R/E/P Community

Please login or register.

Login with username, password and session length
Advanced search  

Pages: 1 ... 19 20 [21] 22 23   Go Down

Author Topic: George....what's the resolution of analog?  (Read 75180 times)

Jon Hodgson

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 1854
Re: George....what's the resolution of analog?
« Reply #300 on: April 02, 2009, 11:28:44 am »

Jay Kadis wrote on Thu, 02 April 2009 15:55

Jon Hodgson wrote on Wed, 01 April 2009 17:30

Actually analogue systems can have quantization errors too. When I met Tim De Paravicini (a man who certainly seems to know his cutting lathes) I was surprised to discover that the error caused by the fact that you can either have a vinyl molecule or not (so the resolution of a cut is +/- half a molecule) was only 90dB below full signal, I'd always assumed that the molecules were too small for that to be a consideration... it seems not.

The fact that you random variations in the matrix results in those errors being random and thus white noise.

It amused me somewhat to discover that vinyl is actually a dithered system with a quantization step size approximately equivalent to a 15 bit system.
I'd like to hear more about the molecular size issue:  the 90 dB figure seems pretty high (or low?)    (The S/N of vinyl in practice falls significantly below 90 dB though.)  As I understand it "vinyl" is actually a vinyl chloride-vinyl acetate copolymer.  Wouldn't such a polymer would consist of particles of varying dimensions?

We have a similar issue in analog tape due to the finite dimensions of the individual magnetic domains.



I haven't looked into it in detail myself to confirm what he said, but I think anyone who's read Tim's bio would probably agree he's someone who knows about vinyl cutting, and anyone who's met him would agree that he can spout numbers like nobody's business... so I have good reason to find it quite likely that he's correct, especially since the figure he's apparantly calculated seems consistent with the sorts of numbers I'd already heard of for SNR on vinyl (he was quoting a very best case, the typical value would be less because you're not using the maximum signal level all the time and because of manufacturing issues and damage post manufacture), but obviously it shouldn't be taken as gospel without some further verification.

As for particles being of varying dimensions, this might be the case, I don't know, but if it is probably reasonable to use the largest size in order to make your calculation of SNR.

I had also thought about the analogue tape issue you mention, but have no figures on that, so I don't know if it is significant in real terms (just as quantization error isn't really significant in a 24 bit A/D conversion)


Logged

Andrew Hamilton

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 573
Re: George....what's the resolution of analog?
« Reply #301 on: June 29, 2009, 10:11:38 pm »

Jon Hodgson wrote on Wed, 01 April 2009 20:30

Actually analogue systems can have quantization errors too.


Errors of truncation, when the lacquer molecules refuse to be cut in two.  Not a quantum "decision," though, since the molecule _can_ be cut in two.   Just a dance between inertia and momentum.  No sampling means no quantization, though, doesn't it?

Jon Hodgson wrote

When I met Tim De Paravicini (a man who certainly seems to know his cutting lathes) I was surprised to discover that the error caused by the fact that you can either have a vinyl molecule or not...


Aren't we talking about cellulose nitrate at the mastering lathe?  Vinyl (or pvc with black pigmentation) would be a pressing concern, no?

Jon Hodgson wrote

The fact that you [have] random variations in the matrix results in those errors being random and thus white noise.

It amused me somewhat to discover that vinyl is actually a dithered system with a quantization step size approximately equivalent to a 15 bit system.


So, the dynamic range is equivalent to that of a 14+ bit LPCM carrier?  No actual steps in analog, though, right?




Andrew
Logged
www.serifsound.com
premastering for CD and DVD-A.  Featuring FTP load in and delivery as well as analog tape transfers.

Jon Hodgson

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 1854
Re: George....what's the resolution of analog?
« Reply #302 on: June 30, 2009, 01:29:00 pm »

Andrew Hamilton wrote on Tue, 30 June 2009 03:11

Jon Hodgson wrote on Wed, 01 April 2009 20:30

Actually analogue systems can have quantization errors too.


Errors of truncation, when the lacquer molecules refuse to be cut in two.  Not a quantum "decision," though, since the molecule _can_ be cut in two.   Just a dance between inertia and momentum.  No sampling means no quantization, though, doesn't it?

Jon Hodgson wrote

When I met Tim De Paravicini (a man who certainly seems to know his cutting lathes) I was surprised to discover that the error caused by the fact that you can either have a vinyl molecule or not...


Aren't we talking about cellulose nitrate at the mastering lathe?  Vinyl (or pvc with black pigmentation) would be a pressing concern, no?

Jon Hodgson wrote

The fact that you [have] random variations in the matrix results in those errors being random and thus white noise.

It amused me somewhat to discover that vinyl is actually a dithered system with a quantization step size approximately equivalent to a 15 bit system.


So, the dynamic range is equivalent to that of a 14+ bit LPCM carrier?  No actual steps in analog, though, right?




Andrew


Ever tried to cut a molecule? I don't think you're going to do it with a chisel (which is effectively what a cutting lathe is).

Certainly Tim was of the opinion that you had a molecule or you didn't, which gives you an instantaneous error of +/- 1/2 a molecule (which is according to him, 90dB below the maximum deviation you would let the needle make).

This is actually a quantization, but you have a huge number of them per second, and they are all slightly off from each other, so can be considered random (I don't know about the distribution of the error, we might guess gaussian).

So there are steps in vinyl if you look at it closely enough, but we can view it as having no steps and some noise.

Which funnily enough is how dither works.
Logged

Andrew Hamilton

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 573
Re: George....what's the resolution of analog?
« Reply #303 on: July 02, 2009, 03:08:19 am »

Jon Hodgson wrote on Tue, 30 June 2009 13:29


Ever tried to cut a molecule? I don't think you're going to do it with a chisel (which is effectively what a cutting lathe is).


The molecules in the lacquer may refuse to be cut by a mere heated wiggling blade...

...more nitro-cellulose? (;

If I add uncorrelated dither to an analog signal, that would make the signal have slightly more hiss, but I hesitate to call the result quantized audio, since the available amplitude modulation values are still continuous in scope.  Any error, as you have said, is heard as part of the noise floor, although at 90 dB into vinyl, I suspect that other surface noises are predominant.

Dither is supposed to be the remedy for the audibility of quantization error.  However, in your analogy, the molecular auto-dither is what's causing the "error."  

The lacquer molecule apparently randomly shunts the signal to heat(?) at the level of the randomly undulating noise floor of the cutting, in a way which emulates redithered digital audio.   I am not trying to keep comparisons  between analog and digital from happening, but, to me, the quantization term is a bit (; generous in this analogy (;.  

On a(n lp) master, for example, it's physically possible for the output signal to be of any arbitrary value within the dynamic range allowed by the lacquer, at any moment in time, in spite of the fact that the noise floor that is added to it is random, by +/- 1/2 lacquer molecule, when examined at its deep floor level, because the first term in the expression (analog audio) is continuously variable, unlike in quantized audio, so even if the second term in the expression (molecule/no molecule) is a constant, the result is still continuously variable.  

Whereas, in  quantized audio systems, such as LPCM, or DSD, the peak sample level of the digitized signal can only be up or down by one or more of the same, finite  amount(s) from  where it was at any given time prior.  By the time it no longer is digital, it's no longer quantized.  At least that's my layman's usage of the terminology.  





Respectfully,
    Andrew

Logged
www.serifsound.com
premastering for CD and DVD-A.  Featuring FTP load in and delivery as well as analog tape transfers.

Jon Hodgson

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 1854
Re: George....what's the resolution of analog?
« Reply #304 on: July 02, 2009, 06:26:04 am »

Andrew Hamilton wrote on Thu, 02 July 2009 08:08

Jon Hodgson wrote on Tue, 30 June 2009 13:29


Ever tried to cut a molecule? I don't think you're going to do it with a chisel (which is effectively what a cutting lathe is).


The molecules in the lacquer may refuse to be cut by a mere heated wiggling blade...

...more nitro-cellulose? (;

If I add uncorrelated dither to an analog signal, that would make the signal have slightly more hiss, but I hesitate to call the result quantized audio, since the available amplitude modulation values are still continuous in scope.  Any error, as you have said, is heard as part of the noise floor, although at 90 dB into vinyl, I suspect that other surface noises are predominant.

Dither is supposed to be the remedy for the audibility of quantization error.  However, in your analogy, the molecular auto-dither is what's causing the "error."


You're misunderstanding me a little.

The one molecule limit to accuracy is the quantization, if you had a perfect matrix of molecules and the time axis of your groove was along one axis of the matrix, your output would have steps one molecule high.

However the matrix isn't perfect, and the groove wouldn't be travelling straight across it if it was, so the quantization levels vary, in a random way, the dither is the variation in that quantization point. Normally we use dither by having a fixed quantization point and adding noise to our input signal, but mathematically it would work exactly the same if we left the input signal clean and randomly varied the quantization level.


Logged

Jay Kadis

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 2165
Re: George....what's the resolution of analog?
« Reply #305 on: July 02, 2009, 10:38:10 am »

The dimensions of a vinyl chloride molecule are on the nanometer scale.  Polymerized vinyl chloride chains may consist of hundreds to thousands of monomer units and complex side-chains may form producing a very complex molecule.  How the surface of a groove may interact with a stylus depends on the particular composition of the polyvinyl chloride and would be quite complex to analyze, requiring scanning electron microscopy to image the surface.

Given the dimensions of a stylus (microns), the molecular size of the particles themselves (nanometers) would not matter as much as the way the molecules pack together as there's a thousand-fold difference in size between the stylus and the surface molecular arrangement.  I doubt the stylus could follow individual molecules and would rather respond to the macro-level surface that would depend on the aggregation of vinyl molecules.

Jon Hodgson

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 1854
Re: George....what's the resolution of analog?
« Reply #306 on: July 02, 2009, 09:44:32 pm »

Jay Kadis wrote on Thu, 02 July 2009 15:38

The dimensions of a vinyl chloride molecule are on the nanometer scale.  Polymerized vinyl chloride chains may consist of hundreds to thousands of monomer units and complex side-chains may form producing a very complex molecule.  How the surface of a groove may interact with a stylus depends on the particular composition of the polyvinyl chloride and would be quite complex to analyze, requiring scanning electron microscopy to image the surface.

Given the dimensions of a stylus (microns), the molecular size of the particles themselves (nanometers) would not matter as much as the way the molecules pack together as there's a thousand-fold difference in size between the stylus and the surface molecular arrangement.  I doubt the stylus could follow individual molecules and would rather respond to the macro-level surface that would depend on the aggregation of vinyl molecules.


As I said, I haven't checked the numbers I was given myself, I was surprised by them but then the guy who gave them to me knows his cutting lathes, vinyl production and turntables and is full of facts and numbers... perhaps when he said molecule it was a misnomer for a larger particle?

In fact, if I'm being honest I couldn't swear in court if he said "molecule" or "particle", it was a very long day (and a very interesting one). The rest of his theory I'm pretty certain on.


Logged

johnR

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 923
Re: George....what's the resolution of analog?
« Reply #307 on: July 03, 2009, 07:04:56 am »

Jay Kadis wrote on Thu, 02 July 2009 15:38


Given the dimensions of a stylus (microns), the molecular size of the particles themselves (nanometers) would not matter as much as the way the molecules pack together as there's a thousand-fold difference in size between the stylus and the surface molecular arrangement.  I doubt the stylus could follow individual molecules and would rather respond to the macro-level surface that would depend on the aggregation of vinyl molecules.

Excellent points, but bear in mind that a thousand-fold difference in linear movement is only 60dB.
Logged

Jay Kadis

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 2165
Re: George....what's the resolution of analog?
« Reply #308 on: July 03, 2009, 04:11:05 pm »

johnR wrote on Fri, 03 July 2009 04:04

Jay Kadis wrote on Thu, 02 July 2009 15:38


Given the dimensions of a stylus (microns), the molecular size of the particles themselves (nanometers) would not matter as much as the way the molecules pack together as there's a thousand-fold difference in size between the stylus and the surface molecular arrangement.  I doubt the stylus could follow individual molecules and would rather respond to the macro-level surface that would depend on the aggregation of vinyl molecules.

Excellent points, but bear in mind that a thousand-fold difference in linear movement is only 60dB.
Which is close to the dynamic range of the phonograph.

KAyo

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 345
Re: George....what's the resolution of analog?
« Reply #309 on: November 13, 2009, 02:35:50 am »

Bob Olhsson wrote on Wed, 07 September 2005 20:19

I think Bob Katz treats this issue very well.

Since I can't rebuild most of the gear I have to use, I simply take the time to find the sweetest sounding gain structure for the system I'm using.


Brilliantly put!
That approach, is key to Analogue heaven. Dig deep to find the sweetest operations mode, and then fire all guns.

KAyo

Logged
http://www.Kantabiz.com
Business Video Directory

Dusk Bennett

  • Full Member
  • ***
  • Offline Offline
  • Posts: 108
Re: George....what's the resolution of analog?
« Reply #310 on: November 18, 2009, 05:01:56 pm »

sfdennis wrote on Fri, 14 May 2004 13:35

I'm surprised the thread got this far without anyone mentioning this:

Digital Word Length <=> Analog Noise Floor
Digital Sample Rate <=> Analog Bandwidth

So for example, in an ideal converter system at 96kHz would correspond with an analog bandwidth of 0-48kHz. A digital word length of 24 bits would correspond with an analog noise floor of about -132dBFS, notwithstanding dither.

-Dennis



Dennis,
I want to hit this point harder (and I may repost this separately since this is what I came here to ask). I got involved in an academic discussion today with a student regarding proper digital recording technique. He claims his professor (a Pohlmann student who's quite knowledgeable in her own right) stated that to obtain the highest resolution possible it is a MUST to record your sample as high up against 0dBfs as possible. (Which is a really bad idea IMHO). It's as if he's suggesting amplitude and "Word Length" are mutually exclusive (and you imply they are).

I have argued with the students here that there is no reason to "over record" in digital because the medium is much more forgiving than analog, and analog hardware cannot handle such hot levels coming out of digital converters. So I recommend that they print near 0VU (-20dBfs). There are many benefits to this in my mind (which I'll get into later), aside from the obvious.

So what is it? Is it really necessary in digital to print all the way up to 0dBfs to maximize your word length? Are amplitude and word length in digital truly related in that way? What do the facts say in digital theory on this?

The problem I have with recording so hot going into the box (especially if you end up interfacing with the outside analog world) is how much gain reduction one must apply to actually get your plug-ins and inserts to actually avoid distortion. Recording a guitar at -2dbfs leaves you no head room to compress and add eq after the fact without altering the "word length" of your inserts by reducing it's input to compensate for a very hot output. At that point aren't you just trading one set of problems for another?

I still believe (based on alot of practical experience) that there is no reason to print so fricking hot to digital. As long as you fall within the range of a typical VU meter your headroom in digital will allow for more than enough gain processing after the fact and not destroy the quality of the signal.

How would you respond to this?
Logged
----
Dusk Bennett
Artist Development/Production
Audio Engineering
Los Angeles CA
www.duskbennett.com

Tomas Danko

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 4733
Re: George....what's the resolution of analog?
« Reply #311 on: November 20, 2009, 06:33:19 am »

Dusk Bennett wrote on Wed, 18 November 2009 22:01

sfdennis wrote on Fri, 14 May 2004 13:35

I'm surprised the thread got this far without anyone mentioning this:

Digital Word Length <=> Analog Noise Floor
Digital Sample Rate <=> Analog Bandwidth

So for example, in an ideal converter system at 96kHz would correspond with an analog bandwidth of 0-48kHz. A digital word length of 24 bits would correspond with an analog noise floor of about -132dBFS, notwithstanding dither.

-Dennis



Dennis,
I want to hit this point harder (and I may repost this separately since this is what I came here to ask). I got involved in an academic discussion today with a student regarding proper digital recording technique. He claims his professor (a Pohlmann student who's quite knowledgeable in her own right) stated that to obtain the highest resolution possible it is a MUST to record your sample as high up against 0dBfs as possible. (Which is a really bad idea IMHO). It's as if he's suggesting amplitude and "Word Length" are mutually exclusive (and you imply they are).

I have argued with the students here that there is no reason to "over record" in digital because the medium is much more forgiving than analog, and analog hardware cannot handle such hot levels coming out of digital converters. So I recommend that they print near 0VU (-20dBfs). There are many benefits to this in my mind (which I'll get into later), aside from the obvious.

So what is it? Is it really necessary in digital to print all the way up to 0dBfs to maximize your word length? Are amplitude and word length in digital truly related in that way? What do the facts say in digital theory on this?

The problem I have with recording so hot going into the box (especially if you end up interfacing with the outside analog world) is how much gain reduction one must apply to actually get your plug-ins and inserts to actually avoid distortion. Recording a guitar at -2dbfs leaves you no head room to compress and add eq after the fact without altering the "word length" of your inserts by reducing it's input to compensate for a very hot output. At that point aren't you just trading one set of problems for another?

I still believe (based on alot of practical experience) that there is no reason to print so fricking hot to digital. As long as you fall within the range of a typical VU meter your headroom in digital will allow for more than enough gain processing after the fact and not destroy the quality of the signal.

How would you respond to this?


In the theoretial world she would be correct.

But when looking at the implementations that we can use in reality, she would have to weigh the importance of getting "all the dynamic range" versus all the negative issues that arises when recording close to full scale. And there are many such negative issues, that will have a lot more negative impact on the audio compared to the initial concern.

A further study in the world of reality will also show that "full dynamic range" is somewhat of a can of worms considering the room noise floor, electronic noise floor of the signal chain including the A/D front end which could for instance indicate that only 80% or so of the theoretical "full dynamic range" is actually of practical interest.


It's not really part of the initial topic, but when also considering how much we futz with the dynamics (i.e. compression, limiting etc) to make an album of music it's even less important for the audio engineer to be able to capture the total full dynamic range.
Logged
http://www.danko.se/site-design/dankologo4s.gif
"T(Z)= (n1+n2*Z^-1+n2*Z^-2)/(1+d1*z^-1+d2*z^-2)" - Mr. Dan Lavry
"Shaw baa laa raaw, sidle' yaa doot in dee splaa" . Mr Shooby Taylor

tamasdragon

  • Full Member
  • ***
  • Offline Offline
  • Posts: 107
Re: George....what's the resolution of analog?
« Reply #312 on: December 12, 2009, 12:22:04 pm »

Although I know it's a bit offtopic, but I'm really interested in it.
This is a question for all (George M., Bob O., Bob K., Paul F., etc.), do you think that today's digital audio is every bit as good as today's analog? Sonically and in every aspect?

Geoff Emerick de Fake

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 348
Re: George....what's the resolution of analog?
« Reply #313 on: December 12, 2009, 03:01:05 pm »

Dusk Bennett wrote on Wed, 18 November 2009 16:01


So what is it? Is it really necessary in digital to print all the way up to 0dBfs to maximize your word length? Are amplitude and word length in digital truly related in that way? What do the facts say in digital theory on this?
It is a fact: 24 bit offer ca. 144dB dynamic range. Everytime you drop the level by 6dB, you loose 1 bit of resolution. Recording with a maximum INSTANTANEOUS level of -18dBfs uses only 21 of the available bits; the three MSB's will be constantly at 0.
Quote:


The problem I have with recording so hot going into the box (especially if you end up interfacing with the outside analog world) is how much gain reduction one must apply to actually get your plug-ins and inserts to actually avoid distortion. Recording a guitar at -2dbfs leaves you no head room to compress and add eq after the fact without altering the "word length" of your inserts by reducing it's input to compensate for a very hot output. At that point aren't you just trading one set of problems for another?
Different answers for outboard and for plug-ins. All reasonably designed plug-ins have built-in bit-depth (resolution) conversion that allows pre-attenuation, in an amount that will allow the most extreme processing to be handled nicely.
When using outboard, you have to consider the practicalities. If you use standard +4dBu equipment and your converter is set at +4dBu for -20dBfs, you should operate at -20dBfs, which leaves you with a real headroom of about 10db (depending on the type of meters), which is really one-and-a-half wasted bits. Different converters and different outboard gear woul give different numbers.
Quote:


I still believe (based on alot of practical experience) that there is no reason to print so fricking hot to digital.
"Hot" is an analog concept, tied to the progressive dirtying of signals when the level increases.  There is no such concept in the digital world. The only rule is to make sure that digital clipping is not reached, and that enough resolution is used for minimization of noise and quantizing distortion. Recording with 0.2 dB of headroom is not hot, it's just right. (I don't mean having a meter with the needle constantly at -0.2, I mean having the peak of the waveform reaching never higher than 0.2)
Quote:

 As long as you fall within the range of a typical VU meter your headroom in digital will allow for more than enough gain processing after the fact and not destroy the quality of the signal.

First of all, a VU meter is definitely not the right tool to monitor levels when tracking. The slow response leaves fast transients completely unnoticed; that's why you need to record conservatively low. Fortunately, the dynamic range of a 24bit system is so huge you can afford to loose 20dB (more than 3 bits) and get away with it. it doesn't mean it is good practice. The use of ultra-fast meters gives a much better control of recording level.
Logged

Tomas Danko

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 4733
Re: George....what's the resolution of analog?
« Reply #314 on: December 13, 2009, 09:31:02 am »

Geoff Emerick de Fake wrote on Sat, 12 December 2009 20:01

Dusk Bennett wrote on Wed, 18 November 2009 16:01


So what is it? Is it really necessary in digital to print all the way up to 0dBfs to maximize your word length? Are amplitude and word length in digital truly related in that way? What do the facts say in digital theory on this?
It is a fact: 24 bit offer ca. 144dB dynamic range. Everytime you drop the level by 6dB, you loose 1 bit of resolution. Recording with a maximum INSTANTANEOUS level of -18dBfs uses only 21 of the available bits; the three MSB's will be constantly at 0.



But when we step out of the world of theory and into the practical world of technical implementation and the actual digital tools that we have in order to record audio the 144 dB range does not really exist (i.e. several sources of noise). And when looking at the analog front end of most AD converters, they will start sounding bad a long time before reaching full scale.

We don't need more than 20-22 bits, and we don't need opamp distortion. Downstreams there is a lot said already about keeping levels far from full scale when mixing inside a DAW.

So all in all, advocating staying as close to full scale based on theory will only hurt actual recordings due to the way it's been practically implemented.
Logged
http://www.danko.se/site-design/dankologo4s.gif
"T(Z)= (n1+n2*Z^-1+n2*Z^-2)/(1+d1*z^-1+d2*z^-2)" - Mr. Dan Lavry
"Shaw baa laa raaw, sidle' yaa doot in dee splaa" . Mr Shooby Taylor
Pages: 1 ... 19 20 [21] 22 23   Go Up