Home |
Search |
Today's Posts |
#41
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On Tue, 12 Feb 2013 18:35:04 +0000, Dick Pierce wrote:
Edmund wrote: On Tue, 12 Feb 2013 14:50:01 +0000, Andrew Haley wrote: Edmund wrote: I don't know if you are wrong but there are at least other claims. http://www.channld.com/vinylanalysis1.html I don't dispute that spectrogram, but I do doubt that it's an accurate copy of the master tape. What's to say that the upper harmonics aren't distortion in the vinyl replay chain? It's hard to know without the original master tape, but I'd lay a small wager on it. I don't know either and I could not find more info on the internet about what is or was possible with recording in the whole chain. Examining the spectrogram reveals little in terms of a definite diagnosis: the extended information is clearly harmonically related to the below 20 kHz signal, so it could just as likely be due to simple non-linearity in either the recording or playback process. Phone cartridges, for example, are far from the perfect linear devices we might like to think of them as. Some people say there is vinyl with 35 and even 50 kHz on it. I can say with some certainty that there IS information at those frequencies on almost every LP I have ever examined, and I can say with equal cerainty that those signals have little if anything to do with the original signal AND they were very likely NOT present when fed to the cutter head. They consist of noise, distortion artifacts and the like. Do you know if there are records with real music recorded in high(er) frequencies to what frequency and is there is such information somewhere to be found on the internet? What about the half speed cut records? What is true or not I don't know but I do know that even today I have a hard time finding even a mic that is able to record those frequencies. Yup, that's right. But ANY nonlinearity ANYWHERE in the chain is going to produce high-frequency artifacts, that might SEEM to be correlated with the signal, but are, in fact, added by the reproduction chain and NOT present in the original suignal. As for the journalist, although it is philosophic more then anything he is right about the "infinite" ( not really ) amount of information on the analog recording and he is completely right that any information lost by the AD conversion is lost forever. In what sense can he be right? The amount of information isn't infinite or virtually infinite. I know the information isn't infinite, it just hasn't the hard limit that a digital master has given an certain sample rate. You are mistaking simple "bandwidth" with information. You are ignoring, it seems, the role of dynamic range. You are also, it seems, assuming that all information is useful. Noise is information, but it's arguably not useful in and of itself. With a "perfect" analog recording -that doesn't exist either I know that too- there is no such hard limit. The limit is an awful lot harder than you, and at least one other poster, might think. Yes, there might be ONE part of the change that has maybe has a 6 dB or 12 dB/octave rolloff, but you have to account for them all. And when you do, the "ugly" Nyquist limiting looks very nice and neat by comparison. Consider the typical microphone, which has a series of complex resonances and cancellations and the like. Now consider the mic preamps and the electronics associated with that. Next, let's look at the tape recorder, for which the definition for the equalization curves beyond 15-20 kHz doesn't exist, now, look at the phenomenon of head cancellation as the wavelength approaches the dimensions of the gap, and self- erasure and a similar set of issues on playback, and the forced limitation of the bandwidth being fed to the cutter head to prevent its self-immolation and the physical, limits of the cutting stylus itself, and on and on and on and on. Please, the effective "analog rolloff" is not some simple, nice 6- or 12-dB/octave, it's MUCH greater than that and VERY messy. I appreciate what you are saying, really but I wonder how and why some analog recordings sound so good... Edmund |
#42
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On Sunday, February 10, 2013 4:37:52 PM UTC-5, Scott wrote:
I just read Fremer's overview of the box set. So I don't think it was him. He got the facts right for starters "Clearly the engineers feel that digitizing analog at high sampling and bit rates is essentially transparent to the source or they might not have done it. And once they had the music captured at 192/24 bit they also felt down-converting it to 44.1/24 wouldn't diminish the sonic quality." Pity you missed the very next sentence: "Here I definitely differ with the producers!" Fremer has long held that vinyl offers higher resolution than CD. He hasn't changed, and he will never get it right. bob |
#43
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
Scott wrote:
On Wednesday, February 13, 2013 11:54:15 AM UTC-8, Dick Pierce wrote: Scott wrote: Maybe in some other neck of the woods. But all too often I see some folks dragging out Shannon/Nyquist and saying "see digital IS perfect." I would like to see a direct quote from someone who made this assertion. If you are asking for one I'll just show you one from one of my favorite sources of misinformation. "The Nyquist theorem (which is mathematically proven) says that the exact waveform can be reproduced if the original signal is frequency limited to less than half the sampling frequency." http://www.hydrogenaudio.org/forums/...yquist+perfect IMO "exact" and "perfect" are synonymous as used here. I can find more but you only asked for one. Fine, several poiunts. The quote you supplied does NOT say that "digital is perfect." Second, exactly what is the myth, misinformation, whatever, in statement you quoted, as you quoted it? Do you have reason to believe that the Shannon/Nyquist sampling theorem is incorrect? -- +--------------------------------+ + Dick Pierce | + Professional Audio Development | +--------------------------------+ |
#44
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
Edmund wrote:
On Tue, 12 Feb 2013 18:35:04 +0000, Dick Pierce wrote: I can say with some certainty that there IS information at those frequencies on almost every LP I have ever examined, and I can say with equal cerainty that those signals have little if anything to do with the original signal AND they were very likely NOT present when fed to the cutter head. They consist of noise, distortion artifacts and the like. Do you know if there are records with real music recorded in high(er) frequencies to what frequency and is there is such information somewhere to be found on the internet? What about the half speed cut records? Sure, it's possible that it's the one cannot a priori discount that possibility. But two points: first, the RIAA curve is undefined past 20 kHz, and any number of preamps throw another pole or two in to reduce the bandwidth even further. Second, it's REALLY, REALLY hard to get the stylus tip resonance much above 20 kHz, and that resonance represents yet another 12 dB/octave low-pass filter. So, it's on there, so what, it's not likely you can get it off. Please, the effective "analog rolloff" is not some simple, nice 6- or 12-dB/octave, it's MUCH greater than that and VERY messy. I appreciate what you are saying, really but I wonder how and why some analog recordings sound so good... First, define "so good." I might hazard to say that "so good" means "I like it a lot." A perfectly legitimate definition, but one that has definiable objective context behind it as it stands. Secondly, you seem to be equating extended bandwith with "so good", as if this were the sole or at least principle criteria defining what "sounds good." It's not. There are ALL sorts of phenomenon that exist well within the audio bandwidth that could be the source of such an evocation. Have you eliminatd all of those as a possibility. What if the technical properties of the recording were, in fact, abysmal, yet the music, the performance, even the album cover, overwhelmed the objective sound attributes? Third, where is the data suggesting even a correlation between those analog recordings that "sound so good" and extended bandwidth well above 20 kHz. Can you, in fact, show that the recordings that "sound so good" have this information on them. Conversely, can you show that recordings that have information extending well beyond 20 kHz necessarily "sound so good?" I think the Ansermet Beethoven 7th is one of those recordings that "sound so good," yet I also know that the top end is fairly limited (I've, in fact, measured it). It drops like a stone above about 16 kHz, and, indeed, except for noise, the specturm looks like it came out of a digital channel running at 38 kHz with a fast but sloppy 16 kHz anti-aliasing filter. Yet it sounds so good, so much so that I'd rather listen to the LP than and CD, simply because, to me, it is far an away the best performance, to me. When and if I ever find a CD of THAT performance, my opinion may change. But it is MY opinion, dammit! -- +--------------------------------+ + Dick Pierce | + Professional Audio Development | +--------------------------------+ |
#45
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On Wednesday, February 13, 2013 2:10:07 PM UTC-8, Scott wrote:
On Wednesday, February 13, 2013 11:54:15 AM UTC-8, Dick Pierce wrote: Scott wrote: snip All analog media are finite in their resolution AND THEREFORE LOSE INFORMATION, Yeah I said that. Not sure why you feel the need to shout it. and all A/D converters lose some information. No real point in rehashing the audio myths that say otherwise. They are rehashed here and in the high-end audio press over and over again. Indeed they are. But if we are *not* doing it (which it appears neither of us are) then there is no need for us to rehash them in this thread. If someone decides to breach that then of course we should set them straight. Yes, the beat goes on. The current Editor of another high-end rag, In answering a letter from a reader who was advising that said Editor go on the Internet and read an (unspecified in the letter) article exploding the myth that 24-bit sounds no better than 16-bit. That Editor's response was that they get letters from people all the time who have read some "anti-audiophile propaganda" asserting that 16/44.1 digital audio is perfect, or that all amps sound the same and that there is no difference between different cables or interconnects. . He goes on to say that the (unspecified) article asserts that not only is 24-bit no better than 16-bit but actually sounds slightly INFERIOR to 16/44.1 or 16/48 and merely uses 6X the storage space for no improvement of the sound. He then goes on to say that said article author seems credible on the surface, but is actually naive at best. This Editor went on to explain that the resolution of a digital audio system isn't determined by the number of bits in a system, but rather by the number of bits BEING USED at any given moment. He then gives an example that a very low-level passage might be encoded at -80dBFS. on a 16-bit system, such a low-level signal is only being encoded by THREE bits and with 24-bits it's being encoded with ELEVEN bits. That's just GOT TO BE better, right, ,says he, ignoring dither in a 16-bit system altogether. He goes on to say those who insist that 24-bits offer no advantage over 16-bits in a system based upon a 20 KHz bandwidth fails to address the "Well Documented" problem of the time-domain distortion introduced by steep filters with cut-off frequencies close to the audio passband. In his explanation, faster sample rates take care of the problem by allowing for filters with gentler slopes and cut-off frequencies well above the audio band which do less damage to the signal. He finishes by reiterating that finally, anyone who listens to a 16/44.1 version of a recording vs, say, a 24/178.4 version can simply come to no other conclusion than that hi-res sounds significantly better. I know this editor and have worked with him in the past. He is technically savvy (or rather he should be, given his background), but this mixture of fact and mythology misses so many known real-world issues that to call it a competent or digitally savvy dissertation on why 24-bit is better than 16-bit would be stretching the definition of competent to the point of failure. Yes, 24-bit MASTERING can yield better results than 16-bit mastering (if the recording engineer knows what he is doing) but not for the reasons given in this apologetic response to a reader's query. |
#46
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On Wednesday, February 13, 2013 4:41:10 PM UTC-8, Edmund wrote:
On Tue, 12 Feb 2013 18:35:04 +0000, Dick Pierce wrote: Edmund wrote: On Tue, 12 Feb 2013 14:50:01 +0000, Andrew Haley wrote: Edmund wrote: I don't know if you are wrong but there are at least other claims. http://www.channld.com/vinylanalysis1.html I don't dispute that spectrogram, but I do doubt that it's an accurate copy of the master tape. What's to say that the upper harmonics aren't distortion in the vinyl replay chain? It's hard to know without the original master tape, but I'd lay a small wager on it. I don't know either and I could not find more info on the internet about what is or was possible with recording in the whole chain. Examining the spectrogram reveals little in terms of a definite diagnosis: the extended information is clearly harmonically related to the below 20 kHz signal, so it could just as likely be due to simple non-linearity in either the recording or playback process. Phone cartridges, for example, are far from the perfect linear devices we might like to think of them as. Some people say there is vinyl with 35 and even 50 kHz on it. I can say with some certainty that there IS information at those frequencies on almost every LP I have ever examined, and I can say with equal cerainty that those signals have little if anything to do with the original signal AND they were very likely NOT present when fed to the cutter head. They consist of noise, distortion artifacts and the like. Do you know if there are records with real music recorded in high(er) frequencies to what frequency and is there is such information somewhere to be found on the internet? What about the half speed cut records? As I said in an earlier post, Most cutting heads made after about 1975 definitely have frequency responses that can get to 35 or even 50 KHz due to their necessity to cut CD-4 quadraphonic discs with a 50 KHz sub-carrier. And any stereo cutter head should be able to do so when half speed mastering is employed. After all, a cutting master tape running at 7.5 ips when it had been recorded at 15 ips, feeding a cutter head cutting a groove into a blank disc running at 16.6 RPM gives the cutter head the ability to inscribe TWICE it's full-speed cut-off frequency (when played back at the full 33.3 RPM. That doesn't mean that the head can actually do, say 40 KHz or better, because the signal it's being fed has only half the high-frequency that the tape would have at full speed. Also, keep in mind that while half-speed mastering doubles the available high-frequency content of a vinyl record, it halves its low end. I.E. if the cutter head has a low-frequency limit of 20 Hz at normal mastering speeds, at half- speed that limit becomes 40 Hz. |
#47
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On 2/13/2013 3:10 PM, Scott wrote:
On Wednesday, February 13, 2013 11:54:15 AM UTC-8, Dick Pierce wrote: Scott wrote: Maybe in some other neck of the woods. But all too often I see some folks dragging out Shannon/Nyquist and saying "see digital IS perfect." I would like to see a direct quote from someone who made this assertion. If you are asking for one I'll just show you one from one of my favorite sources of misinformation. "The Nyquist theorem (which is mathematically proven) says that the exact waveform can be reproduced if the original signal is frequency limited to less than half the sampling frequency." http://www.hydrogenaudio.org/forums/...yquist+perfect IMO "exact" and "perfect" are synonymous as used here. I can find more but you only asked for one. You seem to have left out the clear caveat in the very next sentence; "The word "exact" gets a little shaky if the initial assumptions aren't met (example: each sample is taken exactly on time.)" from which it is abundantly clear that the OP was decidedly Not implying "digital is perfect", merely that if "done perfectly" - an impossibility - the resulting waveform would be perfect, which IS clearly supported by information theory. And of course, you excised the context of the statement as well, in that it was a response to the ludicrous claim that "...converting an analog signal into a discrete-time one (as it happens when converting from analog to digital) destroys the phase information in the two top octaves of the resulting spectrum. In a CD-standard digital recording, all phase information are lost from 5.5kHz up to 22kHz," I don't see how the quote you provided has ANY relevance to your claim. Keith |
#48
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On Wednesday, February 13, 2013 4:49:59 PM UTC-8, wrote:
On Sunday, February 10, 2013 4:37:52 PM UTC-5, Scott wrote: I just read Fremer's overview of the box set. So I don't think it was him. He got the facts right for starters "Clearly the engineers feel that digitizing analog at high sampling and bit rates is essentially transparent to the source or they might not have done it. And once they had the music captured at 192/24 bit they also felt down-converting it to 44.1/24 wouldn't diminish the sonic quality." Pity you missed the very next sentence: "Here I definitely differ with the producers!" Fremer has long held that vinyl offers higher resolution than CD. He hasn't changed, and he will never get it right. I didn't miss it. It simply had nothing to do with whether or not Fremer had authored the misinformation about the actual source for the LPs or made the claim of infinite resolution of analog. So I was just staying on subject. |
#49
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On Wednesday, February 13, 2013 7:41:41 PM UTC-8, Dick Pierce wrote:
Scott wrote: On Wednesday, February 13, 2013 11:54:15 AM UTC-8, Dick Pierce wrote: Scott wrote: Maybe in some other neck of the woods. But all too often I see some folks dragging out Shannon/Nyquist and saying "see digital IS perfect." I would like to see a direct quote from someone who made this assertion. If you are asking for one I'll just show you one from one of my favorite sources of misinformation. "The Nyquist theorem (which is mathematically proven) says that the exact waveform can be reproduced if the original signal is frequency limited to less than half the sampling frequency." http://www.hydrogenaudio.org/forums/...yquist+perfect IMO "exact" and "perfect" are synonymous as used here. I can find more but you only asked for one. Fine, several poiunts. The quote you supplied does NOT say that "digital is perfect." In effect it does. Second, exactly what is the myth, misinformation, whatever, in statement you quoted, as you quoted it? Do you have reason to believe that the Shannon/Nyquist sampling theorem is incorrect? No. But this represents exactly what I was talking about. It is in reference to actual digitization of an actual analog signal. So it is exactly the myth I claim is constantly dragged out. The myth that one can cite Shannon/Nyquist in support of the incorrect belief that "the EXACT WAVEFORM CAN BE REPRODUCED if the ORIGINAL (analog) SIGNAL is frequency limited to less than half the sampling frequency. So while I don't believe there is a problem with Shannon/Nyquist theorem I do see a problem with the claim that it is proof that one can "exactly reproduce" the "original signal" digitally. I hope that clears things up. |
#50
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On Thu, 14 Feb 2013 03:43:05 +0000, Dick Pierce wrote:
Edmund wrote: On Tue, 12 Feb 2013 18:35:04 +0000, Dick Pierce wrote: I can say with some certainty that there IS information at those frequencies on almost every LP I have ever examined, and I can say with equal cerainty that those signals have little if anything to do with the original signal AND they were very likely NOT present when fed to the cutter head. They consist of noise, distortion artifacts and the like. Do you know if there are records with real music recorded in high(er) frequencies to what frequency and is there is such information somewhere to be found on the internet? What about the half speed cut records? Sure, it's possible that it's the one cannot a priori discount that possibility. But two points: first, the RIAA curve is undefined past 20 kHz, and any number of preamps throw another pole or two in to reduce the bandwidth even further. Second, it's REALLY, REALLY hard to get the stylus tip resonance much above 20 kHz, and that resonance represents yet another 12 dB/octave low-pass filter. So, it's on there, so what, it's not likely you can get it off. Please, the effective "analog rolloff" is not some simple, nice 6- or 12-dB/octave, it's MUCH greater than that and VERY messy. I appreciate what you are saying, really but I wonder how and why some analog recordings sound so good... First, define "so good." I might hazard to say that "so good" means "I like it a lot." A perfectly legitimate definition, but one that has definiable objective context behind it as it stands. To my knowledge there is no unambiguous way to measure the quality of reproduced music so I mean it in a pure subjective manner. Secondly, you seem to be equating extended bandwith with "so good", as if this were the sole or at least principle criteria defining what "sounds good." It's not. I really just think it is at least ONE of the things that matter. There are ALL sorts of phenomenon that exist well within the audio bandwidth that could be the source of such an evocation. Have you eliminatd all of those as a possibility. What if the technical properties of the recording were, in fact, abysmal, yet the music, the performance, even the album cover, overwhelmed the objective sound attributes? Third, where is the data suggesting even a correlation between those analog recordings that "sound so good" and extended bandwidth well above 20 kHz. Can you, in fact, show that the recordings that "sound so good" have this information on them. I do have heard a few SACD recordings from the concert hall in Amsterdam which sound really good in to ears but I have no means to measure what is really recorded. Conversely, can you show that recordings that have information extending well beyond 20 kHz necessarily "sound so good?" No I cannot, I am curious and I like to know. And as a matter of fact I am planning to do some research. I think the Ansermet Beethoven 7th is one of those recordings that "sound so good," yet I also know that the top end is fairly limited (I've, in fact, measured it). It drops like a stone above about 16 kHz, and, indeed, except for noise, the specturm looks like it came out of a digital channel running at 38 kHz with a fast but sloppy 16 kHz anti-aliasing filter. Yet it sounds so good, so much so that I'd rather listen to the LP than and CD, simply because, to me, it is far an away the best performance, to me. When and if I ever find a CD of THAT performance, my opinion may change. But it is MY opinion, dammit! OK thanks Edmund |
#51
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On Wednesday, February 13, 2013 8:05:12 PM UTC-8, KH wrote:
On 2/13/2013 3:10 PM, Scott wrote: On Wednesday, February 13, 2013 11:54:15 AM UTC-8, Dick Pierce wrote: Scott wrote: Maybe in some other neck of the woods. But all too often I see some folks dragging out Shannon/Nyquist and saying "see digital IS perfect." I would like to see a direct quote from someone who made this assertion. If you are asking for one I'll just show you one from one of my favorite sources of misinformation. "The Nyquist theorem (which is mathematically proven) says that the exact waveform can be reproduced if the original signal is frequency limited to less than half the sampling frequency." http://www.hydrogenaudio.org/forums/...yquist+perfect IMO "exact" and "perfect" are synonymous as used here. I can find more but you only asked for one. You seem to have left out the clear caveat in the very next sentence; "The word "exact" gets a little shaky if the initial assumptions aren't met (example: each sample is taken exactly on time.)" Seriously? You think that makes it correct? You think that is all it takes? from which it is abundantly clear that the OP was decidedly Not implying "digital is perfect", merely that if "done perfectly" - an impossibility - the resulting waveform would be perfect, which IS clearly supported by information theory. Really? So you don't believe in quantization error? And of course, you excised the context of the statement as well, in that it was a response to the ludicrous claim that "...converting an analog signal into a discrete-time one (as it happens when converting from analog to digital) destroys the phase information in the two top octaves of the resulting spectrum. In a CD-standard digital recording, all phase information are lost from 5.5kHz up to 22kHz," I don't see how the quote you provided has ANY relevance to your claim. People often see what they want to see. |
#52
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On Wednesday, February 13, 2013 7:43:05 PM UTC-8, Dick Pierce wrote:
Edmund wrote: On Tue, 12 Feb 2013 18:35:04 +0000, Dick Pierce wrote: I can say with some certainty that there IS information at those frequencies on almost every LP I have ever examined, and I can say with equal cerainty that those signals have little if anything to do with the original signal AND they were very likely NOT present when fed to the cutter head. They consist of noise, distortion artifacts and the like. Do you know if there are records with real music recorded in high(er) frequencies to what frequency and is there is such information somewhere to be found on the internet? What about the half speed cut records? Sure, it's possible that it's the one cannot a priori discount that possibility. But two points: first, the RIAA curve is undefined past 20 kHz, and any number of preamps throw another pole or two in to reduce the bandwidth even further. Second, it's REALLY, REALLY hard to get the stylus tip resonance much above 20 kHz, and that resonance represents yet another 12 dB/octave low-pass filter. So, it's on there, so what, it's not likely you can get it off. Please, the effective "analog rolloff" is not some simple, nice 6- or 12-dB/octave, it's MUCH greater than that and VERY messy. I appreciate what you are saying, really but I wonder how and why some analog recordings sound so good... First, define "so good." I might hazard to say that "so good" means "I like it a lot." A perfectly legitimate definition, but one that has definiable objective context behind it as it stands. Secondly, you seem to be equating extended bandwith with "so good", as if this were the sole or at least principle criteria defining what "sounds good." It's not. There are ALL sorts of phenomenon that exist well within the audio bandwidth that could be the source of such an evocation. Have you eliminatd all of those as a possibility. What if the technical properties of the recording were, in fact, abysmal, yet the music, the performance, even the album cover, overwhelmed the objective sound attributes? Third, where is the data suggesting even a correlation between those analog recordings that "sound so good" and extended bandwidth well above 20 kHz. Can you, in fact, show that the recordings that "sound so good" have this information on them. Of course not. In fact, even if they did, it would be irrelevant because humans can't hear that high. There is some evidence to support that young female children, below the age of puberty, can hear 23 -25 KHz reliably, and as young adults (if they haven't had their hearing damaged by long exposure to high-levels of sound pressure) can hear 20 KHz and can keep their acuity to those frequencies longer than can young men. OTOH, such good high frequency hearing doesn't seem to make females susceptible to being audiophiles, which leads me to believe that such hearing does not make music sound any better to young women that it does for young men. But more importantly, even if some recordings do have response beyond 20 KHz, there's damn little actual information up there. This is because the recording systems used to capture and mass produce music simply does not capture information much beyond 20 KHz because (A) it's difficult, and (B) humans can't hear it anyway, even if musical instruments do actually produce it. Conversely, can you show that recordings that have information extending well beyond 20 kHz necessarily "sound so good?" No. I think the Ansermet Beethoven 7th is one of those recordings that "sound so good," yet I also know that the top end is fairly limited (I've, in fact, measured it). It drops like a stone above about 16 kHz, and, indeed, except for noise, the specturm looks like it came out of a digital channel running at 38 kHz with a fast but sloppy 16 kHz anti-aliasing filter. I don't doubt it. Older condenser microphones have a peak at around 14-16 KHz and fall off like that proverbial stone after that peak. The tape recorder used to capture the Ansermat Beethoven 7th's round-trip frequency response likewise probably falls off very quickly above 15 KHz for reasons of both practicality and physics. Yet it sounds so good, so much so that I'd rather listen to the LP than and CD, simply because, to me, it is far and away the best performance, to me. Yes, I have a copy and I agree that it is a marvelous performance, but to me there is another just as good - Bruno Walter and the New York Philharmonic. I too have performance that to me sound better on LP than on CD (I've mentioned the Mercury Classic Records reissue of Stravinsky's "Firebird" with Dorati and the LSO pressed on single-side 200 gram vinyl and 45 RPM playback speed many times.It is the best sounding commercial recording that I have ever heard - bar none. The CD is absolutely anemic by comparison and both were mastered by the original producer (Wilma Cozert-Fine) from the same master tape!) When and if I ever find a CD of THAT performance, my opinion may change. That's on British Decca (London Records) label isn't it? Most of Decca's classical fare has been released to CD. I'm surprised it's not available. You got me interested, I will pause here while I go look. Be back in a mo... OK I found it. From Arkive Music http://tinyurl.com/d9x3abd Label: London Eloquence - Catalog: 4800394. Release Date: 08/04/2009. Number of Discs: 2 Composer: Ludwig van Beethoven "Symphony 5 through 8" Conductor: Ernest Ansermet Orchestra/Ensemble: Suisse Romande Orchestra $15.99 But it is MY opinion, dammit! I don't disagree (try Ansermet's "Three Cornered Hat". another of the great sounding Suisse Romande recordings). The late 50's Decca stereo recordings (made with the "Decca tree" three microphone arrangement) all sound spectacular. Simple stereo mike setup, straight to tape with no fiddling from electronic enhancements, just the honest truth. That's why classical music audio enthusiasts still buy these 50+ year-old recordings from the likes of Mercury, RCA Red Seal, London, Columbia, etc. No recordings of these works ever sounded better (and most not as good, in spite of the "advances" in recording technology in the ensuing decades). Add to that the great conductors from the early part of the 20th century such as Ansermet, Ormandy, Walter, Szell, Boult, Furtwangler, Von Karajan, Reiner, etc. and you have the best recordings technically and performance-wise. |
#53
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
Scott wrote:
On Wednesday, February 13, 2013 7:41:41 PM UTC-8, Dick Pierce wrote: Scott wrote: "The Nyquist theorem (which is mathematically proven) says that the exact waveform can be reproduced if the original signal is frequency limited to less than half the sampling frequency." The quote you supplied does NOT say that "digital is perfect." In effect it does. To you. It does not to me. It simply, to me, states that when the nyquist criteria is met, and that means the signal must be limited to less than half the sampling rate, samplig does NOT lose any information needed to reproduce an exact replica of the signal meeting the criterion. Second, exactly what is the myth, misinformation, whatever, in statement you quoted, as you quoted it? Do you have reason to believe that the Shannon/Nyquist sampling theorem is incorrect? No. You proceed to contradict yourself: But this represents exactly what I was talking about. It is in reference to actual digitization of an actual analog signal. So it is exactly the myth I claim is constantly dragged out. The myth that one can cite Shannon/Nyquist in support of the incorrect belief that "the EXACT WAVEFORM CAN BE REPRODUCED if the ORIGINAL (analog) SIGNAL is frequency limited to less than half the sampling frequency. Please, with some degree of rigor, demonstrate why you claim this is a myth. So while I don't believe there is a problem with Shannon/Nyquist theorem Then it can reproduce the waveform. I do see a problem with the claim that it is proof that one can "exactly reproduce" the "original signal" digitally. I hope that clears things up. It does. With all due respect, and with no intent to cast insult, it seem to clarify that either you do not understand the sampling theorem, or that you are inarticulate about your belief in its shortcomings that your view is not getting across. -- +--------------------------------+ + Dick Pierce | + Professional Audio Development | +--------------------------------+ |
#54
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
Audio_Empire wrote:
On Wednesday, February 13, 2013 7:43:05 PM UTC-8, Dick Pierce wrote: I think the Ansermet Beethoven 7th is one of those recordings that "sound so good," yet I also know that the top end is fairly limited (I've, in fact, measured it). It drops like a stone above about 16 kHz, and, indeed, except for noise, the specturm looks like it came out of a digital channel running at 38 kHz with a fast but sloppy 16 kHz anti-aliasing filter. I don't doubt it. Older condenser microphones have a peak at around 14-16 KHz and fall off like that proverbial stone after that peak. The tape recorder used to capture the Ansermat Beethoven 7th's round-trip frequency response likewise probably falls off very quickly above 15 KHz for reasons of both practicality and physics. I think you missed the point of the context I introduced. I like this recording exclusively because of the performance. The technology details and the technological failings and limitations are, to me, utterly irrelevant. The performance transcends all that. I can think of a number of other examples, Bernstein's Shostakovich 5th: an absolutely exhilarating performance cmopletely unequaled by any other (although in this case, the one pressing I have is so BADLY mastered it is sometimes difficult difficult to listen through: it was a perfect macth for an old Quad 33 with it's nice tone controls). to sit through) When and if I ever find a CD of THAT performance, my opinion may change. That's on British Decca (London Records) label isn't it? Most of Decca's classical fare has been released to CD. I'm surprised it's not available. You got me interested, I will pause here while I go look. Be back in a mo... Actually, I was also inspired to go look, and I found the complete Ansermet Beethoven Symphony set and bought it: $0 bucks and it will be here by Saturday! But it is MY opinion, dammit! I don't disagree (try Ansermet's "Three Cornered Hat". another of the great sounding Suisse Romande recordings). The late 50's Decca stereo recordings (made with the "Decca tree" three microphone arrangement) all sound spectacular. Simple stereo mike setup, straight to tape with no fiddling from electronic enhancements, just the honest truth. That's why classical music audio enthusiasts still buy these 50+ year-old recordings from the likes of Mercury, RCA Red Seal, London, Columbia, etc. No recordings of these works ever sounded better (and most not as good, in spite of the "advances" in recording technology in the ensuing decades). Add to that the great conductors from the early part of the 20th century such as Ansermet, Ormandy, Walter, Szell, Boult, Furtwangler, Von Karajan, Reiner, etc. and you have the best recordings technically and performance-wise. I have been forced on a number of occasions to sit through the Berlioz Symphonie Fantastique, which I found to be, well, silly at best (if you can't do a fugue, please don't try). Late Sunday, on WGBH, on a program called "The BSO on record," they played what was one of the last recordings made by Charles Munsch with the BSO: it was of the Berlioz, and was at once a thrilling and sensitive performance and actually sounded quite nice as well. Enough so that I may actually get a copy. Now, to put this into some perspective, if I acquire such a disk, my Berlioz section will grow to 1/6th the size of my Jan Pieterszoon Sweelink section, and perhaps 1/20th my Diederich Buxtehude section (quick anecdote, I was at the Harvard Coop and grabbed a boxed set of the complete organ works by Bustehude performed by Andre Isoir: 7 disks and their scanner insisted it was one. Lovely set for 85% off) -- +--------------------------------+ + Dick Pierce | + Professional Audio Development | +--------------------------------+ |
#55
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
Scott wrote:
On Wednesday, February 13, 2013 8:05:12 PM UTC-8, KH wrote: On 2/13/2013 3:10 PM, Scott wrote: from which it is abundantly clear that the OP was decidedly Not implying "digital is perfect", merely that if "done perfectly" - an impossibility - the resulting waveform would be perfect, which IS clearly supported by information theory. Really? So you don't believe in quantization error? Hold it, part of your confusionm is clear. Several points: 1. Sampling and quantization are two different processes, that can be treated completely separate from one another. Sampling does not case quantization errors. It can create errors IF the original signal is improperly band limited before sampling. If properly band limited, it will not create errors. Now, you can argue whether or not the original filtering is the issue, but that's a separate argument, in fact. 2. Quantization without at least simple dithering, yes, does create quantization errors, but NO ONE in their right minds (high end community excepted, of course) does such a thing. And propoer dithering or noise shaping eliminates (not, not masks, not covers up, rather ELIMINATES as in "makes it go away) quantization error. WHat you are left with is a simple noise floor with resolution extending well below the least significant bit of resoltuion. And of course, you excised the context of the statement as well, in that it was a response to the ludicrous claim that "...converting an analog signal into a discrete-time one (as it happens when converting from analog to digital) destroys the phase information in the two top octaves of the resulting spectrum. In a CD-standard digital recording, all phase information are lost from 5.5kHz up to 22kHz," I don't see how the quote you provided has ANY relevance to your claim. People often see what they want to see. Yes, apparently they do. -- +--------------------------------+ + Dick Pierce | + Professional Audio Development | +--------------------------------+ |
#56
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On Thu, 14 Feb 2013 22:14:07 +0000, Dick Pierce wrote:
Scott wrote: On Wednesday, February 13, 2013 7:41:41 PM UTC-8, Dick Pierce wrote: Scott wrote: "The Nyquist theorem (which is mathematically proven) says that the exact waveform can be reproduced if the original signal is frequency limited to less than half the sampling frequency." The quote you supplied does NOT say that "digital is perfect." In effect it does. To you. It does not to me. It simply, to me, states that when the nyquist criteria is met, and that means the signal must be limited to less than half the sampling rate, samplig does NOT lose any information needed to reproduce an exact replica of the signal meeting the criterion. So if I offer you an analog signal, limited to 40kHz you can sample that at 80 kHz and 4 bit, you can recreate the input signal exactly? Well, would you like to prove that? Edmund |
#57
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On Thursday, February 14, 2013 2:14:07 PM UTC-8, Dick Pierce wrote:
Scott wrote: =20 On Wednesday, February 13, 2013 7:41:41 PM UTC-8, Dick Pierce wrote: =20 Scott wrote: =20 "The Nyquist theorem (which is mathematically proven) says=20 =20 that the exact waveform can be reproduced if the original =20 signal is frequency limited to less than half the sampling frequency.= " =20 The quote you supplied does NOT say that "digital is perfect."=20 =20 In effect it does. =20 =20 =20 To you. It does not to me. It simply, to me, states that =20 when the nyquist criteria is met, and that means the =20 signal must be limited to less than half the sampling rate, =20 samplig does NOT lose any information needed to reproduce =20 an exact replica of the signal meeting the criterion. =20 =20 =20 Second, exactly what is the myth, misinformation, whatever, =20 in statement you quoted, as you quoted it? Do you have =20 reason to believe that the Shannon/Nyquist sampling theorem =20 is incorrect? =20 =20 =20 No.=20 =20 =20 =20 You proceed to contradict yourself: =20 =20 =20 But this represents exactly what I was talking about.=20 =20 It is in reference to actual digitization of an actual =20 analog signal. So it is exactly the myth I claim is =20 constantly dragged out. The myth that one can cite =20 Shannon/Nyquist in support of the incorrect belief that =20 "the EXACT WAVEFORM CAN BE REPRODUCED if the ORIGINAL =20 (analog) SIGNAL is frequency limited to less than half the =20 sampling frequency.=20 =20 =20 =20 Please, with some degree of rigor, demonstrate why you =20 claim this is a myth. =20 =20 =20 So while I don't believe there is a problem with =20 Shannon/Nyquist theorem =20 =20 =20 Then it can reproduce the waveform. =20 =20 =20 I do see a problem with the claim that it is proof that one =20 can "exactly reproduce" the "original signal" digitally. =20 =20 =20 I hope that clears things up.=20 =20 =20 =20 It does. With all due respect, and with no intent to cast insult, =20 it seem to clarify that either you do not understand the sampling =20 theorem, or that you are inarticulate about your belief in its =20 shortcomings that your view is not getting across. =20 Didn't we just go over quantization error? That is the one and only answer = to all your questions above. The FACT is no digital system can "exactly" re= produce an analog signal and dragging out Nyquist in support of the notion = that any digital system *can* "exactly" reproduce an analog signal is an a= udio myth.=20 Now I am done with this little side bar.=20 |
#58
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On Thursday, February 14, 2013 3:36:42 PM UTC-8, Dick Pierce wrote:
Audio_Empire wrote: On Wednesday, February 13, 2013 7:43:05 PM UTC-8, Dick Pierce wrote: I think the Ansermet Beethoven 7th is one of those recordings that "sound so good," yet I also know that the top end is fairly limited (I've, in fact, measured it). It drops like a stone above about 16 kHz, and, indeed, except for noise, the specturm looks like it came out of a digital channel running at 38 kHz with a fast but sloppy 16 kHz anti-aliasing filter. I don't doubt it. Older condenser microphones have a peak at around 14-16 KHz and fall off like that proverbial stone after that peak. The tape recorder used to capture the Ansermat Beethoven 7th's round-trip frequency response likewise probably falls off very quickly above 15 KHz for reasons of both practicality and physics. I think you missed the point of the context I introduced. I like this recording exclusively because of the performance. The technology details and the technological failings and limitations are, to me, utterly irrelevant. The performance transcends all that. I can think of a number of other examples, Bernstein's Shostakovich 5th: an absolutely exhilarating performance cmopletely unequaled by any other (although in this case, the one pressing I have is so BADLY mastered it is sometimes difficult difficult to listen through: it was a perfect macth for an old Quad 33 with it's nice tone controls). to sit through) Well. if I did miss your point, it was because you stated it poorly. You said that it was a great sounding recording in spite of the fact that it had little content above about 16 KHz. That looks to me like you were talking about the SOUND of the record, not the performance. I understand the idea of the sound being utterly irrelevant, to your enjoyment of the performance, but you didn't say that. I have a bunch of 78's (transcribed to CD of course, the 78's themselves are too much trouble to listen to) that have great performances that I listen to in spite of the limitations of the medium. (Vaughan Williams' Oboe concerto on early 1950's Mercury 78's with the oboe solo played by Mitch Miller (yes, THAT Mitch Miller) with the Cleveland Orchestra, Louis Lane conducting comes immediately to mind, here). When and if I ever find a CD of THAT performance, my opinion may change. That's on British Decca (London Records) label isn't it? Most of Decca's classical fare has been released to CD. I'm surprised it's not available. You got me interested, I will pause here while I go look. Be back in a mo... Actually, I was also inspired to go look, and I found the complete Ansermet Beethoven Symphony set and bought it: $0 bucks and it will be here by Saturday! But it is MY opinion, dammit! I don't disagree (try Ansermet's "Three Cornered Hat". another of the great sounding Suisse Romande recordings). The late 50's Decca stereo recordings (made with the "Decca tree" three microphone arrangement) all sound spectacular. Simple stereo mike setup, straight to tape with no fiddling from electronic enhancements, just the honest truth. That's why classical music audio enthusiasts still buy these 50+ year-old recordings from the likes of Mercury, RCA Red Seal, London, Columbia, etc. No recordings of these works ever sounded better (and most not as good, in spite of the "advances" in recording technology in the ensuing decades). Add to that the great conductors from the early part of the 20th century such as Ansermet, Ormandy, Walter, Szell, Boult, Furtwangler, Von Karajan, Reiner, etc. and you have the best recordings technically and performance-wise. I have been forced on a number of occasions to sit through the Berlioz Symphonie Fantastique, which I found to be, well, silly at best (if you can't do a fugue, please don't try). Late Sunday, on WGBH, on a program called "The BSO on record," they played what was one of the last recordings made by Charles Munsch with the BSO: it was of the Berlioz, and was at once a thrilling and sensitive performance and actually sounded quite nice as well. Enough so that I may actually get a copy. I love WGBH/WCRB radio. It is on my Squeezebox Touch all the time and I listen to the station almost exclusively for classics. I think it is probably the best classical music station in America. Love the BSO and I also love the archival BSO recordings that they play. Some are from record/CD others are past live concerts either from Symphony Hall in the winter and Tanglewood in the summer. Great stuff! a real American treasure. So glad it's available via Internet "radio". I don't even have an FM tuner in my stereo system any more. My Yamaha T-85 (one of the best "digital" tuners ever made), just sits in my closet, unused. Now, to put this into some perspective, if I acquire such a disk, my Berlioz section will grow to 1/6th the size of my Jan Pieterszoon Sweelink section, and perhaps 1/20th my Diederich Buxtehude section (quick anecdote, I was at the Harvard Coop and grabbed a boxed set of the complete organ works by Bustehude performed by Andre Isoir: 7 disks and their scanner insisted it was one. Lovely set for 85% off) Berlioz fails to be one of my favorite composers as well. In fact, I doubt if I have more than a couple Berlioz works in my collection. Yes, I have "Symphony Fantastique" and I think I have a disc of Berlioz overtures, but that's about all I can recall. My tastes run to Beethoven, Dvorak, Rachmaninoff, Richard Strauss, Wagner (orchestra) Tchaikovsky, Ravel, Vaughan Williams Holst, Debussy, Puccini, Verde, etc. |
#59
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On Thursday, February 14, 2013 2:14:07 PM UTC-8, Dick Pierce wrote:
Scott wrote: On Wednesday, February 13, 2013 7:41:41 PM UTC-8, Dick Pierce wrote: Scott wrote: "The Nyquist theorem (which is mathematically proven) says that the exact waveform can be reproduced if the original signal is frequency limited to less than half the sampling frequency." The quote you supplied does NOT say that "digital is perfect." In effect it does. To you. It does not to me. It simply, to me, states that when the nyquist criteria is met, and that means the signal must be limited to less than half the sampling rate, samplig does NOT lose any information needed to reproduce an exact replica of the signal meeting the criterion. If I might chime in here. That's not exactly correct. It is correct as far as it goes, but I'm sure that you didn't mean to infer that a musical performance recorded in 8-bit 32 KHz sampling rate is going adequately reconstruct the actual original waveform? Even given that the highs would be truncated at about 15 KHz, which was once considered part of the definition of High- Fidelity, the dynamic range of such a quantization would be limited to about 48 dB and distortion would be very high compared to 16-bit, 44.1 KHz. At one time it was postulated that 8-bit, 32 KHz could work as a viable consumer medium IF the analog signal were compressed a la DBX before quantization and then expanded by the same ratio after the D/A conversion on playback. The expander would be a part of the player. Second, exactly what is the myth, misinformation, whatever, in statement you quoted, as you quoted it? Do you have reason to believe that the Shannon/Nyquist sampling theorem is incorrect? No. You proceed to contradict yourself: But this represents exactly what I was talking about. It is in reference to actual digitization of an actual analog signal. So it is exactly the myth I claim is constantly dragged out. The myth that one can cite Shannon/Nyquist in support of the incorrect belief that "the EXACT WAVEFORM CAN BE REPRODUCED if the ORIGINAL (analog) SIGNAL is frequency limited to less than half the sampling frequency. ONLY if the sampling frequency and bit depth were adequate to encompass the bandwidth of the signal being sampled. Without that condition, one could argue that a modern telephone system could reconstruct a symphony orchestra waveform completely and perfectly, but of course, we all know it can't. It was designed to have enough bandwidth and dynamic range to encompass voice, but no more. But if we assume that a digital system is designed to encompass the entire audio spectrum and is used to that end, then Nyquist/Shannon in quite correct in anticipating that the outcome of the applied theorem will be. It is NOT a myth or an overstatement of capability in any way shape or form. |
#60
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On Thursday, February 14, 2013 7:40:47 PM UTC-8, Dick Pierce wrote:
Scott wrote: On Wednesday, February 13, 2013 8:05:12 PM UTC-8, KH wrote: On 2/13/2013 3:10 PM, Scott wrote: from which it is abundantly clear that the OP was decidedly Not implying "digital is perfect", merely that if "done perfectly" - an impossibility - the resulting waveform would be perfect, which IS clearly supported by information theory. Really? So you don't believe in quantization error? Hold it, part of your confusionm is clear. Several points: 1. Sampling and quantization are two different processes, that can be treated completely separate from one another. Sampling does not case quantization errors. It can create errors IF the original signal is improperly band limited before sampling. If properly band limited, it will not create errors. Now, you can argue whether or not the original filtering is the issue, but that's a separate argument, in fact. 2. Quantization without at least simple dithering, yes, does create quantization errors, but NO ONE in their right minds (high end community excepted, of course) does such a thing. And propoer dithering or noise shaping eliminates (not, not masks, not covers up, rather ELIMINATES as in "makes it go away) quantization error. WHat you are left with is a simple noise floor with resolution extending well below the least significant bit of resoltuion. And of course, you excised the context of the statement as well, in that it was a response to the ludicrous claim that "...converting an analog signal into a discrete-time one (as it happens when converting from analog to digital) destroys the phase information in the two top octaves of the resulting spectrum. In a CD-standard digital recording, all phase information are lost from 5.5kHz up to 22kHz," My point has nothing to do with any of that. The fact is any AD converter is going to be making quantization errors when sampling any analog signal. Nyquist theorem assumes no such error. So in practice no AD converter will ever "exactly" encode an analog signal despite the truth of Nyquist theorem. [quoted text deleted -- deb] |
#61
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
Edmund wrote:
On Thu, 14 Feb 2013 22:14:07 +0000, Dick Pierce wrote: Scott wrote: On Wednesday, February 13, 2013 7:41:41 PM UTC-8, Dick Pierce wrote: Scott wrote: "The Nyquist theorem (which is mathematically proven) says that the exact waveform can be reproduced if the original signal is frequency limited to less than half the sampling frequency." The quote you supplied does NOT say that "digital is perfect." In effect it does. To you. It does not to me. It simply, to me, states that when the nyquist criteria is met, and that means the signal must be limited to less than half the sampling rate, samplig does NOT lose any information needed to reproduce an exact replica of the signal meeting the criterion. So if I offer you an analog signal, limited to 40kHz you can sample that at 80 kHz and 4 bit, you can recreate the input signal exactly? Well, would you like to prove that? You just went and changed the conditions. To this point, all the discussion had to do with sampling. One other poster is clearly confused about the difference between sampling in the time domain and quantization in the amplitude domain, and it appears that you have similarly bundled the two together as well. Let's examine your statement, for it contains several errors: "So if I offer you an analog signal, limited to 40kHz you can sample that at 80 kHz First error, the bandwidth must be limited to LESS THAN HALF the sample rate, NOT half the sample rate. "and 4 bit" Second error, just like in the sampling error above, you made an assumption about the signal which may, in fact, not be true. Now IF the dynamic range is limited to 24 dB, then with proper dithering, yes, all of the information present in the original signal WILL be available at the output of the system. Now, the separation between sampling and quantization is NOT some clever symantic trick, rather it is at the very basis of the process. If it helps, you can think of the sampling process as quantization in the time domain, and what many here term "quantization" as quantization in the amplitude domain. Assuming the wo are inextricably tied together is the root of much confusion, as exhibited by your question as one example. Let's rephrase your question slightly, relating it to more practical terms: what is the capability of a sampling system with a sample rate of 88.2 kHz (twice Redbook CD rate) and a signed 16-bit linear integer quantization (redbook CD spec), using adequate dithering (or noise shaping)? Well, IF the input signal is limited to less than half the sample rate, let's say 40 kHz, and the dynamic range of the input signal does not exceed 96 dB, ALL the information in that signal will be captured and be available in the output of the system. NO information in either the time domain or the amplitude domain will be lost. Now, recall where I said that the resolution of the system is defined by the product of the bandwidth and quantization level? That allows us to trade off one form the other, for example, I could double the sample rate (to 176.4 kHz) and, in doing so, gain an extra 3 dB of resolution, and, assuming my signal is STILL limited to 40 kHz, I can now use a 15-bit encoding instead of 16 bit and still achieve that 40 kHz, 96 dB resolution with no information loss. Double it again (352.8 kHz), and I get to throw away another bit and still achieve the same base-band resolution (40 kHz, 96 dB). And, by the way, that 96 dB I stated is the BROADBAND dynamic range: it DOES NOT represent a hard floor like so many assume, it is simply the braodband noise floor of the system: valid signal information is still encoded BELOW that noise floor, so that narrow-band avergaing systems, like spectrum analyzers or human hearing, are QUITE capable of hearing the real signal BELOW that noise floor. -- +--------------------------------+ + Dick Pierce | + Professional Audio Development | +--------------------------------+ |
#62
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On Fri, 15 Feb 2013 15:04:03 +0000, Dick Pierce wrote:
Edmund wrote: On Thu, 14 Feb 2013 22:14:07 +0000, Dick Pierce wrote: Scott wrote: On Wednesday, February 13, 2013 7:41:41 PM UTC-8, Dick Pierce wrote: Scott wrote: "The Nyquist theorem (which is mathematically proven) says that the exact waveform can be reproduced if the original signal is frequency limited to less than half the sampling frequency." The quote you supplied does NOT say that "digital is perfect." In effect it does. To you. It does not to me. It simply, to me, states that when the nyquist criteria is met, and that means the signal must be limited to less than half the sampling rate, samplig does NOT lose any information needed to reproduce an exact replica of the signal meeting the criterion. So if I offer you an analog signal, limited to 40kHz you can sample that at 80 kHz and 4 bit, you can recreate the input signal exactly? Well, would you like to prove that? You just went and changed the conditions. To this point, all the discussion had to do with sampling. One other poster is clearly confused about the difference between sampling in the time domain and quantization in the amplitude domain, and it appears that you have similarly bundled the two together as well. Let's examine your statement, for it contains several errors: "So if I offer you an analog signal, limited to 40kHz you can sample that at 80 kHz First error, the bandwidth must be limited to LESS THAN HALF the sample rate, NOT half the sample rate. I am not sure but If my memory serves me well one need at least twice the highest frequency as sample rate. In your terms LESS OR EQUAL TO HALF THE SAMPLE RATE. Correct me if you are sure that I am wrong. "and 4 bit" Second error, just like in the sampling error above, you made an assumption about the signal which may, in fact, not be true. Now IF the dynamic range is limited to 24 dB, then with proper dithering, yes, all of the information present in the original signal WILL be available at the output of the system. Well you did not mention the dynamic range at all. ---------------- "The Nyquist theorem (which is mathematically proven) says that the exact waveform can be reproduced if the original signal is frequency limited to less than half the sampling frequency." ------------------------ And I don't think that is right. Now, the separation between sampling and quantization is NOT some clever symantic trick, rather it is at the very basis of the process. If it helps, you can think of the sampling process as quantization in the time domain, and what many here term "quantization" as quantization in the amplitude domain. Assuming the wo are inextricably tied together is the root of much confusion, as exhibited by your question as one example. Let's rephrase your question slightly, No lets not do that. Stick to what I asked which is an analog signal. Edmund |
#63
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
Audio_Empire wrote:
On Thursday, February 14, 2013 2:14:07 PM UTC-8, Dick Pierce wrote: Scott wrote: On Wednesday, February 13, 2013 7:41:41 PM UTC-8, Dick Pierce wrote: Scott wrote: "The Nyquist theorem (which is mathematically proven) says that the exact waveform can be reproduced if the original signal is frequency limited to less than half the sampling frequency." The quote you supplied does NOT say that "digital is perfect." In effect it does. To you. It does not to me. It simply, to me, states that when the nyquist criteria is met, and that means the signal must be limited to less than half the sampling rate, samplig does NOT lose any information needed to reproduce an exact replica of the signal meeting the criterion. If I might chime in here. That's not exactly correct. It is correct as far as it goes, but I'm sure that you didn't mean to infer that a musical performance recorded in 8-bit 32 KHz sampling rate is going adequately reconstruct the actual original waveform? Even given that the highs would be truncated at about 15 KHz, which was once considered part of the definition of High- Fidelity, the dynamic range of such a quantization would be limited to about 48 dB and distortion would be very high compared to 16-bit, 44.1 KHz. There seems to be a general,confusion he sampling and quantization ARE |NOT the same thing: they are two separate processes. You cna have one without the other, and you can certainly treat and explore the two separately (despite the fact that in most audio systems, the two work together). As an aside, let me give some examples of one without the other: first, a switched-capacitor filter is a discrete- time sampled, continuous-amplitude system. No quantization takes place. Another example in the audio realm is the old classic bucket-brigade CCD analog delay lines, again, a discrete-time sampled, continuous-amplitude system with no quantization. And, for a very common one, an FM stereo broadcast can be viewed as yet another discrete-time sampled time-division multiplexed, continuous amplitude system. ALL of these systems are sampled systems, but NONE of them have quantization. You don't talk about the "bit-depth" of FM radio (but, you most assuredly could, since their dynamic range has a bit-depth equivalent). And ALL of these systems require the SAME constraints as a classic digital audio system: their input bandwidth MUST be limited to less than half the sample rate. In the case of FM stereo, whose effective sample rate is 38 kHz, the bandwidth of the base channel (which holds the L+R information) and the subchannel (which holds the L-R information are both limited to 15 kHz or so. Look at ANY of the data sheets on analog bucket-brigade delay lines or switched-capacitor filters, they tell you that a low- pass filter is mandatory to prevent aliasing by the sampling process. And NONE of these examples are quantized. Let's explore a system closer to the topic at hand, a classic analog to digital converter, which does the digitization in TWO steps: First, you have your sample-and-hold amplifier. It's job is to simply sit there and listen to, but otherwise ignore the incoming signal voltage. Then at the moment commanded by the sample clock, it captures ("samples") the instantaneous voltage and "holds" on to it. So far, NO QUANTIZATION HAS OCCURED: THERE IS NO "BIT DEPTH" ISSUE. Now, if at this point, the sample-and-hold amplifier is presented with a signal exceeding less than half the sample clock rate, aliasing WILL occur: the Nyquist criteria has been violated. It's too late at this point to deal with out-of-band foldback to the base band, or "aliasing." It should have been taken care of BEFORE the sample-and-hold step. But once again: NO QUANTIZATION HAS YET OCCURRED, therefore the question of "bit depth" is completely irrelevant. Only after the sample and hold amplifier has captured and held a sample, which to this point is still a continuous- amplitude representation, do we start to think about quantization. Now, bit depth becomes an issue, and so does dealing wiuth quantization error. Pick a bit depth which you find suitable, say 16 bits. Now, if you do NOTHING else, you WILL get quantization errors. BBut NO A/D converter used for audio proceeds in this fashion: ALL of them apply dither and/or noise shaping BEFORE quantization. And the application of dither and/or noise chaping ELIMINATES quantization error. It does NOT cover it up, it does not mask it, it does not replace quantization error, IT ELIMINATES it. Now you have a stream of digital data which, if the bit depth exceeds the original dynamic range of the original signal AND the bandwidth is less than 1/2 the sample rate, ALL information in that signal WILL be captured. Nowhere, it should be noted, did I say that "digital audio is perfect", because just like the claims that analog has infinite resolution, that would require infinite sample rate and infinitie bit dpeth. Rather, I said that within the constraints set by the sample rate and the bit depth, a properly implemented PRACTICAL digital system IS capable of capturing ALL information in the signal presented to it. Religious-like, self-contradictory beliefs based on misunderstandings and obstinate opinions notwithstanding. ONLY if the sampling frequency and bit depth were adequate to encompass the bandwidth of the signal being sampled. Without that condition, one could argue that a modern telephone system could reconstruct a symphony orchestra waveform completely and perfectly, but of course, we all know it can't. Hold it, you have already violated the basic constraints, if you make the assumption that the minimum bandwidth required is 15 kHz or greater. No one said a telephone could, and Shannon and Nyquist tell us why. Is this a strawman? -- +--------------------------------+ + Dick Pierce | + Professional Audio Development | +--------------------------------+ |
#64
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On 2/14/2013 8:06 AM, Scott wrote:
On Wednesday, February 13, 2013 8:05:12 PM UTC-8, KH wrote: On 2/13/2013 3:10 PM, Scott wrote: On Wednesday, February 13, 2013 11:54:15 AM UTC-8, Dick Pierce wrote: Scott wrote: Maybe in some other neck of the woods. But all too often I see some folks dragging out Shannon/Nyquist and saying "see digital IS perfect." I would like to see a direct quote from someone who made this assertion. If you are asking for one I'll just show you one from one of my favorite sources of misinformation. "The Nyquist theorem (which is mathematically proven) says that the exact waveform can be reproduced if the original signal is frequency limited to less than half the sampling frequency." http://www.hydrogenaudio.org/forums/...yquist+perfect IMO "exact" and "perfect" are synonymous as used here. I can find more but you only asked for one. You seem to have left out the clear caveat in the very next sentence; "The word "exact" gets a little shaky if the initial assumptions aren't met (example: each sample is taken exactly on time.)" Seriously? You think that makes it correct? You think that is all it takes? Well, yes. In fact that is all that is required to make it "correct". It still says *nothing* about "digital is perfect". from which it is abundantly clear that the OP was decidedly Not implying "digital is perfect", merely that if "done perfectly" - an impossibility - the resulting waveform would be perfect, which IS clearly supported by information theory. Really? So you don't believe in quantization error? In "sampling"? No. And sampling is what that statement relates to. I believe Dick Pierce has sufficiently addressed that. And of course, you excised the context of the statement as well, in that it was a response to the ludicrous claim that "...converting an analog signal into a discrete-time one (as it happens when converting from analog to digital) destroys the phase information in the two top octaves of the resulting spectrum. In a CD-standard digital recording, all phase information are lost from 5.5kHz up to 22kHz," I don't see how the quote you provided has ANY relevance to your claim. People often see what they want to see. Yes, they do. And no one, with even a superficial objective reading, would construe the post you cited as saying "digital is perfect". Keith |
#65
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On Friday, February 15, 2013 5:36:20 PM UTC-8, Dick Pierce wrote:
Audio_Empire wrote: On Thursday, February 14, 2013 2:14:07 PM UTC-8, Dick Pierce wrote: Scott wrote: On Wednesday, February 13, 2013 7:41:41 PM UTC-8, Dick Pierce wrote: Scott wrote: "The Nyquist theorem (which is mathematically proven) says that the exact waveform can be reproduced if the original signal is frequency limited to less than half the sampling frequency." The quote you supplied does NOT say that "digital is perfect." In effect it does. To you. It does not to me. It simply, to me, states that when the nyquist criteria is met, and that means the signal must be limited to less than half the sampling rate, samplig does NOT lose any information needed to reproduce an exact replica of the signal meeting the criterion. If I might chime in here. That's not exactly correct. It is correct as far as it goes, but I'm sure that you didn't mean to infer that a musical performance recorded in 8-bit 32 KHz sampling rate is going adequately reconstruct the actual original waveform? Even given that the highs would be truncated at about 15 KHz, which was once considered part of the definition of High- Fidelity, the dynamic range of such a quantization would be limited to about 48 dB and distortion would be very high compared to 16-bit, 44.1 KHz. There seems to be a general,confusion he sampling and quantization ARE |NOT the same thing: they are two separate processes. You cna have one without the other, and you can certainly treat and explore the two separately (despite the fact that in most audio systems, the two work together). As an aside, let me give some examples of one without the other: first, a switched-capacitor filter is a discrete- time sampled, continuous-amplitude system. No quantization takes place. Another example in the audio realm is the old classic bucket-brigade CCD analog delay lines, again, a discrete-time sampled, continuous-amplitude system with no quantization. And, for a very common one, an FM stereo broadcast can be viewed as yet another discrete-time sampled time-division multiplexed, continuous amplitude system. Yes, I understand this difference. When I use the term quantization I mean the general process of converting an analog AC signal to a digital one. This is irrespective of the sampling rate which is driven by the bandwidth that needs to be quantized. ALL of these systems are sampled systems, but NONE of them have quantization. You don't talk about the "bit-depth" of FM radio (but, you most assuredly could, since their dynamic range has a bit-depth equivalent). And ALL of these systems require the SAME constraints as a classic digital audio system: their input bandwidth MUST be limited to less than half the sample rate. In the case of FM stereo, whose effective sample rate is 38 kHz, the bandwidth of the base channel (which holds the L+R information) and the subchannel (which holds the L-R information are both limited to 15 kHz or so. I also believe that Shannon had some input into the allocation of bandwidth for FM after the Second World War. as well. Look at ANY of the data sheets on analog bucket-brigade delay lines or switched-capacitor filters, they tell you that a low- pass filter is mandatory to prevent aliasing by the sampling process. And NONE of these examples are quantized. Understood. Like I said, I'm not confused. I was using the term to describe the digitizing of a finite analog AC signal and was not confusing quantization with sampling at all. If I expressed that poorly, then mia culpa, but I was not at all confused when I wrote it. |
#66
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On Friday, February 15, 2013 8:16:30 PM UTC-8, KH wrote:
On 2/14/2013 8:06 AM, Scott wrote: =20 On Wednesday, February 13, 2013 8:05:12 PM UTC-8, KH wrote: =20 On 2/13/2013 3:10 PM, Scott wrote: =20 On Wednesday, February 13, 2013 11:54:15 AM UTC-8, Dick Pierce wrote: =20 Scott wrote: =20 =20 Maybe in some other neck of the woods. But all too often I =20 see some folks dragging out Shannon/Nyquist and saying "see digital =20 IS perfect." =20 =20 I would like to see a direct quote from someone who =20 made this assertion. =20 =20 If you are asking for one I'll just show you one from one of my =20 favorite sources of misinformation. =20 "The Nyquist theorem (which is mathematically proven) says that =20 the exact waveform can be reproduced if the original signal is =20 frequency limited to less than half the sampling frequency." =20 =20 http://www.hydrogenaudio.org/forums/...3D98761&hl=3D= nyquist+perfect =20 =20 IMO "exact" and "perfect" are synonymous as used here. I can find =20 more but you only asked for one. =20 =20 You seem to have left out the clear caveat in the very next sentence; =20 "The word "exact" gets a little shaky if the initial assumptions aren'= t =20 met (example: each sample is taken exactly on time.)" =20 =20 Seriously? You think that makes it correct? You think that is all it ta= kes? =20 =20 =20 Well, yes. In fact that is all that is required to make it "correct".=20 =20 It still says *nothing* about "digital is perfect". No "in fact" that is not all it takes. If that were all it took then bit de= pth would be irrelevant to resolution.=20 =20 =20 =20 =20 =20 from which it is =20 abundantly clear that the OP was decidedly Not implying "digital is =20 perfect", merely that if "done perfectly" - an impossibility - the =20 resulting waveform would be perfect, which IS clearly supported by =20 information theory. =20 =20 Really? So you don't believe in quantization error? =20 =20 =20 In "sampling"? No. And sampling is what that statement relates to. I=20 =20 believe Dick Pierce has sufficiently addressed that. Well that seems to be the problem. The statement is limited to sampling rat= es and ignores the fact that quantization error is also a factor when it co= mes to Nyquist. There can be no quanitization error for Nyquist to give us = an "exact" copy of an analog signal. There is always some quantization erro= r. So the claim "The Nyquist theorem (which is mathematically proven) says = that the exact waveform can be reproduced if the original signal is frequen= cy limited to less than half the sampling frequency." Is an "audio" myth.It= doesn't happen in real AD conversion. Nyquist works perfectly and gives ex= act waveforms on a mathematical level not on a practical level. but heck, a= sine wave has infinite resolution on a mathematical level. On a mathematic= al level both *audio myths* are actually true. Now I look forward to the ar= guments that a sine wave doesn't have infinite resolution on a mathematical= level. That will be fun.=20 =20 =20 =20 =20 And of course, you excised the context of the statement as well, in th= at =20 it was a response to the ludicrous claim that "...converting an analog =20 signal into a discrete-time one (as it happens when converting from =20 analog to digital) destroys the phase information in the two top octav= es =20 of the resulting spectrum. In a CD-standard digital recording, all pha= se =20 information are lost from 5.5kHz up to 22kHz," =20 =20 I don't see how the quote you provided has ANY relevance to your claim= .. =20 =20 =20 People often see what they want to see. =20 =20 =20 Yes, they do. And no one, with even a superficial objective reading,=20 =20 would construe the post you cited as saying "digital is perfect". =20 =20 You really don't get to speak for anyone other than yourself.=20 |
#67
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On Saturday, February 16, 2013 6:37:47 AM UTC-8, Scott wrote:
snip In "sampling"? No. And sampling is what that statement relates to. I believe Dick Pierce has sufficiently addressed that. Well that seems to be the problem. The statement is limited to sampling rates and ignores the fact that quantization error is also a factor when it comes to Nyquist. There can be no quanitization error for Nyquist to give us an "exact" copy of an analog signal. There is always some quantization error. So the claim "The Nyquist theorem (which is mathematically proven) says that the exact waveform can be reproduced if the original signal is frequency limited to less than half the sampling frequency." Is an "audio" myth.It doesn't happen in real AD conversion. Nyquist works perfectly and gives exact waveforms on a mathematical level not on a practical level. but heck, a sine wave has infinite resolution on a mathematical level. On a mathematical level both *audio myths* are actually true. Now I look forward to the arguments that a sine wave doesn't have infinite resolution on a mathematical level. That will be fun. I suspect that you are putting too much emphasis on quantization error. It's been a long time since I studied the nuts and bolts of digital audio, but it seems to me that unless quantization error is greater than one LSB (Least Significant Bit), that it really has no effect on the reconstructed waveform. Modern ADCs and DAC chips are laser trimmed so that quantization error is kept low to non-existent, IIRC. [quoted text deleted -- deb] |
#68
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
Audio_Empire wrote:
On Saturday, February 16, 2013 6:37:47 AM UTC-8, Scott wrote: snip In "sampling"? No. And sampling is what that statement relates to. I believe Dick Pierce has sufficiently addressed that. Well that seems to be the problem. The statement is limited to sampling rates and ignores the fact that quantization error is also a factor when it comes to Nyquist. There can be no quanitization error for Nyquist to give us an "exact" copy of an analog signal. There is always some quantization error. So the claim "The Nyquist theorem (which is mathematically proven) says that the exact waveform can be reproduced if the original signal is frequency limited to less than half the sampling frequency." Is an "audio" myth.It doesn't happen in real AD conversion. Nyquist works perfectly and gives exact waveforms on a mathematical level not on a practical level. but heck, a sine wave has infinite resolution on a mathematical level. On a mathematical level both *audio myths* are actually true. Now I look forward to the arguments that a sine wave doesn't have infinite resolution on a mathematical level. That will be fun. I suspect that you are putting too much emphasis on quantization error. It's been a long time since I studied the nuts and bolts of digital audio, but it seems to me that unless quantization error is greater than one LSB (Least Significant Bit), that it really has no effect on the reconstructed waveform. Modern ADCs and DAC chips are laser trimmed so that quantization error is kept low to non-existent, IIRC. Uh, no. First, modern A/D converters used for audio are almost universally delta-signa converters, different than the typical combination of DAC, successive approximation register and comparator. The DAC in such a case would typically use an R-2R ladder for bit weighting and COULD benefit from laser trimming for linearity. But conversion linearity (e.g., the steps are equidistant and the conversion is monotonic and no missing codes, etc.) is not quiet what "qauntization error" is typically considered in this context. The quantization arizes from the fact that a once continuous (which is NOT the same as "infinite resolution) signal is now, well, quantized. If that's all the converter does, two attributes are noted: 1. the quantization process is essentially like a modulation (gee, that's why it's called PCM, or "pulse code modulation") and modulation creates spurious information. This is the so-called "stair-step" problem that so many in the high-end considered to be such a terrible disease, and 2. your converter design is defective. What is done is immediately BEFORE the quantization step, a process known (in the simplest case) as "dither" is performed. This is essentially the addition of a small amount of random noise, on the order of an LSB, to the signal BEFORE the quantizer. It's function is to essentially randomize the statistics of the quantizer. The result is that you ELIMINATE (that means, you make it go away) the quantization error in the qauntizer. ALL A/D converters used in audio have at least dither and, in most cases a process called "noise shaping" where, in essence, the error is fed back through the quantizer, with appropriate weighting, so that the resulting dither signal is added back into the input so as also to eliminate the quantization error (notice again, "eliminate"). Noise shaping is, in effect, different than straight dither because the noise spectrum of random dither is essentially white (equal energy at every frequency), whereas you chose the weighting of your noise shaper right, and you stick the energy at places where the ear is less sensitive to it. You, in essence, increase the noise floor at, oh, above 15 kHz, but reduce it at, say, 2 kHz - 7 kHz, where the ear is most sensitive. Now, intuitively, it might seem like magic, and those that are in to faith-based as opposed to fact-based audio simply will refuse to accept it. But a simple example will illustrate how it works, and the principle is quite straightforward. It depends upon the fact that the ear (and most measuring instruments) do not hear "instantaneous" signals, but rather detect the average properties of the filter (averaging is another word for filtering, and the ear is basically a spectrum ananlyzer, i.e. a series of narrow-band filters spaced out in frequency. By taking advantage of this averaging, dither can encode signals well below the least significant bit quantization level. Let's consider a sinple DC signal for our example. The smallest value ("least significant bit") our gedanken quantizer can encode is 1. Any value smaller than that ends up being 0. So, 20 consecutive samples of the value, say, 0.35, end up all being 0: # Signal Quantized 1 0.35 0 2 0.35 0 3 0.35 0 4 0.35 0 5 0.35 0 6 0.35 0 7 0.35 0 8 0.35 0 9 0.35 0 10 0.35 0 11 0.35 0 12 0.35 0 13 0.35 0 14 0.35 0 15 0.35 0 16 0.35 0 17 0.35 0 18 0.35 0 19 0.35 0 20 0.35 0 Averaged over those 20 snapshots, the result is 0. Clearly, the error in the quantizer has lost information in the original signal. Instead, let's add a small random number to the signal, and then quantize the sum: # Signal Dither Sig+dither Quantized 1 0.35 0.086643028 0.436643028 0 2 0.35 0.016667786 0.366667786 0 3 0.35 0.196228521 0.546228521 1 4 0.35 0.104346522 0.454346522 0 5 0.35 0.187354055 0.537354055 1 6 0.35 0.155220547 0.505220547 1 7 0.35 0.093829761 0.443829761 0 8 0.35 -0.015919348 0.334080652 0 9 0.35 0.066064189 0.416064189 0 10 0.35 0.157076587 0.507076587 1 11 0.35 -0.177857024 0.172142976 0 12 0.35 -0.155519641 0.194480359 0 13 0.35 -0.306383986 0.043616014 0 14 0.35 0.446211844 0.796211844 1 15 0.35 -0.351768339 -0.001768339 0 16 0.35 0.226796118 0.576796118 1 17 0.35 -0.022827209 0.327172791 0 18 0.35 0.427503922 0.777503922 1 19 0.35 0.081767267 0.431767267 0 20 0.35 0.049525623 0.399525623 0 THESE samples, which are no longer all 0's, but seem to "randomly" flip back and forth betweem 0 and 1, when averaged over the same set, result in, surprise, 0.35. That's because the flipping of the quantized values is NOT purely random, but rather is "weighted" to flip to 1 about 35% of the time by the signal value of 0.35. One might argue that I picked a trivial example of a single, constant value (a "DC voltage), and that music signals are something different. Yes, it's true I picked a single value, NOT because the math works any better, but simply becasue the example is much more obvious in a character-oriented presentation. Do the same thing with ANY real signal, and look (or, what the hell, listen) to the results, and you find the signal, right there, the quantization distortion is gone. Depending upon the kind of signal, you can here the signal FAR below the "quantization error" floor that at least one respondant insists MUST be there. Easily 20 dB lower for simple dither models (the example I did above uses +- 1/2 LSB of triangular probability distributed or TPD broadband dither). Gee, that would seem to suggest that with proper dither, a 16-bit system really is a 19 1/3 bit system. And, broadband, that's correct, and the limit is NOT the quantization error, but the perceptual noise floor. If "quantization error" were, in fact, happening as at least one respondant insists (with no proof, BTW), then there would be NOTHING below the LSB of the system. Indeed, people see what they want to see. Any number of religions rely on that failing of humans. -- +--------------------------------+ + Dick Pierce | + Professional Audio Development | +--------------------------------+ |
#69
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On Saturday, February 16, 2013 12:11:46 PM UTC-8, Audio_Empire wrote:
On Saturday, February 16, 2013 6:37:47 AM UTC-8, Scott wrote: snip In "sampling"? No. And sampling is what that statement relates to. I believe Dick Pierce has sufficiently addressed that. Well that seems to be the problem. The statement is limited to sampling rates and ignores the fact that quantization error is also a factor when it comes to Nyquist. There can be no quanitization error for Nyquist to give us an "exact" copy of an analog signal. There is always some quantization error. So the claim "The Nyquist theorem (which is mathematically proven) says that the exact waveform can be reproduced if the original signal is frequency limited to less than half the sampling frequency." Is an "audio" myth.It doesn't happen in real AD conversion. Nyquist works perfectly and gives exact waveforms on a mathematical level not on a practical level. but heck, a sine wave has infinite resolution on a mathematical level. On a mathematical level both *audio myths* are actually true. Now I look forward to the arguments that a sine wave doesn't have infinite resolution on a mathematical level. That will be fun. I suspect that you are putting too much emphasis on quantization error. It's been a long time since I studied the nuts and bolts of digital audio, but it seems to me that unless quantization error is greater than one LSB (Least Significant Bit), that it really has no effect on the reconstructed waveform. Modern ADCs and DAC chips are laser trimmed so that quantization error is kept low to non-existent, IIRC. It may be low but it is not non existent. |
#70
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On 2/17/2013 7:42 AM, Scott wrote:
On Saturday, February 16, 2013 12:11:46 PM UTC-8, Audio_Empire wrote: Modern ADCs and DAC chips are laser trimmed so that quantization error is kept low to non-existent, IIRC. It may be low but it is not non existent. Uhm, as Dick Pierce just pointed out, by providing a clear example of the mathematical mechanism involved (thx Dick, a simple and illustrative treatise), it *is* non-existent. How about you provide some evidence to the contrary? And no, that *still* doesn't imply that "digital is perfect". Keith |
#71
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On Sunday, February 17, 2013 10:26:26 AM UTC-8, KH wrote:
On 2/17/2013 7:42 AM, Scott wrote: On Saturday, February 16, 2013 12:11:46 PM UTC-8, Audio_Empire wrote: Modern ADCs and DAC chips are laser trimmed so that quantization error is kept low to non-existent, IIRC. It may be low but it is not non existent. Uhm, as Dick Pierce just pointed out, by providing a clear example of the mathematical mechanism involved (thx Dick, a simple and illustrative treatise), it *is* non-existent. How about you provide some evidence to the contrary? No that is not what he pointed out. He made a big deal about the terminology and pointed out how dither works to *lower* the distortion. Some distortion is still there. If you think it isn't you are simply wrong. |
#72
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
Scott wrote:
On Sunday, February 17, 2013 10:26:26 AM UTC-8, KH wrote: On 2/17/2013 7:42 AM, Scott wrote: On Saturday, February 16, 2013 12:11:46 PM UTC-8, Audio_Empire wrote: Modern ADCs and DAC chips are laser trimmed so that quantization error is kept low to non-existent, IIRC. It may be low but it is not non existent. Uhm, as Dick Pierce just pointed out, by providing a clear example of the mathematical mechanism involved (thx Dick, a simple and illustrative treatise), it *is* non-existent. How about you provide some evidence to the contrary? No that is not what he pointed out. He made a big deal about the terminology and pointed out how dither works to *lower* the distortion. Some distortion is still there. If you think it isn't you are simply wrong. He wasn't asking for more assertions, but evidence. I'd like to recommend that you read Resolution Below the Least Significant Bit in Digital Systems with Dither by Vanderkooy and Lip****z. This very famous paper (available on the Internet if you search for it) comes to the same conclusion as Dick Pierce: dither effectively turns [all of] the signal distortion caused by quantization into wide-band noise. If you can find any fault in that paper, it would be interesting to see you present it here. They say: We feel that the audio community in general does not yet understand the nature of quantization error in digital systems, and in particular the beneficial effects of adding an appropriate amount of dither. We shall show that dither really does remove the "digital" aspects of quantization error, leaving an equivalent analog signal with high resolution and some benign wide-band noise. Andrew. |
#73
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
"Scott" wrote in message
... On Sunday, February 17, 2013 10:26:26 AM UTC-8, KH wrote: On 2/17/2013 7:42 AM, Scott wrote: On Saturday, February 16, 2013 12:11:46 PM UTC-8, Audio_Empire wrote: Modern ADCs and DAC chips are laser trimmed so that quantization error is kept low to non-existent, IIRC. It may be low but it is not non existent. Uhm, as Dick Pierce just pointed out, by providing a clear example of the mathematical mechanism involved (thx Dick, a simple and illustrative treatise), it *is* non-existent. How about you provide some evidence to the contrary? No that is not what he pointed out. He made a big deal about the terminology and pointed out how dither works to *lower* the distortion. Some distortion is still there. If you think it isn't you are simply wrong. I think we're on the brink of an exceptional claim. Properly applied TPDF dither can be mathematically proven to competely decorrelate the first moment of the quantization error (AKA distortion amplitude) and make it statistically independent of the signal. In layman's terms that means "nonlinear distortion amplitude sums to zero". Furthemore, properly applied TPDF dither can be mathematically proven to competely decorrelate the second moment of the quantization error (AKA distortion power) and make it statistically independent of the signal. In layman's terms that means "nonlinear distortion power sums to zero". I can see the correlation summing to zero and still be non-zero at times, but for the square of the correlation to sum to zero, AFAIK the correlation has to be solidly zero at all times. The only case where the sum of the square of an error signal sums to a non-zero number that comes to mind is when the error signal is imaginary. ;-) As a practical matter it can be observed that there are many different kinds of dither including TPDF dither that randomize the quantization error to the point where electrical and mathematical means of detecting it are completely frustrated. This information has been known and widely disseminated without controversy for several decades. I would be very interested in knowing what more recent authority could be formally cited that would change that situation. |
#74
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
Scott wrote:
No that is not what he pointed out. He made a big deal about the terminology and pointed out how dither works to *lower* the distortion. Some distortion is still there. If you think it isn't you are simply wrong. Please support your assertion with the same rigor and used by Shannon, Nyquist, Lip****z, Vanderkooy, and others. -- +--------------------------------+ + Dick Pierce | + Professional Audio Development | +--------------------------------+ |
#75
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
Andrew Haley wrote:
Scott wrote: On Sunday, February 17, 2013 10:26:26 AM UTC-8, KH wrote: On 2/17/2013 7:42 AM, Scott wrote: It may be low but it is not non existent. Uhm, as Dick Pierce just pointed out, by providing a clear example of the mathematical mechanism involved (thx Dick, a simple and illustrative treatise), it *is* non-existent. How about you provide some evidence to the contrary? No that is not what he pointed out. He made a big deal about the terminology and pointed out how dither works to *lower* the distortion. Some distortion is still there. If you think it isn't you are simply wrong. He wasn't asking for more assertions, but evidence. But wait: assertions are pretty much all high-end audio has to live on any more, since facts have been taken away after it starting poking itself in the eye with them. -- +--------------------------------+ + Dick Pierce | + Professional Audio Development | +--------------------------------+ |
#76
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On Tuesday, February 19, 2013 6:41:31 AM UTC-8, Andrew Haley wrote:
He wasn't asking for more assertions, but evidence. I'd like to recommend that you read Resolution Below the Least Significant Bit in Digital Systems with Dither by Vanderkooy and Lip****z. This very famous paper (available on the Internet if you search for it) comes to the same conclusion as Dick Pierce: dither effectively turns [all of] the signal distortion caused by quantization into wide-band noise. If you can find any fault in that paper, it would be interesting to see you present it here. They say: We feel that the audio community in general does not yet understand the nature of quantization error in digital systems, and in particular the beneficial effects of adding an appropriate amount of dither. We shall show that dither really does remove the "digital" aspects of quantization error, leaving an equivalent analog signal with high resolution and some benign wide-band noise. Isn't that "benign wide-band noise" essentially below the threshold of audibility? I would think that it would be. Can someone address this question? |
#77
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
On 2/18/2013 8:58 PM, Scott wrote:
On Sunday, February 17, 2013 10:26:26 AM UTC-8, KH wrote: On 2/17/2013 7:42 AM, Scott wrote: On Saturday, February 16, 2013 12:11:46 PM UTC-8, Audio_Empire wrote: Modern ADCs and DAC chips are laser trimmed so that quantization error is kept low to non-existent, IIRC. It may be low but it is not non existent. Uhm, as Dick Pierce just pointed out, by providing a clear example of the mathematical mechanism involved (thx Dick, a simple and illustrative treatise), it *is* non-existent. How about you provide some evidence to the contrary? No that is not what he pointed out. He made a big deal about the terminology and pointed out how dither works to *lower* the distortion. Some distortion is still there. Well, no, as you're fond of saying, you don't get to speak for other people. Mr. Pierce said, and I quote: "What is done is immediately BEFORE the quantization step, a process known (in the simplest case) as "dither" is performed. This is essentially the addition of a small amount of random noise, on the order of an LSB, to the signal BEFORE the quantizer. It's function is to essentially randomize the statistics of the quantizer. The result is that you ELIMINATE (that means, you make it go away) the quantization error in the qauntizer." Note the use of "ELIMINATE" - not lower, not reduce, eliminate quantization error. Your characterization is flat wrong. If you think it isn't you are simply wrong. Assertion is not evidence. How about some citations? Keith |
#78
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
"KH" wrote in message
... On 2/18/2013 8:58 PM, Scott wrote: No that is not what he pointed out. He made a big deal about the terminology and pointed out how dither works to *lower* the distortion. Some distortion is still there. If the point is that someone on this forum said something like that, it is meaningless. With all due respect for our esteemed moderator(s), nothing posted here goes through a proper editorial review by a duely credentialed editorial board that is poart of an internationally respected a professional organization such as the AES. Thease are all casual conversations. Well, no, as you're fond of saying, you don't get to speak for other people. Mr. Pierce said, and I quote: "What is done is immediately BEFORE the quantization step, a process known (in the simplest case) as "dither" is performed. This is essentially the addition of a small amount of random noise, on the order of an LSB, to the signal BEFORE the quantizer. It's function is to essentially randomize the statistics of the quantizer. The result is that you ELIMINATE (that means, you make it go away) the quantization error in the qauntizer." Note the use of "ELIMINATE" - not lower, not reduce, eliminate quantization error. Your characterization is flat wrong. Agreed. If you think it isn't you are simply wrong. He is certainly correct. Assertion is not evidence. How about some citations? I'm sure that anecdotal evidence from any number of musical artists, recording engineers and mastering engineers can be provided. This speaks to the fact that these people are primarily artists, and are not world-class technical experts even though their professional endeavors may involve the exercise of certain non-trivial amounts of technical expertise. |
#79
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
"Audio_Empire" wrote in message
... On Tuesday, February 19, 2013 6:41:31 AM UTC-8, Andrew Haley wrote: He wasn't asking for more assertions, but evidence. I'd like to recommend that you read Resolution Below the Least Significant Bit in Digital Systems with Dither by Vanderkooy and Lip****z. This very famous paper (available on the Internet if you search for it) comes to the same conclusion as Dick Pierce: dither effectively turns [all of] the signal distortion caused by quantization into wide-band noise. If you can find any fault in that paper, it would be interesting to see you present it here. They say: We feel that the audio community in general does not yet understand the nature of quantization error in digital systems, and in particular the beneficial effects of adding an appropriate amount of dither. We shall show that dither really does remove the "digital" aspects of quantization error, leaving an equivalent analog signal with high resolution and some benign wide-band noise. Isn't that "benign wide-band noise" essentially below the threshold of audibility? "It depends" I would think that it would be. Can someone address this question? This is controversial, it depends on who you believe. If you believe Fielder, he said that 120 dB dynamic range is an absolute requirement. If you believe Krueger, he says that 88 dB suffices. If you believe Vanderkooy and Lipchitz, 16 bit media can have an effective perceived dynamic range on the order of 120 dB. I say that at least two facts support Krueger: (1) Three well-funded attempts have made to raise the performance of mainstream prerecorded media to 93 or 96 dB/ They have all had enough time to prove themselves in the marketplace. They all failed to gain even a tiny fraction of critical mass in the mainstream marketplace. (2) All three attempts included legacy sources with 93-96 dB actual dynamic range, and nobody made a specific complaint based on "Just listening". Technical measurements proved the existence of the lapses in up to 50% of the so-called hi rez media. |
#80
Posted to rec.audio.high-end
|
|||
|
|||
Some People Haven't a Clue
ScottW wrote:
On Feb 19, 7:49 pm, KH wrote: On 2/18/2013 8:58 PM, Scott wrote: No that is not what he pointed out. He made a big deal about the terminology and pointed out how dither works to *lower* the distortion. Some distortion is still there. Well, no, as you're fond of saying, you don't get to speak for other people. Mr. Pierce said, and I quote: "What is done is immediately BEFORE the quantization step, a process known (in the simplest case) as "dither" is performed. This is essentially the addition of a small amount of random noise, on the order of an LSB, to the signal BEFORE the quantizer. It's function is to essentially randomize the statistics of the quantizer. The result is that you ELIMINATE (that means, you make it go away) the quantization error in the qauntizer." Is "eliminate" the right description? My understanding is (and I'd be happily corrected if this is wrong) is that dither decorrelates the quanitization error from the original signal...a necessary requirement for distortion is that it correlates with the original signal. Dither decorrelates and therefore essentially replaces quantization distortion with noise. The beneifts are that with noise shaping the added noise can be placed where it is least offensive to the human ear and the original low level signal on the threshold of dynamic range is still there and we can hear signals below noise better than a distorted low level signal. Again, I'd be happily corrected if my simple understanding is incorrect. well, you're right, but incomplete. Talking about things like decorrelating the quantization error" whiule technically correct, fails to convey to the audience at hand the basic concept of what's happening and why it's good. Look at it from a different viewpoint: if you were to draw a graph of the input to a perfect, noiseless linear "analog" system vs its ouput, it should be a straight line at 45 degrees from horizontal, from lower left to upper right, assuming the system is perfect, linear and noiseless. Now, draw the same graph for the perfect "analog" system, but with a little noise. Averaged over a long enough period, the graph would be a straight line, but over a short period, you'd see a reasonable approximation of that straight line, only noisy. Now, draw the same graph of an undithered digital linear PCM system. Your graph would be a series of equi-spaced stairsteps, the classical, naive view of a digital system. Another step: dither the input to your digital system and now plot the graph. Averaged over a long enough period, the graph would a straight line, but obver a short period, you'd see a reasomable approximation of that straight line, only noisy. Now, compare to ofthose graphs, first the perfect analog system but with a little noise: Averaged over a long enough period, the graph would be a straight line, but over a short period, you'd see a reasonable approximation of that straight line, only noisy. The properly dithered digital system: Averaged over a long enough period, the graph would be a straight line, but obver a short period, you'd see a reasonable approximation of that straight line, only noisy. Yes, in the latter case, the noise is certainly decorrelating the quantixation products from the signal, but looking at it in such a narrow (but correct) way loses the essence of what dithering is doing: it is linearizing the system. And, over the limits of its dynamic range, it is perfectly linear, the residual is noise, only noise. Yes, that noise is decorrleated from the signal. But it's noise, and the system is linear. Linear system don't have distortion, despite what one person or another may want to assert, no matter how vigorously. Now, said persons may want to accuse me of vigorous assertions: fine, have at it. But I'm willing to refer said parties to the following: Blesser, B. A., "Elementary and Basic Aspects of Digital Audio," AES Digital Audio Collected Papers, 1983 Blesser, B. A., "Digitization of Audio: A Comprehensive Examination of Theory, Implementation and Current Practice," JAES, vol 26, no. 10, 1978 Oct. Lip****z, Wannamaker and Vaderkooy, "Quantization and Dither: A Theoretical Sruvey," JAES, vol. 40, no. 5, 1992 May Schuman, L., "Dither Signals and Their Effect on Quantization Noise," IEEE Trans Comm. Tech., COM-12 1964 Dec. Vanderkooy and Lip****z, "Dither in Digital Audio," JAES, vol 35, no. 12, 1987 Dec. Vanderkooy and Lip****z, "Resolution Below the Least Significant Bit in Digtial Audio Systems with Dither," JAES, vol. 32, no. 3, 1984 March. Blesser and Locanthi, "The Applicaion of Narrow-Band Dither Operating at the Nyquist Frequency in Digital Systems to Provide Improved Signal-to-Noise Ratio over Conventional Dithering," JAES, vol. 25, no. 6, 1987 June Jayant and Rabiner, "The Application of Dither to the Quantization of Speech Signals," Bell Sys. Tech J., vol 51, 1972 Shannon, C. E., "Communications in the Presence of Noise," Proc. IRE, vol. 37, 1949. Shannon, C.E., "A mathematical Theory of Communication," Bell Sys. Tech. J., vol. 27, 1948 Oct. So, to those who say assert things like: "No that is not what he pointed out. He made a big deal about the terminology and pointed out how dither works to *lower* the distortion. Some distortion is still there." or "[quantization distortion] It may be low but it is not non existent." or "There is always some quantization error." or "The fact is any AD converter is going to be making quantization errors when sampling any analog signal." To said person(s), I have provided 10 peer-reviewed, rigorous (certainly by ANY standards of the high-end press) articles that would tend to refute such assertions. How about coming up with 10 references that, with equal rigor, supports these assertions? How about 10? 3? And, as one in this thread said, "People often see what they want to see." Most especially when they aren't looking. -- +--------------------------------+ + Dick Pierce | + Professional Audio Development | +--------------------------------+ |
Reply |
Thread Tools | |
Display Modes | |
|
|