Home |
Search |
Today's Posts |
#81
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
|
#82
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
"Tommi" wrote in message
wrote in message ... Justin Ulysses Morse wrote: While it's true that the additional bits tack your extended resolution onto "the bottom" of the dynamic range, it clearly increases the resolution at all levels. You can have a -100dB component to a -1dB signal, and you still want to hear it. Is the ear even capable of hearing the -100 component against the much louder -1? I thought masking pervented this. Masking, it is frequency-dependent. However, this leads to thinking about the fact that the human ear actually compresses dynamics at higher sound pressures. My understanding is that we have roughly 80dB's worth of dynamic range at a time, which we then move according to the sound pressure levels of the sound sources. Probably less than 80 dB, more like 60 or 70 dB. The actual number depends on the kinds of sounds you use to establish the 0 dB point. There's a big difference in what happens if you set the 0 level with a sustained tone like that from a piano, or a percussive tone, like that from a castanet or triangle. For example, if you'd be listening something at 110dB SPL for 5 minutes, after that you couldn't hear the same sound with 2dB SPL for a while. Yes, at sustained levels of 110 dB most people would experience quite a bit of threshold shift. It works the other way round too: If you're listening something at 5dB spl for a while, and then suddenly the same sound source produces a 120dB spl sound, your ear would compress it lower(by stretching the eardrum, moving the hammer away from it etc) in order to protect your hearing mechanism. This, however isn't true with very short peaks because your protection mechanism takes some time to wake up. Agreed. |
#83
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
"Garthrr" wrote in message
... Thanks Chris, So is this point a matter of contention or is this agreed upon by all? If it is agreed upon then is the argument on the "24 bit sounds no better than 16 bit" side that the effects of the ambiguity are inherently negligable or perhaps that interpolation or something else "repairs" the ambiguity adequately? I hope I have framed thew question well enough to be understood. Garth~ 24 bit sounds no better or may sound better simply because of the analog front end of the converter and the quality of the converter itself in terms of thermal noise. But beyond that, 24 bits is not greater resolution, and I think that's what throws people off. It's greater bit depth, depth being the representative word meaning it does down to further quiet (ideally -144 dBFS, but thermal noise plays a part along with the front end implimentation). Ostensibly this only means that RECORDINGS might sound better because there's more available headroom above noise to fart around with, but great quality digital recordings have been made on 16 bit machines, so it's not inherent that 24 bits is better. It is inherent that one can be less involved with watching levels on a session because now we can have peaks that eek into -18 or so without worrying that one of the major hits is going to freak into the clipping range. Otherwise, if you calibrate your system where 0 dB VU = -20 dBFS or so then you still have the same headroom regardless of bit depth. The problem can lie in the fact that the noise floor is higher with 16 bit, but it's more likely that almost any room (project studio concept) is going to be far noisier than the theoretical lower limit of a 16 bit converter anyway, so where's the great boost from 24 bit? Again, it comes down to noise floor, and noise floor only. In the basic 16 bit world we just went through one would be lucky to get a noise floor of -90 dBFS, but the realworld range is more like -80 to -82 dBFS (cassette and vinyl were 10 to 25 dB noiser). In a 24 bit world, that figure drops maybe 20 dB or so, as I often see about -104 dBFS for decent converters, and something like -110 dBFS on really good converters. Not even close to the theoretical lower end that 24 bit converters have. But it is quieter, and this may make a difference. It also may not. It is, by no means, the telling tale on recording. It's simply a fact of the converters bit depth. Better ones are quieter than cheaper ones. It's even possible that really good 16 bit converters can sound better than really bad 24 bit ones. However, DSD has been described as someone standing at a light switch, turning it on and off. If they turn it on only once a second, then your ability to resolve the room becomes haphazard has you try to scan the entire contents of the room. The faster the room light is turned on and off, the more perception there is that it's on and you have the greater resolving power to see the contents. Before it was a chair here, a table there, etc. Now as it goes faster and faster you see, not only the chair, but the slipper protruding from underneath it that might trip you were you walking around in the dark. As it goes even faster you see that there's little bits of paper on the floor or dust or cat hair or cat hairball, etc. I realize that it's not a technically correct description of what scanning an input at 2.8 mHz is really like, but even in it's simple form it's easy to see that, relative to the content, the faster the light goes off and on, the more like being on it is. The more it seems like it's on, the more you can see, or in the case of DSD, the more you can hear. Whether you NEED to have all the content exposed is another question. If I don't intend to walk across the room and kill the spider crawling up the wall, do I really need the resolution to see it in the first place? Or do I just need to know the layout of the room? -- Roger W. Norman SirMusic Studio Purchase your copy of the Fifth of RAP CD set at www.recaudiopro.net. See how far $20 really goes. "I think the fact that music can come up a wire is a miracle." Ed Cherney |
#84
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
"Jay - atldigi" wrote in message
... In article , (Garthrr) wrote: Actually I was under the impresion that Jay was on the other side of the fence--that he was saying that 24 bit was really no better than 16 bit for any sort of real world audio. Absolutely not. I know there's a lot of posting going on, and I've written a lot in this thread, but I know I've stated several times that the above is not what I'm saying. Instead of adding even more confustion, please try to go back and read my posts again. Hopefully in retrospect they'll make more sense. Thanks, Well, you have to think in terms of the statement "real world audio", then it's possible that 24 isn't better than 16 if real world audio recording is accomplished in a room where quiet is represented by about -55 dB of room noise, which would be a lot of live situations (or worse) and in a lot of home/project studios where thousands haven't been spent in acoustic taming. At that point, I believe there's a point where, on mic'd instruments, masking starts taking place if one is playing that quietly, and it's questionable whether having 24 bit converters actually does anything for you. In your real world audio, Jay, as a mastering engineer, you obviously should work at the highest resolution one can give you, hopefully allowing enough real world audio headroom to do your job without having to step on it. In 16 bit products it was normal to see -.1 dBFS levels with absolutely no room for a mastering engineer to do anything but bring it down in order to bring it back up. 24 bits doesn't even stop this stupid waste of bandwidth, but at least it allows some level of having real headroom over noise than 16 bits did. Does that matter on the above scenario when one is working in a noisy room? Probably not. Does it matter if the 24 bit's analog front end sucks and the 16 bit's analog front end doesn't? Probably. I guess the real world problem is that one cannot define "real world audio" that sticks for everyone. Personally I'd start at the bottom and say if the room is live and going to be noisy with A/C turning on and off, people shuffling their feet or programs or coughing, then 24 bit isn't a necessity in that "real world audio" environment. Somewhat akin to the idea of having EQ boosted in the HF range on a bass track where there's no content. Stupid waste of space. Or dollars. -- Roger W. Norman SirMusic Studio Purchase your copy of the Fifth of RAP CD set at www.recaudiopro.net. See how far $20 really goes. |
#85
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
In article , "Roger W. Norman"
wrote: "real world audio", then it's possible that 24 isn't better than 16 if real world audio recording is accomplished in a room where quiet is represented by about -55 dB of room noise, There certainly are a number of situations - probably a majority if we're talking about most popular music styles, where more than 16 bits most likely won't help. That's true enough. And also. there are no situations where an ADC at 24 bits is strictly necessary since there aren't any converters that can reproduce 144 dB dynamic range. 20 would be a practical way to go, with the exceedingly rare converter perhaps approaching the equivalent of 21 bits and change. But hey, with a 24 bit chip you know that it can convey the best the converter electronics can throw at it, and those are the chips available, so it makes sense that they would be used by designers. And there are some recording situations where you can use more than 16 bits, so I have no problem having it available and using it since storage is cheap. That's quite different from saying you NEED it always, or you need it for final delivery on the new Metallica record. One needs to understand all the details of digital audio to know when it helps, when it doesn't, and why. There's not a "one size fits all" answer. In your real world audio, Jay, as a mastering engineer, you obviously should work at the highest resolution one can give you, I can tell you that the minute you string a bunch of processors together, especially if one has tubes, the noise floor in any mastering room will likely be at least a little above 16 bits. however, there is still useful and audible signal that can happen within the noise, so it certainly pays to convert with more than 16 bits to be as certain as possible you're getting everything you can. As far as processing goes, you want as many bits as you can throw at it. That's different criteria. In 16 bit products it was normal to see -.1 dBFS levels with absolutely no room for a mastering engineer to do anything but bring it down in order to bring it back up. Yeah, that's a pain. Well, not really, but one more unneccesary step. -- Jay Frigoletto Mastersuite Los Angeles promastering.com |
#86
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
"Jay - atldigi" wrote in message ... "increased resolution" on a full-scale signal is nothing more than the added ability to resolve the quietest overtones, then he's right and is actually in full agreement with Arny. At least somebody understands me, but I thought I had already said this somewhere in the thread. It's those quieter components that you are getting from the extra bits. The louder components aren't represented any better. In the end, it can be a more precise and better sounding recording (provided the source is of a quality to benefit), but it's because of the little things you can now record, not that the big ones are better. Still, isn't it so that the number of possibilities doubles each time we add a new bit, thus a 24 bit converter has 16, 777, 217 values to choose from when converting the voltage to numbers. This means, that the most-significant-bit of a 24 bit converter has (16,777,217 / 2) 8, 388,608 values to choose from when it's giving a number to any signal in the region of roughly -6dB to 0 dB Full Scale, right? That leads to the following: A 24 bit system has quantized a loud component somewhere between -6 to 0 dB FS a lot more accurately than a 16 bit system would with the same signal, since it's rounded the original voltage more precisely. Also, if you were recording the same source with a 24 bit system peaking at -48dB FS _and_ a 16 bit system peaking at 0dB FS, after normalizing the 24bit file to 0 dB, you would essentially have two identical files, _identically_ quantized, since the 24 bit system had used its 16 least-significant-bits. So, louder components are also represented better in a 24 bit system. Are THESE aforementioned things something we can ALL agree on? |
#87
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
Jay - atldigi wrote in
: But that's exactly my point: only the -100 component is what you've gained. The -1 component is not rendered any better than it was before. Jay, as I am a person steeped in computer bits but weak on audio theory, I ask you to explain that statement. A 24-bit signal offers +/- 8,388,608 possible volume levels. A 16-bit signal supports only +/- 32,768 volume levels. Only +/- 128 levels (48 dB) of the 24-bit signal are less than the lowest bit of the 16-bit (providing lower threshold). The other bits provide 256 more possible levels between each of the 16- bits' levels. Do you disagree with any of the above? It's just math. Is your argument that this higher precision is inaudible? |
#88
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
Tommi wrote: Still, isn't it so that the number of possibilities doubles each time we add a new bit, thus a 24 bit converter has 16, 777, 217 values to choose from when converting the voltage to numbers. This means, that the most-significant-bit of a 24 bit converter has (16,777,217 / 2) 8, 388,608 values to choose from when it's giving a number to any signal in the region of roughly -6dB to 0 dB Full Scale, right? That's right. That leads to the following: A 24 bit system has quantized a loud component somewhere between -6 to 0 dB FS a lot more accurately than a 16 bit system would with the same signal, since it's rounded the original voltage more precisely. Also, if you were recording the same source with a 24 bit system peaking at -48dB FS _and_ a 16 bit system peaking at 0dB FS, after normalizing the 24bit file to 0 dB, you would essentially have two identical files, _identically_ quantized, since the 24 bit system had used its 16 least-significant-bits. Right again. So, louder components are also represented better in a 24 bit system. Are THESE aforementioned things something we can ALL agree on? Yes. What must be remembered, however, is how the inaccuracy is perceived. Many think that the increased resolution results in less perception of some kind of stairstep effect. That is not the case. The preceived situation with an N bit converter done properly and going through the A/D and then the D/A process is _exactly_ the same as an infinite resolution conversion at both stages with a digital adder in between just adding in a noise signal comprised of a random variable with values of 0 or 2^-N at each sample time. What is heard is additive noise and only that iff the conversion is done without correlation between the value of that bit and the value of the sample. This is practically achievable. Bob -- "Things should be described as simply as possible, but no simpler." A. Einstein |
#89
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
Jay - atldigi wrote:
with a 24 bit chip you know that it can convey the best the converter electronics can throw at it, and those are the chips available, so it makes sense that they would be used by designers. And there are some recording situations where you can use more than 16 bits, so I have no problem having it available and using it since storage is cheap. Also: due to the way computers store information, 18 bit PCM takes the same amount of disk space as does 24 bit PCM (and often the same amount as does 32 bit floating point PCM data.) |
#90
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
In article , Carey Carlan
writes: The other bits provide 256 more possible levels between each of the 16- bits' levels. Do you disagree with any of the above? It's just math. This the question I keep trying to get an answer to but after trying a number of times over several years I have not gotten one. Its like the question just bounces right off. Garth~ "I think the fact that music can come up a wire is a miracle." Ed Cherney |
#91
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
Right. There are more discrete 24-bit sample values available between
those provided by 16-bit depth, just as there are below the lowest 16-bit (non-zero) sample. Much of the discussion of sample size focuses on absolute dynamic range and seems to ignore the additional resolution between 16-bit values over the entire range. And the absolute dynamic range is somewhat irrelevant because no analog circuitry exists with 144 dB of dynamic range. Though it does buy a LOT of headroom (or slop room). Illustration: Near zero, showing increased range: 16-bit: 00000000 00000001 (00000000) - the lowest 0's are all you get 24-bit: 00000000 00000000 (00000000-11111111) - you get the whole range The same applies between any two arbitrary 16-bit sample values: 16-bit: 01010101 10101011 (00000000) 24-bit: 01010101 10101011 (00000000-11111111) - the actual value can be defined 256 times more closely. Carey Carlan wrote: Jay - atldigi wrote in : But that's exactly my point: only the -100 component is what you've gained. The -1 component is not rendered any better than it was before. Jay, as I am a person steeped in computer bits but weak on audio theory, I ask you to explain that statement. A 24-bit signal offers +/- 8,388,608 possible volume levels. A 16-bit signal supports only +/- 32,768 volume levels. Only +/- 128 levels (48 dB) of the 24-bit signal are less than the lowest bit of the 16-bit (providing lower threshold). The other bits provide 256 more possible levels between each of the 16- bits' levels. Do you disagree with any of the above? It's just math. Is your argument that this higher precision is inaudible? |
#92
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
Carey Carlan wrote:
Jay - atldigi wrote in : But that's exactly my point: only the -100 component is what you've gained. The -1 component is not rendered any better than it was before. Jay, as I am a person steeped in computer bits but weak on audio theory, I ask you to explain that statement. A 24-bit signal offers +/- 8,388,608 possible volume levels. A 16-bit signal supports only +/- 32,768 volume levels. Only +/- 128 levels (48 dB) of the 24-bit signal are less than the lowest bit of the 16-bit (providing lower threshold). The other bits provide 256 more possible levels between each of the 16- bits' levels. Do you disagree with any of the above? It's just math. Is your argument that this higher precision is inaudible? I thought this was the question we just got hammered out. I'm going to use a "completely" unrelated analogy to illustrate a very similar misunderstanding that should hopefully make this clear (but may just muddle things up more). Bear with me. ----analogy---- When we're talking about SAMPLE RATE, people who don't entirely understand digital or analog audio get the mistaken impression that higher sample rates (more samples) will result in a better representation of the audio due to its documentation of information in-between samples. Those who understand how it works try to explain that the additional information "in between the samples" consists NOTHING other than higher-frequency information. If you were to "subtract" the lower Fs audio from the higher Fs audio, all that would remain would be frequencies above the Nyquist frequency of the lower Fs. I know you understand this one. ----analogy---- Okay, so let's think about the bit depth again. Suppose you have a loud-ass signal represented by 16 bits. If you move to 24 bits, you are now able to "more accurately" represent that loud-ass signal because you can represent an actual signal that falls in-between the bits in the 16-bit system. But this additional information is NOTHING more than low-level information. If you were to "subtract" the 16-bit audio from the 24-bit audio, all that would remain would be a signal at below -96dBFS. Can you see how these are similar situations in different domains? It's true that higher sample rates give a more accurate reproduction, but that accuracy is nothing but higher-frequency information. It's true that higher bit depth gives a more accurate reproduction, but that accuracy is nothing but lower-level information. If you want to understand this more intuitively, you should try it. Start with a very clean 24-bit recording. make a copy of this data, and truncate it to 16 bits. Then convert the 16-bit file back to 24-bits. You now have two 24-bit files that are identical in the top 16 bits but different in the bottom 24. One file has music down there, the other has silence. Now create a difference file from these two (invert one and then sum them). You now have nothing but the bottom 8 bits of the original 24-bit file. Listen to it. It probably won't even sound like the music it came from because all the "loud" stuff is gone. All that remains is the low-level information that was riding on those taller waves. I know you're asking Jay and not me, but I'm confident we both understand the issue and he just finished explaining that he doesn't disagree with what you wrote above, and he hasn't said that the higher precision is inaudible. What you need to realize about Jay's statement (which was a reference to my explanation) is that the theoretical -100dB and -1dB signals we're talking about are *simultaneous* and are components of a single sound that can be considered separately for the sake of analysis in attempting to understand this question of bit depth, dynamic range, and resolution. Does that make things any more clear? ulysses |
#93
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
In article , Justin Ulysses
Morse writes: Okay, so let's think about the bit depth again. Suppose you have a loud-ass signal represented by 16 bits. If you move to 24 bits, you are now able to "more accurately" represent that loud-ass signal because you can represent an actual signal that falls in-between the bits in the 16-bit system. But this additional information is NOTHING more than low-level information. This is the first time I have heard this. Now this makes more sense to me. I still dont understand why the info that falls between the 16 bits would necessarily be low level information but I dont doubt that its true. Perhaps someone could explain. Thanks Ulysses, thats a step forward in my understanding. Garth~ "I think the fact that music can come up a wire is a miracle." Ed Cherney |
#94
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
Garthrr wrote:
This is the first time I have heard this. Now this makes more sense to me. I thought it was about the 4th or 5th time I said it in these threads over the past 2 days, and I thought I was repeating myself. But I'm glad to hear it's starting to gel. I still dont understand why the info that falls between the 16 bits would necessarily be low level information but I dont doubt that its true. Because the space between those bits is very tiny. Remember, we're not talking about the whole 24-bit sample or the whole 16-bit sample. We're talking about the DIFFERENCE between those two. Think about a 16-bit sample as simply a 24-bit sample with 8 zeroes on the end. So the difference between 24-bit audio and 24-bit audio truncated to 16 bits is simply those last 8 bits dancing around. It doesn't matter what the first 16 bits are doing because they're doing it the same in both cases. We're only discussing what's in one sample that's NOT in the other. Now, those last 8 bits can dance around as rambunctiously as they like, but they'll never represent a DEVIATION from the 16-bit sample of more than -96dBFS. Suppose you have an input signal whose voltage at some arbitrary point in time is 3.26534263219541623 volts. Now, off the top of my head I estimate that the best approximation of this voltage you can represent with 24 bits is maybe 3.2653426 volts. And 16 bits would round it off to around 3.26534. So what's going on in the 24-bit audio that's missing from the 16-bit audio? A signal in the neighborhood of 2.6 microvolts. Which is pretty dang low-level if you ask me. Even though the signal we're listening to is up in the ballpark of full scale, the "missing detail" we're talking about is down below -96dBFS. Next question? ulysses |
#95
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
In article , Carey
Carlan wrote: Jay - atldigi wrote in : But that's exactly my point: only the -100 component is what you've gained. The -1 component is not rendered any better than it was before. First up, don't get hung up on this -1 and -100 example that was just in reponse to another post where the example was given. The original thought was useful enough conceptually, but shouldn't be thought of as a real world practical example. Jay, as I am a person steeped in computer bits but weak on audio theory, I ask you to explain that statement. A 24-bit signal offers +/- 8,388,608 possible volume levels. A 16-bit signal supports only +/- 32,768 volume levels. Only +/- 128 levels (48 dB) of the 24-bit signal are less than the lowest bit of the 16-bit (providing lower threshold). The other bits provide 256 more possible levels between each of the 16- bits' levels. Do you disagree with any of the above? It's just math. Is your argument that this higher precision is inaudible? That's the basic math for an ideal quantizer which is not what we use in audio. Ideal (noiseless) quantizers are an excellent and necessary learning aid, but aren't the ultimate implimentation we use in our audio ADCs as they are non-linear and cause distortion (this non-linear behavior being what you guys are giving examples of and worrying about). When you look at a digital audio system as a whole, not just the isolated parts (quantizer, sampler, filter etc.), it actually behaves a little differently from how it's isolated parts first look on paper. We deliberately use non-ideal quantizers to achieve linearity. The key here, and some might not want to hear it, is dither. Dither is not only used in the familiar requantizing. An example of requantizing would be when you have a 24 bit file on your computer and "dither it" down to 16 bits to go on a CD. In digital audio, all quantizers need to be dithered. The quantizer in your ADC needs to be dithered to behave properly. Here's the important sentence taken right from Watkinson: "The dither has resulted in a form of duty cycle modulation and the resolution of the system has been extended indefinitely instead of being limited by the size of the steps." To explain all of this fully, the post would get very, very long. It may be more effiecient to order a few AES papers (I could offer appropriate references) and get Watkinson's and Pohlmann's books. However, I can offer a few points that may be helpful. The transfer function really does become linear with dither, at the expense of a noise floor. Even in an ideal (noiseless) quantizer, loud, complex signals' quantizing error will manifest as random noise. The numerical values for the samples in such audio are so widely varying that the quantizing errors will be independent or each other and will distribute themselves in such a random pattern according to probability that it will essentially behave as broadband noise (yet not of the same gaussian distribution as thermal noise in an analog circuit). Smaller or less complex signals, however, are more of a problem. This is where the error becomes correlated to the signal and you get distortion. What's worse, the distortion happens after the anti-aliasing filter, and the harmonics of the distortion above Nyquist will alias causing "birdies". Correct application of dither decorrelates the quantizing error and linearizes the transfer function at the expense of a noise floor. The resolution becomes, in effect, infinite. The limitation is that the noise floor can be plainly audible in low resolution systems and obscure low level detail in addition to being distracting. However, within it's available dynamic range, the system is infinite. You can't get better than that, even if you throw more bits at it. Taken as a whole, the system works better than it would appear from the initial learning aid of the ideal quantizer on paper. Add more bits, the noise floor drops, and you can resolve smaller (quieter) details. -- Jay Frigoletto Mastersuite Los Angeles promastering.com |
#96
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
|
#97
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
In article , "Tommi"
wrote: "Jay - atldigi" wrote in message ... "increased resolution" on a full-scale signal is nothing more than the added ability to resolve the quietest overtones, then he's right and is actually in full agreement with Arny. At least somebody understands me, but I thought I had already said this somewhere in the thread. It's those quieter components that you are getting from the extra bits. The louder components aren't represented any better. In the end, it can be a more precise and better sounding recording (provided the source is of a quality to benefit), but it's because of the little things you can now record, not that the big ones are better. Still, isn't it so that the number of possibilities doubles each time we add a new bit, thus a 24 bit converter has 16, 777, 217 values to choose from when converting the voltage to numbers. This means, that the most-significant-bit of a 24 bit converter has (16,777,217 / 2) 8,388,608 values to choose from when it's giving a number to any signal in the region of roughly -6dB to 0 dB Full Scale, right? That leads to the following: A 24 bit system has quantized a loud component somewhere between -6 to 0 dB FS a lot more accurately than a 16 bit system would with the same signal, since it's rounded the original voltage more precisely. Also, if you were recording the same source with a 24 bit system peaking at -48dB FS _and_ a 16 bit system peaking at 0dB FS, after normalizing the 24bit file to 0 dB, you would essentially have two identical files, _identically_ quantized, since the 24 bit system had used its 16 least-significant-bits. So, louder components are also represented better in a 24 bit system. Are THESE aforementioned things something we can ALL agree on? I'm going to repeat my last post in case you missed it or just glanced over it because it's really the same question as Carey's. To preface: what you see on paper with an ideal quantizer is not what you see in actual practice with digital audio's intentionally non-ideal quantizers. Also, we're ultimately dealing with voltages coming in and going out, and the digits viewed at each isolated intermediate step don't tell the whole story. Here goes: -- That's the basic math for an ideal quantizer which is not what we use in audio. Ideal (noiseless) quantizers are an excellent and necessary learning aid, but aren't the ultimate implimentation we use in our audio ADCs as they are non-linear and cause distortion (this non-linear behavior being what you guys are giving examples of and worrying about). When you look at a digital audio system as a whole, not just the isolated parts (quantizer, sampler, filter etc.), it actually behaves a little differently from how it's isolated parts first look on paper. We deliberately use non-ideal quantizers to achieve linearity. The key here, and some might not want to hear it, is dither. Dither is not only used in the familiar requantizing. An example of requantizing would be when you have a 24 bit file on your computer and "dither it" down to 16 bits to go on a CD. In digital audio, all quantizers need to be dithered. The quantizer in your ADC needs to be dithered to behave properly. Here's the important sentence taken right from Watkinson: "The dither has resulted in a form of duty cycle modulation and the resolution of the system has been extended indefinitely instead of being limited by the size of the steps." To explain all of this fully, the post would get very, very long. It may be more effiecient to order a few AES papers (I could offer appropriate references) and get Watkinson's and Pohlmann's books. However, I can offer a few points that may be helpful. The transfer function really does become linear with dither, at the expense of a noise floor. Even in an ideal (noiseless) quantizer, loud, complex signals' quantizing error will manifest as random noise. The numerical values for the samples in such audio are so widely varying that the quantizing errors will be independent or each other and will distribute themselves in such a random pattern according to probability that it will essentially behave as broadband noise (yet not of the same gaussian distribution as thermal noise in an analog circuit). Smaller or less complex signals, however, are more of a problem. This is where the error becomes correlated to the signal and you get distortion. What's worse, the distortion happens after the anti-aliasing filter, and the harmonics of the distortion above Nyquist will alias causing "birdies". Correct application of dither decorrelates the quantizing error and linearizes the transfer function at the expense of a noise floor. The resolution becomes, in effect, infinite. The limitation is that the noise floor can be plainly audible in low resolution systems and obscure low level detail in addition to being distracting. However, within it's available dynamic range, the system is infinite. You can't get better than that, even if you throw more bits at it. Taken as a whole, the system works better than it would appear from the initial learning aid of the ideal quantizer on paper. Add more bits, the noise floor drops, and you can resolve smaller (quieter) details. -- Jay Frigoletto Mastersuite Los Angeles promastering.com |
#98
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
"Garthrr" wrote in message
In article , Carey Carlan writes: The other bits provide 256 more possible levels between each of the 16- bits' levels. Do you disagree with any of the above? It's just math. This the question I keep trying to get an answer to but after trying a number of times over several years I have not gotten one. Its like the question just bounces right off. It's doesn't bounce off everybody. I for one strongly affirm what Carey said above. The relevant facts are very compelling to me. It's simply how things work. |
#99
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
"Kurt Albershardt" wrote in message
... Also: due to the way computers store information, 18 bit PCM takes the same amount of disk space as does 24 bit PCM (and often the same amount as does 32 bit floating point PCM data.) Well, my point wasn't that using 24 bit has anything wrong with it, nor, with the cost of storage today, is there a problem with using 24 bit for any project. In fact, the processing aspects of digital certainly require that you use 24 bit, and save it as 32 bit floating point if you can. I was just trying to make it clear that there are technical aspects of 24 bit converters that wouldn't necessarily make them better than 16 bit in given circumstances. Pretty much the concept of choosing one's tools to fit the job. When I go out and do location recordings, I still use Tascam DA38s and I've not had one client that wasn't happy with the recordings. They might not have liked the performances, but that's a different story. My last submission to A Fifth of RAP was recorded onto DA38, and it sounds pretty good to me (obviously biased) with plenty of dynamics. The room lacked something and 24 bits wouldn't necessarily have bought me anything more. The same with Scott Dorsey's extremely dynamic recording. And again, Tonebarge absolute knocks one out of the park with his Mackie/Adat combination (although he uses different pres for tracking). I just didn't want any lurkers to think that only 24 bit converters are a solution to any recording problems they might have. Again, it's the use of the tools rather than the tools themselves. -- Roger W. Norman SirMusic Studio Purchase your copy of the Fifth of RAP CD set at www.recaudiopro.net. See how far $20 really goes. |
#100
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
"Garthrr" wrote in message
In article , Justin Ulysses Morse writes: Okay, so let's think about the bit depth again. Suppose you have a loud-ass signal represented by 16 bits. If you move to 24 bits, you are now able to "more accurately" represent that loud-ass signal because you can represent an actual signal that falls in-between the bits in the 16-bit system. But this additional information is NOTHING more than low-level information. This is the first time I have heard this. Now this makes more sense to me. I still dont understand why the info that falls between the 16 bits would necessarily be low level information but I dont doubt that its true. Perhaps someone could explain. Thanks Ulysses, thats a step forward in my understanding. The information that falls between the pairs of 16 bit values is obviously very small, so I guess its fair to call it "low level information". Imagine two 16 bit signals. One is a 1 KHz sine wave and one is a 10 KHz sine wave. Both are integers that vary between 0 and 65535. Now, let's take both sine waves and convert them to 24 bit numbers by adding trailing zeros. They are now both numbers that vary between 0 and 16,777,215. Actually, since the numbers were created by adding 8 zeroes to the end, no number goes above 16,776,960. Both sets of numbers are a bit unusual, in that every number ends with 8 zeroes. Let's take the 10 KHz sine wave and divide it by 65536. It becomes very low-level signal. It becomes a number that varies between 0 and 255. Now, lets add the 10 KHz sine wave to the 1 KHz sine wave. Finally, we have a set of numbers that don't all end in 8 zeroes! The samples whose last 8 bits aren't zero represent the 10 KHz low level signal, don't they? So the low level detail is, in a manner of speaking, in the samples that are "in between". If we want to separate the 1 KHz and 10 KHz signals, we have two possible strategies: (1) Do frequency domain filtering (2) Take the top 16 bits of every sample which will give us back the original 1 KHz sine wave, and take the bottom 8 bits every sample which will give us back an 8 bit 10 KHz sine wave. |
#101
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
In article , Jay -
atldigi writes: The transfer function really does become linear with dither, at the expense of a noise floor. Ok, there's another important piece of the puzzle (at least for me). So if I understand correctly, its a situation where what would seem to be ideal on paper is problematic as far as implementation because of the side effects of distortion and aliasing. So dither cures these problems at the expense of the noise floor by somehow either correcting the quantization error or rendering it harmless. So what would be stairstep errors in voltage are smoothed over by dither and thus the resolution of the system in the higher levels becomes, for all practical purposes, perfect. Is that an essentially correct oversimplification? Garth~ "I think the fact that music can come up a wire is a miracle." Ed Cherney |
#102
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
In article , Justin Ulysses
Morse writes: I thought it was about the 4th or 5th time I said it in these threads over the past 2 days, and I thought I was repeating myself. But I'm glad to hear it's starting to gel. You may well have said it and I could have either missed the post or missed the point. Either way it is starting to make sense to me even though its still a little blurry. Suppose you have an input signal whose voltage at some arbitrary point in time is 3.26534263219541623 volts. Now, off the top of my head I estimate that the best approximation of this voltage you can represent with 24 bits is maybe 3.2653426 volts. And 16 bits would round it off to around 3.26534. So what's going on in the 24-bit audio that's missing from the 16-bit audio? A signal in the neighborhood of 2.6 microvolts. Which is pretty dang low-level if you ask me. Here is a bblluurrrryy moment for me. Is it that there is a _signal_ which is 2.6 microvolts or... is it that there is an error of 2.6 microvolts in the reproduction of a signal which is the above 3.26534263219541623 volts? To me this seems like a qualitative difference (no matter how insignificant the quantity in question may be). I think its difficult for someone who knows a lot about a thing to explain it to someone who knows nothing about it because the one who knows is apt to assume certain understandings on the part of the one who doesnt. Garth~ "I think the fact that music can come up a wire is a miracle." Ed Cherney |
#103
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
In article , "Arny Krueger"
writes: The information that falls between the pairs of 16 bit values is obviously very small, so I guess its fair to call it "low level information". Oh I never thought about it in that way. I took "low level" to mean low dB level as in -100dB or something of that sort. I guess instead it was meant as "insignificant". Garth~ "I think the fact that music can come up a wire is a miracle." Ed Cherney |
#104
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
In article , Jay -
atldigi writes: what you see on paper with an ideal quantizer is not what you see in actual practice with digital audio's intentionally non-ideal quantizers. I think this is really important for anyone who is struggling to get this stuff as I am. Until this morning I was not aware of this although it doesnt surprise me. Garth~ "I think the fact that music can come up a wire is a miracle." Ed Cherney |
#105
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
In article , Jay -
atldigi writes: There are probably a few other helpful things that you may have missed or that didn't quite make sense the first time through. Good lord! You have a singular gift for understatement! Garth~ "I think the fact that music can come up a wire is a miracle." Ed Cherney |
#106
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
In article , "Arny Krueger"
writes: "Garthrr" wrote in message In article , Carey Carlan writes: The other bits provide 256 more possible levels between each of the 16- bits' levels. Do you disagree with any of the above? It's just math. GW This the question I keep trying to get an answer to but after trying a number of times over several years I have not gotten one. Its like the question just bounces right off. AK It's doesn't bounce off everybody. I for one strongly affirm what Carey said above. The relevant facts are very compelling to me. It's simply how things work. I guess what I'm trying to find out is whether others do not agree with what Carey said--whether this is in contention or not. From something I just read I'm beginning to think that dither is somehow responsible for negating the advantage of that extra resolution at least in higher level signals. Garth~ "I think the fact that music can come up a wire is a miracle." Ed Cherney |
#107
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
"Roger W. Norman" wrote in message
"Kurt Albershardt" wrote in message ... Also: due to the way computers store information, 18 bit PCM takes the same amount of disk space as does 24 bit PCM (and often the same amount as does 32 bit floating point PCM data.) Well, my point wasn't that using 24 bit has anything wrong with it, nor, with the cost of storage today, is there a problem with using 24 bit for any project. In fact, the processing aspects of digital certainly require that you use 24 bit, and save it as 32 bit floating point if you can. I was just trying to make it clear that there are technical aspects of 24 bit converters that wouldn't necessarily make them better than 16 bit in given circumstances. Pretty much the concept of choosing one's tools to fit the job. Agreed. One thing to realize is that most if not all of the major chip makers have unflinchingly released 24/96 and 24/192 converters with worse measured dynamic range than some of their earlier 16/44 converters. There's a part of the market that is all about numbers. When I go out and do location recordings, I still use Tascam DA38s and I've not had one client that wasn't happy with the recordings. They might not have liked the performances, but that's a different story. My last submission to A Fifth of RAP was recorded onto DA38, and it sounds pretty good to me (obviously biased) with plenty of dynamics. The room lacked something and 24 bits wouldn't necessarily have bought me anything more. The same with Scott Dorsey's extremely dynamic recording. And again, Tonebarge absolute knocks one out of the park with his Mackie/Adat combination (although he uses different pres for tracking). I just didn't want any lurkers to think that only 24 bit converters are a solution to any recording problems they might have. Again, it's the use of the tools rather than the tools themselves. I think that this is one of the most important messages that a group like this has to present to newbies. |
#108
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
"Garthrr" wrote in message
In article , "Arny Krueger" writes: "Garthrr" wrote in message In article , Carey Carlan writes: The other bits provide 256 more possible levels between each of the 16- bits' levels. Do you disagree with any of the above? It's just math. GW This the question I keep trying to get an answer to but after trying a number of times over several years I have not gotten one. Its like the question just bounces right off. AK It's doesn't bounce off everybody. I for one strongly affirm what Carey said above. The relevant facts are very compelling to me. It's simply how things work. I guess what I'm trying to find out is whether others do not agree with what Carey said--whether this is in contention or not. On Usenet, *anything* can be in contention! ;-) From something I just read I'm beginning to think that dither is somehow responsible for negating the advantage of that extra resolution at least in higher level signals. Not if dither is used correctly, and these days it often is used correctly. Usually, dither is sized to match the actual size of the smallest quantization step. This number has to be chosen carefully because the smallest step being quantized by just about all 24-bit converters isn't the theoretical 1 sixteen-millionth of full scale. It's something vastly bigger, and it varies quit a bit from converter to converter. But the education process has worked and most chip designers know about this. Dither makes the size of the quantization steps much more sonically palatable, no matter how big or small they are. From the standpoint of technical accuracy, dither doesn't negate any advantages, and it doesn't level any playing fields. These days, *everybody* uses dither in their quantizers, and IME *everybody* makes pretty good use of it. However a converter with a relatively coarse step size is still going to be noisier than one that has a smaller step size. |
#109
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
"Garthrr" wrote in message
In article , Justin Ulysses Morse writes: I thought it was about the 4th or 5th time I said it in these threads over the past 2 days, and I thought I was repeating myself. But I'm glad to hear it's starting to gel. You may well have said it and I could have either missed the post or missed the point. Either way it is starting to make sense to me even though its still a little blurry. Suppose you have an input signal whose voltage at some arbitrary point in time is 3.26534263219541623 volts. Now, off the top of my head I estimate that the best approximation of this voltage you can represent with 24 bits is maybe 3.2653426 volts. And 16 bits would round it off to around 3.26534. So what's going on in the 24-bit audio that's missing from the 16-bit audio? A signal in the neighborhood of 2.6 microvolts. Which is pretty dang low-level if you ask me. Here is a bblluurrrryy moment for me. Is it that there is a _signal_ which is 2.6 microvolts or... is it that there is an error of 2.6 microvolts in the reproduction of a signal which is the above 3.26534263219541623 volts? Both. You can think of 3.26534 volts as 3.26534263219541623 volts with an approximately 2.6 microvolt error, or you can think of .26534 volts as a 3.26534263219541623 volt signal with an approximately 2.6 microvolt error voltage added. To me this seems like a qualitative difference (no matter how insignificant the quantity in question may be). It is a small qualitative difference. It's an error that is about 120 dB down in the presence of a signal which is close to full scale (0 dB), which means don't worry about it. If the size of a significant signal was -100 dB, then it would be worth worrying about. |
#110
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
"Garthrr" wrote in message
In article , Jay - atldigi writes: The transfer function really does become linear with dither, at the expense of a noise floor. Ok, there's another important piece of the puzzle (at least for me). So if I understand correctly, its a situation where what would seem to be ideal on paper is problematic as far as implementation because of the side effects of distortion and aliasing. So dither cures these problems at the expense of the noise floor by somehow either correcting the quantization error or rendering it harmless. Dither clearly doesn't correct quantization error. Dither does change quantization error into something that is sonically more benign. What dither does is sort of like turning sewage into beer. The sewage hasn't been exactly been turned into chemically pure water. But, the sewage has been turned into beer, which is something that is more palatable to most of us than sewage. So what would be stairstep errors in voltage are smoothed over by dither and thus the resolution of the system in the higher levels becomes, for all practical purposes, perfect. Is that an essentially correct oversimplification? The stairstep errors don't get smoothed, they become randomized. Think of quantization errors as being splats of paint on the wall. Without dither the splats of paint would tend to be grouped, so they would look like bigger messier splats. Dither makes the splats more uniformly distributed so they appear to be smaller and actually more like some nicely shaded, more neutral color. BTW, dithering pretty much works this way when it's applied to computer graphics. |
#111
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
Garthrr wrote:
Here is a bblluurrrryy moment for me. Is it that there is a _signal_ which is 2.6 microvolts or... is it that there is an error of 2.6 microvolts in the reproduction of a signal which is the above 3.26534263219541623 volts? To me this seems like a qualitative difference (no matter how insignificant the quantity in question may be). It's exactly the same thing either way you think about it. The 16-bit signal is the what you'd end up with if you started with the 24-bit signal and added (or subtracted) the error signal. Since we're "rounding" (kind of), sometimes the 16-bit value will be a little bigger, some times it'll be a little smaller than the 24-bit value. So when you look at the error "signal" over the course of time, it's constantly bouncing up and down, not quite randomly but sort of arbitrarily because the music doesn't care where those 16-bit quanitizations are. Of course I'm ignoring the topic of dither for the sake of theory, but as Jay points out Dither changes everything. Dither is another "error" signal added to the 24-bit signal just before you do the "rounding." The dither signal, which averages just under -96dB, plus those 8 lost bits which average under -96dB, add up to a "music plus noise" signal which averages just OVER -96dB, so it is able to show up in the 16-bit signal. ulysses |
#112
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
Arny Krueger wrote:
Now, lets add the 10 KHz sine wave to the 1 KHz sine wave. Finally, we have a set of numbers that don't all end in 8 zeroes! I like this example because it equates to something musically meaningful. That 10kHz sine wave is a high-order even harmonic of the 1kHz sine wave. Though our example created it synthetically, it could just as easily have been a distortion component in an amplifier, or an overtone in an acoustic instrument. It contributes to the tonal character of the louder 1kHz tone. Removing this harmonic, either by filtering out the high frequencies or by swamping it with noise or quantization error, will change the musical character of the music just like adding a harmonic that wasn't originally there would. ulysses |
#113
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
Carey Carlan writes:
The other bits provide 256 more possible levels between each of the 16- bits' levels. Garthrr wrote: I guess what I'm trying to find out is whether others do not agree with what Carey said--whether this is in contention or not. I haven't heard anybody disagree. I don't think anybody will. From something I just read I'm beginning to think that dither is somehow responsible for negating the advantage of that extra resolution at least in higher level signals. It uses the statistical average of the dither noise to bring the "extra resolution" up into the range of the smaller sample size. This is really a whole new can of worms though. It complicates the "bits" question, which really does make sense even before you start talking about dither. ulysses |
#114
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
Justin Ulysses Morse wrote in
m: Think about a 16-bit sample as simply a 24-bit sample with 8 zeroes on the end. So the difference between 24-bit audio and 24-bit audio truncated to 16 bits is simply those last 8 bits dancing around. It doesn't matter what the first 16 bits are doing because they're doing it the same in both cases. We're only discussing what's in one sample that's NOT in the other. Now, those last 8 bits can dance around as rambunctiously as they like, but they'll never represent a DEVIATION from the 16-bit sample of more than -96dBFS. Carey wrote: Is your argument that this higher precision is inaudible? and your answer is "Yes". That's all I needed to know. |
#115
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
"Jay - atldigi" wrote in message ... Correct application of dither decorrelates the quantizing error and linearizes the transfer function at the expense of a noise floor. The resolution becomes, in effect, infinite. The limitation is that the noise floor can be plainly audible in low resolution systems and obscure low level detail in addition to being distracting. However, within it's available dynamic range, the system is infinite. You can't get better than that, even if you throw more bits at it. Taken as a whole, the system works better than it would appear from the initial learning aid of the ideal quantizer on paper. Add more bits, the noise floor drops, and you can resolve smaller (quieter) details. If it is so, then it..umm..is so. So this is what happens in the real world, but in theory 24 bits represents the original signal more accurately than 16 bits. Thank you Jay for your information! |
#116
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
|
#117
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
Jay - atldigi wrote in
: In article , Carey Carlan wrote: Is your argument that this higher precision is inaudible? excerpts of Jay's reply : The transfer function really does become linear with dither. The limitation is that the noise floor can be plainly audible in low resolution systems. However, within it's available dynamic range, the system is infinite. You can't get better than that, even if you throw more bits at it. Therefore, your answer is "No". Dithering smooths out the differences between the 65,535 steps, making them as smooth as the 16 million steps. Add more bits, the noise floor drops, and you can resolve smaller (quieter) details. So, while the bits between the higher samples aren't necessary because dithering smooths out the differences, the bottom 256 values between 16-bit 1 and 16-bit 0 allow you to use finer, quieter dithering that drops the noise floor by 48 dB (ideally). Thank you! I think I understand now. |
#119
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
On Thu, 20 Nov 2003 16:04:43 GMT, Carey Carlan
wrote: Dithering smooths out the differences between the 65,535 steps, making them as smooth as the 16 million steps. Dither is just noise, but noise has a special property in this case. Although it can't smooth out differences, it can remove errors. The process of quantizing has fundamental errors estimating the smallest bit, errors which track the signal itself. Dither randomizes this error completely, leaving only noise. A 24 bit conversion of the signal with the dither level of a 16 bit conversion would have no more (or less) information than a 16 bit conversion. So, while the bits between the higher samples aren't necessary because dithering smooths out the differences, the bottom 256 values between 16-bit 1 and 16-bit 0 allow you to use finer, quieter dithering that drops the noise floor by 48 dB (ideally). Perzactly! Chris Hornbeck "That is my theory, and what it is too." Anne Elk |
#120
|
|||
|
|||
16 bit vs 24 bit, 44.1khz vs 48 khz <-- please explain
Carey Carlan wrote:
Carey wrote: Is your argument that this higher precision is inaudible? and your answer is "Yes". That's all I needed to know. What? Where do you get that? I've written a novel or two explaining why it's theoretically audible, even though in practice it will be covered up by noise most of the time. To say my answer is "yes" is to miss the point which I have over-articulated here. ulysses |
Reply |
Thread Tools | |
Display Modes | |
|
|
Similar Threads | ||||
Thread | Forum | |||
Explain me this | Audio Opinions | |||
TS/TRS balanced/unbalanced can someone explain | General | |||
Can you explain this 50Hz hum?? | Pro Audio | |||
Reverb & EQ and "damping" etc .. please explain .. | Pro Audio |