respects to NHK, testing for subtle perceived differences in audio systems is very tricky stuff. A seemingly well-designed a/b/x test can end up demonstrating that a 1 7/8 ips Dolby cassette is indistinquishable from the source. The differences between converters and sample rates can be a lot less discernable than that. I'd like to know more about NHK's testing methodology before accepting the results without question.

 

Thanks Tweaker for your insights. I for one whole heartedly agree with your statements.

 

I hear a big difference in the layers here in the studio. Is the 16/44 layer a dub down of the SACD layer or a separate mastering job?

 

I'm not sure what Nyquist has to do with proving cds are good enough. All Nyquist said was the input frequency has to be less than half of the sampling rate.

Hi res (both pcm and dsd) isn't just about bandwith, it's about more data per given sample. In terms of pcm: for every bit you add, you double the amount of data per given sample. That's what Steve is talking about.

 

Not exactly. The crux of Nyquist is that a sin waves of up to 1/2 the sampling frequency will be perfectly reproduced.

 

I don't know of any such tests of cassettes vs. source, so that statement strikes me as an exaggeration. I'd be happy to know I'm wrong.

Subtle differences matter a lot to all of us, but if they're so subtle that they can't be heard on premium equipment under controlled conditions, I wonder if they'll ever be heard in the field, given routine manufacturing variations at least.

Angel reports that she hears a big difference between the CD and SACD layers of Steve's Willie and the Poor Boys. I do too. I can't believe a well-designed a/b/x test wouldn't reveal a difference of that magnitude.

 

I find it difficult to relate to forum members, who are dismissing a professinal audio engineer. Because, it 'sounds better'.
I would also love to see the dirty little secrets of DVD-A list. Can you say; Watermark?
Tubes, typically, produce double digit distortion.
Tubes, 'sound better'.
I don't believe that in anyway, dismisses tubes high distortion levels.
Just because, one prefer's double digit second order distortion, too solid state hundredths of a percent first order distortion.
I would hope, we all can learn something from 'The List'.
And yet, our beloved endeavor lives on with the knowledge we acquire from our listening experience.

Thank's SH, for the listening 'test'!

 

I've been away for a while and I'm glad to see that this string has picked up again.

Could you tell me who is selling these? I didn't make it to AES this year so I may have missed something.

As far as the Nyquist thing; Tweaker is right, Nyquist assumes perfect boxes and we don't have any. Both approaches to high-res are, in part, about giving those imperfect boxes a bit (sorry) more room for error (we've discussed pass, transition, and stop bands here already).

 

In a sampled system, the sampling theorem must hold.

So assuming *perfect* conversion, ABSOLUTELY.

_________
Thom

 

This is analogous to saying all Einstein said was E=MC2. We are talking about a man who, as early as the 1920's, formulated one of the basic tenants of the sampling theorem. So please, show a little respect for the man, will ya!

For every bit you add you increase the dynamic range.

Look at this way, let us say we go from 16bits to 24bits, what the added 8 bits gives you is the potential of a noise floor that is some 48dB lower.

So you are in effect increasing the dynamic range, i.e. you are increasing the *potential* for communicating more real information.

In a properly functioning sampling system, with a given dynamic range, it is the bandwidth ALONE that determines the accuracy of the representation of the waveform.

___________
Thom

 

I see your point. So, instead of developing hi-rez formats the industry should concentrate on developing better converters for 44.1kHz/16-bit PCM, which is perfectly adequate for reproducing everything within the audible range.

 

Except for that annoying difference between pure mathematics and engineering. Nyquist requires that no frequencies above 1/2 sampling enter the system. Got a brickwall filter up your sleeve that's 120dB down at 22k but flat at 20k with no ripple, ringing, or pre/post echo in the audible band? Going out to 96k relaxes the filtering requirement nearest the audible frequencies and leaves most of the distortion at frequencies outside the range of hearing. Then there is the matter of impulse response which seems to improve by going out to 192k or DSD. And since there is no free lunch, 192k and DSD have their own problems. So, all good design balances compromises for the best audible results.

 

1Fs DACs in the "real-world" are a thing of the past. The "real-world" has moved on to upsampling.

In which "real world" A/D converter systems have you witnessed this? I would scrap it immediately.

A properly designed "real-world" A/D converter system can exhibit deviations in FR to less than +- 1/4 dB up to cutoff, as well as phase errors well within +- 3 deg.

Try to achieve that with analog.

________
Thom

 

I have been following this thread and line of reasoning since day one, but I feel that the whole thing with extened top end is a very important factor. Of course our very astute host has stated that only a dog can hear the upper two octaves provided by an extended frequency music medium, but when we are talking about real world perception and the human ability to "sub consciously" sense the higher frequency timbre of certain musical instruments, I believe that we all gain by allowing this timbre to be reproduced. Just as brick wall filters are detremental to high fidelity musical reproduction, the human aural perception system does not have a brick wall upper limit, rather our ability to percieve high frequency sound rolls off with frequency response, so that a natural reproduction of high frequency timbre would increase our psycho-acoustic enjoyment of music. Of course, this is just my humble opinion.

 

I'm afraid I was a bit hurried in posting my last response and may have ended up appearing to endorse an erroneous idea. Nyquist does not claim that frequencies from the Nyquist frequency (0.5F(s)) on down will be accurately sampled, he says that frequencies above the Nyquist frequency will not. As a worst case, imagine a situation where samples are taken exactly at the zero crossings of a Nyquist tone resulting in a data stream indistinguishable from black. Take the previous example, simply stretch the wave, and picture how the data points still fall close to the zero-crossings yielding little information about that particular signal component's amplitude. Reconstruction filters have come a long way, but I'm not sure they are that good yet. So it is difficult to claim, even if we had phase-perfect, ripple-proof, brickwall filters available, that Nyquist frequencies as close to audibility as we have with red-book would be optimal.

 

This is not correct. The signal is captured, but at 44.1KHz sampling rate it barely looks like 18KHz....

Nyquist's theory does say the signal can theoretically be captured, but at higher sampling rates the signal does resemble its analog parent a lot more.

 

Here's proof that Nyquist's theorem does describe the max theoretical limit of the representation of a certain frequency, but cannot assure us of the faithfulness of such representation.

Here's an 18KHz sinewave, 100%, sampled @ 44.1KHz.

 

Here's an 18KHz sinewave, 100%, sampled at 48KHz.

Neither one looks like a sinewave, right? But it's obviously apparent that higher sampling rates do, in fact, help resolution all over the audible bandwidth.

 

Here's another 'golden oldie' (from 2000) technical explanation of the problems SACD introduces:

http://www.stereophile.com/features/374/

(Assumes firm stentorian Rush Limbaugh voice: )

Folks! ... It ... just ... doesn't ... work!! Heh heh! (slaps desk)




FFF

(just MHO, as always!)

 

Yes, a 18kHz wave is represented by 2,45 samples per period when sampled in 44.1. That could cause some errors when reconstructing the signal.

Javier, where are these pics from?

 

But here's the deal. Many listeners think it does work, and at its best it works very well indeed. Many engineers think it can work well. Some of our favorite mastering engineers think it works well. All these folks could be wrong, sure, but their testimony strongly suggests that DSD's virtues far outweigh its problems and represent a significant improvement over redbook.

Some folks claim that all SACDs sound bad. That position seems extreme to me, and flies in the face of a lot of expert testimony. Some say that most sound bad, with a few exceptions. Logically, those exceptions mean not that SACD is fundamentally flawed, but that most implementations of it are flawed. That is a much less extreme position, but it also contradicts a lot of expert testimony.

Expert testimony's there to be disproved, of course, but more trustworthy experts have lined up behind SACD as a true audiophile medium and a step forward in high fidelity than I EVER saw lining up behind redbook. Bob Ludwig, Doug Sax, and lots of others were cool or hostile to redbook from the get go. Both think SACD/DSD is a significant improvement that puts digital at or much nearer the excellence of analog reproduction.

None of this is to say that SACD/DSD have no flaws, or that DVD-A doesn't sound good or that PCM is the devil. I don't believe any of those things.

Edit: The conclusion of the article linked to above states "It is likely that SACD was brought to market not as a way of bringing audiophiles closer to the music, but of making digital audio more difficult to copy." In that case, we're positing that all of Sony/Philips' claims about improved sound are not just wrong but lies meant to cover their true motives. Those gorgeous-sounding George Szell/Cleveland Orchestra SACDs don't really sound gorgeous, and Sony's figured out a way to keep me from making bootlegs of them? Seems like quite a stretch to me.

 

My goodness, Javier, you have just proven Nyquist was wrong, Shannon was wrong and science was wrong.

Of *all* the millions of EE's out there, no one, not *one*, has had the balls or the brains to challenge the sampling theorem......until now.

And now Markus seems pumped up and ready to go as well.

Good luck, guys!

__________
Thom

 

Relax, Thom. Here's a pic of a 18kHz signal sampled at 44.1kHz. Looks fine to me.