Is there a way to detect if DSD is up-sampled PCM?

[wgscott]

How would one go about testing this?

[Julf]

How would one go about testing this?

 

DSD to PCM is detectable, PCM to DSD is less so, as long as we assume the sample rate of the original PCM was high enough not to show the typical 22-KHz brickwall filtering (and even that tends to be obscured by the DSD noise shaping).

 

I can't think of any measurable characteristic changes in the resulting DSD file between a straight-from-analog DSD recording and an analog-to-hires-PCM-to-DSD chain, unless you use very special artificial input signals that might trigger artifacts and interference patterns. If you looked at individual waveforms, you might theoretically see some signs of the PCM reconstruction filter, but the DSD process probably hides those anyway.

[wgscott]

That's too bad. Our collective experience does not leave me at all confident.

[testikoff]

Check out the spectrogram and/or spectrum of the DSD material?

 

The below examples appear to be stemmed from PCM sources:

 

- Oasis - Hello [track 1 of 2003 (What's The Story) Morning Glory? Sony SACD, 2.0 area]:

 

- Norah Jones - Chasing Pirates [track 1 of 2012 The Fall AP SACD]:

[wgscott]

as long as we assume the sample rate of the original PCM was high enough not to show the typical 22-KHz brickwall filtering

 

I guess that is an example!

[Julf]

I guess that is an example!

 

Indeed. What you see is not the PCM encoding as such, but the brick wall filtering.

[christian u]

Thanks,very informative!

[jzahr]

Indeed. What you see is not the PCM encoding as such, but the brick wall filtering.

 

Check out the spectrogram and/or spectrum of the DSD material?

 

The below examples appear to be stemmed from PCM sources:

 

- Oasis - Hello [track 1 of 2003 (What's The Story) Morning Glory? Sony SACD, 2.0 area]:

[ATTACH=CONFIG]7535[/ATTACH]

 

- Norah Jones - Chasing Pirates [track 1 of 2012 The Fall AP SACD]:

[ATTACH=CONFIG]7534[/ATTACH]

 

I can see some difficulties in trying to deduce provenance from interpretation of the above images: in the audacity screenshots can be seen 88 khz 32 bits, but from DSD files one cannot directly produce spectrograms in audacity, so DSD files must be first converted to PCM. This conversion can be done in two ways:

 

(1) DSD file -> [conversion algorithm in a computer] -> 88.2/32 PCM file

 

(2) DSD file -> DSD DAC -> analog signal -> ADC -> 88.2/32 PCM file

 

If first method was used, how can we be sure that the marked 22 khz brickwalling seen in the images is solely due to a prior life as a low-res 44.1 PCM, and not introduced by the "conversion algorithm in a computer" used for spectrogram representation?

 

Am I missing something?

 

I remember an article in Stereophile (years ago, read it in paper, don't know if available online) where a reviewer of a Norah Jones reissue in SACD declared that he was absolutely unable to hear a difference w/r to the RBCD version. The reviewer gave the disk to JA who converted a track of stereo SACD layer to high res PCM via a method similar to number (2) of the above: analog out of a SACD player to line in of a digital recorder.

 

Frequency analysis in audio editor (think it was cool edit, now adobe audition) showed no audio content above 22 khz, except for the usual ultrasonic noise inherent to DSD, which they judged "inconsequential" (quoting from memory). In the end, they valued the SACD as providing no-improvement in audio quality over the previous stereo RBCD, but with the added benefit of extra material in the multichanel layer of the SACD.

 

In the context of this thread and with regard to the concern expressed above, I first thought that, for spectral analysis, method (2) is probably better than method (1). Then however, I remembered many of Miska's posts in which he reminds us that majority of ADCs these days (and for sure the ACD used in that stereophile article) are delta-sigma based.

 

So, in the end, even in method (2) there is a form of "algorithmic" conversion from sigma-delta to high-res PCM, a conversion that takes place inside the ADC chip. So the concern still applies here.

 

Probably, if there is some brickwall filter in the sigma-delta to high-res PCM conversion needed to plot spectrograms for DSD files, it surely is located at higher than 22 khz (probably 44.1 in the images by testikof which, as said, were made at 88.2). If this is the case, then the 22 khz limit seen in these images is indeed a result of a prior life as low-res PCM.

 

However, as the exact specification of the sigma-delta to high-res PCM conversion is not generally available for us, end users trying to assess provenance issues, we must be careful in interpreting spectrograms obtained by these methods.

 

Also, what if a track was produced and mastered natively at 88.2 khz and then upsampled to DSD64 for comercial distribution? In this case, frequency analysis in audacity for a 88.2 PCM derived by the end user from the DSD purchased by him will tell almost nothing.

 

Probably, the correct way would be to produce spectrograms directly from the DSD data, but I don't know of any readily available tool for this. Don't even know if it is technically possible.

 

Regards, and sorry for the long post

Jorge

[Boris75]

I

Probably, the correct way would be to produce spectrograms directly from the DSD data, but I don't know of any readily available tool for this. Don't even know if it is technically possible.

Thanks for a very interesting post. On the question you raised, that should be possible by converting the DSD stream losslessly to a multi-bit 2.8224MHz PCM stream and working on the latter.

[realafrica]

Thanks for a very interesting post. On the question you raised, that should be possible by converting the DSD stream losslessly to a multi-bit 2.8224MHz PCM stream and working on the latter.

 

What to use for such a conversion to PCM. Saracon will only convert DSD up to 384kHz PCM It does not offer 2.8224MHz PCM.

[audiventory]

How would one go about testing this?

 

If wideband PCM converted to DSD with wideband filter, impossibly distinguish it and analog signal recorded to DSD.

Physically it is very same. PCM to DSD converter is mathematical model of analog to DSD converter.

 

If signal was edited/mixed, it was converted from PCM to DSD.

 

Probably, the correct way would be to produce spectrograms directly from the DSD data, but I don't know of any readily available tool for this. Don't even know if it is technically possible.

 

It is possible because from point of spectrum analyzis PCM and DSD is similar.

Band 100 kHz is enough for analyzis because upper frequencies consumed by modulation noise primarily.

 

But there no answer: it is signal recorded in DSD or result of PCM to DSD conversion.

 

On the question you raised, that should be possible by converting the DSD stream losslessly to a multi-bit 2.8224MHz PCM stream and working on the latter.

 

In general, DSD to PCM conversion without resampling is lossless operation.

 

But playback of such PCM in non-filtered mode may cause serious audible issues (noise).