Redbook:Yet another new magic bullet arrives?

[redshifter]

thanks, dreamslacker. that's just the info i was looking for. one more question, i am shopping for a dvd rw drive. any suggestions for one that burns cd's well? right now i have a sony cd rw drive.

 

[Wodgy]

Quote:

Originally Posted by The_Mac In theory, 16bits and 44.1 khz is enough to reproduce EVERY CONCIEVABLE WAVEFORM up to 20khz and with a dynamic range of about 96dB. If you are concerned with music that has less than a 96dB dynamic range (which is most music) and has no information over 20khz, then redbook records all the possible information, and there's no possible theoretical gain from going to any other medium, vinyl, SACD, DVD-A.

Meh. This is not true at all. You need to take more advanced courses in signal processing (and your Intro to Signal Processing prof was lacking).

What you're saying is true only for static, periodic waveforms. If only music had no changing time-domain component, things would be grand, but it doesn't work that way. Fourier Analysis 101 is far from sufficient in being a perfect representation of real signals, as any study of transient analysis will show. (You're also neglecting the need for dithering to compensate for time-domain quantization, which drops the reproduceable dynamic range, but that's a side issue.)

Glassman's points are also very valid, though they focus on the engineering angle, rather than the theoretical angle.

 

[KZEE]

Quote:

Originally Posted by Megaptera So let me see if I have this straight: The information is on the CD. The CDP can't get all the information because of the nature of how it's physically encoded on the disc. Rip it and recopy it to a CDR and your CDP can get at the information. I guess my question is: if the CDP can't get that information off, why can the CD-ROM? Sounds a bit like wishful thinking to me.

Along with around 300 commercial CD's, I also have about 100 CDR's, and on my reasonably resolving playback system - Musical Fidelity X-Ray v3, McCormack Micro Drive amp, home-rolled Dynaudio loudspeakers, Senn 580's, and a recently added MPX3 - I've never noticed a difference between the two types of CD's.

 

[WmAx]

Quote:

Originally Posted by Wodgy What you're saying is true only for static, periodic waveforms. If only music had no changing time-domain component, things would be grand, but it doesn't work that way.

44.1khz has a higher precision time domain than the rate implies. The temporal precision is a [1]product of the bandwidth and signal to noise ratio, not simply the bandwidth. As a result, 44.1/16 PCM has the nesecarry precision to capture time domian components that are at the [2]detectable thresholds of humans even under special test signals designed to maximize sensativity.

Quote:

(You're also neglecting the need for dithering to compensate for time-domain quantization, which drops the reproduceable dynamic range, but that's a side issue.)

As you know, with proper perceptual dithering, the weighted result is a SNR that can at minimum meet the 96dB specification under ideal circumstances. That was the purpose of dithering afterall -- to overcome the quantization problem in the early years of digital audio. A SNR of [3]80dB is evidenced to be sufficient in demanding situations for music program using headphones(75dB range is deemed suitable for loudspeaker playback). How many commercial recordings even achieve an average -80dB noisefloor after mastering?

-Chris

Footnotes

[1] Perception of mid-frequency and high frequency intermodulation distortion in loudspeakers, and it's relation to hi-definition audio
David Griesinger
Lexicon Labarotory


[2] Signal-to-Noise Ratio Requirement for Digital Transmission Systems
Spikofski, Gerhard
AES Preprint: 2196

[3] Binaural Time Discrimination
Jan O. Nordmark
J. Acoust. Soc. Am., Vol. 60, No. 4, October 1976, pages 870-879

 

[Wodgy]

Quote:

Originally Posted by WmAx 44.1khz has a higher precision time domain than the rate implies. The temporal precision is a [1]product of the bandwidth and signal to noise ratio, not simply the bandwidth.

Yes, but The_Mac was claiming that 44.1kHz was mathematically perfect (= infinite precision), which is not true. His argument was a common but erroneous one that introductory students often make. I've heard it many times before.

Quote:

As a result, 44.1/16 PCM has the nesecarry precision to capture time domian components that are at the [2]detectable thresholds of humans even under special test signals designed to maximize sensativity.

As you probably know there are peer-reviewed papers arguing one side or another of this issue or closely related issues at nearly every AES conference. I'm not really interested in arguing it here, as there is no possible ultimate resolution until a broad scientific consensus emerges.

From an engineering rather than theory standpoint, I tend to dislike "brittle" designs (where "brittle" means "probably adequate if our current scientific understanding does not change but still a little too close for comfort if everything goes right"), of which 44.1kHz/16 bit is one such design. Civil engineers design bridges to handle several times their rated load so that there is reasonable room for error in any one component. High-resolution PCM is valuable in my mind if only for analogous reasons -- you can get closer to perfection in the 20Hz-20kHz range while being relatively sloppy at every stage in the DAC, particularly the digital filter.

Quote:

How many commercial recordings even achieve an average -80dB noisefloor after mastering?

Too few, regrettably.

 

[JaZZ]

Quote:

Originally Posted by WmAx 44.1khz has a higher precision time domain than the rate implies. The temporal precision is a [1]product of the bandwidth and signal to noise ratio, not simply the bandwidth. As a result, 44.1/16 PCM has the nesecarry precision to capture time domain components that are at the [2]detectable thresholds of humans even under special test signals designed to maximize sensitivity.

Temporal precision is highly affected by the CD format's usual steep low-pass filter.

The graph shows the signal curves resulting from pure sine waves sampled at 44.1 kHz before low-pass filtering (in a simplified representation). A time-optimized filter (as favored e.g. from Wadia) will leave the format-inherent amplitude modulation intact, measuring as a ~3.5 dB drop-off at 20 kHz. A filter optimized for linear frequency response up to 20 kHz (with sharp roll-off) will eliminate the amplitude modulation by introducing a filter resonance smoothing all transients near the corner frequency, with the AM representing a transient.

In the case where something like the displayed amplitude modulation is the actual signal to be reproduced, common frequency-optimized filters will not allow this, instead they'll make a continuous sine wave out of it. In the case where a continuous sine wave ist wanted, the time-optimized filter will fail.

In any case the inaccuracy provided by the CD format is enough to be above the hearing threshold.

 

[boodi]

interesting graphs Jazz, thanks for posting

 

[WmAx]

Quote:

Originally Posted by JaZZ The graph shows the signal curves resulting from pure sine waves sampled at 44.1 kHz before low-pass filtering (in a simplified representation).

Of course, it is not proper to sample data without an anti-alias filter; this will result in spurious aliasing errors.This graph not exactly represenative of actual A-D devices or the net A-D-A process. With the DVD-A image at the right bottom, it makes me think the source of this graph(i did not follow the link) may be biased.

Here is an informative paper that is based on more realistic circumstances on this issue:

http://www.lavryengineering.com/docu...ing_Theory.pdf


Quote:

In the case where something like the displayed frequency modulation is the actual signal to be reproduced, common frequency-optimized filters will not allow this, instead they'll make a continuous sine wave out of it. In the case where a continuous sine wave ist wanted, the time-optimized filter will fail.

Since the only data of demonstrated interest are sine waves up to approx. 20khz frequency for human hearing; it's not known to be important to be able to reliably sample the spectral content of a transient that exceeds the human audible range.


-Chris

 

[WmAx]

Quote:

Originally Posted by Wodgy Yes, but The_Mac was claiming that 44.1kHz was mathematically perfect (= infinite precision), which is not true. His argument was a common but erroneous one that introductory students often make. I've heard it many times before.

I understand; I was pointing out that the commonly assumed termporal resolution is higher then many falsely assume. I did not intend to reply or address anything that Mac had stated.

Quote:

From an engineering rather than theory standpoint, I tend to dislike "brittle" designs (where "brittle" means "probably adequate if our current scientific understanding does not change but still a little too close for comfort if everything goes right"), of which 44.1kHz/16 bit is one such design.

I understand and agree. But it's the 'too close for comfort' part that makes the issue so fun to discuss.

-Chris

 

[JaZZ]

Quote:

Originally Posted by WmAx Of course, it is not proper to sample data without an anti-alias filter; this will result in spurious aliasing errors. So this graph not exactly represenative of actual A-D devices. With the DVD-A image at the right bottom, it makes me think the source of this graph (i did not follow the link) may be biased.

There's no aliasing in play since the concerned frequencies are below the Nyquist frequency. Showing the curves resulting from sampling without low-pass filtering is the clue of the whole thing. Any possible bias of the creator of the graph has nothing to do with the fact that the curves without filtering look like this (more precisely: the amplitude values).


Quote:

Since the only data of demonstrated interest are sine waves up to a 20 khz frequency for human hearing; it's not known to be important to be able to reliably sample transients that have an equivalent higher frequency content.

We're not talking of transients with fourier contents above 20 kHz. We're talking of errors such as amplitude modulation -- or the smoothing of comparable signal shapes, resp. -- in the audible frequency range.

 

[WmAx]

Quote:

Originally Posted by JaZZ There's no aliasing in play since the concerned frequencies are below the Nyquist frequency. Showing the curves resulting from sampling without low-pass filtering is the clue of the whole thing. Any possible bias of the creator of the graph has nothing to do with the fact that the curves without filtering look like this (more precisely: the amplitude values).

The graph is misleading. The waveforms demonstrated are not represenative of what is output from an actual standard A-D-A process used for audio recording that sampled the frequencies specified. The graph is a theoretical one that does not account or specify a realistic situation.

-Chris

 

[Dreamslacker]

Quote:

Originally Posted by redshifter thanks, dreamslacker. that's just the info i was looking for. one more question, i am shopping for a dvd rw drive. any suggestions for one that burns cd's well? right now i have a sony cd rw drive.

I have no idea. I'm not in the DVD recording bandwagon yet. I should guess from user feedback locally that the TDK drives are pretty good.

 

[abrichr]

Quote:

Originally Posted by gaboo The error rate for Redbook was deemed acceptable more than 20 years ago when the standard was drafted. If we were to believe certain legends, their goal was to get the 74mins required to fit Beethoven's 9th symphony on one disc.

This is absolutely fascinating! I read what was to be read at the link at the top of this thread, very interesting. However, I know virtually nothing about all of this. I did a little bit of research on what Red Book is, but that's about it. Anyone know any good links or books on the topic of whether bits are/aren't bits?

 

[eyeteeth]

Quote:

Originally Posted by abrichr This is absolutely fascinating! I read what was to be read at the link at the top of this thread, very interesting. However, I know virtually nothing about all of this. I did a little bit of research on what Red Book is, but that's about it. Anyone know any good links or books on the topic of whether bits are/aren't bits?

I'm like you just figuring things out. I just read a basic look at the issue at 6moons.

http://www.6moons.com/industryfeatures/eac/eac.html

 

[Glassman]

Quote:

Originally Posted by WmAx The graph is misleading. The waveforms demonstrated are not represenative of what is output from an actual standard A-D-A process used for audio recording that sampled the frequencies specified. The graph is a theoretical one that does not account or specify a realistic situation. -Chris

that's right, A/D conversion is done with heavily oversampled delta sigma converters, there's virtualy no Nyquist limitation, the single bit signal is then low pass filtered and decimated to high resolution PCM, edited in such format and finaly low pass filtered and decimated to 44.1kHz.. so how would graphs of these frequencies look like in the end of real world recording process?