Should we put a ASR welcome pack together?
1st item prayer Matt... Any other ideas?
As we are all sheep should we include free sheep dip? Free yearly hair cuts?...
-
WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!
Is DSD superior or just the audio file du jour?
- Thread starter Phelonious Ponk
- Start date
1st item prayer Matt... Any other ideas?
As we are all sheep should we include free sheep dip? Free yearly hair cuts?...
Exactly. I've watched many youtube videos of some of the professional mixing guys doing major label work, and often their sessions are running at 48 or 44.1 kHz. My own blind testing has shown that I can record acoustic guitar/vocals/etc. at 44.1 kHz and maintain an audibly transparent recording chain. Since I'll usually be mastering to 44.1 kHz anyway, it makes sense. I see the value in recording at 24 bits, but am not convinced of the need for anything higher than 16 bits as a distribution format.
dallasjustice
Major Contributor
I know there have been tests done which show there's no audible difference between DSD and PCM. But I think the way DSD upsampling is being done now, it would be worthwhile to revisit the issue to determine whether "better" digital upsampling/reconstruction filtering can be audible. My suspicion is that digital filters can be audible. But I have no evidence to the contrary. Having said that, there's really no evidence to prove they aren't audible.
Last edited:
Fitzcaraldo215
Major Contributor
- Joined
- Mar 4, 2016
- Messages
- 1,440
- Likes
- 633
OK. I get what you mean now. Yes, in the digital domain, the PCM representation is more like the analog waveform. And, it is not truly stair steps, but individual sample values at discrete points in time. Scope displays artificially extend each sample's timing to the next sample, creating the fictional stair step.
I have never been able to relate in my feeble brain how a string of 1-bit DSD samples(PDM) relates to the analog waveform. It is far less intuitive to me. But, again, the samples are at points of time, not stair steps.
The PCM representation lends itself to DSP, including EQ and such, whereas the more cryptic DSD bit representation does not. This is important in making recordings, and, for most, in applying room EQ, crossovers, etc. in playback.
But, technical differences aside, as Joe Whip and others have said, they can both sound terrific. I have great recorded examples in both formats in hi rez.
I have to think this is something that has AES publications.
Not related to the filters, Lipshitz/ Vanderkooy have an AES paper "Why 1 Bit Delta Sigma Conversion is Unsuitable for High Quality Applications" (2001, May 12-15 AES convention). Entire paper can be found by Googling, not sure if this was one of the papers freely available, so I'll refrain from linking it.
Last edited:
Not to drift too far off topic or start any side debates but I'm just wondering what definition you would apply to "affordable digital" ?
TIA
Ken Newton
Active Member
- Joined
- Mar 29, 2016
- Messages
- 190
- Likes
- 47
Hi, Fitzcaraldo, following, are a few points of clarification.
That some PCM converter's output holds the current sample value until moving directly to the next value has nothing to do with viewing it on an oscilloscope. This 'sample and hold' behavior is termed an, zero-order hold and is how the most DACs function. The hold period creates a treble roll-off side effect, which is typically EQ'd back to flat within the digital filter. Which is why NOS (digital filterless) DACs exhibit this treble-roll off unless EQ'd in the analog domain.
Whether an PCM DAC's analog output produces theoretically ideal impulses, or is instead stair-stepped, the output is fully analog. It's only an illusion that it is not analog due to the lack of image filtering at that point. The images are a series of spectrally repeating replicas of the desired analog signal band. The desired signal band is itself present along with the images, and both are analog. Seen together on an o-scope, the desired signal and it's repeating images gives the illusion that the D/A output is discretely discontinuous, and not yet analog. Strip away (filter) all of the images and the desired analog signal band is revealed in it's expected smoothly continuous form when viewed on a o-scope.
An PCM data steam only resembles the original analog signal after it has been converted to the analog domain. PCM viewed in the digital domain does not look anything like the orginal signal, which is to be expected, as PCM is a (numerical) code modulation. The data stream of DSD, on the other hand, does sort of appear to vary in density in proportion to the original analog signal. Which, as I mentioned earlier, doesn't necessarily have any consequence for the subjective sound, but it is interesting to note.
The truth is, much of digital signal processing science is not intuitive. Often, it's actually counter-intuitive. It may help you to picture 1-bit as pulse density modulation, which it what it really is. The more binary ones that are arranged in time together the more postive will go the analog signal. The more binary zeroes that are arranged in time together the more negative will go the signal. A continuous string of alternating ones and zeroes would have an average output level of bipolar zero. In effect, the analog signal's amplitude will essentially be proportional to how densely arranged in time are the binary ones versus zeroes. I hope this helps.
That some PCM converter's output holds the current sample value until moving directly to the next value has nothing to do with viewing it on an oscilloscope. This 'sample and hold' behavior is termed an, zero-order hold and is how the most DACs function. The hold period creates a treble roll-off side effect, which is typically EQ'd back to flat within the digital filter. Which is why NOS (digital filterless) DACs exhibit this treble-roll off unless EQ'd in the analog domain.
Whether an PCM DAC's analog output produces theoretically ideal impulses, or is instead stair-stepped, the output is fully analog. It's only an illusion that it is not analog due to the lack of image filtering at that point. The images are a series of spectrally repeating replicas of the desired analog signal band. The desired signal band is itself present along with the images, and both are analog. Seen together on an o-scope, the desired signal and it's repeating images gives the illusion that the D/A output is discretely discontinuous, and not yet analog. Strip away (filter) all of the images and the desired analog signal band is revealed in it's expected smoothly continuous form when viewed on a o-scope.
An PCM data steam only resembles the original analog signal after it has been converted to the analog domain. PCM viewed in the digital domain does not look anything like the orginal signal, which is to be expected, as PCM is a (numerical) code modulation. The data stream of DSD, on the other hand, does sort of appear to vary in density in proportion to the original analog signal. Which, as I mentioned earlier, doesn't necessarily have any consequence for the subjective sound, but it is interesting to note.
The truth is, much of digital signal processing science is not intuitive. Often, it's actually counter-intuitive. It may help you to picture 1-bit as pulse density modulation, which it what it really is. The more binary ones that are arranged in time together the more postive will go the analog signal. The more binary zeroes that are arranged in time together the more negative will go the signal. A continuous string of alternating ones and zeroes would have an average output level of bipolar zero. In effect, the analog signal's amplitude will essentially be proportional to how densely arranged in time are the binary ones versus zeroes. I hope this helps.
Last edited:
Fitzcaraldo215
Major Contributor
- Joined
- Mar 4, 2016
- Messages
- 1,440
- Likes
- 633
Ken - thanks for trying to explain, but I have a pretty fair picture in my mind of how the two work. Incidentally, I have no scope, so I do not look at scope displays of the two.
However, I fundamentally disagree with some of your conclusions. A PCM sample is simply a multibit word containing a numerical value corresponding to the analog waveform's level (voltage), + or -, at an instant in time. The bigger that numerical value the proportionately greater the displacement from the axis = 0 volts. As an old computer guy, I can look at a binary sample word, interpret its scalar magnitude and sign quite intuitively by understanding powers of 2 and most vs. least significant bits in the binary word.
I, for one, cannot do that at all intuitively in looking at a train of 1-bit pulses, not words, each one corresponding to a single sample and then interpreting that in term of "pulse density". I find that much more abstract. A single sample means nothing resembling the waveform in DSD except in the context of the other samples around it, whereas each single PCM sample completely and unambiguously represents the instaneous signal level at that point in time.
These are key reasons that DSP, especially something like EQ which is just controlled signal level manipulation, is comparatively easier to do in PCM, sample by sample, but very difficult to do in DSD.
As I said, I have not looked at scope traces, but I keep pondering the example of the analog waveform vs. the succession of DSD samples here:
https://en.m.wikipedia.org/wiki/Pulse-density_modulation
I just do not see a close resemblance to the waveform in the 1's and 0's or how the digital samples are derived from it at all intuitively. But, maybe that is just me.
I do concede based on my understanding that DSD DACs are conceptually simpler than PCM DACs in converting digital back to the analog waveform. But, I have scratched my head looking at photos of the innards of some DSD DACs, PS Audio for example, and seen a very complex circuit, though much of that may be tied to its PCM-DSD conversion and its use of FPGA.
All of this, of course, leads nowhere in telling us anything about perceived sound quality. I am quite happy with both DSD and PCM on that score, and my DAC handles both. But, in truth, I most always play DSD via conversion to PCM in order to gain the advantages of DSP, notably Room EQ, bass management and speaker distance correction in Mch. Those offer much more sonic improvement to me than does pure DSD playback. But, if I had all those DSP tools readily available to me in DSD, I would happily use them and play in the originally recorded format.
By the way, I know a little bit about HQPlayer's DSP in DSD capabilities. I am curious, but I have seen nowhere near enough about it to want to explore it further. That, and it is incompatible with the Dirac Live EQ I now use.
However, I fundamentally disagree with some of your conclusions. A PCM sample is simply a multibit word containing a numerical value corresponding to the analog waveform's level (voltage), + or -, at an instant in time. The bigger that numerical value the proportionately greater the displacement from the axis = 0 volts. As an old computer guy, I can look at a binary sample word, interpret its scalar magnitude and sign quite intuitively by understanding powers of 2 and most vs. least significant bits in the binary word.
I, for one, cannot do that at all intuitively in looking at a train of 1-bit pulses, not words, each one corresponding to a single sample and then interpreting that in term of "pulse density". I find that much more abstract. A single sample means nothing resembling the waveform in DSD except in the context of the other samples around it, whereas each single PCM sample completely and unambiguously represents the instaneous signal level at that point in time.
These are key reasons that DSP, especially something like EQ which is just controlled signal level manipulation, is comparatively easier to do in PCM, sample by sample, but very difficult to do in DSD.
As I said, I have not looked at scope traces, but I keep pondering the example of the analog waveform vs. the succession of DSD samples here:
https://en.m.wikipedia.org/wiki/Pulse-density_modulation
I just do not see a close resemblance to the waveform in the 1's and 0's or how the digital samples are derived from it at all intuitively. But, maybe that is just me.
I do concede based on my understanding that DSD DACs are conceptually simpler than PCM DACs in converting digital back to the analog waveform. But, I have scratched my head looking at photos of the innards of some DSD DACs, PS Audio for example, and seen a very complex circuit, though much of that may be tied to its PCM-DSD conversion and its use of FPGA.
All of this, of course, leads nowhere in telling us anything about perceived sound quality. I am quite happy with both DSD and PCM on that score, and my DAC handles both. But, in truth, I most always play DSD via conversion to PCM in order to gain the advantages of DSP, notably Room EQ, bass management and speaker distance correction in Mch. Those offer much more sonic improvement to me than does pure DSD playback. But, if I had all those DSP tools readily available to me in DSD, I would happily use them and play in the originally recorded format.
By the way, I know a little bit about HQPlayer's DSP in DSD capabilities. I am curious, but I have seen nowhere near enough about it to want to explore it further. That, and it is incompatible with the Dirac Live EQ I now use.
Ken Newton
Active Member
- Joined
- Mar 29, 2016
- Messages
- 190
- Likes
- 47
You have before indicated, and continue to indicate, that you cannot picture how 1-bit quantization works. In addition, you have suggested that DAC sample-and- hold operation is only an illusion due to viewing the output via a scope. Then you submit the above puzzling comment. Huh? Perhaps, I misunderstand what you have been saying.
I'll attempt to make this clear this final time. In vewing a PCM serial data stream (pulse train) it is impossibe to gain any feeling for how the digital signal relates to the analog signal it represents because the PCM stream is a pulse coded series of rapidly changing numbers. If those numbers were translated to decimal form and printed to hard copy, then, yes, you could gain an accurate sense of the analog signal represented. My main point, however, is that with 1-bit you CAN visually gain an feeling for the analog signal that is being represented via the serial digital data stream. If you've never viewed the serial pulse trains for both PCM and DSD on a scope then you may have difficulty picturing this.
Correct, a single DSD sample means nothing on it's own. However, as I've repeatedly indicated, it's the density of ones versus zeroes, relative to each other in time, which carries information.
Yes, much of that circuitry has to do with transcoding PCM to DSD. Circuitry such as oversampling digital filters and sigma-delta modulators plus noise-shaping logic.
Last edited:
Fitzcaraldo215
Major Contributor
- Joined
- Mar 4, 2016
- Messages
- 1,440
- Likes
- 633
Ken - I admire your persistence. I am quite sure nobody else at this point cares about our discussion of DSD vs. PCM in how they are represented in theory or on scope displays. What is more important is how they sound and the implications of their format/representation on sound quality, DSP processing, etc. as I have tried to discuss.
But, to your points, I had to search for this excellent video on PCM, which dispels the stair step idea in the first 5 minutes:
http://www.wired.com/2013/02/sound-smart-watch-this-excellent-primer-on-digital-audio/
Yes, as it says, stair steps are created by sample/hold on displays and also for use internally by DACs. But, that does not change the theory of PCM sampling, which is discrete and instantaneous, and it does not affect the analog output from the DAC in the real world of finite, non-instantaneous time. DSD sampling is only theoretically discrete and instantaneous, too, for that matter, and it must also work in finite time. DSD bits would also look like a convoluted up/down one single stair step in a sample/hold display of the digital bits on a scope.
By the way, having worked with statistical number series, including time series financial and economic data, in the computer world for a long time, I am quite comfortable with a representation of a stream of numbers or magnitudes at discrete time intervals. That is exactly analogous to audio signal and to how it is represented in PCM. I can look at the numbers and understand what is going on, but a graph can be helpful with a very long series.
Sure, I prefer a decimal representation, like in a spreadsheet, but I understand powers of 2 so that I can interpret values and magnitudes in binary, too. A lot slower in binary, but I do not see your point at all about a succession of magnitudes in time for successive PCM samples with signed magnitudes being "impossible" to get any feeling for how it relates to the analog signal, even without a graph, even without a scope, just from the stream of magnitudes = levels=signal voltages. There was also no problem with that in the PCM-oriented video on the scope graphical displays.
If if you are looking at PCM data one bit at a time, rather than a sample word at a time, I can see that you would have a problem. We need to look at it one sample at a time, which is a 16,24,32 bit word typically in PCM. Serial pulse trains are meaningless to me. Why would we look a PCM that way rather than as numeric magnitudes?
It is DSD that is one bit = one sample at a time, which is a totally different ballgame, and the succession of bits, the pulse train, is important there. And, sorry, once again without scope, I do not get any good intuitive feel for the analog signal, its level or even its slope at any point in time from the Wiki PDM examples I previously cited. I cannot look at the succession of bits and figure out exactly at what time the + - peaks or zero axis crossings occur in the analog signal just from the succession of PDM/DSD bits. If you can find a way to describe that to me, great. I d not think it is easy or intuitive.
But, as I said, all this is neither here nor there. It is not what we humans can more/less intuitively sense from the a display of the digital data. We normally do not look at the bits nor do we listen to the bits directly, but only after conversion to analog. All that bit interpretation and conversion to analog for audio is engineered and hard-wired into the audio stuff, DACs, etc., we plug into our systems. That stuff does not care if we humans can interpret the digital bits or not. It just plays the music in either format.
But, to your points, I had to search for this excellent video on PCM, which dispels the stair step idea in the first 5 minutes:
http://www.wired.com/2013/02/sound-smart-watch-this-excellent-primer-on-digital-audio/
Yes, as it says, stair steps are created by sample/hold on displays and also for use internally by DACs. But, that does not change the theory of PCM sampling, which is discrete and instantaneous, and it does not affect the analog output from the DAC in the real world of finite, non-instantaneous time. DSD sampling is only theoretically discrete and instantaneous, too, for that matter, and it must also work in finite time. DSD bits would also look like a convoluted up/down one single stair step in a sample/hold display of the digital bits on a scope.
By the way, having worked with statistical number series, including time series financial and economic data, in the computer world for a long time, I am quite comfortable with a representation of a stream of numbers or magnitudes at discrete time intervals. That is exactly analogous to audio signal and to how it is represented in PCM. I can look at the numbers and understand what is going on, but a graph can be helpful with a very long series.
Sure, I prefer a decimal representation, like in a spreadsheet, but I understand powers of 2 so that I can interpret values and magnitudes in binary, too. A lot slower in binary, but I do not see your point at all about a succession of magnitudes in time for successive PCM samples with signed magnitudes being "impossible" to get any feeling for how it relates to the analog signal, even without a graph, even without a scope, just from the stream of magnitudes = levels=signal voltages. There was also no problem with that in the PCM-oriented video on the scope graphical displays.
If if you are looking at PCM data one bit at a time, rather than a sample word at a time, I can see that you would have a problem. We need to look at it one sample at a time, which is a 16,24,32 bit word typically in PCM. Serial pulse trains are meaningless to me. Why would we look a PCM that way rather than as numeric magnitudes?
It is DSD that is one bit = one sample at a time, which is a totally different ballgame, and the succession of bits, the pulse train, is important there. And, sorry, once again without scope, I do not get any good intuitive feel for the analog signal, its level or even its slope at any point in time from the Wiki PDM examples I previously cited. I cannot look at the succession of bits and figure out exactly at what time the + - peaks or zero axis crossings occur in the analog signal just from the succession of PDM/DSD bits. If you can find a way to describe that to me, great. I d not think it is easy or intuitive.
But, as I said, all this is neither here nor there. It is not what we humans can more/less intuitively sense from the a display of the digital data. We normally do not look at the bits nor do we listen to the bits directly, but only after conversion to analog. All that bit interpretation and conversion to analog for audio is engineered and hard-wired into the audio stuff, DACs, etc., we plug into our systems. That stuff does not care if we humans can interpret the digital bits or not. It just plays the music in either format.
Ken Newton
Active Member
- Joined
- Mar 29, 2016
- Messages
- 190
- Likes
- 47
Fitzcaraldo, admire no longer. My persistence has ended.
- Joined
- Feb 23, 2016
- Messages
- 20,696
- Likes
- 37,433
Well I know a few examples that are at $400 as good as you need. Some at $200 that should be.
The theory of digital playback says everything is inaudible; the practical implementation of the analogue side of that playback, using the typical components and methodologies of the moment, means that a significant amount is audible.
OK, thanks for the response and clarification.
I would agree that implementation falls short of theory. If it's audible or not is something to be tested using the proper controls. A lot of the "problems" audiophiles claim to hear are down -120 dB or more when measured. When that is the case, I'm certainly skeptical that it's audible.
Of course, I don't think that community would ever admit that well-designed digital equipment has reached the point of transparency. If they did, the audiophile pursuit of always "tweaking" the system would be much less fun! That's why they misunderstand how digital works and prefer to treat it like analog in the first place.
<Not directed at anyone in particular, just my general comments after skimming a bit here and elsewhere (other fora) this morning. It's Sunday, I've put in about 85 work hours this week and still going, so I'm cranky.>
Good try Ken, but much of the theory falls into the "explaining color to the blind" category. Not only do many audiophiles have no basis for the technical explanations, but many don't care and in fact become incensed when theory and practical data contradicts their world view. You'll never be able to explain something to someone who is convinced of the opposite and in their own infallibility (and equally convinced of your fallibility, ignorance, or whatever).
What's vexing and a little sad is that a few minutes with a 'scope and some appropriate test signals makes it pretty easy to visually distinguish general PCM from DSD, or PWM, etc. Much easier in person.
Audibility is a whole 'nuther ball game.
Good try Ken, but much of the theory falls into the "explaining color to the blind" category. Not only do many audiophiles have no basis for the technical explanations, but many don't care and in fact become incensed when theory and practical data contradicts their world view. You'll never be able to explain something to someone who is convinced of the opposite and in their own infallibility (and equally convinced of your fallibility, ignorance, or whatever).
What's vexing and a little sad is that a few minutes with a 'scope and some appropriate test signals makes it pretty easy to visually distinguish general PCM from DSD, or PWM, etc. Much easier in person.
Audibility is a whole 'nuther ball game.
Similar threads
