Does DSD sound better than PCM?

[Don Hills]

Except that DSD doesn't just have a two level output signal, like a 1-bit PCM. It's a delta modulation scheme. That single bit is used to chip away at the signal slope, very frequently.

No. You have described delta modulation. DSD is sigma-delta modulation.
Delta-Sigma Modulation

 

[captain paranoia]

I was being simplistic, to make the point that it's not a 1-bit PCM.

 

[pjug]

White paper by Bruno Putzeys and Eelco Grimm starts with "DSD is another word for 1-bit PCM sampled at 2.8224MHz"

https://www.grimmaudio.com/site/assets/files/1088/dsd_faq-1.pdf

 

[bothandneither33]

This is such a weird argument to me. First it depends on what kind of system you have. There's gonna be a threshold at which it doesn't matter. Second......Its been a very long time since I took calculus but I know that increasing the amount of samples gives you higher accuracy when measuring the area under the curve. This means the reproduction of the wave is much more accurate the more samples you take. Third, I don't get quite how sample rate is tied to the frequencies we hear but point number 2 makes this argument moot so I don't need to know.

This is really all there is to it. How could more information be bad? You just need a system that can make use of the massive bandwith. If you're using a set of bose through a integrated amp you probably shouldn't be part of the conversation.

 

[andreasmaaan]

Second......Its been a very long time since I took calculus but I know that increasing the amount of samples gives you higher accuracy when measuring the area under the curve. This means the reproduction of the wave is much more accurate the more samples you take. Third, I don't get quite how sample rate is tied to the frequencies we hear but point number 2 makes this argument moot so I don't need to know.

Are you familiar with digital sampling?

100% accuracy is achieved for all frequencies less than half the sample rate. There is just no question about this, sorry. This is a fundamental pillar of digital audio; it just would not work at all if this were not the case.

 

[bothandneither33]

Are you familiar with digital sampling?

100% accuracy is achieved for all frequencies less than half the sample rate. There is just no question about this, sorry. This is a fundamental pillar of digital audio; it just would not work at all if this were not the case.

I don't think you fully understood my post. Do you know what an integral is?

Have you simply listened to music that is upsampled to DSD? On a pair of emerald physics 2.8's with a LKS 004 DAC the difference in SQ between a flac file and that same flac file upsampled to DSD128 is massive. Unquestionably different. DSD128 vs DSD512 gets harder to tell the difference. I think I can but haven't done a blind test yet because ...I guess I'd rather spend the time just enjoying the music.

edit: I'm just gonna be a snob and say that if your system is sub 5k$ its most likely not gonna make a difference. That is a total shot in the dark. Like I said the system needs to be able to make use of the massive amount of information. Not all of them can.

 

[andreasmaaan]

I don't think you fully understood my post. Do you know what an integral is?

I certainly understood it a lot better back in school, it's been a while though

I recall enough about it however to know it's not relevant to this question. Yes, a higher sample rate increases accuracy. However, it increases accuracy only at frequencies above the Nyquist frequency (1/2Fs). It simply doesn't increase the accuracy of frequencies below Nyquist.

 

[bothandneither33]

I certainly understood it a lot better back in school, it's been a while though

I recall enough about it however to know it's not relevant to this question. Yes, a higher sample rate increases accuracy. However, it increases accuracy only at frequencies above the Nyquist frequency (1/2Fs). It simply doesn't increase the accuracy of frequencies below Nyquist.

Integrals are incredibly relevant to the question. They kind of are the question. You definitely should google it.

 

[andreasmaaan]

Integrals are incredibly relevant to the question. They kind of are the question. You definitely should google it.

They are certainly not relevant to the question of how accurately a digital signal is encoded below the Nyquist frequency.

I suggest you read this (or similar - there's plenty of these introductory texts online) and then come back to the discussion.

 

[March Audio]

They are certainly not relevant to the question of how accurately a digital signal is encoded below the Nyquist frequency.

I suggest you read this (or similar - there's plenty of these introductory texts online) and then come back to the discussion.

When in doubt always refer the reader to this

 

[andreasmaaan]

When in doubt always refer the reader to this

Should have remembered it - this guy is awesome

 

[Purité Audio]

He is, has he made any more, I am only aware of two videos?
Keith

 

[March Audio]

Should have remembered it - this guy is awesome

I think its a video that should be supplied to audiophiles (made forcibly to watch ) with every cd/dac purchase.

 

[bothandneither33]

FFT. But I should learn about Nyquist. Integrals turn a continuous wave into a discrete wave so you definitely can't ignore them or not know what they are.

The people critical of DSD. Have you heard the difference yourself? I haven't tried anything other than HQPlayer. And theres only a consistent difference with upsampling. There are many DSD recordings that sound no better than flac files. And again, you gotta do it on a beefy system.

When I found the difference, I barely knew what I was doing. I wasn't sold on anything because I didn't buy anything. I was just trying out all the free and trial options I had in access to.

 

[graz_lag]

For what it is worth within this discussion as the below refers to my personal experience, so it is a 100% subjective feedback :

I have a quite large collection of SACDs, with some of them also in the CD format.
Each time I compare the same album playing the discs on the same player (Oppo 105) I clearly identify more details for the SACD version.
It might well be that some of my SACDs are high resolution PCM recorded SACDs.
Some labels declare whether the SACD was DSD or analog recorded SACD, others not.

I also have a bunch of DSD digital album versions of music within a much larger Hi-Res FLAC/WAV collections, each time I play the same music via the same DAC (Topping DX7S), here again, I clearly identify more details for the DSD version.

I do not know what the technical reason justifying all that might be, but that is the reality of the sound rendering to my ears with my setups.

However, I cannot "survive" to listening to either SACD or digital DSD for any longer than a couple of albums, as it becomes too much fatiguing to my ears.
At that point I feel I need to switch to 16/44.1 CDs/FLACs.

 

[DonH56]

The integral argument is non sequitur IMO. If you integrate (filter) the output of a non-oversampling DAC you get the right answer. If you instead use a single-bit stream with rate sufficient to produce the same resolution and integrate you get the same answer, but in the real world pick up extra noise due to the wider bandwidth. The simple comparison without including resolution does not work; you are taking a multi-level output and comparing to a bi-level output. Oversampling can reduce quantization noise but does not improve resolution. Changing from e.g. a 16-bit output at 44 kS/s to a single-bit output at whatever MS/s does not improve the reconstructed waveform assuming the proper anti-image filter is applied to each.

There are some articles in the technical area of this site to explain sampling and such.

 

[bennetng]

On a pair of emerald physics 2.8's with a LKS 004 DAC the difference in SQ between a flac file and that same flac file upsampled to DSD128 is massive.

Why don't just send your LKS 004 to Amir and getting it measured in both PCM and DSD in different rates?

 

[swanlee]

To me DSD has different qualities than PCM, DSD sounds warmer, has deeper bass, better time decay, cymbals in particular have a more solid sound than the thinner sound of PCM.

One thing I think PCM has that is better is top end Freq, highs tend to be more bright and detailed than DSD. Overall though I prefer the sound signature of DSD. Been listening to both since high res and DSD was introduced to the public in the early 2000's and these are the generalizations I have come away with.

 

[Miska]

Changing from e.g. a 16-bit output at 44 kS/s to a single-bit output at whatever MS/s does not improve the reconstructed waveform assuming the proper anti-image filter is applied to each.

You just stated what it's all about. I don't assume something. I measure which input format to a particular DAC (chip and entire device) gives best reconstruction of the signal... How to squeeze out best performance in practice, not from theoretical point of view.

And I don't need to consider any computing power limits when producing what ever format to send to the DAC. I just put both CPU and GPU at work, some work for ~40 billion transistors.

 

[DonH56]

You just stated what it's all about. I don't assume something. I measure which input format to a particular DAC (chip and entire device) gives best reconstruction of the signal... How to squeeze out best performance in practice, not from theoretical point of view.

And I don't need to consider any computing power limits when producing what ever format to send to the DAC. I just put both CPU and GPU at work, some work for ~40 billion transistors.

Then you are assuming anti-image filters are inadequate and are assuming your definition of "best reconstruction" is the best there is. I am not sure it is possible to not assume something, somewhere, at a very basic level.

I said nothing about computing limitations; not my field, decades since my digital filter design classes, and I had no thought that modern CPUs and/or DSPs couldn't handle the load.

I definitely need to stay off the 'net for a while...