why bother with 24 bit

[Farview]

If time slices is not the issue then what about the claims of 1-bit recording with mega high sample rates

http://www.korg.com/service/downloadinfo.asp?DID=1253

Or is it to do with the reconstruction of the sample back into analogue – or are manufactures playing on our (at least some of us) naivety? I understand the significance of reducing bit rate has a lot to do with improved quality

Snow lizard did a great job explaining this.

But I can not understand how sample rates are not relevant to the quality even at higher bit rates – surely having 20 pictures of a wave form is better than 10 when it comes to reproducing the analogue sound.

It isn't. You can't get more accurate than accurate. It's like being dead, you can't make someone more dead.

I’m less concerned about the number of pictures being taken and more interested in how the software deals with the gaps – particularly with any sort of distortion where a multitude of frequencies are present.

You are assuming that the analog sound has more detail to capture. Remember, even if the reconstrution worked the way you think it does (it doesn't), all of the distortion and noise from the 'gaps' would be happening at the sample frequency (44.1k in this instance). Which is too high to hear and gets filtered out anyway.

If there were any extra detail on a 20k wave, it would have to be at a (much) higher frequency. You wouldn't hear it in the first place, your mic probably wouldn't have picked it up, your monitors wouldn't reproduce it, and the filters would have filtered it out.

 

[SouthSIDE Glen]

But I can not understand how sample rates are not relevant to the quality even at higher bit rates – surely having 20 pictures of a wave form is better than 10 when it comes to reproducing the analogue sound. Basically with less pictures you are leaving bigger gaps for the software to interpret.

Everybody keeps coming back to the photo analogy.

The problem is, the sample rate of a converter is not analogous to the pixel density of a CCD. Though "common sense" wants to tell us that increasing the sample rate of digital audio is analogous to increasing the resolution of a digital image, that is in fact a completly false analogy.

Converter sample rate is instead analogous to the functional bandwidth of the CCD. An increase in audio sample rate increases the range of wavelengths that can be resolved; the higher the sample rate, the shorter the wavelength "resolution". Or to put it another way, the higher the frequency of the sampling, the higher the frequencies that can be resolved.

Going beyond a 44.1 kHz sample rate is directy analogous to going beyond violet into ultraviolet in the bandwidth of the CCD. Taking the sample rate to 192kHz is equally analogous to pushing the functional bandwidth of the CCD into the extreme ultraviolet; it ain't gonna do a damn thing to increase the visible light quality of the picture you took. It also has no effect whatsoever on the picture's actual resolution.

As far as higher sample rates more accurately filling in the gaps, this is another common misconception borne of a lack of understanding of how analog waves are actually reconstructed from digital data. While not a tutorial on that, the Lavery article that NL5 referenced a couple of different times in this thread does illustrate the process at work.

I really back up NL5's recommendation that everyone give that article a look, even if they do not completely understand it all (I get lost in some of the denser math parts myself). The important thing that one may get from it, though, is that reconstruction of an anlog wave from digital data is not a matter of connecting the dots between samples.

The fact remains that the Nyquest therom is rock solid in this regard: the amount of information obtained at a sample rate of just over twice that of the target frequency is enough information to exactly reproduce that frequency. The extra information obtained by sample rates over the Nyquest frequency does not make that reproduction of that target frequency any more accurate; i.e. it does not "fill in the gaps" any more accurately at that frequency. All the extra information carried by the increased sample rate does is allow the filters to reconstruct higher frequencies equally as accurate as the original target frequnecy.

It reminds me of quantum physics

Sample rate is certainly similar to quantum theory in one sure way: It takes an actual understanding of how the physics actually works to know that the classical pictures our brain conjures up (lke sample rate = resolution) are just wrong.

G.

 

[cusebassman]

It utterly amazes me that this topic is still going, especially since it isn't even about the original post anymore

 

[MCI2424]

Face it man, some people just DON'T get it. They are the kind of people that all the 192K hardware sellers are counting on. ....and there's alot of them.

Do any of you know who Dan Lavry is? Have you read his credentials? Did you read the fucking article he wrote?

If you discount the inherent shortcomings of the converters, 44.1 is COMPLETELY accurate in reproducing the AUDIO band. Once you have complete accuracy, there is nothing more. Sorry, but that is a fact. Now, if you read Dan Lavry's article, there are reasons for going above 44.1, but it has NOTHING to do with the ability to reproduce the wave with 44.1khz sampling - it is to make up for the converters. According to Mr. Lavry, everything after about 60khz is a waste, and 192 is a joke, and can actually be WORSE.

Read the fucking article - PLEASE.

edit - and go listen to Joshua Judges Ruth by Lyle Lovett - recorded on crappy 16 bit 44.1khz converters - no tape, no 96khz, no 192.

There are 192K types of people and each one understands a bit.

Get it?

 

[SouthSIDE Glen]

It utterly amazes me that this topic is still going, especially since it isn't even about the original post anymore

It sure beats a thread arguing about stupid shit like whether one shoud mix from the perspective of the keyboardist or the roadie.

G.

 

[snow lizard]

It sure beats a thread arguing about stupid shit like whether one shoud mix from the perspective of the keyboardist or the roadie.

G.

Or the groupie for that matter.


sl

 

[TerraMortim]

Everybody keeps coming back to the photo analogy.

The problem is, the sample rate of a converter is not analogous to the pixel density of a CCD. Though "common sense" wants to tell us that increasing the sample rate of digital audio is analogous to increasing the resolution of a digital image, that is in fact a completly false analogy.

Converter sample rate is instead analogous to the functional bandwidth of the CCD. An increase in audio sample rate increases the range of wavelengths that can be resolved; the higher the sample rate, the shorter the wavelength "resolution". Or to put it another way, the higher the frequency of the sampling, the higher the frequencies that can be resolved.

Going beyond a 44.1 kHz sample rate is directy analogous to going beyond violet into ultraviolet in the bandwidth of the CCD. Taking the sample rate to 192kHz is equally analogous to pushing the functional bandwidth of the CCD into the extreme ultraviolet; it ain't gonna do a damn thing to increase the visible light quality of the picture you took. It also has no effect whatsoever on the picture's actual resolution.

As far as higher sample rates more accurately filling in the gaps, this is another common misconception borne of a lack of understanding of how analog waves are actually reconstructed from digital data. While not a tutorial on that, the Lavery article that NL5 referenced a couple of different times in this thread does illustrate the process at work.

I really back up NL5's recommendation that everyone give that article a look, even if they do not completely understand it all (I get lost in some of the denser math parts myself). The important thing that one may get from it, though, is that reconstruction of an anlog wave from digital data is not a matter of connecting the dots between samples.

The fact remains that the Nyquest therom is rock solid in this regard: the amount of information obtained at a sample rate of just over twice that of the target frequency is enough information to exactly reproduce that frequency. The extra information obtained by sample rates over the Nyquest frequency does not make that reproduction of that target frequency any more accurate; i.e. it does not "fill in the gaps" any more accurately at that frequency. All the extra information carried by the increased sample rate does is allow the filters to reconstruct higher frequencies equally as accurate as the original target frequnecy.

Sample rate is certainly similar to quantum theory in one sure way: It takes an actual understanding of how the physics actually works to know that the classical pictures our brain conjures up (lke sample rate = resolution) are just wrong.

G.

hmm.. That's not really correct. It IS the same as increasing the resolution, and at the same time increasing the niquist. The sampling rate is basically how many times a second the A/D chip is surveying the audio and making a representation of what the wave is doing at that time. Just like in an image, if you have a low pixel depth, the pixels will have to be more of guesswork on the computers part to fill in the blanks in order to recreate the image. It might look pretty close, but there will always be digital artifacts (hmmm, strange...similar to audio.) This is the same with sample rates. If you sample at 44.1kHz or having the A/D take 44,100 readings of the analog wave per second, that is the resolution. Just as if you have 71 pixels per inch (71 differently coloured dots within one inch of physical space to trick us into thinking we see what the CCD chip has captured), that is the resolution. If you increase that resolution, you will be given a closer representation of what the original energy, whether it be light energy or acoustic energy, was. So if you increase it to 96kHz, it's sampling this same wave 96,000 times per second. Just as in an image, when you increase the pixel count, the digitized versions of the former analog medium will be much closer to what it originally was. It's not related to frequency as in pitch as much when you get to that high of sampling rates (most speakers can't reproduce much higher than the niquist of 44.1 or 48kHz), as also we can only hear so high with our ears, and the niquist frequency for 44.1 is TECHNICALLY above what most people can hear. It's missing the point to assume that's the only thing happening here. It's completely true that it makes things mesh together, increases the resolution, JUST like a digital picture. All digital data is just a representation, 0s and 1s at the core of it, of something that's physical, whether it's light/film, or audio/tape. Simply, the more imformation that is given by the A/D or CCD the less guesswork the computer is doing to make complete waves, and therefor the increased success of recreating the waveform as perfectly as can be. The fact is totally correct, the closer together you sample something audio OR visual, the better resemblance it'll have to the original, the further away you sample, the more the computer has to do to fill in the blanks, which sometimes it does correctly, but sometimes it doesn't (it's entirely logical guesswork at that point). I don't remember the Niquist therum saying anything about perfect represention. From what I remember, It just states what the highest frequency recorded can be. If it does, then well...exact is subjective... It reminds me maybe of when computers had less than a meg of ram, and everyone thought, hey this is about as good as it could get, who the hell would need more, it's simply just overkill to go beyond this powerhouse! I mean, I can say if I take a picture of a tree with a camera, it looks exactly like it, but it doesn't, because the camera lacks the ability to gather the details, it resembles it, but there are major differences to actually seeing it. However, when we make the resolution higher, a higher pixel count, the closer it will be to looking like it actually looked. Audio is the same way, even all the way up to 192kHz, while close, is still not a perfect representation of the waveform. There will always be a certain degree of stepping in the wave and guesswork by the chips, sometimes right sometimes wrong, even though at that high of a sample rate (which for most purposes is probably beyond what is practical, with what gear we're using today) it becomes neglegable, and not really important at all in my opinion. Even if the audio is digital, it is converted back to analog to reach our ears, and in order to smooth out the wave so that it resembles the way sound waves are naturally, it has to guess how it thinks the wave should be, between all of the sample points. There really is an audible difference in samplerates, A/B it. AND NOT WITH GUITAR (guitar is by nature a really low-fi instrument, so you will almost never hear a difference with that). Try with some complex synths, and throw nice reverbs on it (make sure the plugins actually can operate at highe samplerates ...some still work at 44.1, or whatever ther maximum is..which is not as common as it used to be.. even if your project is set higher...) There will be a ton of detail in the synth sound that wasn't present before. Entire textures, and the reverb tail will tend not to wash out the sound as much, will be a bit richer, and much more interesting sounding. There will be a certain sheen in the sound that is missing from so many digital recordings. The difference IS there, and there are many people the world over that can hear that difference. There are some that can't, but there are also people that can't pick out compression settings, or certain colourations of different pieces of gear from hearing a recording as well. That doesn't mean that it's impossible to pick that stuff out, or that the differences don't exist. It all boils down to training your ears to hear the fine details within a recording. If you still have your hearing intact, and you train yourself to listen close enough, it's there. It's not always HUGE, but subleties are usually much more important than the obvious things right in front of your face when dealing with emotional mediums (music/visual art/etc...)

 

[TerraMortim]

It isn't. You can't get more accurate than accurate. It's like being dead, you can't make someone more dead.

true, but there doesn't exist anything analog OR digital that can completely record the sound perfectly accurately. There are however different degrees of accuracy.

 

[cusebassman]

The fact remains that the Nyquest therom is rock solid in this regard: the amount of information obtained at a sample rate of just over twice that of the target frequency is enough information to exactly reproduce that frequency. The extra information obtained by sample rates over the Nyquest frequency does not make that reproduction of that target frequency any more accurate; i.e. it does not "fill in the gaps" any more accurately at that frequency. All the extra information carried by the increased sample rate does is allow the filters to reconstruct higher frequencies equally as accurate as the original target frequnecy.

G.

But the fact remains that if you take slightly more than 2x the highest frequency in "pictures" of the wave of sound, you are accurately capturing every possible wavelength below that. So taking 4 pictures of a wave tha can already be reproduced with 2 pictures is pointless. Or so it would seem -if that were completely concrete, we wouldn't all be throwing text at each other.

I am inclined to agree with Glen on this one...

 

[Robert D]

You know, to paraphrase Tom "scientology" Cruz from risky business, "Sometimes you just gotta say fuckit and get on with making music".

 

[cusebassman]

You know, to paraphrase Tom "scientology" Cruz from risky business, "Sometimes you just gotta say fuckit and get on with making music".

Another excellent point. It doesn't necessarily solve the issue at hand, but in the end, both 16-bit and 24-bit can produce decent sounding recordings, and 44.1kHz sampling does just fine. If you can really hear the difference recording at 1024kHz, go right ahead. It'll get beat the Hell back down again anyway when it gets converted to mp3, burned to a CD, etc.

I think between the two, 16 vs 24 yields a noticeable difference moreso than recording 44.1/48/96 does... I can hear a diff. between recordings at 16 and 24 - but comparing the same thing at 48 and 96 kHz? Eh, not so much.

 

[Farview]

Just like in an image, if you have a low pixel depth, the pixels will have to be more of guesswork on the computers part to fill in the blanks in order to recreate the image.

NO! The pixels equate to the bit depth. The number of bits is the resolution, the sample rate only affects frequency response.

 

[Farview]

true, but there doesn't exist anything analog OR digital that can completely record the sound perfectly accurately. There are however different degrees of accuracy.

Now you are talking about the quality of the converter, not the conversion process. Two completely different things.

 

[Mindset]

NO! The pixels equate to the bit depth. The number of bits is the resolution, the sample rate only affects frequency response.

Sample rate also effects amplitude & time if I'm correct right?

 

[Farview]

Sample rate also effects amplitude & time if I'm correct right?

Amplitude is what the bits do. The more bits, the more dynamic range. (amplitude).

Anything that happens below the nyquist frequency will happen with enough samples to be replicated. The timing issue is affected by jitter more than samplerate.

 

[eraos]

Amplitude is what the bits do. The more bits, the more dynamic range. (amplitude).

Anything that happens below the nyquist frequency will happen with enough samples to be replicated. The timing issue is affected by jitter more than samplerate.

I completely understand that at 44.1kHz, all of the frequency information below 22.05kHz is captured. And I understand that "resolution" is due to bit depth.

But doesn't quantization occur at each sample and hold period? So, wouldn't the amplitude be referenced 96000 times a second at 96kHz or 48000 times a second at 48kHz? Or is this not true because of oversampling?

I've read through half of the Lavry article and this is the only I could find that might address my question :

For example, most front ends of modern AD (the modulator section) work at rates between 64 and 512 faster than a basic 44.1 or 48KHz system. This is 16 to 128 times faster than 192KHz. Such speedy operation yields only a few bits. Following such high speed low bits intermediary outcome is a process called decimation, slowing down the speed for more bits. There is a tradeoff between speed and accuracy. The localized converter circuit (few bits at MHz speeds) is followed by a decimation circuit, yielding the required bits at the final sample rate.

I don't really know what decimation is.

 

[Farview]

The quantization error happens at the sample frequency (ie. 44.1k per second) so it is higher than we can hear and it is filtered out. A higher sample rate would just change the number of times a second (frequency) that the error occurs. It's still higher than anyone can hear and it still gets filtered out.

 

[SouthSIDE Glen]

The fact is totally correct, the closer together you sample something audio OR visual, the better resemblance it'll have to the original

We're Getting The Bandwidth Back Together
This is true only when assuming an unlimited bandwidth. When there is a limit to the required bandwidth, then there is a limit to the required sample rate in order to 100% reproduce it accurately.

If we assume a 20kHz upper limit to the audible bandwidth (more on that number in a minute), then there is an upper limit to the sample rate required to 100% accurately reproduce it That limit is the Nyquist frequency. An 88kHz sample rate does not resolve to a more accurate representation of a 20kHz sound than a 4kHz samle rate does. The "blanks" in a 20khz sample rate are filled in equally at either sample rate; the resulting 20kHz wave is identical either way.

Do Not Connect The Dots, Do Not Collect $200
Once again, reconstrucing analog waveforms from digital is not done by connecting points between samples. There are no actual "blanks" to fill in at that point. There is no "drawing of lines" from time slice to time slice. Instead the waveform is built up via a summing of mathematical analog waveform functions. The wave is - in a fashon - "grown" from the ground up, with it's analog contours intact the entire way; the final shape determined by the number of component functions that are summed together. It's pure math and pure waveform theory at that point, and of a nature that has nothing to do with connecting points or filling in apparent gaps between sample rates.

What I think throws people off in that regard is the *visual* representation of time slicing itself. It deeply implies that the conversion of digital to analog is indeed a conversion of stairsteps into slopes. This implies the connection of dots or the filling in of gape between the stairsteps, and that a higher sample rate gives a finer stairstep resolution. This may be an easy way to explain it on a "D/A 101" teaching level, but it's actually a VERY erroneous description of the actual process that's taking place.

Why You Should Not Connect The Dots
For proof, look at it by trying to prove the connect-the-dots, sample rate as resolution idea. Everybody agrees that 44.1kHz gives at least a fairly accurate representation of a 20kHz signal right? One may (wrongly, but we'll ride along with that for a minute) believe that a higher sample rate may yield a slightly "finer" 20khz signal, but nonetheless, 44.1kHz is at least adequate, right?

Not according to the connect the dots/resolution view of sample rates. According to that viewpoint, there's no way that a 44.1kHz sample could even come close to accurately representing 20kHz. How could it? It's only taking samples (slightly more than) twice per entire wave cycle. It's almost infinitely probable that those samples will fall elsewhere on the wave than the points where the wave crests. If we were to just draw lines between those sampled points, we'd have a waveform of entirely incorrect amplitude - possibly even zero amplitude if the two sample points happened to fall on the zero crossing points of the full wave cycle. Not only that, we'd have a triangle wave instead of a sine. The resulting reconstruction would be so incorrect and so inadequate as to be totally useless.

In fact, by that theory, a 44.1kHz sample rate would be pretty awful even for resolving a 4khz signal. There'd only be about 10 sample per wave. That's only 5 samples per crest and trough. Assuming one sample each at each peak, that leaves only two samples on each slope of the wave with which to reconstruct the wave. That's an awfully low quality reconstruction that would be audibly distorted. It would be twice as bad at 8khz and 4 times as bad at 16khz. Let's face it; a 44.1kHz sample rate would be pretty useless in terms of fidelity reconstruction for the majority of the audio spectrum.

Yet that's not the case. 44.1kHz can make pretty darn good reconstructions all the way up to 20kHz. Nobody disputes that. So what's the problem? The problem is that the connect the dots idea of waveform reconstruction is simply a false idea. It's just not done that way. This also renders as equally invalid the idea of "resolution" meaning anything other than frequency response.

Y20k?
Now you mention that 20kHz may not be a fair upper limit. I addressed this in a earlier post. Even those that fall on the side of thinking that there is an elusive "air" to be obtained by resolving some frequencies above 20kHz or so - and that is an honest debate which not even the "experts" agree upon - don't believe that frequencies anything above 30khz are useful. That would mean a sample rate of 66.15kHz or so. The fact is, that even for those golden ears and audiophiliacs, 88.2kHz is overkill already by a good 22kHz. A 96kHz rate is just piling on, and 192kHz is ridiculously obscene.

So perhaps, PERHAPS, one might argue that the next steps up from 44.1kHz - namely 48kHz and 88.2kHz might have some worthiness in audiophile-class recordings. But that is certainly debatable - the experts debate it all the time - and certianly not solidly backed up by expirimental data from actual controlled listening tests. Test after test has yield, at best, conflicting or inconclusive results. It is also not backed up when looking at the limitations of the rest of the signal chain from microphone to loudspeaker, which simply are not designed to give a rat's ass about those higher frequencies.

G.

 

[Freddy]

Synchronicity - I see Dan Lavry has been posting (in this case about whether up-sampling is worth doing, but basically about this stuff ) on gearslutz .. http://www.gearslutz.com/board/high-end/118517-mix-96k-not-3.html#post1234224

 

[Farview]

BRAVO MR. GLENThank you for taking the time.