How do you master a DSD recording?

[71 dB]

In terms of the claims about reproduction and its perfection, I am looking at videos by https://youtube.com/channel/UCb_NEjjKOXV9pilaSOjlkZA to understand how PCM works. Right now trying to figure out how the nynquistic theorem makes sense. I found it to be really interesting when it says 2x and a bit more sometimes of a sampling rate and perfectly capture and reproduce an analog signal. Just can’t seem to get over the two sampling points can construct the sound wave completely part.

Listen to music,
J

It is all about band-limited signals. A properly band-limited signal can't "move" between the sampling points in any other way, because it would need frequencies above the Nyquist frequency to do that, but a properly executed digital system filters all those frequencies away so that what remains is the correct band-limited signal.

Digital audio can be a bit (pun intended) unintuitive sometimes, but it is built on proven mathematical facts one can learn to understand and accept even on intuitive level with some effort. Without this solid mathematical foundation, the development of digital audio to REPLACE analog audio would have been a fools errand.

 

[gregorio]

Back to the main topic, so far I have seen this statement: “Then there are the differences in the ways DAC chips work. Most modern DAC chips are Delta-Sigma which decode native DSD. R-2R DAC chips decode native PCM. In order for you to play PCM files on a Delta-Sigma DAC or DSD files on an R-2R DAC the files have to be converted in real time.”
- https://www.mojo-audio.com/blog/dsd-vs-pcm-myth-vs-truth/

Again, you need to be careful what you read on the internet. That article should really be called: “DSD vs PCM, myth vs nonsense I’ve just made up”. It contains almost no truth at all, even the basic history is false, so is the 1 photograph and so are most of the stated facts. Some of it should be obvious, such as wax cylinders effectively being the highest fidelity and every development since being a further reduction in quality. Most of it though wouldn’t be obvious unless you had a reasonable knowledge/understanding of the basic facts.

Just can’t seem to get over the two sampling points can construct the sound wave completely part.

Yep, it’s not intuitive if you think in terms of the “analogue” concept, which of course pretty much everyone does, and it’s this fact that’s exploited by marketing BS. An analogy might help:

Let’s say we have a perfect circle and we want to store and reproduce it. If we measure that circle using 2 points, store those 2 points and then reproduce it by joining them together again, say with a printer, we’d actually get a straight line, not a circle. If we used 3 points, the result would be more circle like but it still wouldn’t be a circle, it would be a triangle. 4 points would give us a square, 5 a pentagon and so on. As each additional point gets us closer to a circle it’s therefore simple logic that the more points the better, although we could never quite reproduce a perfect circle because we would need an infinite number of points. This is the “analogue concept”. The “digital concept” would be based on the fact that we can easily define all circles with a fairly simple mathematical equation. So, we program our printer to ONLY print perfect circles and then all we need to measure and store our original circle is 2 points, because anything the printer prints which bisects these 2 points must be a perfect circle identical to our original. More than 2 points makes no difference, we’re going to get the exact same result whether we use two or a million points. This is obviously a gross over simplification and to translate it to digital audio you have to understand or just accept some mathematical axioms. For example, those discovered/proven by Fourier 200 years ago, which show that all sound waves are made of combinations of sine waves (so perfect sine waves instead of perfect circles) and those discovered/proven by Shannon in 1948. This acceptance is made easier by the fact that after all this time, some mathematician would have made a huge name for him/herself by disproving it but more tellingly, so much of our modern world relies on these mathematical axioms that it simply wouldn’t exist if they were wrong. For example, the modern world would be quite different if someone had proved that one plus one does not equal two.

About mastering, I like classical music. Which I believe is light on mixing and very focused on mastering considering the remastered releases there are. This is just an inference, I have not got to the stage of understanding mastering yet in my research. Please do shed some light on this subject if you guys don’t mind!

This is a big subject area that spans many decades, much technology and subjective decisions, both of which have evolved significantly over that period. So it’s easier to respond to specific questions than explain all about mixing and mastering.

If we take the term “mixing” literally, then classical music contains a lot of it. An orchestra for example is typically recorded with at least 20 mics and possibly as many as 50 or more, all of which have to be mixed together (and balanced and “panned”/positioned relative to each other) to form our 2 channels of Stereo, so that’s a lot of mixing. Even a solo, unaccompanied piano would typically use at least 3 mics and often more. However, the term “mixing” doesn’t only include mixing the sources (mic outputs) together, it also includes the application of “effects”, such as EQ, reverb, compression and many dozens of others. In classical music we tend to only use those first 3 and typically quite subtly compared to all the “popular” music genres.

Mastering is the process of taking the final mix produced in the recording studio and applying effects to make it sound good during reproduction on consumer systems, rather than only in the studio/s in which it was created. In general, classical music is processed less during mastering than popular genres. Unfortunately, the use of the term “re-mastering” can be a bit misleading because releases that have actually been remixed and remastered are sometimes just called “remasters”.

G

 

[Chimmy9278]

Again, you need to be careful what you read on the internet. That article should really be called: “DSD vs PCM, myth vs nonsense I’ve just made up”. It contains almost no truth at all, even the basic history is false, so is the 1 photograph and so are most of the stated facts. Some of it should be obvious, such as wax cylinders effectively being the highest fidelity and every development since being a further reduction in quality. Most of it though wouldn’t be obvious unless you had a reasonable knowledge/understanding of the basic facts.

Yep, it’s not intuitive if you think in terms of the “analogue” concept, which of course pretty much everyone does, and it’s this fact that’s exploited by marketing BS. An analogy might help:

Let’s say we have a perfect circle and we want to store and reproduce it. If we measure that circle using 2 points, store those 2 points and then reproduce it by joining them together again, say with a printer, we’d actually get a straight line, not a circle. If we used 3 points, the result would be more circle like but it still wouldn’t be a circle, it would be a triangle. 4 points would give us a square, 5 a pentagon and so on. As each additional point gets us closer to a circle it’s therefore simple logic that the more points the better, although we could never quite reproduce a perfect circle because we would need an infinite number of points. This is the “analogue concept”. The “digital concept” would be based on the fact that we can easily define all circles with a fairly simple mathematical equation. So, we program our printer to ONLY print perfect circles and then all we need to measure and store our original circle is 2 points, because anything the printer prints which bisects these 2 points must be a perfect circle identical to our original. More than 2 points makes no difference, we’re going to get the exact same result whether we use two or a million points. This is obviously a gross over simplification and to translate it to digital audio you have to understand or just accept some mathematical axioms. For example, those discovered/proven by Fourier 200 years ago, which show that all sound waves are made of combinations of sine waves (so perfect sine waves instead of perfect circles) and those discovered/proven by Shannon in 1948. This acceptance is made easier by the fact that after all this time, some mathematician would have made a huge name for him/herself by disproving it but more tellingly, so much of our modern world relies on these mathematical axioms that it simply wouldn’t exist if they were wrong. For example, the modern world would be quite different if someone had proved that one plus one does not equal two.

This is a big subject area that spans many decades, much technology and subjective decisions, both of which have evolved significantly over that period. So it’s easier to respond to specific questions than explain all about mixing and mastering.

If we take the term “mixing” literally, then classical music contains a lot of it. An orchestra for example is typically recorded with at least 20 mics and possibly as many as 50 or more, all of which have to be mixed together (and balanced and “panned”/positioned relative to each other) to form our 2 channels of Stereo, so that’s a lot of mixing. Even a solo, unaccompanied piano would typically use at least 3 mics and often more. However, the term “mixing” doesn’t only include mixing the sources (mic outputs) together, it also includes the application of “effects”, such as EQ, reverb, compression and many dozens of others. In classical music we tend to only use those first 3 and typically quite subtly compared to all the “popular” music genres.

Mastering is the process of taking the final mix produced in the recording studio and applying effects to make it sound good during reproduction on consumer systems, rather than only in the studio/s in which it was created. In general, classical music is processed less during mastering than popular genres. Unfortunately, the use of the term “re-mastering” can be a bit misleading because releases that have actually been remixed and remastered are sometimes just called “remasters”.

G

That article was fishy, I knew there was something wrong with it. Glad I posted it here to be peer reviewed. Maybe a re read after watching those videos will show me how bad it really is.

I can get over the fact that it can create one sin wave, but what about the added sin waves? It just seems to me that the local maximum and minimums cannot be faithfully captured. But I do trust Fourier’s math. At the end of the day I don’t think I know enough advanced math to fully understand their theorems, so for now I will just accept it. A tough piece to swallow.

Lastly, I have read somewhere that there is not much mixing regarding classical music, but as you said and I recall from previous experiences, the engineers do indeed pan and balance the channels. So in reality a large number of the remastered releases and actually remixed and remastered.
So in a broad sense, I should keep in mind different releases will have a different mix causing a larger difference than the format it is released in.

Therefore the testing method for further blind tests should indeed be done by converting it to PCM and back to DSD to eliminate these variables.

Reading,
J

 

[71 dB]

The “digital concept” would be based on the fact that we can easily define all circles with a fairly simple mathematical equation. So, we program our printer to ONLY print perfect circles and then all we need to measure and store our original circle is 2 points, because anything the printer prints which bisects these 2 points must be a perfect circle identical to our original. More than 2 points makes no difference, we’re going to get the exact same result whether we use two or a million points.
G

Actually you need 3 points to store the information of an arbitrary circle. For example circles

x² + y² = 1

and

x² + ( y - 1 )² = 2

both share the coordinate points (-1, 0) and (1, 0), but only the former also contains the point (0, 1). So if I give you points (-1, 0), (1, 0) and (0, 1) and say they represent a circle, you (should) know I mean the circle x² + y² = 1. If I only give you points (-1, 0) and (1, 0), you have infinite options for the circle:

x² + ( y - 𝛼 )² = 1 + 𝛼²

where 𝛼 is a free parameter. Nitpicking perhaps, but lets make the math correct, since this is supposed to be the science sub-forum.

 

[Chimmy9278]

Actually you need 3 points to store the information of an arbitrary circle. For example circles

x² + y² = 1

and

x² + ( y - 1 )² = 2

both share the coordinate points (-1, 0) and (1, 0), but only the former also contains the point (0, 1). So if I give you points (-1, 0), (1, 0) and (0, 1) and say they represent a circle, you (should) know I mean the circle x² + y² = 1. If I only give you points (-1, 0) and (1, 0), you have infinite options for the circle:

x² + ( y - 𝛼 )² = 1 + 𝛼²

where 𝛼 is a free parameter. Nitpicking perhaps, but lets make the math correct, since this is supposed to be the science sub-forum.

Ignoring the math, for now. A sin wave only needs 2 right?

 

[71 dB]

Ignoring the math, for now. A sin wave only needs 2 right?

Sine wave needs three points of data: Frequency, phase and amplitude

 

[sander99]

Sine wave needs three points of data: Frequency, phase and amplitude

Now you risk confusing him!

 

[Chimmy9278]

Sine wave needs three points of data: Frequency, phase and amplitude

I don’t think phase is needed when digitizing music? Regardless, I must be missing something here since what you are saying is at odds of the nynquistic theorem.

 

[sander99]

Sine wave needs three points of data: Frequency, phase and amplitude

I don’t think phase is needed when digitizing music? Regardless, I must be missing something here since what you are saying is at odds of the nynquistic theorem.

I am not really familiar with all the math details of the sampling theorem, but the trick will be something like this I think:
If you know nothing else then you need those three points of data (which is something else than three sample points by the way) to define a sine wave. But the sampling theorem is about the situation where you have more information besides the sampled points. For example you know there are no frequencies above half the sampling frequency.

 

[gregorio]

Maybe a re read after watching those videos will show me how bad it really is.

It will hopefully show you that it’s worse than “fishy” but not how bad it really is. You need to know the history of recording and probably a bit more knowledge of analogue and digital technology then those videos can provide. A list of the accurate/truthful facts would be much shorter than a list of the incorrect ones and even with the correct ones, he usually draws incorrect conclusions from them. Whole sections are therefore just complete nonsense, it really is terrible!

I can get over the fact that it can create one sin wave, but what about the added sin waves? It just seems to me that the local maximum and minimums cannot be faithfully captured. But I do trust Fourier’s math.

What Fourier did, was provide the equations that allow us to take a waveform (of any complexity) and break it down into all it’s component sine waves and vice versa (a Fourier Transform and an Inverse Fourier Transform). What Shannon did, was provide the equations that would allow the perfect capture of all those sine waves, under the condition that those sine waves have an audio frequency less than the sample rate divided by 2.

So in a broad sense, I should keep in mind different releases will have a different mix causing a larger difference than the format it is released in.

A far, far larger difference. The different lossless formats were specifically chosen to be transparent to human hearing. In the case of CD, a sampling rate of 44.1kHz which therefore allows for sine waves up to 22.05kHz, which is beyond human hearing and a bit depth of 16bit which puts quantisation error/noise below the level of audibility. Subsequent formats specified either higher sample rates or more more bits and sometimes both, which results in them being at least equally transparent to human hearing. So, there are differences between formats but they are only measurable, not audible. This is the exact opposite of different masters/remasters/remixes which are specifically designed to have differences that ARE audible! Why would anyone pay a mastering engineer to create a different master that sounds exactly the same as a master they’ve already paid for? The whole point is for it to sound different.

Therefore the testing method for further blind tests should indeed be done by converting it to PCM and back to DSD to eliminate these variables.

Or just converting the DSD master to PCM and then comparing these two masters that are identical except for the format. If you compare the DSD version to say a CD version or other supplied version, even if they are from the same disk, you could actually be comparing different masters, in which case you should be able to hear a difference.

You might find this peer reviewed published study by Meyer and Moran interesting, because it was large scale, has often been quoted and discussed in great detail and deals specifically with the issues.

G

 

[71 dB]

Now you risk confusing him!

I know, but if he ever wishes to understand these things well, he needs to get confused temporarily. He needs to think his way out of the confusion.

I don’t think phase is needed when digitizing music? Regardless, I must be missing something here since what you are saying is at odds of the nynquistic theorem.

Depents on how you think of this. The digitized data MUST contain all information of the digitized band-limited analog signal, or the reproduction of the original analog signal isn't possible (at least 100 % accurately). What I say is not at odds of the Nyquist theorem, because what do you mean by 2 or 3 points of data? Sampling a bandlimited signal is different from giving functions in parametric form.

 

[71 dB]

I am not really familiar with all the math details of the sampling theorem, but the trick will be something like this I think:
If you know nothing else then you need those three points of data (which is something else than three sample points by the way) to define a sine wave. But the sampling theorem is about the situation where you have more information besides the sampled points. For example you know there are no frequencies above half the sampling frequency.

Since digital audio deals with band-limited signals, the signals can't start and stop immediately. This obscures the amount of "data triples" needed, because the signal is spread in time a little bit. Very short busts can't exist in digital audio, because that requires infinite bandwidth. So, you never deal with just 2 samples. That doesn't make sense in digital audio. In a way there is a "lower" Nyquist frequency fs/3 and above that frequency the accuracy of sampling is dependent on the length of the signal, but in practice this doesn't matter. Our mind just tricks us to think about the signal between fs/3 and fs/2 in ways that aren't allowed with bandlimited signals.

 

[gregorio]

Nitpicking perhaps, but lets make the math correct, since this is supposed to be the science sub-forum.

Yes, you are correct, both in really needing 3 references and in nitpicking! I was deliberately trying to avoid the math and provide an analogy that was easy to visualise. A simple understanding that at least doesn’t contradict the basic facts of digital audio is best to start with and then we can get into the actual scientific details later.

Ignoring the math, for now. A sin wave only needs 2 right?

Technically you need slightly more, say an average of 2.1 would be fine. To put it another way, a sample rate of 44.1kHz can perfectly capture sine waves up to but not including or beyond 22.05kHz. Also, bare in mind we’re not talking about sine waves only, we’re talking about ALL continuous (time variant) wave forms. A digital audio signal is just a sequence of zeros and ones, that represent the amplitude of the analogue waveform. That’s it, there’s nothing else! The timing/phase are implicit in the sample rate and frequency is derived from that time and amplitude information.

Regardless, I must be missing something here since what you are saying is at odds of the nynquistic theorem.

Yes, you/we are talking about sample points and @71 dB is talking about reference points, not quite the same thing. I do NOT agree that more confusion is a good thing at this point!

Since digital audio deals with band-limited signals, the signals can't start and stop immediately. This obscures the amount of "data triples" needed, because the signal is spread in time a little bit…

You’re really not helping at this point, you’re doing the opposite!

G

 

[71 dB]

Yes, you are correct, both in really needing 3 references and in nitpicking! I was deliberately trying to avoid the math and provide an analogy that was easy to visualise. A simple understanding that at least doesn’t contradict the basic facts of digital audio is best to start with and then we can get into the actual scientific details later.

Unfortunately the logic jump from circles to sampling analog signals is not that straighforward. If the sampling theorem was easy to understand, mankind would have figured it out centuries earlier.

You’re really not helping at this point, you’re doing the opposite!

G

This is debatable. I think the worst thing is people thinking they understand something when they don't (very common with digital audio!). My tactic is to scare people away from thinking they understand these things or be motivated to do the work required to learn this stuff.

 

[gregorio]

Unfortunately the logic jump from circles to sampling analog signals is not that straighforward.

Before one can make that logic jump one has to have a basic concept of the logic, which is what my analogy attempted to achieve. Once that basic concept is there, then you can start going deeper.

If the sampling theorem was easy to understand, mankind would have figured it out centuries earlier.

It took mankind until 1948 because it’s very difficult to figure out and prove but chimmy doesn’t need to figure out and prove it, he just needs a usable simple, layman’s understanding at this stage. If he wants an in depth mathematical understanding at a later stage, then your comments would be useful but until that time they’re not, they’re just adding more confusion.

G