Home |
Search |
Today's Posts |
#1
Posted to rec.audio.tech
|
|||
|
|||
Nyquist Sampling Theorem
For some years now the Nyquist Sampling Theorem has been bugging me
and I've decided to come public because none of what I've read adresses my concern - not directly anyway. As you all know, the Nyquist theorem states that "Exact reconstruction of a continuous-time baseband signal from its samples is possible if the signal is bandlimited and the sampling frequency is greater than twice the signal bandwidth". Now if I sample a 1 Hz waveform with a sampling rate of just over 40 kHz then I'd expect to see just over 40 thousand samples for that single-cycle wave - 20 thousand samples for the positive half and another 20 thousand for the negative half. Now go right up to 20 kHz. I get ONE sample in the positive half and ONE sample in the negative half! That's not too impressive is it? Just how are you going to reconstruct a 20 kHz wave accurately with just one sample per half-cycle. That sample could have landed anywhere on the sine-curve - so how do you know what the maximum amplitude of that 20 kHz wave is? Surely you'd need about four or five samples per half-cycle to even hope to reproduce the 20 kHz sine wave accurately. All that the Nyquist limit guarantees is that you can detect a 20 kHz wave (ie. yes, a 20 kHz wave exists) but at a sampling rate of 40 kHz you certainly can't reproduce the wave accurately. This seems to argue for a sampling rate of 4 or 5 times the Nyquist minimum limit. Perhaps this is why DVD sampling is at 192 kHz? Don't shoot me down in flames please. I'd be interested in the flaws in my reasoning that all :-) Cheers Eddy |
#3
Posted to rec.audio.tech
|
|||
|
|||
Nyquist Sampling Theorem
writes:
[...] For some years now the Nyquist Sampling Theorem has been bugging me and I've decided to come public because none of what I've read adresses my concern - not directly anyway. As you all know, the Nyquist theorem states that "Exact reconstruction of a continuous-time baseband signal from its samples is possible if the signal is bandlimited and the sampling frequency is greater than twice the signal bandwidth". This is correct. Note the term "greater than." Now go right up to 20 kHz. Then the sampling rate isn't "greater than." There is nothing else wrong in your reasoning. -- % Randy Yates % "...the answer lies within your soul %% Fuquay-Varina, NC % 'cause no one knows which side %%% 919-577-9882 % the coin will fall." %%%% % 'Big Wheels', *Out of the Blue*, ELO http://home.earthlink.net/~yatescr |
#4
Posted to rec.audio.tech
|
|||
|
|||
Nyquist Sampling Theorem
Then the sampling rate isn't "greater than." Point taken Randy. I was aware of this when I wrote it but stopped myself saying "just greater than" to keep things simple. Glad to hear you think the reasoning is correct though! I wonder what others will think? Eddy |
#5
Posted to rec.audio.tech
|
|||
|
|||
Nyquist Sampling Theorem
writes:
Then the sampling rate isn't "greater than." Point taken Randy. I was aware of this when I wrote it but stopped myself saying "just greater than" to keep things simple. Glad to hear you think the reasoning is correct though! I wonder what others will think? You're welcome, Eddy. However, what others think is irrelevent. It can be shown mathematically to be so. Chris Bores probably has something about this on his web site on DSP tutorials. Google for him. -- % Randy Yates % "Midnight, on the water... %% Fuquay-Varina, NC % I saw... the ocean's daughter." %%% 919-577-9882 % 'Can't Get It Out Of My Head' %%%% % *El Dorado*, Electric Light Orchestra http://home.earthlink.net/~yatescr |
#6
Posted to rec.audio.tech
|
|||
|
|||
Nyquist Sampling Theorem
|
#7
Posted to rec.audio.tech
|
|||
|
|||
Nyquist Sampling Theorem
|
#8
Posted to rec.audio.tech
|
|||
|
|||
Nyquist Sampling Theorem
writes:
[...] Surely you'd need about four or five samples per half-cycle to even hope to reproduce the 20 kHz sine wave accurately. All that the Nyquist limit guarantees is that you can detect a 20 kHz wave (ie. yes, a 20 kHz wave exists) but at a sampling rate of 40 kHz you certainly can't reproduce the wave accurately. This seems to argue for a sampling rate of 4 or 5 times the Nyquist minimum limit. Perhaps this is why DVD sampling is at 192 kHz? Eddy, I retract my initial response to you - sorry for the confusion! I didn't read all the way through your post. It is helpful to separate theory from practice in issues like these. In theory, you can reconstruct ANY frequency between 0 = f Fs/2. Notice the inequalities - they are important. In theory, you CANNOT reconstruct (in general) a frequency AT Fs/2. In practice, the computational complexity gets higher the closer you want to come to Fs/2. However, even in practice, it is not unreasonable to expect to get to within 80 percent of Fs/2 or even closer. I recently designed a filter (it was a monster, though) that was down 80 dB at within 95 percent of Nyquist. So, no, you can't get exactly TO Fs/2, even theoretically, but also no, you don't have to have 4 or 5 samples per cycle. As far as I'm concerned, the high sample rate stuff (DSD, DVD-A, etc.) is BS. -- % Randy Yates % "Rollin' and riding and slippin' and %% Fuquay-Varina, NC % sliding, it's magic." %%% 919-577-9882 % %%%% % 'Living' Thing', *A New World Record*, ELO http://home.earthlink.net/~yatescr |
#9
Posted to rec.audio.tech
|
|||
|
|||
Nyquist Sampling Theorem
|
#10
Posted to rec.audio.tech
|
|||
|
|||
Nyquist Sampling Theorem
Soundhaspriority wrote:
wrote in message oups.com... [snip] Surely you'd need about four or five samples per half-cycle to even hope to reproduce the 20 kHz sine wave accurately. All that the Nyquist limit guarantees is that you can detect a 20 kHz wave (ie. yes, a 20 kHz wave exists) but at a sampling rate of 40 kHz you certainly can't reproduce the wave accurately. This seems to argue for a sampling rate of 4 or 5 times the Nyquist minimum limit. Perhaps this is why DVD sampling is at 192 kHz? Don't shoot me down in flames please. I'd be interested in the flaws in my reasoning that all :-) Cheers Eddy Eddy, Your concern is practically speaking, correct. The wave is reconstructed by an algorithm or by an analog filter. As the Nyquist limit is approached, the problem acquires infinite complexity. What actually happens is that the reconstruction gradually deteriorates as the Nyquist "brick wall" is encountered. The error manifests as phase shift, which is one of the reasons higher sampling rates are preferred. You will probably find this ignites a "flame war", but stick to your guns. You're right. As I understand it, this is not a problem if oversampling is used. And if it isn't (rare nowadays), it's not a problem *if* the player's filters are well-implemented, in which case the phase shift should be practically inaudible. The technical difficulty/cost-effectiveness of implementing such filtering is the reason why oversampling has been the norm for years now in CD players. ___ -S "As human beings, we understand the world through simile, analogy, metaphor, narrative and, sometimes, claymation." - B. Mason |
#11
Posted to rec.audio.tech
|
|||
|
|||
Nyquist Sampling Theorem
On May 2, 10:58 am, wrote:
As you all know, the Nyquist theorem states that "Exact reconstruction of a continuous-time baseband signal from its samples is possible if the signal is bandlimited and the sampling frequency is greater than twice the signal bandwidth". Now if I sample a 1 Hz waveform with a sampling rate of just over 40 kHz then I'd expect to see just over 40 thousand samples for that single-cycle wave - 20 thousand samples for the positive half and another 20 thousand for the negative half. Now go right up to 20 kHz. I get ONE sample in the positive half and ONE sample in the negative half! That's not too impressive is it? Just how are you going to reconstruct a 20 kHz wave accurately with just one sample per half-cycle. That sample could have landed anywhere on the sine-curve - so how do you know what the maximum amplitude of that 20 kHz wave is? But you just violated the Nyquist criteria by having the sample rate be EXACTLY twice the frequency of interest: your example is broken. Surely you'd need about four or five samples per half-cycle to even hope to reproduce the 20 kHz sine wave accurately. All that the Nyquist limit guarantees is that you can detect a 20 kHz wave (ie. yes, a 20 kHz wave exists) but at a sampling rate of 40 kHz you certainly can't reproduce the wave accurately. Nope, your understanding is faulty. The sampling theorem works because even though there an infinite number of waveforms that can pass through those points, ONE AND ONLY ONE meets the important critera that it's total energy is limited to less than 1/2 the sampling rate. Let's take a somewhete easier example: a 10 kHz sine wave sampled at 40 kHz. that gives 4 points per 10 kHz cycle. But not only will that 10 kHz sine wave fully match all the samples, but so will a30 khz sine wave of the same amplitude, a 50 Khz, 70 Kz, 90 kHz and so on to infinity. In fact, one can imaging a non-sinusoidal waveform that passes precisely through every one of the samples. So why does ONLY the EXACT 10 kHz sine wave that went in also come out? Because that's the ONLY one of the infinite number of possible waveforms that has those particular sample values AND falls within the 20kHz Nyquist bandwidth. All others are eliminated in the reconstruction/anti- imaging filter. IN the case out our complicated non-sinu- soidal waveform, it consitis of a 10 kHz waveform plus lots of hamronics, all of which fall outside of the bandwidth of the system. All those the "images" of the original 10 kHz wave, the ones that can also be desccribed by the sample stream, are all present, all the way up to infinity, in the sample digital stream The original 0-20 kHz base band has an image (frequency inverted) from 20-40 kHz, another (non-inverted) from 40-60 kHz, another inverted one from 60-80 kHz, ad infinitum. ALL of them are present in a discrete time-sampled stream. Any one of them can be considered "valid," and, in fact, represent a means of shifting the signal in frequency. But the ENTIRE point in the anti-imaging filter is to filter out all the "images" that aren't wanted. and to leave the one that is, in this case, the baseband (0-20 kHz). (In fact, if you were to sum ALL the possible waves that could pass through the samples, you'd end up with a train of infinitely narrow pulses: the amplitude of the stream would be exactly 0 between samples, and it would rise or fall in zero time to the sample value. Such a waveform, if you were to look at its frequency content, would have harmonics extending out to infinity, filter out ALL of the harmonics of a complex waveform, leaving only the fundamental, and, poof, out comes only a sine wave) This seems to argue for a sampling rate of 4 or 5 times the Nyquist minimum limit. Nope,, it argues that your model is wrong, and that you're leaving out the imaging and required image filtering portion of the system. Perhaps this is why DVD sampling is at 192 kHz? Nope, it's because it can, and thuc can sell. There is no justification technically for it along the lines you are considering. Don't shoot me down in flames please. I'd be interested in the flaws in my reasoning that all :-) In summary, then, you have two flaws: 1. The first, a somewhat minor on, but one that trips most people up, is that you gave an example where you are sampling at EXACTLY twice the frequency of interest, 2. The second, and the one that is most fundamental, is that you have neglected the vital stage of the anti-image and reconstruction fuilter to eliminate the ambiguities in the stream that you point out. Those ambiguities are, in fact, the out-of-band images that the anti-imaging filter eliminates. |
#12
Posted to rec.audio.tech
|
|||
|
|||
Nyquist Sampling Theorem
On May 2, 12:23 pm, "Soundhaspriority" wrote:
wrote in message oups.com... [snip] Surely you'd need about four or five samples per half-cycle to even hope to reproduce the 20 kHz sine wave accurately. All that the Nyquist limit guarantees is that you can detect a 20 kHz wave (ie. yes, a 20 kHz wave exists) but at a sampling rate of 40 kHz you certainly can't reproduce the wave accurately. This seems to argue for a sampling rate of 4 or 5 times the Nyquist minimum limit. Perhaps this is why DVD sampling is at 192 kHz? Don't shoot me down in flames please. I'd be interested in the flaws in my reasoning that all :-) Cheers Eddy Eddy, Your concern is practically speaking, correct. The wave is reconstructed by an algorithm or by an analog filter. As the Nyquist limit is approached, the problem acquires infinite complexity. What actually happens is that the reconstruction gradually deteriorates as the Nyquist "brick wall" is encountered. The error manifests as phase shift, which is one of the reasons higher sampling rates are preferred. In a word, bullsh*t. It does NOT manifest itself as phase shift. Typical phase shift errorsof a the vast majority of 44.1 kHz sampling systems are less than a dozen or so degeres right up to cutoff. If you assertion were correct, the effects you claim would be trivial to measure. Clearly, you have NOT measured them becuase, if you had, you would have not been honestly able to make these claims. You will probably find this ignites a "flame war", but stick to your guns. You're right. No, he's not, he's mistaken. But at least he ASKED the question instead of spouting nonsense, as you just did. What ignires flame wars is not honest questions, or even wrong guesses as the original poster wrote, but uninformed op9inion pasing as fact, as you just did. |
#13
Posted to rec.audio.tech
|
|||
|
|||
Nyquist Sampling Theorem
Thanks for all your remarkably understanding replies. I have noticed
in the past that sometimes the simplest questions don't get simple answers. I can see now that I should have remembered my Fourier analysis. I don't know how these sine waves are reconstructed but I can see from your responses that, from the infinite number of possible sine wave solutions, there is only one solution, provided bandwidth is limited appropriately. That answers my original concern - which has been bugging me for some 15 years now. It's nice to have that burden off my shoulders at last! Cheers Eddy |
#14
Posted to rec.audio.tech
|
|||
|
|||
Dick Pierce is WRONG
"Soundhaspriority" writes:
wrote in message ups.com... On May 2, 12:23 pm, "Soundhaspriority" wrote: wrote in message oups.com... [snip] Surely you'd need about four or five samples per half-cycle to even hope to reproduce the 20 kHz sine wave accurately. All that the Nyquist limit guarantees is that you can detect a 20 kHz wave (ie. yes, a 20 kHz wave exists) but at a sampling rate of 40 kHz you certainly can't reproduce the wave accurately. This seems to argue for a sampling rate of 4 or 5 times the Nyquist minimum limit. Perhaps this is why DVD sampling is at 192 kHz? Don't shoot me down in flames please. I'd be interested in the flaws in my reasoning that all :-) Cheers Eddy Eddy, Your concern is practically speaking, correct. The wave is reconstructed by an algorithm or by an analog filter. As the Nyquist limit is approached, the problem acquires infinite complexity. What actually happens is that the reconstruction gradually deteriorates as the Nyquist "brick wall" is encountered. The error manifests as phase shift, which is one of the reasons higher sampling rates are preferred. In a word, bullsh*t. It does NOT manifest itself as phase shift. Typical phase shift errorsof a the vast majority of 44.1 kHz sampling systems are less than a dozen or so degeres right up to cutoff. You are WRONG. I think it depends on the architecture used in the A/D. If the A/D runs at the final sample rate (e.g., 44.1 kHz), then the analog filter must be sharp, and there would be significant phase distortion near Fs/2. If the A/D oversamples, then the filter requirement is relaxed and the phase shift may be relatively benign at Fs/2. I have to admit, as much as I respect Dick, that I must agree with you, Bob, in this particular context and discussion. HOWEVER...., the very relevent piece of information that most of the "high-sample-rate advocates" seem to conveniently omit is that the high sample rate is only required the data conversion ends of the chain. That is, the storage medium can use the "sane" sample rates of 44.1 or 48 kHz while the A/D and D/A end-points do the appropriate downconversion and upconversion, respectively, that is necessary to overcome the filtering problems. There is no need for 192 kHz DVD-A. -- % Randy Yates % "Ticket to the moon, flight leaves here today %% Fuquay-Varina, NC % from Satellite 2" %%% 919-577-9882 % 'Ticket To The Moon' %%%% % *Time*, Electric Light Orchestra http://home.earthlink.net/~yatescr |
#15
Posted to rec.audio.tech
|
|||
|
|||
Nyquist Sampling Theorem
Soundhaspriority wrote:
"Steven Sullivan" wrote in message ... Soundhaspriority wrote: wrote in message oups.com... [snip] Surely you'd need about four or five samples per half-cycle to even hope to reproduce the 20 kHz sine wave accurately. All that the Nyquist limit guarantees is that you can detect a 20 kHz wave (ie. yes, a 20 kHz wave exists) but at a sampling rate of 40 kHz you certainly can't reproduce the wave accurately. This seems to argue for a sampling rate of 4 or 5 times the Nyquist minimum limit. Perhaps this is why DVD sampling is at 192 kHz? Don't shoot me down in flames please. I'd be interested in the flaws in my reasoning that all :-) Cheers Eddy Eddy, Your concern is practically speaking, correct. The wave is reconstructed by an algorithm or by an analog filter. As the Nyquist limit is approached, the problem acquires infinite complexity. What actually happens is that the reconstruction gradually deteriorates as the Nyquist "brick wall" is encountered. The error manifests as phase shift, which is one of the reasons higher sampling rates are preferred. You will probably find this ignites a "flame war", but stick to your guns. You're right. As I understand it, this is not a problem if oversampling is used. And if it isn't (rare nowadays), it's not a problem *if* the player's filters are well-implemented, in which case the phase shift should be practically inaudible. The technical difficulty/cost-effectiveness of implementing such filtering is the reason why oversampling has been the norm for years now in CD players. It is a problem at some point near the Nyquist frequency. For 44.1 sampling rate, that's ~22 kHz. Not much of a problem, *really*, and even the theoreticalally audible issues are addressed by oversampling . I leave you to Mr.Pierce for the rest of the dissection. ___ -S "As human beings, we understand the world through simile, analogy, metaphor, narrative and, sometimes, claymation." - B. Mason |
#16
Posted to rec.audio.tech
|
|||
|
|||
Dick Pierce is WRONG
On Wed, 02 May 2007 22:23:08 -0400, Randy Yates
wrote: "Soundhaspriority" writes: wrote in message ups.com... On May 2, 12:23 pm, "Soundhaspriority" wrote: wrote in message oups.com... [snip] Surely you'd need about four or five samples per half-cycle to even hope to reproduce the 20 kHz sine wave accurately. All that the Nyquist limit guarantees is that you can detect a 20 kHz wave (ie. yes, a 20 kHz wave exists) but at a sampling rate of 40 kHz you certainly can't reproduce the wave accurately. This seems to argue for a sampling rate of 4 or 5 times the Nyquist minimum limit. Perhaps this is why DVD sampling is at 192 kHz? Don't shoot me down in flames please. I'd be interested in the flaws in my reasoning that all :-) Cheers Eddy Eddy, Your concern is practically speaking, correct. The wave is reconstructed by an algorithm or by an analog filter. As the Nyquist limit is approached, the problem acquires infinite complexity. What actually happens is that the reconstruction gradually deteriorates as the Nyquist "brick wall" is encountered. The error manifests as phase shift, which is one of the reasons higher sampling rates are preferred. In a word, bullsh*t. It does NOT manifest itself as phase shift. Typical phase shift errorsof a the vast majority of 44.1 kHz sampling systems are less than a dozen or so degeres right up to cutoff. You are WRONG. I think it depends on the architecture used in the A/D. If the A/D runs at the final sample rate (e.g., 44.1 kHz), then the analog filter must be sharp, and there would be significant phase distortion near Fs/2. If the A/D oversamples, then the filter requirement is relaxed and the phase shift may be relatively benign at Fs/2. I have to admit, as much as I respect Dick, that I must agree with you, Bob, in this particular context and discussion. HOWEVER...., the very relevent piece of information that most of the "high-sample-rate advocates" seem to conveniently omit is that the high sample rate is only required the data conversion ends of the chain. That is, the storage medium can use the "sane" sample rates of 44.1 or 48 kHz while the A/D and D/A end-points do the appropriate downconversion and upconversion, respectively, that is necessary to overcome the filtering problems. There is no need for 192 kHz DVD-A. What do you mean by "if" the A/D oversamples (and I presume you include the D/A in that). I can't think of a single example of one that doesn't. As far as I am concerned it is a vital part of the process. As far as the real world is concerned, Dick is spot on. I've seen reconstruction filters that demonstrate virtually no measurable phase shift at 20kHz. d -- Pearce Consulting http://www.pearce.uk.com |
#17
Posted to rec.audio.tech
|
|||
|
|||
Nyquist Sampling Theorem
"Soundhaspriority" wrote in message
wrote in message oups.com... Surely you'd need about four or five samples per half-cycle to even hope to reproduce the 20 kHz sine wave accurately. Just over 2 samples are all that is needed to define a unique sine wave, both amplitude and phase (timing). That's one of those things that you learn from trigonometry, if perchance you study and remember it. All that the Nyquist limit guarantees is that you can detect a 20 kHz wave (ie. yes, a 20 kHz wave exists) but at a sampling rate of 40 kHz you certainly can't reproduce the wave accurately. Yes 40.000 KHz won't work, but 40.0000000000000001 KHz will. All you need is just over 2 samples to define that unique sine wvae. This seems to argue for a sampling rate of 4 or 5 times the Nyquist minimum limit. Not at all. Furthermore, this all works out in the lab. It's not an esoteric theory, it is how things work. Perhaps this is why DVD sampling is at 192 kHz? There is no sonic justification for such high sample rates as 192 KHz. If you try to go the opposite way, you'll find that sampling rates as low as 32 KHz are "music friendly". |
#18
Posted to rec.audio.tech
|
|||
|
|||
Dick Pierce is not WRONG
"Soundhaspriority" wrote in message
wrote in message ups.com... On May 2, 12:23 pm, "Soundhaspriority" wrote: wrote in message oups.com... [snip] Surely you'd need about four or five samples per half-cycle to even hope to reproduce the 20 kHz sine wave accurately. All that the Nyquist limit guarantees is that you can detect a 20 kHz wave (ie. yes, a 20 kHz wave exists) but at a sampling rate of 40 kHz you certainly can't reproduce the wave accurately. This seems to argue for a sampling rate of 4 or 5 times the Nyquist minimum limit. Perhaps this is why DVD sampling is at 192 kHz? Don't shoot me down in flames please. I'd be interested in the flaws in my reasoning that all :-) Cheers Eddy Eddy, Your concern is practically speaking, correct. The wave is reconstructed by an algorithm or by an analog filter. As the Nyquist limit is approached, the problem acquires infinite complexity. What actually happens is that the reconstruction gradually deteriorates as the Nyquist "brick wall" is encountered. The error manifests as phase shift, which is one of the reasons higher sampling rates are preferred. In a word, bullsh*t. It does NOT manifest itself as phase shift. Typical phase shift errorsof a the vast majority of 44.1 kHz sampling systems are less than a dozen or so degeres right up to cutoff. The oft-disproven arrogant intellectual failure spews: You are WRONG. No, Dick Pierce is right. Here is a real-world example that supports his claim - actually makes him look maybe a bit conservative. http://www.pcavtech.com/soundcards/l...644-xfus10.gif 44 KHz sampling, and about 5 degrees off at 20 KHz. |
#19
Posted to rec.audio.tech
|
|||
|
|||
Nyquist Sampling Theorem
Hi,
Although at first thoughts you sound right, what is missing in your vision is the low pass filter which is used during analog reconstruction, which if implemented right, forces your close-to-limit samples into a sinewave. Think about it in terms of frequency-windowed Fourier Transform series, your high freq samples represent the energy for that high frequency... sinewave. The low pass filter will reject anything that would let it become a square or whatever. Regards, -- Stéphane a écrit dans le message de news: ... For some years now the Nyquist Sampling Theorem has been bugging me and I've decided to come public because none of what I've read adresses my concern - not directly anyway. As you all know, the Nyquist theorem states that "Exact reconstruction of a continuous-time baseband signal from its samples is possible if the signal is bandlimited and the sampling frequency is greater than twice the signal bandwidth". Now if I sample a 1 Hz waveform with a sampling rate of just over 40 kHz then I'd expect to see just over 40 thousand samples for that single-cycle wave - 20 thousand samples for the positive half and another 20 thousand for the negative half. Now go right up to 20 kHz. I get ONE sample in the positive half and ONE sample in the negative half! That's not too impressive is it? Just how are you going to reconstruct a 20 kHz wave accurately with just one sample per half-cycle. That sample could have landed anywhere on the sine-curve - so how do you know what the maximum amplitude of that 20 kHz wave is? Surely you'd need about four or five samples per half-cycle to even hope to reproduce the 20 kHz sine wave accurately. All that the Nyquist limit guarantees is that you can detect a 20 kHz wave (ie. yes, a 20 kHz wave exists) but at a sampling rate of 40 kHz you certainly can't reproduce the wave accurately. This seems to argue for a sampling rate of 4 or 5 times the Nyquist minimum limit. Perhaps this is why DVD sampling is at 192 kHz? Don't shoot me down in flames please. I'd be interested in the flaws in my reasoning that all :-) Cheers Eddy |
#20
Posted to rec.audio.tech
|
|||
|
|||
Dick Pierce is WRONG
(Don Pearce) writes:
[...] What do you mean by "if" the A/D oversamples (and I presume you include the D/A in that). Which part of the word "if" don't you understand? I can't think of a single example of one that doesn't. As far as I am concerned it is a vital part of the process. You are probably correct in the audio world. I was not constraining the discussion to audio A/Ds. -- % Randy Yates % "With time with what you've learned, %% Fuquay-Varina, NC % they'll kiss the ground you walk %%% 919-577-9882 % upon." %%%% % '21st Century Man', *Time*, ELO http://home.earthlink.net/~yatescr |
#21
Posted to rec.audio.tech
|
|||
|
|||
Dick Pierce is WRONG
On Thu, 03 May 2007 08:53:31 -0400, Randy Yates
wrote: (Don Pearce) writes: [...] What do you mean by "if" the A/D oversamples (and I presume you include the D/A in that). Which part of the word "if" don't you understand? I understand "if" perfectly. My confusion is in that you would think of using the word - but you have explained below. I think that for the purposes of this group we should stick to audio. Things are confused enough already without straying too far away from sound reproduction. I can't think of a single example of one that doesn't. As far as I am concerned it is a vital part of the process. You are probably correct in the audio world. I was not constraining the discussion to audio A/Ds. d -- Pearce Consulting http://www.pearce.uk.com |
#22
Posted to rec.audio.tech
|
|||
|
|||
Dick Pierce is WRONG
(Don Pearce) writes:
On Thu, 03 May 2007 08:53:31 -0400, Randy Yates wrote: (Don Pearce) writes: [...] What do you mean by "if" the A/D oversamples (and I presume you include the D/A in that). Which part of the word "if" don't you understand? I understand "if" perfectly. My confusion is in that you would think of using the word - but you have explained below. I think that for the purposes of this group we should stick to audio. Things are confused enough already without straying too far away from sound reproduction. For most posts in this group I would agree. However, if we go back through the history of this thread to the original question by Eddy, the scope is necessarily bigger than audio. -- % Randy Yates % "...the answer lies within your soul %% Fuquay-Varina, NC % 'cause no one knows which side %%% 919-577-9882 % the coin will fall." %%%% % 'Big Wheels', *Out of the Blue*, ELO http://home.earthlink.net/~yatescr |
#23
Posted to rec.audio.tech
|
|||
|
|||
Dick Pierce is WRONG
On Thu, 03 May 2007 09:21:11 -0400, Randy Yates
wrote: (Don Pearce) writes: On Thu, 03 May 2007 08:53:31 -0400, Randy Yates wrote: (Don Pearce) writes: [...] What do you mean by "if" the A/D oversamples (and I presume you include the D/A in that). Which part of the word "if" don't you understand? I understand "if" perfectly. My confusion is in that you would think of using the word - but you have explained below. I think that for the purposes of this group we should stick to audio. Things are confused enough already without straying too far away from sound reproduction. For most posts in this group I would agree. However, if we go back through the history of this thread to the original question by Eddy, the scope is necessarily bigger than audio. Sure, but his question was about Nyquist rate and the reconstruction of a sine wave. That was all kind of dealt with, and the thread has moved on. The reconstruction filter phase response sub-thread is decidedly audio-related. d -- Pearce Consulting http://www.pearce.uk.com |
#24
Posted to rec.audio.tech
|
|||
|
|||
Dick Pierce is WRONG
(Don Pearce) writes:
Sure, but his question was about Nyquist rate and the reconstruction of a sine wave. That was all kind of dealt with, and the thread has moved on. The reconstruction filter phase response sub-thread is decidedly audio-related. It is apparent you think so. -- % Randy Yates % "Remember the good old 1980's, when %% Fuquay-Varina, NC % things were so uncomplicated?" %%% 919-577-9882 % 'Ticket To The Moon' %%%% % *Time*, Electric Light Orchestra http://home.earthlink.net/~yatescr |
#25
Posted to rec.audio.tech
|
|||
|
|||
Dick Pierce is WRONG
On Thu, 03 May 2007 09:40:04 -0400, Randy Yates
wrote: (Don Pearce) writes: Sure, but his question was about Nyquist rate and the reconstruction of a sine wave. That was all kind of dealt with, and the thread has moved on. The reconstruction filter phase response sub-thread is decidedly audio-related. It is apparent you think so. Whatever. You seem to be angling for a row about this, but frankly I'm not interested. It just isn't important. d -- Pearce Consulting http://www.pearce.uk.com |
#26
Posted to rec.audio.tech
|
|||
|
|||
Nyquist Sampling Theorem
|
#27
Posted to rec.audio.tech
|
|||
|
|||
Dick Pierce is WRONG
Randy Yates wrote:
"Soundhaspriority" writes: wrote in message roups.com... On May 2, 12:23 pm, "Soundhaspriority" wrote: wrote in message legroups.com... [snip] Surely you'd need about four or five samples per half-cycle to even hope to reproduce the 20 kHz sine wave accurately. All that the Nyquist limit guarantees is that you can detect a 20 kHz wave (ie. yes, a 20 kHz wave exists) but at a sampling rate of 40 kHz you certainly can't reproduce the wave accurately. This seems to argue for a sampling rate of 4 or 5 times the Nyquist minimum limit. Perhaps this is why DVD sampling is at 192 kHz? Don't shoot me down in flames please. I'd be interested in the flaws in my reasoning that all :-) Cheers Eddy Eddy, Your concern is practically speaking, correct. The wave is reconstructed by an algorithm or by an analog filter. As the Nyquist limit is approached, the problem acquires infinite complexity. What actually happens is that the reconstruction gradually deteriorates as the Nyquist "brick wall" is encountered. The error manifests as phase shift, which is one of the reasons higher sampling rates are preferred. In a word, bullsh*t. It does NOT manifest itself as phase shift. Typical phase shift errorsof a the vast majority of 44.1 kHz sampling systems are less than a dozen or so degeres right up to cutoff. You are WRONG. I think it depends on the architecture used in the A/D. If the A/D runs at the final sample rate (e.g., 44.1 kHz), then the analog filter must be sharp, and there would be significant phase distortion near Fs/2. Says who? It very much depends on the chosen type of filter. For elliptic filters (Chebychef, Legendre..) yes; for Bessel types no. Yes, Bessel filters require more orders, but remains at a constant rate, in contrast to elliptic filters, which tend to flaten out again, albeit somewhat down the amplitude. Pls excuse me if I remember incorrectly. -- Kind regards, Mogens V. |
#28
Posted to rec.audio.tech
|
|||
|
|||
Dick Pierce is WRONG
"Mogens V." writes:
Randy Yates wrote: "Soundhaspriority" writes: wrote in message ups.com... On May 2, 12:23 pm, "Soundhaspriority" wrote: wrote in message glegroups.com... [snip] Surely you'd need about four or five samples per half-cycle to even hope to reproduce the 20 kHz sine wave accurately. All that the Nyquist limit guarantees is that you can detect a 20 kHz wave (ie. yes, a 20 kHz wave exists) but at a sampling rate of 40 kHz you certainly can't reproduce the wave accurately. This seems to argue for a sampling rate of 4 or 5 times the Nyquist minimum limit. Perhaps this is why DVD sampling is at 192 kHz? Don't shoot me down in flames please. I'd be interested in the flaws in my reasoning that all :-) Cheers Eddy Eddy, Your concern is practically speaking, correct. The wave is reconstructed by an algorithm or by an analog filter. As the Nyquist limit is approached, the problem acquires infinite complexity. What actually happens is that the reconstruction gradually deteriorates as the Nyquist "brick wall" is encountered. The error manifests as phase shift, which is one of the reasons higher sampling rates are preferred. In a word, bullsh*t. It does NOT manifest itself as phase shift. Typical phase shift errorsof a the vast majority of 44.1 kHz sampling systems are less than a dozen or so degeres right up to cutoff. You are WRONG. I think it depends on the architecture used in the A/D. If the A/D runs at the final sample rate (e.g., 44.1 kHz), then the analog filter must be sharp, and there would be significant phase distortion near Fs/2. Says who? It very much depends on the chosen type of filter. For elliptic filters (Chebychef, Legendre..) yes; for Bessel types no. Yes, Bessel filters require more orders, but remains at a constant rate, in contrast to elliptic filters, which tend to flaten out again, albeit somewhat down the amplitude. Pls excuse me if I remember incorrectly. Point taken. See, e.g., [proakis]. --Randy @BOOK{proakis, title = "{Digital Signal Processing: Principles, Algorithms, and Applications}", author = "John~G.~Proakis and Dimitris~G.~Manolakis", publisher = "Prentice Hall", edition = "third", year = "1996"} -- % Randy Yates % "Midnight, on the water... %% Fuquay-Varina, NC % I saw... the ocean's daughter." %%% 919-577-9882 % 'Can't Get It Out Of My Head' %%%% % *El Dorado*, Electric Light Orchestra http://home.earthlink.net/~yatescr |
#29
Posted to rec.audio.tech
|
|||
|
|||
Dick Pierce is WRONG
Randy Yates wrote:
I think it depends on the architecture used in the A/D. If the A/D runs at the final sample rate (e.g., 44.1 kHz), then the analog filter must be sharp, and there would be significant phase distortion near Fs/2. Suh A/Ds (surely) were superceded by the end of the 80s ?!!! If the A/D oversamples, then the filter requirement is relaxed and the phase shift may be relatively benign at Fs/2. And higher for higher ... I have to admit, as much as I respect Dick, that I must agree with you, Bob, in this particular context and discussion. But not in the ciontext of current and recent practice. geoff |
#30
Posted to rec.audio.tech
|
|||
|
|||
Dick Pierce is WRONG
"Geoff" writes:
Randy Yates wrote: I think it depends on the architecture used in the A/D. If the A/D runs at the final sample rate (e.g., 44.1 kHz), then the analog filter must be sharp, and there would be significant phase distortion near Fs/2. Suh A/Ds (surely) were superceded by the end of the 80s ?!!! Let's not get too tunnel-visioned here, folks. There are other applications of digital other than to audio. So, no, such A/Ds were NOT superceded by the end of the 80s. I have to admit, as much as I respect Dick, that I must agree with you, Bob, in this particular context and discussion. But not in the ciontext of current and recent practice. As I stated, yes, even in current practice. But I'm not constraining my consideration to audio applications. -- % Randy Yates % "Ticket to the moon, flight leaves here today %% Fuquay-Varina, NC % from Satellite 2" %%% 919-577-9882 % 'Ticket To The Moon' %%%% % *Time*, Electric Light Orchestra http://home.earthlink.net/~yatescr |
#31
Posted to rec.audio.tech
|
|||
|
|||
Nyquist Sampling Theorem
On May 3, 7:23 am, "Arny Krueger" wrote:
There is no sonic justification for such high sample rates as 192 KHz. If you try to go the opposite way, you'll find that sampling rates as low as 32 KHz are "music friendly". I think there is a sonic justification, and you actually pointed me in the right direction. Some years ago I got into soundcards and found your site, PCAVTECH. Then I starting browsing the newgroups and learned, from you and others, that a 16 bit soundcard cannot fully reproduce all 16 bits of dynamic range. Even a 24 bit soundcard can't produce 16 bits of dynamic range perfectly, and it's supposed to produce 144 db of S/N or dynamics. My point is that electronics are not perfect and do not act ideally. A 44.1Khz /16 signal being reproduced by imperfect electronics doesn't reproduce the audio signal as perfectly as it should, but is still way better than any analog. So it goes to follow that a 192/24 system based on the same imperfect electronics, will have a much easier time reproducing 16 bits of dynamic range and a frequency response to 20 Khz and somehwat beyond better than a 44.1/16 system will. CD |
#32
Posted to rec.audio.tech
|
|||
|
|||
Nyquist Sampling Theorem
"Walt" wrote in message ... wrote: This seems to argue for a sampling rate of 4 or 5 times the Nyquist minimum limit. Perhaps this is why DVD sampling is at 192 kHz? DVD's have a bitrate of 192k bits per second. That's not the same thing as the sampling rate. He probably meant DVD-A which can have a sampling rate of 192kHz, if for no other reason than it's possible. BTW video DVD is usually higher than 192kbs, but of course you are talking compressed MP3/AAC in that case, not HiFi anyway. Some have an uncompressed PCM alternative, but usually 48kHz sample rate. MrT. |
#33
Posted to rec.audio.tech
|
|||
|
|||
Nyquist Sampling Theorem
"codifus" wrote in message ups.com... I think there is a sonic justification, and you actually pointed me in the right direction. Some years ago I got into soundcards and found your site, PCAVTECH. Then I starting browsing the newgroups and learned, from you and others, that a 16 bit soundcard cannot fully reproduce all 16 bits of dynamic range. Even a 24 bit soundcard can't produce 16 bits of dynamic range perfectly, Sure they can, when used at 24 bits. Arny's tests prove just that! And some come amazingly close to perfection even when used at the 16/44 rate. Check out the Lynx2 tests for proof. If only our auditory system was that good! :-) My point is that electronics are not perfect and do not act ideally. A 44.1Khz /16 signal being reproduced by imperfect electronics doesn't reproduce the audio signal as perfectly as it should, but is still way better than any analog. So it goes to follow that a 192/24 system based on the same imperfect electronics, will have a much easier time reproducing 16 bits of dynamic range and a frequency response to 20 Khz and somehwat beyond better than a 44.1/16 system will. Of course, simply select how much overkill you are prepared to pay for, your ears won't tell the difference though. MrT. |
#34
Posted to rec.audio.tech
|
|||
|
|||
Dick Pierce is not WRONG
"Don Pearce" wrote in message ... What do you mean by "if" the A/D oversamples (and I presume you include the D/A in that). I can't think of a single example of one that doesn't. As far as I am concerned it is a vital part of the process. "IF" seems quite appropriate. Not all audio A/D-D/A used oversampling in the past, and he never limited his statement to what is currently being made, so why the need for argument? MrT. |
#35
Posted to rec.audio.tech
|
|||
|
|||
Nyquist Sampling Theorem
"Soundhaspriority" wrote in message ... wrote in message oups.com... [snip] Surely you'd need about four or five samples per half-cycle to even hope to reproduce the 20 kHz sine wave accurately. All that the Nyquist limit guarantees is that you can detect a 20 kHz wave (ie. yes, a 20 kHz wave exists) but at a sampling rate of 40 kHz you certainly can't reproduce the wave accurately. This seems to argue for a sampling rate of 4 or 5 times the Nyquist minimum limit. Perhaps this is why DVD sampling is at 192 kHz? Don't shoot me down in flames please. I'd be interested in the flaws in my reasoning that all :-) Cheers Eddy Eddy, Your concern is practically speaking, correct. The wave is reconstructed by an algorithm or by an analog filter. As the Nyquist limit is approached, the problem acquires infinite complexity. What actually happens is that the reconstruction gradually deteriorates as the Nyquist "brick wall" is encountered. The error manifests as phase shift, which is one of the reasons higher sampling rates are preferred. A Finite Impulse Response digital filter is phase linear, and it can be as "brick wall" as you want to make it. It just takes more registers (and more time to compute). |
#36
Posted to rec.audio.tech
|
|||
|
|||
Nyquist Sampling Theorem
"Soundhaspriority" wrote in message ... "Steven Sullivan" wrote in message ... Soundhaspriority wrote: wrote in message oups.com... [snip] Surely you'd need about four or five samples per half-cycle to even hope to reproduce the 20 kHz sine wave accurately. All that the Nyquist limit guarantees is that you can detect a 20 kHz wave (ie. yes, a 20 kHz wave exists) but at a sampling rate of 40 kHz you certainly can't reproduce the wave accurately. This seems to argue for a sampling rate of 4 or 5 times the Nyquist minimum limit. Perhaps this is why DVD sampling is at 192 kHz? Don't shoot me down in flames please. I'd be interested in the flaws in my reasoning that all :-) Cheers Eddy Eddy, Your concern is practically speaking, correct. The wave is reconstructed by an algorithm or by an analog filter. As the Nyquist limit is approached, the problem acquires infinite complexity. What actually happens is that the reconstruction gradually deteriorates as the Nyquist "brick wall" is encountered. The error manifests as phase shift, which is one of the reasons higher sampling rates are preferred. You will probably find this ignites a "flame war", but stick to your guns. You're right. As I understand it, this is not a problem if oversampling is used. And if it isn't (rare nowadays), it's not a problem *if* the player's filters are well-implemented, in which case the phase shift should be practically inaudible. The technical difficulty/cost-effectiveness of implementing such filtering is the reason why oversampling has been the norm for years now in CD players. It is a problem at some point near the Nyquist frequency. Remember that the filters are implemented by analytic functions. An analytic function doesn't behave nice near a point and then go singular at that point. The closer to the Nyquist frequency, the more the construction falls apart. Since music is not composed of a continuous sine wave, there is the very real problem of how to fit a finite number of points with the appropriate continuous function. It falls apart as phase shift, in some neighborhood of the Nyquist frequency, no matter how thorough one has been at designing the filter. It doesn't matter if the music is composed of a continuous sine wave. If the 22.05KHz filter can reconstruct 20KHz perfectly, it can reconstruct anything and everything below that, perfectly as well, whether it is a mixture of frequencies or not, as long as none of them exceeds 22.05 KHz. |
#37
Posted to rec.audio.tech
|
|||
|
|||
Nyquist Sampling Theorem
We've had FM stereo for 40 years and nobody said a word about brick wall
filters or the Nyquist sampling theorem, but as soon as the compact disc came out, everyone was suspicious of sampled technology. FM stereo uses a 38 KHz suppressed subcarrier, with a 19 KHz pilot tone, phase locked at the Nyquist frequency (effectively a zero-beat with the suppressed carrier). The analog filters of the 60s and 70s were neither phase linear nor particularly brick wall (they started rolling off at around 15 KHz, and they were not down all that far at 19 KHz). Now, I realize that most audiophiles will sneer in derision at FM as an audio source, but FM done right (which happens all too rarely) actually sounds very good. Most of the sound quality goes out the window at the point where it hits the AGC/multiband compressor/multiband limiter/clipper/composite clipper (all intended to give the station the phattest sound on the dial -- whooptee-doo). The main problem is with the standards and practices, not the technology. Today's stereo generators (without all of the audio processing) could probably deliver nearly 18 KHz of phase-linear audio with near zero aliasing. I doubt the receivers have improved much in the last 30 years, though. |
#38
Posted to rec.audio.tech
|
|||
|
|||
Nyquist Sampling Theorem
On Fri, 04 May 2007 06:59:34 GMT, "Karl Uppiano"
wrote: We've had FM stereo for 40 years and nobody said a word about brick wall filters or the Nyquist sampling theorem, but as soon as the compact disc came out, everyone was suspicious of sampled technology. FM stereo uses a 38 KHz suppressed subcarrier, with a 19 KHz pilot tone, phase locked at the Nyquist frequency (effectively a zero-beat with the suppressed carrier). The analog filters of the 60s and 70s were neither phase linear nor particularly brick wall (they started rolling off at around 15 KHz, and they were not down all that far at 19 KHz). Now, I realize that most audiophiles will sneer in derision at FM as an audio source, but FM done right (which happens all too rarely) actually sounds very good. Most of the sound quality goes out the window at the point where it hits the AGC/multiband compressor/multiband limiter/clipper/composite clipper (all intended to give the station the phattest sound on the dial -- whooptee-doo). The main problem is with the standards and practices, not the technology. Today's stereo generators (without all of the audio processing) could probably deliver nearly 18 KHz of phase-linear audio with near zero aliasing. I doubt the receivers have improved much in the last 30 years, though. Just one problem here. FM Stereo isn't sampled, so it doesn't have a Nyquist frequency. I guess that could be why nobody was talking about Nyquist sampling theorem? d -- Pearce Consulting http://www.pearce.uk.com |
#39
Posted to rec.audio.tech
|
|||
|
|||
Nyquist Sampling Theorem
On May 3, 11:51 pm, "Mr.T" MrT@home wrote:
"codifus" wrote in message ups.com... I think there is a sonic justification, and you actually pointed me in the right direction. Some years ago I got into soundcards and found your site, PCAVTECH. Then I starting browsing the newgroups and learned, from you and others, that a 16 bit soundcard cannot fully reproduce all 16 bits of dynamic range. Even a 24 bit soundcard can't produce 16 bits of dynamic range perfectly, Sure they can, when used at 24 bits. Arny's tests prove just that! And some come amazingly close to perfection even when used at the 16/44 rate. Check out the Lynx2 tests for proof. Yes, I've heard of theLynx2, and also the DAC1 Not a soundcard but a very well regarded D/A converter. If only our auditory system was that good! :-) That's just my point. SOME soundcards do very well. Can you say that any soundcard that isn't a Lynx is of inferior electronic design? Perhaps they are. And maybe 10 years from now when any Lynx patent might run out all the soundacrd manufacturers from Creative to EchoAudio may run out and copy the Lynx design and everything would be perfect. My point is that electronics are not perfect and do not act ideally. A 44.1Khz /16 signal being reproduced by imperfect electronics doesn't reproduce the audio signal as perfectly as it should, but is still way better than any analog. So it goes to follow that a 192/24 system based on the same imperfect electronics, will have a much easier time reproducing 16 bits of dynamic range and a frequency response to 20 Khz and somehwat beyond better than a 44.1/16 system will. Of course, simply select how much overkill you are prepared to pay for, your ears won't tell the difference though. Maybe, maybe not. Fact is, CDs were "perfect sound forever" and we've realized just how flawed it was at the beginning. It took about 20 years for CD to mature to the very good state that it's in now. Because it all depends, I wouldn't count out the higher rez formats as un-necessary or overkill. MrT. CD |
#40
Posted to rec.audio.tech
|
|||
|
|||
CD quality
codifus wrote:
On May 3, 11:51 pm, "Mr.T" MrT@home wrote: "codifus" wrote in message ups.com... I think there is a sonic justification, and you actually pointed me in the right direction. Some years ago I got into soundcards and found your site, PCAVTECH. Then I starting browsing the newgroups and learned, from you and others, that a 16 bit soundcard cannot fully reproduce all 16 bits of dynamic range. Even a 24 bit soundcard can't produce 16 bits of dynamic range perfectly, Sure they can, when used at 24 bits. Arny's tests prove just that! And some come amazingly close to perfection even when used at the 16/44 rate. Check out the Lynx2 tests for proof. Yes, I've heard of theLynx2, and also the DAC1 Not a soundcard but a very well regarded D/A converter. If only our auditory system was that good! :-) That's just my point. SOME soundcards do very well. Can you say that any soundcard that isn't a Lynx is of inferior electronic design? Perhaps they are. And maybe 10 years from now when any Lynx patent might run out all the soundacrd manufacturers from Creative to EchoAudio may run out and copy the Lynx design and everything would be perfect. My point is that electronics are not perfect and do not act ideally. A 44.1Khz /16 signal being reproduced by imperfect electronics doesn't reproduce the audio signal as perfectly as it should, but is still way better than any analog. So it goes to follow that a 192/24 system based on the same imperfect electronics, will have a much easier time reproducing 16 bits of dynamic range and a frequency response to 20 Khz and somehwat beyond better than a 44.1/16 system will. Of course, simply select how much overkill you are prepared to pay for, your ears won't tell the difference though. Maybe, maybe not. Fact is, CDs were "perfect sound forever" and we've realized just how flawed it was at the beginning. It took about 20 years for CD to mature to the very good state that it's in now. Because it all depends, I wouldn't count out the higher rez formats as un-necessary or overkill. MrT. CD And yet.... CDs now are generally of poorer audio quality than 10-15 years ago due to excessive processing at the mastering stage reducing dynamic range and introducing deliberate clipping to increase loudness. Some of the early CDs were poor due to ignorance of the new medium, when masters equalised and compressed for Vinyl were used to master CDs. It was soon realised that for CD new masters had to be produced, but this lesson was learned quite early on, say, buy around 1985-6. This means that we had some 10 years in which CD was as close to "pure perfect sound" as was possible at the time, and has rarely been bettered since. The new higher resolution formats are, at least for me, overkill for domestic sound reproduction, but nevertheless, it's good to have them if only for archival purposes. S. -- http://audiopages.googlepages.com |
Reply |
Thread Tools | |
Display Modes | |
|
|
Similar Threads | ||||
Thread | Forum | |||
On sampling, SACD, etc. | Audio Opinions | |||
88.2 vs. 96 K sampling rate | Pro Audio | |||
RADAR 24 Nyquist system | Pro Audio | |||
RADAR 24 Nyquist system | Pro Audio | |||
Why 24/96 sampling isn't necessarily better-sounding than 24/44 sampling | Pro Audio |