To DSD or not to DSD?

[EuroDriver]

Miska is the person on CA I trust the most. AFAIK, his knowledge and experience regarding PCM and DSD is without peer. His contributions on CA are the most important for me and they confirm my opinion about the superiority of DSD128+ over PCM.

 

KR

Matt

 

and DSD256 is significantly better than DSD128 :-)

[PeterSt]

Well, many recordings are done in PCM (and most at 44.1/48kHz at that!) so no wonder he is advocating upsampling them to 2x DSD. If all recordings were done at 2x DSD they could simply be sent to his 2x DSD DACs as is (without downsampling/decimation/digital LPF/upsampling/brick wall fiters/SRC).

 

 

This is not real life graphing; it's a snapshot.

 

This is :

 

And since I'm a nice person, I also took photos of my 100% analog scope for TEAC UD-501.

 

1k square at DSD128:

[ATTACH=CONFIG]9645[/ATTACH]

 

20k sine at DSD128:

[ATTACH=CONFIG]9646[/ATTACH]

 

 

Edit : above links of course don't work well, so now here :

 

[ATTACH=CONFIG]14771[/ATTACH]

 

And since this latter is Miska's own it should be OK to show it. But what I will add is that this is very noisy. For that compare the upgoing slope's (vertical) line thickness with how the level sustains (horizontal's thickness) in the square wave. This with the notice that the vertical slope is even subject to jitter (but the scope will not be fast enough to show the variation (thicker line) of this, while the "frequency" that sustains the level (horizontal) is relatively very thick). Here too the jitter can not be shown, hence it is jot jitter causing that (or otherwise vertical would be as thick) and thus this is "random" noise (which the fluorescent display keeps visible).

 

Looking back at Hiro's picture(s), this would be reality for PCM and on the fluorescent display it would look the same (with the assumption that it is again not fast enough to show jitter).

 

In my view you can go two directions with this :

 

1. The noise on DSD is random and thus de-correlated;

2. The frequency required to sustain such square wave is not a steady one and thus is the signal also not.

 

Undoubtedly Miska will explain it towards #1.

I pick #2 though because the "noise" as I tried to put it forward myself is not the noise we talk about (always and ever) but the impossibility for DSD to sustain any steady level. The noise will have its influence too, but it is fact that this level can not be sustained which is noise as such in itself. This can be read as the high frequency required to form such a square is subject to very high THD.

 

Is this important ? I don't judge. But "high frequency" (or square waves) don't look as nice as Hiro's picture tries to tell.

Here's a more famous one from (too !) long ago, but taken with a high frequency analog scope (to my belief same as I have myself) :

 

 

So this emphasizes what I mean to say.

[matthias]

and DSD256 is significantly better than DSD128 :-)

 

Indeed

 

Matt

[Hiro]

Is this important ? I don't judge. But "high frequency" (or square waves) don't look as nice as Hiro's picture tries to tell.

 

Peter, the measurements I posted come from Miska/Signalyst and Ed Meitner/EMM Labs. And they clearly show that DSD can be vastly superior to what your pic would suggest.

 

I believe that your picture is misleading in that it uses a 10kHz square, while DSD sampling rate is limited to dsd64 in this case.

 

It's a typical tactic of DSD bashers to set the parameters of their measurements in such a way that the subsequent results of those tests will present DSD in the worst light possible. For this reason they will commonly pit DSD64 against 24/192 and 32/358 that use much higher bit rates. Take note that 32/358 has a bitrate equal to that of DSD256.

[Maldur]

But "high frequency" (or square waves) don't look as nice as Hiro's picture tries to tell.

Here's a more famous one from (too !) long ago, but taken with a high frequency analog scope (to my belief same as I have myself) :

 

[ATTACH=CONFIG]14770[/ATTACH]

 

So this emphasizes what I mean to say.

 

Yes indeed. Both Hiro and Miska shows pictures of DSD128, but You showed picture of DSD64 and then ""high frequency" (or square waves) don't look as nice as Hiro's picture tries to tell"".

 

Picture with DSD128 will be better argument.

 

Edit: I found that site where the DSD64 picture comes - Craigman Digital - PCM vs DSD

[marslo]

Well, I am not a techi guy so I do not care how the music is processed before I can listen it at home. I like very much the analogue sounding dsd but the availibility of 128 and 256 dsd is very limited and require fast internet connections to download . Then it needs much more space on hd.

The other issue is the lack of download manager in many " net dsd suppliers" .

So I basicly listen to my sacd ripps which sounds wonderful over lampi dsd dac.

And after few days of dsd only music I switch a dac and listen with pleasure my La Fontaine Dac and reference recordings pcm files which are just fantastic . And much more. In my opinion PCM is just more dynamic and excels dsd as far as deep bass and high frequencies are concerned.

So what bad with listening to both pcm and dsd native files?

And then to switch back to TT and vinyl?

[PeterSt]

Peter, the measurements I posted come from Miska/Signalyst and Ed Meitner/EMM Labs. And they clearly show that DSD can be vastly superior to what your pic would suggest.

 

I believe that your picture is misleading in that it uses a 10kHz square, while DSD sampling rate is limited to dsd64 in this case.

 

And I believe that you don't read very well.

I also believe that this forum tends to remove thumbnails of pictures at inconvenient moments. :)

Lastly I believe that you did not try to open those links or otherwise you wouldn't have said what you did.

 

Heck, I too showed pictures from Miska (read my text).

Next I showed a "too old" picture to EMPHASIZE what I mean.

 

It's a typical tactic of DSD bashers to set the parameters of their measurements in such a way that the subsequent results of those tests will present DSD in the worst light possible.

 

Are you addressing me ?? Can you quote me from ANYWHERE that I do so ?

 

Is this important ? I don't judge.

 

At least in my last post that's the only thing I said.

 

Peter, the measurements I posted come from Miska/Signalyst and Ed Meitner/EMM Labs. And they clearly show that DSD can be vastly superior to what your pic would suggest.

 

Now please read again my post, open at least that one link under my "Edit" (that works) and see how I myself also "clearly show something". Damn it, here, re-try :

 

 

 

 

And since this latter is Miska's own it should be OK to show it. But what I will add is that this is very noisy. For that compare the upgoing slope's (vertical) line thickness with how the level sustains (horizontal's thickness) in the square wave. This with the notice that the vertical slope is even subject to jitter (but the scope will not be fast enough to show the variation (thicker line) of this, while the "frequency" that sustains the level (horizontal) is relatively very thick). Here too the jitter can not be shown, hence it is not jitter causing that (or otherwise vertical would be as thick) and thus this is "random" noise (which the fluorescent display keeps visible).

 

Looking back at Hiro's picture(s), this would be reality for PCM and on the fluorescent display it would look the same (with the assumption that it is again not fast enough to show jitter).

[Hiro]

Now please read again my post, open at least that one link under my "Edit" (that works) and see how I myself also "clearly show something". Damn it, here, re-try :

 

 

 

[ATTACH=CONFIG]14772[/ATTACH]

 

And since this latter is Miska's own it should be OK to show it. But what I will add is that this is very noisy.

 

I don't think that DSD128 is "very noisy", as it has ~ 120dB of DR up to 50kHz. I think you're reading too much to this picture (but then again, you said it yourself: "Is this important ? I don't judge.").

 

To me the DSD128 graph looks very much like a square wave, much more so than, as you said, the "more famous" pic you posted.

 

It's important to keep in mind the fact that the measured DSD performance (SNR, thd, jitter, etc) is dependent on delta sigma modulator design (modulator order, sampling rate). So no one picture (above of the teac ud-501) is representative of DSD capability as a whole. What I can say for sure by looking at these measurements, be it of the Teac or EMM Labs is that things are constantly being improved in this field.

[PeterSt]

Edit: I found that site where the DSD64 picture comes - Craigman Digital - PCM vs DSD

 

Yes, and it is from 2005 IIRC.

 

Yes indeed. Both Hiro and Miska shows pictures of DSD128, but You showed picture of DSD64 and then ""high frequency" (or square waves) don't look as nice as Hiro's picture tries to tell"".

 

If someone thinks this is important then he does not understand;

Regard the horizontal "line" you see as a fast sine frequency which needs to stay "level" in order to sustain the level (haha) straight. It does not matter whether 2,3,4 or 667 sines show there (the more the higher the frequency of the square wil be), but they must stay put. Each time the wave cycle is overlaid it has to appear at the same position. Not only the larger scale square you see, but also the elements it is built from. That are those high frequency sines.

 

If you take one snapshot of such a cycle (which is typically what a digital scope will do without "sustainment" also called digital phosphor, then you will get what Hiro showed. Now THAT is fake. Or ... it is PCM because PCM will do it like that. It is steady as a rock to that respect unless the filter is wacky.

An analog scope will show inherently more than one wave cycle (on the same position) because the fluorescent display works like that (the trace dies out in a relative long time). This is how Miska's picture shows reality.

 

That's all and I hope it is clear better now.

[davide256]

Well, I am not a techi guy so I do not care how the music is processed before I can listen it at home. I like very much the analogue sounding dsd but the availibility of 128 and 256 dsd is very limited and require fast internet connections to download . Then it needs much more space on hd.

The other issue is the lack of download manager in many " net dsd suppliers" .

So I basicly listen to my sacd ripps which sounds wonderful over lampi dsd dac.

And after few days of dsd only music I switch a dac and listen with pleasure my La Fontaine Dac and reference recordings pcm files which are just fantastic . And much more. In my opinion PCM is just more dynamic and excels dsd as far as deep bass and high frequencies are concerned.

So what bad with listening to both pcm and dsd native files?

And then to switch back to TT and vinyl?

 

Agree on the lack of reference native DSD content. Would be lovely if some of the other symphony orchestras could emulate what the San Francisco Symphony has done in self publishing their own native DSD performances. Suspect that might fall afoul of commercial agreements for the most well known orcestras but less published orchestras might succeed, especially in conjunction with an audiophile publishing label. Given how competitive it is to play in a symphony orchestra I have no doubt these would be excellent recordings for sonics, just not the degree of conducting interpretation that discophiles seek.

[spdif-usb]

Look at your post #1067.

 

Matt

So, it is only your "AFAIK" against mine. The final thing that matters is that D. Reefman and P. Nuijten were also using measurements to "prove" such and such, but they were later proven wrong by S.P. Lipshitz and J. Vanderkooy, and that is obviously one of the very perfect reasons why measurements alone don't cut it. Furthermore, I don't just want to hear that it's possible for DSD to be properly dithered nowadays. Instead, I want to hear from you how it is possible when there exists rock solid mathematical proof of the fact it can never be possible. Ever.

[PeterSt]

I don't think that DSD128 is "very noisy", as it has ~ 120dB of DR up to 50kHz. I think you're reading too much to this picture (but then again, you said it yourself: "Is this important ? I don't judge.").

 

Hiro,

 

I tried to explain that too ... this is NOT about the "always and ever" talked about noise. It is just about how DSD can't sustain a steady level. We can almost say it was made for that (well, if it is explained like Miska can nicely do it : try to stand on a rope without moving and you'll fall off).

 

The importance of my shouting could be that this in itself is a sheer reason how high frequency is nice, but if it can not be rendered well (hence high THD) then what. In the end something else will happen though and this is randomizing of the inherently "sharp" signal. Call it analogue if you like.

 

One more notice, because this goes quite beyond what we (me too !) are used to :

The THD I talk about is not about any "sine" THD from test signals. It is from signals beyond the audio band. So, if a square (like) signal of say 12KHz requires 36Khz of sine frequency to be rendered like intended, then there will be a high THD on that 12KHz square signal because of the 36KHz sine showing the bad THD. But

a. we don't measure 36KHz test signals in order to judge how squares in the audio band will behave and

b. we don't measure the THD of squares in the first place.

Still it will do something to audibility and let's say it is an easy explanation how DSD128 (and 256 etc.) will do better here than DSD 64. Still it is only partly about that because all what really happens is that the *deviation* to what the signal is supposed to be, is more quickly corrected. Say that the stick you hold being on that rope can be shorter and still works well.

 

And on that other hand, while PCM does fine here, it can't exhibit the frequency in the first place. Of course there's 100 things more pro and con to say about this, but just showing that I am not bashing any format. Explicitly not.

[Hiro]

The importance of my shouting could be that this in itself is a sheer reason how high frequency is nice, but if it can not be rendered well (hence high THD) then what.

 

Grimm AD1 has THD at -129dB, is that high THD to you?

[spdif-usb]

Indeed

 

Matt

Indeed, DSD256 is better than DSD128, but that phact says nothing about the differences between DSD and PCM. Also indeed.

[spdif-usb]

Hiro,

 

I tried to explain that too ... this is NOT about the "always and ever" talked about noise. It is just about how DSD can't sustain a steady level. We can almost say it was made for that (well, if it is explained like Miska can nicely do it : try to stand on a rope without moving and you'll fall off).

 

The importance of my shouting could be that this in itself is a sheer reason how high frequency is nice, but if it can not be rendered well (hence high THD) then what. In the end something else will happen though and this is randomizing of the inherently "sharp" signal. Call it analogue if you like.

 

One more notice, because this goes quite beyond what we (me too !) are used to :

The THD I talk about is not about any "sine" THD from test signals. It is from signals beyond the audio band. So, if a square (like) signal of say 12KHz requires 36Khz of sine frequency to be rendered like intended, then there will be a high THD on that 12KHz square signal because of the 36KHz sine showing the bad THD. But

a. we don't measure 36KHz test signals in order to judge how squares in the audio band will behave and

b. we don't measure the THD of squares in the first place.

Still it will do something to audibility and let's say it is an easy explanation how DSD128 (and 256 etc.) will do better here than DSD 64. Still it is only partly about that because all what really happens is that the *deviation* to what the signal is supposed to be, is more quickly corrected. Say that the stick you hold being on that rope can be shorter and still works well.

 

And on that other hand, while PCM does fine here, it can't exhibit the frequency in the first place. Of course there's 100 things more pro and con to say about this, but just showing that I am not bashing any format. Explicitly not.

Peter,

 

Next time, just ask Hiro if he has any affiliation with Sony because then we can all laugh.

[Hiro]

Peter,

 

Next time, just ask Hiro if he has any affiliation with Sony because then we can all laugh.

 

Ah, the great argument. You like DSD, and you don't subscribe into the PCM-only world of mega-buck converters, therefore you are affiliated with Sony. SMH .....

[PeterSt]

Originally Posted by PeterSt

The importance of my shouting could be that this in itself is a sheer reason how high frequency is nice, but if it can not be rendered well (hence high THD) then what.

 

Grimm AD1 has THD at -129dB, is that high THD to you?

 

Are you always quoting so the context is out of it ?

 

So show me at 36KHz because that was what I was talking about.

I feel that you miss a few things ...

[PeterSt]

Peter,

 

Next time, just ask Hiro if he has any affiliation with Sony because then we can all laugh.

 

Don't need to ask that.

But funny ... from how many years back would that be ?

Nah, just kidding. But still funny.

 

Let's keep in mind : I personally think that "we" progressed quite a lot on the noise issue and I am not afraid to give Miska the credits for that but with the notice that the Philips guys (or some of them) in the last few years achieved quite similar on paper. And or but, you won't hear me complain about that noise much because I myself am in a quite similar league. But there is much more going on for differences and it makes it even more hard to compare.

[Hiro]

Are you always quoting so the context is out of it ?

 

So show me at 36KHz because that was what I was talking about.

I feel that you miss a few things ...

 

Peter,

 

36kHz are not high frequencies, they are inaudible frequencies.

[PeterSt]

Hiro, I am completely sure now : you don't read a thing of what I write. Read back ...

Edit : OK ?

[Hiro]

Let's keep in mind : I personally think that "we" progressed quite a lot on the noise issue and I am not afraid to give Miska the credits for that

 

I'm glad we can agree on that.

 

And I am too impressed by Miska's work in optimizing delta sigma modulators.

[Hiro]

PCM won't be "properly" done as long as it happens inside a tiny $10 chip. Very likely not the DSD either...

 

Direct input for SDM on a chip is a nice way to bypass the often inferior oversampling filters in DAC chips, isn't it?

[Miska]

I want to hear from you how it is possible when there exists rock solid mathematical proof of the fact it can never be possible. Ever.

 

It cannot be proven exhaustively for a non-linear random system because you cannot cover all time-value space possibilities. You cannot even fully model the system to contain the output value space for a known modulator design.

 

Speaking of mathematical proofs, it is easy to prove that you can never correctly reconstruct PCM in real world, because for that you would have to be able realize infinity in real world.

[spdif-usb]

Don't need to ask that.

But funny ... from how many years back would that be ?

Nah, just kidding. But still funny.

 

Let's keep in mind : I personally think that "we" progressed quite a lot on the noise issue and I am not afraid to give Miska the credits for that but with the notice that the Philips guys (or some of them) in the last few years achieved quite similar on paper. And or but, you won't hear me complain about that noise much because I myself am in a quite similar league. But there is much more going on for differences and it makes it even more hard to compare.

You won't hear me complain about that noise much either because I myself typically prefer the groove noise of the vinyl LP.

[PeterSt]

You won't hear me complain about that noise much either because I myself typically prefer the groove noise of the vinyl LP.

 

Oh. Well, I only like the smell of it. And the visuals.

Otherwise it is lacking all the dynamics in the world. By now it does, haha. But this is for other many threads.