Time resolution of Red Book <=45ns

doh, sorry again Im going too quickly for myself, but as you stated it, the test is not valid because it does not compare the time resolution of what is stored against the origonal, it compares the time resolution of what is stored against the time resolution of the origonal pre-simpified to the same standard as what is stored.
To be fair, B has to be compared to the origonal not 'A' ....

You want to tell people the formats reproduction capabilities are near perfect and to do so the demonstration is the formats capability to repoduce *itself* -it is a trick of technical sophistry.

Why dont you realise that? Do you have the same standards for other equipment?, could you be sold on a 'high spacial resolution' television with only a 10x10 pixel screen on the knowledge the extremely blurry image can be shifted by any arbitrary measure in any direction (using near identical math actualy)?

This is crazy.

 

ahem, excuse me again for communicating my frustration here - terribly undignified of me.

 

Well, you're comparing apples to oranges, but in a sense, it can. That's the point. Record a video of a black square moving across a white background. The edge pixels won't simply just change from black to white, they will transition from white to shades of gray to black. A 10 pixel wide screen will have more than 10 distinct horizontal locations.

 

Well, there's a slightly more sinister explanation for that wording than you may think. I wrote that wiki page

I think you have a point. That answer could be improved a bit. I think the best way to explain this, though, is to point out that the answer is going to be quite different for any particular question one would want to ask about a signal. And very few of those questions have answers that involve precisely a single sample of accuracy.

This can be explained just as easily in the time domain. At 44khz, a 19khz tone does not look at all like the original analog signal. When it's reconstructed, though, it looks pretty much exactly like the original. That's because the reconstruction filter looks at lots of nearby signals to construct the output.

It all depends on the system in question, and what you're actually trying to measure. This system dependence was one of my original points in this thread. If you have a reconstruction filter accurate enough, you can accurately reconstruct a signal with a phenomenally small delay compared to an original signal. You're obviously never going to get that kind of precision from your run of the mill 256-tap (for instance) filter on most DACs. The format itself, analyzed in isolation, can do things that many people do not realize are possible. As well, the usual implementations are also generally better than many people realize.

Terminology is another basic issue here. "Time resolution" is just a crappy term that can be equivocated to mean a wide range of different concepts, since "resolution" is not a fundamental concept or definition of signal processing theory. It's whatever people want it to mean. It can mean the window size of an FFT, or the sampling rate, "how well" it matches the original signal, or (googling up different meanings) the statistical variance of a time measurement in scientific timing experiements, etc.

I abused this ambiguation when making this thread, but I don't see that as necessarily incorrect or wrong. In terms of what people think "time resolution" can mean, I disproved a point. I'll admit it could mean a bunch of other unrelated things.

As you've seen, it's surprisingly easy to shoot down (or support) many assertions with the right measurements. And we have progressed much, much farther in this debate by grounding it in terms of numeric results instead of waxing philosophical about abstract concepts that are mutually unintelligible.

In short, I don't think "time resolution" should ever be used in a technical context, and I used it in this thread to be deliberately provocative. It is far more productive to use terms with precise textbook definitions, or to abandon small phrases entirely and talk in terms of hypotheses and simulations.

On the contrary, if you intend to publish your results or release implementations at some point, you'll probably want to have to have answers that make sense to the engineering populace.

 

Its an almost identical comparison, I made it because I know it can 'in a sense' -the exact same sense that details can be shifted across a waveform very precisely. Any detail that can be represented on that screen in the terms of the standard, but now be aware that everything on that screen is fuzzy, there can be no sharp changes between one point in space and the other, because it must be interprated as a bandlimited image to maintain the arbitrary shifting ability. So where is the spacial resolution? When everything HAS to be fuzzy?
The spacial resultion, is the distance between the center of one pixel and its neighbour, not the theoretical shifting ability of the bandlimited image.
Of course you can always redefine the language employed, saying 'a shift' is a distance therefore we have 'that' resolution in space, but then you are just trying to kid yourself and for the sake of maintaining your original position.

Precise spatial resolution of a fuzzy blob, anyone?

 

Oh hiya Ax*

Anyway, the more in the dark the engineering populace are about my own developed methods, the better for me actualy. Thats a whole other subject I cant go into in public though.

You can talk your way around the criticisms presented, but you would do better just to admit that stating 'time resolution' of fuzzy moments is needlessly artificial and misleading.
So you conclude the term 'time resolution' shouldnt really be used, when it seems the perfect casual term to use for the humble 'sampling period', rather than correcting yourself. Well thats your perogative in a free world, and Id fight to defend it

 

Again.. it all depends on what you're measuring.

I wouldn't expect the peak to move 1 sample. I would expect it to move much further than one sample - perhaps as much as 5-10 samples. Does that mean that 44.1khz has only 0.2ms of accuracy?

 

Sigh.

 

ps. what is the resolution of peoples monitors set to?

 

Oh, actualy you guys are right, Ive got a monitor here with a spacial resolution of at least 3.672 micrometers. Anyone interested?

 

This is wrong, actually, since the "upsampling" inherent in a CRT is not exactly upsampling in the usual sense. There's a lowpass operation in effect that causes the phosphors to change gradually and blur, and also causes ringing/overshoot on high contrast images. But it's a purely analog operation and doesn't take into account the neighboring pixels, like a FIR filter would. The lowpass filter simply is not powerful enough to do that. If you have a black square moving across the screen, you're still going to get blockiness in the movement for that reason.

And of course, if you have a 10x10 LCD, all bets are off.

 

Wrong again eh? Who mention CRT? No flatscreen digital TVs in your neck of the woods? Gimme a break man.

 

My point was simply that what you see isn't limited to 1 pixel increments...

 

oh it was lukpak who was wrong!

Lukpak sort it out! Think about what your posting before you post it will you

 

Ah, yeah, you're right. I didn't even quote the snippet I was trying to correct

 

Anyway, my ultra high spacial definition monitor/display driver combo is properly interpolated courtesy of an ATI Radeon card with a plethora of quality levels to choose from. Top quality levels allow spacial manipulation of the image smaller than a bacteriums peephole. Any offers?

 

Who was wrong?

 

This is a very good theoretical discussion from some learned folk. However, it is also valuable to think about the time distortion from an implementation standpoint and how it affects sound.

As I have written before, with respect to jitter often there are two good ways to go for the best music capture:

1. Use masterclocks to lower jitter.
2. Use very recent ADC and DACs that have inherently low time distortion.
3. Use your ears and compare to the live performance for accuracy. In my experience listening to live acoustic music with transients and high frequencies is where some of the gear is best differentiated.

 

well im certainly not taking the shame this time

 

Very good, however I dont believe you need to worry about jitter which is less than your desired sampling interval, since the effects of this jitter will not survive the implicit lowpass.

 

...but I expect that will take some time to convince...

Another rash statement, the gap between my ears is suggesting subsample jitter could manifest as a small amount of noise distributed most in the highest frequencies and least in the lowest.

An experiment would be in order.
Ill take a break.

cheers'
fe