DSD-2-PCM -- proof of concept, test sample and source code here
- No over 30 sec clips of copyrighted music. Cite properly and never more than necessary for the discussion.
- No copyrighted software without permission.
- Click here for complete Hydrogenaudio Terms of Service
DSD-2-PCM -- proof of concept, test sample and source code here
Oct 8 2005, 23:55
Joined: 20-March 04
From: Göttingen (DE)
Member No.: 12875
The ZIP contains 4 files:
- dsd2pcm.jar (the converter written in Java, outputs raw 24/88 PCM, intel byte order)
- info.txt (contains some infos)
- test2822k.dsd (14 seconds, mono DSD, 5 megs)
- test44k.mp3 (conversion result)
Edit-2011: For the latest C/C++ source code see https://code.google.com/p/dsd2pcm/ It is released under the new BSD license. The attatchment to this post is still the first Java release.
This post has been edited by SebastianG: Dec 4 2011, 17:45
dsd2pcm.zip ( 3.56MB ) Number of downloads: 2940
Dec 7 2009, 17:55
Joined: 23-August 09
Member No.: 72571
Ok finally my warning was lifted after a little situation was addressed .....
Thanks for your information (2Bdecided and rpp3po). Very much appreciated. That's what I'm here for, I got a lot to learn, and I appreciate you sharing your knowledge, which should never be discouraged here (within reasonableness ofc - if it's something that's been discussed to death it would be annoying). Someone like me should not be scared to ask questions in general (and usually it's ones based on uncommon topics like DXD which have certainly not been discussed hardly at all before) and I don't feel that way anyway.
The problem is, unless you've found better sources than me, these sources are just marketing.
e.g. these are sources that also show SACD pulse response being almost infinitely short, while CD is really long (a sinc pulse). This is all-but-nonsense because to get that pulse response from SACD, you have to remove the output filter. If you remove the output filter, you have full scale digital noise - any pulse in there is invisible*. Run this unfiltered signal through an amp and speakers and they'll blow up.
* - so how did they generate these graphs? I can think of three possibilities:
1. They simulated DSD sample frequency with a multibit signal and no noise shaping.
2. They averaged millions of "real" 1-bit DSD "pulse" signals in the multi-bit or floating point domain to remove the noise so that you can actually see the pulse.
3. They drew it with a pen.
I'm not getting how the output filter (a low-pass filter in the analog domain) lessens/makes moot the analog-like impulse response of the digital signal feeding it?... And I'm not sure that the pulse response needs to apply to the ultra high frequencies anyway? If you filter out the noise, you still have the narrow pulse response for the audible band, right?
Secondly: Not that it's a substitute for my own thoughts, and not that I'm defending DSD/DXD (trust me, I'm not, but by the same token I'm not on the pcm-ONLY side either, I'm 100% agnostic and investigative), but this reminds me a bit of the AES papers back and forth about DSD like this and this). Scientifically peer-reviewed papers (am I right in saying that? Are these papers not necessarily peer-reviewed by objective scientifically-minded audio engineers?), arguing against each other (if not well then I'd like to remind, not that it's hugely relevant to my main point right now, about these two scientific studies which plainly contradict each other...). I'm not sure this is _just_ a matter of marketing (which yes I acknowledge, and when I see some of it I laugh too), and what the DSD proponents say, vs. the truth. E.g. in the AES paper defending DSD, they have quite an accurate-looking graphs of impulse response. Is this AES article misleading marketing too? It doesn't look like it's hand-drawn to me.
Btw, I'm not saying higher impulse response is important anyway, just that it was there in the first place and that it appears to be higher in DXD as well...
- - - - -
Now to the DXD matter:
Thanks for the info and those spectrograms!
What I wonder, is how much noise 24-bit 352.8kHz "normal" PCM would have (whatever "normal" is), compared to these 2L "DXD" PCM files? And also, would there be a difference in impulse/frequency response?
The question is, is DXD just a marketing term, and there’s no strict guidelines on what can be called DXD, or does DXD actually have a "format specification" (so to speak), governing exactly what sort of anti-aliasing filter it must have (to be a true DXD encoded signal)? This is why I questioned the claim that a 24-bit 352.8kHz output file from DSD2PCM is a "DXD file".
Of interest is also whether the architecture of the ADC in question has an effect on how much noise there will be....Not that these "optimal" ADCs in the pyramix, AX24 etc DAWs produce low-noise 352.8kHz files (lol). But I once read that the higher the sampling rate, or at least when you start to go above 96kHz, some noise/distortion starts to occur - but that there have however been "advances" in high sampling rate ADC architecture whereby this noise is somewhat reduced - eliminating some past criticisms of 192kHz ADCs. And that the AX24 (one of the few DXD-capable ADCs) is one of those ADCs. Is this correct?
So in the end, the question is whether (true) DXD can only be generated from specific alti-alias settings which match that in the architecture of DXD ADCs manufactured by Digial Audio Denmark (afaik they're all made by the same company, just re-branded into the other DAWs).
Any high-SR PCM allows for more relaxed filtering compared to Redbook.
DSD (1bit/2.82MHz) was initially meant as an archiving format, not a production format. Then some genius decided consumers should be offered DSD in a delivery format (SACDs). But you can't do common production moves -- e.g., mixing, edits-- in DSD. You have to go to PCM. Hence kludges like intermediate conversion to DXD (i.e., 24bit or 32bit/352.8kHz PCM) by Pyramix, or Sadie's 'DSD-Wide' (waggishly called 'PCM Narrow') which is 8bit/2.82Mhz PCM, followed by final (re)conversion to DSD.
Note that all of the SRs are integer multiples of Redbook's 44.1 (1fs). DSD (8fs) itself was designed to be easily transcoded to PCM in the first place.
The final amusing aspect to all this is that Scarlet Book spec recommends that SACD players incorporate some low-pass filtering (either 50 kHz or 100kHz) to lessen the amount of ultrahigh frequency content the downstream gear would have to (perhaps badly) deal with.
Btw, DXD is not 'new', it's been around since at least early 2004. DSD-Wide's been around even longer.
DXD in the end, is not meant as an editing format for DSD recordings, but as an original master recording format...
They like to record DXD for SACD production because it has almost the same impulse response and frequency response to DSD (within the guidelines of what is "good enough" to them - they say the benefits of DXD outweigh the slight inferiority to DSD in the pulse/freq response departments, and that it sounds better anyway...:|).
And when you compare DXD to PCM/DSD, it IS quite new. Especially in terms of exposure, it's very "new". That's what I meant there.
Dec 7 2009, 18:33
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409
I'm not getting how the output filter (a low-pass filter in the analog domain) lessens/makes moot the analog-like impulse response of the digital signal feeding it?... And I'm not sure that the pulse response needs to apply to the ultra high frequencies anyway? If you filter out the noise, you still have the narrow pulse response for the audible band, right?Time and frequency are inextricably linked. The pulse length is inversely proportional to the bandwidth. The pulse that's shown for CD - that's the shortest pulse you can get if you filter out everything beyond the audible band. Or, to put it another way, that is a "narrow pulse response for the audible band".
If you remove high frequencies, the pulse response is longer / slower / fatter.
...Are these papers not necessarily peer-reviewed by objective scientifically-minded audio engineers? ...No. AES conference papers are not peer reviewed. They don't even ask to see the paper before accepting you (just the abstract).
btw, there was a typo in you URL - the pro DSD paper was...
The graph there (figure 9) seems real. The text on the previous page explains it carefully, and I'm fairly sure I would get the same if I repeated the simulation.
It compares a reasonably sharp (but not brick wall) symmetric (linear phase) FIR filter with a gentle, possibly not symmetric filter.
The graph clearly shows the noise of DSD that gets through the filter. The noise is about 30dB down, but still easily visible. A 30dB SNR is nothing to be proud of.
The text says "as the noise floor contains only high frequency components which are uncorrelated with the audio, they are not perceptible."
The 48kHz PCM audio contains "ringing at a -30 dB level approximately 1 ms before the click, which is very audible."
So ultrasonic noise is inaudible, while ultrasonic ringing is audible?! What a strange conclusion - one without a shred of psychoacoustic evidence to back it up. No psychoacoustic theory to support it. No listening tests to verify it.
Worse still, with a 192kHz sampled system, you could make the filter ringing far less severe than what is shown - and of course, you'd never have the problem of ultrasonic noise.
More generally, the pro-DSD paper was produced by someone employed by a company selling DSD, while the anti-DSD paper was produced by academics.
The anti-DSD paper is technically sound, from two fantastically well respected researchers in the field of digital audio. (The author of the pro-DSD paper is also otherwise well respected AFAIK).
The authors of the anti-DSD paper would readily admit that all the artefacts they reveal should be inaudible - but since the benefits of DSD should also be inaudible, that's hardly an excuse!
What I wonder, is how much noise 24-bit 352.8kHz "normal" PCM would have (whatever "normal" is), compared to these 2L "DXD" PCM files?If you restrict your self to real world signals, it will have as little noise as the most noise-free signal you can find. Theoretical signals: -144dB FS. With a practical converter, it'll have the noise of the converter - assuming the quietest real world signal will come from short-circuiting the inputs of the converter with copper wire.
This isn't a silly answer. Even a single resistor "has" more noise than a 24-bit digital audio signal. It really is overkill.
This post has been edited by 2Bdecided: Dec 7 2009, 18:35
|Lo-Fi Version||Time is now: 2nd March 2015 - 13:59|