16-Bits was never enough

[Jon Hodgson]

Originally Posted by karyoky****

hi,

that is not considered to be correct. reliance on input noise or component noise to dither is not recommended. quantizers emplioy methods to effectively dither when sampling analog signals.

It is not possible to create a "correct" analog dither for a 24 bit analog quantizer, thermal noise components will completely swamp any attempt to create the perfect amplitude of TPDF noise.

I'm not so sure about the sigma delta situation.

However this is not a problem, because any analysis of the quantization error of such a relatively large amount of gaussian noise will show an even spectral spread and random distribution... the signal is ALWAYS high in amplitude and chaotic with respect to the quantization levels.

[Jon Hodgson]

Originally Posted by karyoky****

hi,

no, no. the non-error component is inaccurate to the extent of the error component. by inaccurate i mean that it deviates from the actual analog signal.

the non-error component deviates from the original analog signal by the amount of the error component.

the non-error component is more accurate [the original signal's amplitude is described more accurately] when a higher bit depth quantizer is employed.

You're talking in circles now.

NON Error component, i.e. the component that has no error.

The non error component IS the original analog signal, the TOTAL output signal differs from the input by the error component.

[Ethan Winer]

Originally Posted by wado1942

your arguments are fidelity related, not resolution.

And the difference is?

Would you say a 96KHz 24-bit converter with a frequency reponse up to 18KHz is lower resolution than a 44.1KHz 16-bit converter with a 20KHz response?

Who cares? 18 KHz is not enough for "high fidelity" so that's the end of it.

If you can prove that 1/4" tape only has 4,096 descrete levels of audio, I'll believe that it has the same resolution as 12-bit digital.

Magnetic particles have a finite size.

There are only four parameters that affect everything related to audio specs and fidelity:

Frequency response
Signal to noise
Distortion
Time-based errors

That's it! Period, done. So if my $25 SoundBlaster card is better than analog tape in all of those parameters - and it is! - then the fidelity and resolution and quality are better too. More on audio parameters and their subsets here:

Audiophoolery

I'm honestly not trying to be argumentative, but I'm a real stickler about certain things. Like people defining a polarity inversion as a 180 degree phase shift. They're not the same thing even if they look similar on paper.

Just discussing this stuff doesn't have to be argumentative! I'm all for a good discussion as long as it remains civil and respectful. As for phase versus polarity, I agree 100 percent. That drives me nuts too.

--Ethan

[narcoman]

Originally Posted by karyoky****

hi,

show off. yeah, i'll do a few.

e...

yeah it's a long gig with a short deadline - if you know what i mean !!

[mobius.media]

Originally Posted by Jon Hodgson

No, they're the same thing, the error signal is the difference between the correct signal and the actual signal, if it's a non error component, it is by definition accurate.

Yes, but again that error can manifest in two ways - noise floor or distortion - right? At any point it manifests as distortion it becomes detrimental to fidelity, not just dynamic range.

[Jon Hodgson]

Originally Posted by mobius.media

Yes, but again that error can manifest in two ways - noise floor or distortion - right? At any point it manifests as distortion it becomes detrimental to fidelity, not just dynamic range.

Yes, but if it's perceptible as distortion... someone's doing something wrong.

[wado1942]

Originally Posted by Ethan Winer

And the difference is?

Let me put it this way, in a 12-bit signal, there's only 4096 possible levels. No information exists between those points and below -72dBfs, there's NO information at all. Now on my Otari, my nominal levels are 0VU with a good 16dB of headroom and the noise floor is -60VU. To you, that would incinuate a 13-bit equivelant resolution. However, detail can be heard a good 20dB below the nosie floor before it's completely masked by the noise. However, if there were such thing as a 13-bit medium, that detail that fell between the steps would be lost.

Going back to a visual standpoint, just to show that resolution and fidelity are not the same thing, people can see somewhere between 4,000 and 6,000 lines on 35mm film. Yet a lot of people will claim stuff shot on a 2,000 line digital camera looks sharper.

Who cares? 18 KHz is not enough for "high fidelity" so that's the end of it.

What mics and monitors do you use?

Magnetic particles have a finite size.

Yes, but you have virtually infinite combinations of magnetic particles in various states of flux.

There are only four parameters that affect everything related to audio specs and fidelity:

Frequency response
Signal to noise
Distortion
Time-based errors

I think the human ear (and mind) is smarter than any chart or graph. Those perameters you mentioned are very small windows in an otherwise huge physical world. They may help you understand WHY something is the way it is, but it's still limited.

So if my $25 SoundBlaster card is better than analog tape in all of those parameters - and it is! - then the fidelity and resolution and quality are better too.

Yet most people will agree that my Otari sounds better. Now "better" can be subjective but when something is consistently subjectively better, it's an indication that there's holes in our definition of objectivly "better". Ever since I put that thing into service, not a single client has chosen digital mixing over 1/4". Yes, I can emulate the frequency response and THD component fairly easily in the box, but my clients (and me) still hear an advantage with the real thing. That suggests the engineers that decided what makes a good sound have missed something. I don't want to get into another analogue vs digital debate so I won't comment on the subject any further.

Just discussing this stuff doesn't have to be argumentative! I'm all for a good discussion as long as it remains civil and respectful. As for phase versus polarity, I agree 100 percent. That drives me nuts too.

--Ethan

I'm glad, I tend to be a total bunghole without even knowing it half the time.

[Teddy Ray]

Information for prospective students

http://www.physics.sc.edu/kunchur/papers/FAQs.pdf

[Ethan Winer]

Originally Posted by wado1942

that would incinuate a 13-bit equivelant resolution.

Okay, 13 bits it is.

However, detail can be heard a good 20dB below the nosie floor before it's completely masked by the noise. However, if there were such thing as a 13-bit medium, that detail that fell between the steps would be lost.

Why do you think digital noise is different than tape noise? The two types of noise may have a slightly different spectrum, but we can hear stuff below the noise floor of either medium. Not that I want to listen to any noise while music plays. What you describe is more of an intelligibility issue rather than music fidelity.

Going back to a visual standpoint, just to show that resolution and fidelity are not the same thing, people can see somewhere between 4,000 and 6,000 lines on 35mm film. Yet a lot of people will claim stuff shot on a 2,000 line digital camera looks sharper.

I don't get your point. I'm not interested in subjective opinion of what "people" claim. Further, if anything your example makes my point! Film grain is sort of like analog tape grain, in that it's not clearly defined. So if we accept that 4,000 to 6,000 "lines" are needed with film to get the same perceived quality as only 2,000 lines on digital, that means the digital wins hands down.

What mics and monitors do you use?

I have a DPA 4090 microphone that goes past 20 KHz, though my big-ass old school JBL 4430s don't go quite that high. Not that it matters to me because my 60 year old ears don't go that high either. But just talking specs, I would not accept any format that falls short of what 44/16 digital can do.

Yes, but you have virtually infinite combinations of magnetic particles in various states of flux.

It doesn't matter. The proof is in the noise floor and distortion measurements!

I think the human ear (and mind) is smarter than any chart or graph.

Well, smarter is definitely not the right word, but I know what you mean. None the less, it has been shown again and again that specs correspond closely to what we hear. Within the limits of hearing. Which is far worse than specs can measure. We can measure artifacts 120 dB below the music, but nobody could hear anything that soft while music is playing too.

most people will agree that my Otari sounds better.

I sure would not! If you like 1 to 3 percent distortion that's fine, but it's not better fidelity in any way one could possible assess "fidelity."

I can emulate the frequency response and THD component fairly easily in the box, but my clients (and me) still hear an advantage with the real thing.

Unless you test this with a true DBT you can't really know that for sure.

--Ethan

[Waltz Mastering]

Originally Posted by Ethan Winer

I'm not interested in subjective opinion of what "people" claim.

... a little ironic?

[wado1942]

What you describe is more of an intelligibility issue rather than music fidelity.....Why do you think digital noise is different than tape noise?

It's a RESOLUTION ISSUE! If you added dither to that 13-bit signal, yes, you could heard detail below -76dB but assuming it's undithered, you'd hear nothing. You're basically saying what I am, that fidelity and resolution are not the same thing. Digital is theoretically noiseless. What you percieve as noise is simply the inability to resolve detail. Nothing exists between those levels whereas in noise, the signal is still there, just with something else added to it.

if we accept that 4,000 to 6,000 "lines" are needed with film to get the same perceived quality as only 2,000 lines on digital, that means the digital wins hands down.

I didn't say that those lines were "needed" to get the same percieved quality as digital. I work quite heavily in analogue and digital media of all kinds so..... I think it's hilarious that people are insisting on 10MP digital cameras now when the lenses attached to them can often only allow 1,000 lines to pass to the CCD. I've worked with a lot of "HD" video cameras that resolve less than even a good SD camera and think to myself "why did they pay double for having those 2 letters written on the side of the camera?" Yeah, the resolution on paper is 1280x720 but the reality is the fidelity is far less than that. One of the popular "HD" cameras I used had about the same image quality and cost twice as much as my old Canon GL1. Another camera I tested could show 250 lines, THAT'S VHS QUALITY! Yet both those cameras got reviews saying they were great low-cost HD cameras. In this case, it's the lens that determines quality over everything else. Most 35mm cameras have lenses that are no better than those on digital cameras. But digital cameras have this great little trick for when a lighter pixel is next to a darker pixel. They slightly bump up the brightness of the lighter pixel and darken the darker one. It creates the illusion of added sharpness that doesn't really exist.

Again, I'm not interested in an analogue vs digital debate. I'm interested in resolution not being the same as fidelity. You seem to agree with that idea more or less even though we have a conflict of terms.

[Ethan Winer]

Originally Posted by wado1942

I'm interested in resolution not being the same as fidelity. You seem to agree with that idea more or less even though we have a conflict of terms.

I guess I just don't see the distinction. Low resolution is low fidelity and vice versa. Resolution with audio relates directly to distortion and noise for bit depth, and frequency response for sample rate. I guess I see the distinction you're making, but I don't understand why it's relevant.

As for comparisons with lines and video, video quality is much more than lines. A big problem with VHS, as least for consumer decks, is the output comes through a composite wire. So the colors are all washed out and edges are blurry. Versus having access to the original RGB signal.

Whatever, I don't think we disagree much, except maybe that I used to own an Otari tape recorder and I'd never that that over modern digital.

--Ethan

[Ethan Winer]

Originally Posted by Waltz Mastering

... a little ironic?

Hell no!

Seriously Tom, lots of people claim lots of things, and many of those claims are not true. Since it's futile to argue about subjective stuff, I focus on provable fact.

--Ethan

[Waltz Mastering]

Partially joking Ethan, and with all do respect... Ricky Bobby style .; ).

There are some ppl who repeatably try to convince others that the lowest common denominator is all that's needed and anything more is pure bullocks.

I'm no scientist, but a lot of the reasoning seems very subjective.

[Jon Hodgson]

Originally Posted by Waltz Mastering

Partially joking Ethan, and with all do respect... Ricky Bobby style .; ).

There are some ppl who repeatably try to convince others that the lowest common denominator is all that's needed and anything more is pure bullocks.

I'm no scientist, but a lot of the reasoning seems very subjective.

It's the highest common denominator that's all that's needed, which is often a lot lower than some people imagine.

[Jon Hodgson]

Originally Posted by karyoky****

hi,

here something that david rick posted a while ago that talks about sigma delta:


"Dynamic Range = 6.02 * Bits + 1.76 dB

This formula is based on the assumption of a full-scale sinusoidal input signal. If the input signal is a full-scale square wave, then the answer is 3 dB higher. If the input signal is less than full scale, then the signal-to-noise ratio is correspondingly reduced. (I have the feeling I'm being overly pedantic here; perhaps I didn't understand your real question.)

Quote:
is it contended [theoretically] that, when quantizing a signal during sampling, the quantizer's output is:

1.) a perfect linear representation of the input signal's amplitude, plus a signal representing the quantization error, or,

2.) a signal that is defective to the extent of the amount of the quantization error plus a signal representing the quantization error?
You're speaking in terms of the additive noise model, again. In that model, we have:

output = input + error

for each sample. In the undithered case, we've noticed that the error depends on the input. But if we add proper dither, it doesn't. Actually that's a lie. For RPDF dither, the mean value of the error is independent of the input value (no distortion). For TPDF dither, the power in the error is signal is also independent of the input (no noise modulation). Humans apparently can't hear higher statistical moments, so TPDF dither suffices.

A noise shaper works differently. In that case, we wrap some kind of error feedback around from the output of the quantizer. The quantizer input is then some combination of the input signal and the error history. If we do everything right, we can shape the spectrum of the quantization error without changing the spectrum of the audio signal. This idea is the basis of modern delta-sigma audio converters.

BTW, it's possible to do noise shaping with or without dither. But if we leave out the dither, we can still get those pathologically colored truncation artifacts I was talking about earlier.

David L. Rick"

You've emphasized a line referring to noise shaping, not dither. Noise shaping is a seperate thing from dither, as David says just afterwards.

In fact the problems with correctly dithering a 1 bit system are the basis of Lipshitz and Vanderkooy's conclusion that DSD is inherrently unsuitable for high quality audio (they say it is impossible to dither correctly).

In addition I'm not talking about the theoretical desirability of dither, but the practicality of implementing it in the analog domain for a given circumstance. With a non oversampling converter (reads all 24 bits at once) I can't see how you're going to be able to create a nice bit of TPDF noise with the correct amplitude (johnson noise in the circuitry to create the TPDF noise will completely swamp it), with a sigma delta converter you have bigger quantization steps. so you're going to be able to get closer, but I haven't looked into how the imperfection affects the result.

Certainly off the top of my head I can't remember seeing analog dither mentioned in sigma delta ADC datasheets (though I wasn't looking for it).

[Brad Blackwood]

Originally Posted by Cellotron

Folks -
PLEASE - before posting on the limitations or downsides of PCM - or creating posts on how it "works" -
can we just ask that you read the following book and that you spend at least a little bit of time trying to understand what you are reading??

Amazon.com: Principles of Digital Audio: Ken Pohlmann: Books

Thanks very much in advance!!!!

QFT.

[Ethan Winer]

Originally Posted by Waltz Mastering

There are some ppl who repeatably try to convince others that the lowest common denominator is all that's needed and anything more is pure bullocks.

Well, if a higher resolution than needed for transparency were free I'd have no objection. But hard drive space is not free, even though it is pretty darn cheap these days. FTP bandwidth is certainly not free, nor is the time to send all the individual tracks for an entire CD at 24/96 trivial.

I'd never record music using lossy compression, or anything else that really does compromise the fidelity. But at some point the bit depth and sample rate are high enough that nobody can tell. That's where I draw the line. But I'm always up for discussions of where the line is, especially when they're backed by examples and proof and null tests etc.

--Ethan

[Brad Blackwood]

Originally Posted by karyoky****

hi,

hmmm. q-tips for torpedoes? quantize for timeliness? queer foreign trampolines?


wow, this one is hard. i am going to have to set my secret decoder ring on 48 bit 300kHz.

Quoted for truth.