usefull conclusion
So if NOS DAC (so without digital filter chip before or embeded in the dac chip itself) it could be translate by:
1. You can oversample on the fly with some playback softs as Audirvana and alike, digital filter is performed by the personal computer and is a function of such perfectioned playback softaware. Such way is usefull (but you may not like the result depends on the digital filter chosed)
2&3. It's not usefull to oversample off line your library then playback it with a simple playback software which just pass through the NOS DAC the sample rate of the library materials ( i.e. materials that were oversampled off line at higher rate for instance with Sox or having already since the recording a high sample rate). Those simpliest playback softaware don't provide digital filter function on the fly.
If 1 option : analogic filter after the dac chip must be avoided (low pass filter at Nyquist frequency, i.e 22 Khz) because it could waste the digtal filter proceded by the computer ?
If 2&3 : if an oversampling was made off line and playbacked through the NOS dac without digital filtering then an analog filter (steep?) at 22 Khz Nyquist frequency may help a little because the image amplitude was increased ?
Sorry if noob question, I try to translate it for handy use to avoid snake oil in my playback system.
I have sometimes high sample rate material in my library played trough my Redbook NOS DAC with a pass trough playback software à la RaspBerryPi so your scenario 2 or 3: looks like such materials have more details but subjectivly not beiing so natural: details but light sound (biased opionion?)
So if NOS DAC (so without digital filter chip before or embeded in the dac chip itself) it could be translate by:
1. You can oversample on the fly with some playback softs as Audirvana and alike, digital filter is performed by the personal computer and is a function of such perfectioned playback softaware. Such way is usefull (but you may not like the result depends on the digital filter chosed)
2&3. It's not usefull to oversample off line your library then playback it with a simple playback software which just pass through the NOS DAC the sample rate of the library materials ( i.e. materials that were oversampled off line at higher rate for instance with Sox or having already since the recording a high sample rate). Those simpliest playback softaware don't provide digital filter function on the fly.
If 1 option : analogic filter after the dac chip must be avoided (low pass filter at Nyquist frequency, i.e 22 Khz) because it could waste the digtal filter proceded by the computer ?
If 2&3 : if an oversampling was made off line and playbacked through the NOS dac without digital filtering then an analog filter (steep?) at 22 Khz Nyquist frequency may help a little because the image amplitude was increased ?
Sorry if noob question, I try to translate it for handy use to avoid snake oil in my playback system.
I have sometimes high sample rate material in my library played trough my Redbook NOS DAC with a pass trough playback software à la RaspBerryPi so your scenario 2 or 3: looks like such materials have more details but subjectivly not beiing so natural: details but light sound (biased opionion?)
Last edited:
Hi DF96, thank you for the comment.
What I’m not really sure about is this: As far as I understand the digital filter theory, when the cut off frequency of digital filter is several times higher than nyquist frequency of the original source, it does not do any calculation that affects the original signal. We can't detect any difference between NOS and upsampling DAC below nyquist in this regard. Am I correct?
What I’m not really sure about is this: As far as I understand the digital filter theory, when the cut off frequency of digital filter is several times higher than nyquist frequency of the original source, it does not do any calculation that affects the original signal. We can't detect any difference between NOS and upsampling DAC below nyquist in this regard. Am I correct?
https://www.analog.com/media/cn/training-seminars/tutorials/MT-017.pdf
Rather than going round and round, how about a very nice document from one of AD's mixed mode designers*? It hits most all the expressed confusion and answers the op's question better than anything thus written here. I'm more surprised the op didn't find this document before asking his question, given the very keywords in his post.
*Or I think; Scott can correct me
Rather than going round and round, how about a very nice document from one of AD's mixed mode designers*? It hits most all the expressed confusion and answers the op's question better than anything thus written here. I'm more surprised the op didn't find this document before asking his question, given the very keywords in his post.
*Or I think; Scott can correct me
Last edited:
Strawman.
Do you really think that a digital recording and reproduction chain can't be tested for measurable, both objective and subjective, flaws?
Wow.
This answers!
"90% of the books and websites" are limited to repeat that "the main advantage of oversampling is to allow a less critical analog filter: without oversampling it is necessary to cut sharp at 22.05 kHz, with 8x oversampling we have to cut at 176.4 kHz", therefore in a larger window; which is in fact right.
Unfortunately, "90% of books and websites" forget to mention that this is just because there is a digital filter before doing the work and cutting out the main quantization noise.
This cheated me.
Therefore, I would say that the primary advantage of oversampling is the use of a digital filter, and not the use of a smooth type of analog filter, that is just the secondary consequence of having first a digital filter that does the dirty job in its place!
This answers!
"90% of the books and websites" are limited to repeat that "the main advantage of oversampling is to allow a less critical analog filter: without oversampling it is necessary to cut sharp at 22.05 kHz, with 8x oversampling we have to cut at 176.4 kHz", therefore in a larger window; which is in fact right.
Unfortunately, "90% of books and websites" forget to mention that this is just because there is a digital filter before doing the work and cutting out the main quantization noise.
This cheated me.
Therefore, I would say that the primary advantage of oversampling is the use of a digital filter, and not the use of a smooth type of analog filter, that is just the secondary consequence of having first a digital filter that does the dirty job in its place!
Perhaps there's something lost in translation between what is said and what you understood, but the description you gave here does not match up with the basics of signal processing. Please start with the PDF I provided above, and at least the effects of zero padding, which is technically considered a form of interpolation, convenient to integer multiples of the original sampling rate, and benefitting from process gain.
Oversampling/upsampling can be done in 3 ways, if understood correctly: by inserting "zero" samples or repeating each digit multiple times or by adding interpolated samples between the original source samples.
I was just referring to the first two modes, for which the original quantization noise images are maintened and still need to be filtered sharply (like in NOS but by using a "digital" filter).
Oversampling. Upsampling Audio [How to Convert Sample Rate Explained]
Your written refers to the third one (by interpolation) which in effect moves the alias image much ahead. Does this clarify?
Yep, but they're all of one type, albeit some are more computationally simple than others. In zero order hold there's a baked in rectangular filter block that is more explicitly needed in a zero pad/interpolation.
Edited from first order hold, ie triangular and zero order hold, ie rectangular padding. I meant the latter.
Edited from first order hold, ie triangular and zero order hold, ie rectangular padding. I meant the latter.
Last edited:
This subject takes a bit of study to properly comprehend. Our attempts to provide short answers here are necessarily too brief to give you an comprehensive understanding, and, as has become apparent, added to your confusion.
Digital audio signal interpolation is essentially an low-pass filtering operation. So, to recover the original analog signal, essentially all that we need do is filter away all of the ultrasonic image bands. Over/upsampling of the native input sample rate to some higher rate is necessary for opening bandwidth where to relocate the image bands. They're not exactly fully removed. They are relocated higher in frequency.
'Zero stuffing' is an signal interpolation method for producing an oversampled, low-pass filtered digital signal. As has already been said multiple times, you cannot produce an interpolated over/upsampled digital signal without recourse to some digital filtering method. No digital filter, no true over/upsampled digital signal. This is, of course, an highly oversimplified explanation.
Simply repeating successive samples IS NOT a form of interpolation. The digital signal remains EXACTLY the same as before the repetition. Hopefully, your question has been satisfied by now.
Last edited:
Inserting N-1 zero samples between the original samples itself does not provide any kind of interpolation, the interpolation part is done by the properly designed digital low pass filter.
Let´s take 4 times oversampling as a example; it is usually done by inserting 4-1 = 3 zero samples between the original samples.
That alone with the higher replay rate (4 x Fsampleoriginal) provides shorter pulses (in the place of the original samples, each followed by a time-sequence of zero output) and therefore would result in a better amplitude weighting function as the shorter pulses are a better approximation of the Dirac pulses that should have been used (but couldn´t be used because of their nonreality nature).
But still the same kind of steep reconstruction low pass filter at the output is needed as before.....
Let´s take 4 times oversampling as a example; it is usually done by inserting 4-1 = 3 zero samples between the original samples.
That alone with the higher replay rate (4 x Fsampleoriginal) provides shorter pulses (in the place of the original samples, each followed by a time-sequence of zero output) and therefore would result in a better amplitude weighting function as the shorter pulses are a better approximation of the Dirac pulses that should have been used (but couldn´t be used because of their nonreality nature).
But still the same kind of steep reconstruction low pass filter at the output is needed as before.....
Last edited:
Unfortunately the site that you´ve linkes above is using an incorrect representation, as the multiple image spectra (around the multiples of the original sampling frequency) are not a result of the interpolation process but originate from the original sampling process.
Let´s assume that one uses an original sampling rate of 44.1 kHz, and samples a 1 kHz sinus signal, then after the sampling process you´ll have in the frequency domain a single spectral line at 1 kHz (idealized version for simplicity) and you´ll have a spectral line at (Fsample - 1 kHz = 43.1 kHz) and another spectral line at (Fsample + 1 kHz = 45.1 kHz.). The next spectral lines will be at the multiples of the original sampling frequency.
Means at (2xFsample - 1 kHz = 87.1 kHz) and (2xFsample + 1 kHz = 89.1 kHz) and the same around 3xFsample and 4xFsample and so on.
Reconstruction of the original signal means that you have to get rid of all of these image spectra lines at the multiples (of Fsample) and that can be done by a perfect analog "brickwall" low pass filter (not realizable in reality) or by a perfect digital low pass filter (not realizable in reality) in the oversampling/upsampling process, followed by a more gentle analog low pass filter.
Last edited:
To ensure complete and utter confusion, we could start a little side discussion about (by necessity oversampled) sigma-delta DACs with quasi-NOS mode. This is how I guess they work:
https://www.diyaudio.com/forums/digital-line-level/328003-dacs-contain-op-amp-14.html#post5561036
https://www.diyaudio.com/forums/digital-line-level/328003-dacs-contain-op-amp-14.html#post5561036
Ok thank you.
Let clarify my last doubt: oversampling allows simpler and less critical lowpass analog filter because:
1) it has to act in a larger frequency window (two due to the fact that oversampling by x-times has moved undesirably noise to x-times higher frequency)
2) oversampling performs filtering mainly in the digital domain, lightening the requirements of the analog filter
3) both 1) and 2)
?
Let clarify my last doubt: oversampling allows simpler and less critical lowpass analog filter because:
1) it has to act in a larger frequency window (two due to the fact that oversampling by x-times has moved undesirably noise to x-times higher frequency)
2) oversampling performs filtering mainly in the digital domain, lightening the requirements of the analog filter
3) both 1) and 2)
?
Last edited:
Combination of 1 and 2.
A digital signal with sample rate fs necessarily has a repetitive spectrum that repeats every fs. By stuffing N - 1 zero samples in between the actual samples, the sample rate goes from fs to N fs, but the spectrum still stays the same, that is, it still repeats every fs. A digital low-pass filter can then remove all spectral copies except those around multiples of N fs. Put this into a DAC and the analogue reconstruction filter only needs to suppress the rubbish around multiples of N fs.
A digital signal with sample rate fs necessarily has a repetitive spectrum that repeats every fs. By stuffing N - 1 zero samples in between the actual samples, the sample rate goes from fs to N fs, but the spectrum still stays the same, that is, it still repeats every fs. A digital low-pass filter can then remove all spectral copies except those around multiples of N fs. Put this into a DAC and the analogue reconstruction filter only needs to suppress the rubbish around multiples of N fs.
I don't know which 90% you read, but it was always obvious in anything I read about oversampling that the whole point of it is to use a digital filter in order to get a near-ideal reconstruction and a much relaxed specification for the analogue filter. Perhaps what threw you was the idea that oversampling and reconstruction filtering were somehow separate processes when in fact they are done in the same digital filter.ygg-it said:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.