24/176.4 PCM vs. SACD

[Sik_Lescinovid]

In light of the many discussions on listening tests and differences in digital audio storage I thought that some of you might find this paper interesting:

 

http://old.hfm-detmold.de/eti/projekte/diplomarbeiten/dsdvspcm/aes_paper_6086.pdf

 

[esldude]

Very nice. Good quality dCS converters used. And in part of the test Stax headphones. Having owned some Stax Lamda Pro headphones I would think if you don't hear it over the Stax then you aren't going to hear it.

 

Would have been even better if they had tested at higher sample rates rather than down-converting everything to 44.1 khz.

 

I don't enjoy listening over headphones as much as good speakers. But Stax headphones leave no detail hidden when listening critically. I later owned early Koss ESP9 electrostatics and consider them more musical while still very revealing. Currently I use Beyer DT880 headphones which are probably a bit more revealing than old Koss ESL's and still pretty musical. But I don't think they are as revealing as Stax

 

[Miska]

One question not answered there is how the dCS converts PCM vs DSD into analog? Same method used for both? Does it convert DSD to PCM?

 

Previous similar test I've seen was using multiformat player that had Mediatek's internal 88.2 kHz DSD-to-PCM conversion.

 

Would be interesting to see comparison between native multibit ladder PCM DAC vs true native DSD DAC. It would be also interesting to see comparison with various different DSD and PCM A/D converters.

 

This test only tells that with that dCS hardware combination it is hard to tell the two different.

 

 

[jonmarsh]

It makes me feel even better about the choice for using 24/176.4 for conversion of my SACD library to PCM, as opposed to 24/88.2. It was more of a hail mary, let's go the extra mile choice when made, not much empirical data I was aware of at that time to back it up.

 

In general, the results of a comparison like this WILL be quite dependent on the playback gear; with dCS, you've got a certain baseline that most folks can agree on. Regrettably, even some fairly expensive SACD players don't, in my experience, compare to the cream of the crop in DAC's in the 5-7K price range.

 

[elcorso]

The only problem I can see is, from May 2004 to November 2011, there are seven years on software and hardware development, and this could be light years in the digital domain.

 

Also I do agree with @Miska questions, maybe not important on year 2004, but a lot in 2011.

 

What I have on my music library, being from the same excellent analog source, and then converted to PCM 24/176.4 or 24/192 from DVD-Audio, or DSD 64 from SACD, I prefer with no doubt DSD native playback. But I don't have a lot of this music tracks for a proper and universal comparison, but I can guess DSD will be the winner, with the proper gear.

 

I also agree with @jonmarsh to make the DSD conversion to PCM 24/176.4 instead to 24/88.2, if your DAC can play this sample rate.

 

Roch

 

[agentsmith]

Curiously, other people claim to hear differences and preferences between lossless compressed and WAV/AiFF, others even hear beween Aiff and WAV. I don't think there has ever been a "scientific" test like this one though.

 

I would think there should be more differences between 1 bit and 24 bit encoding formats, but even that apparently is infinitismal.

 

But then again wasn't theresimilar studies between 320kbps and WAV, and people couldn't tell the difference either?

 

Makes you wonder, doesnt it?

 

[elcorso]

I really don't know if the words "scientific" and "infinitesimal" should be used in the computer audio listening, or even on all audio listening.

 

It's a very subjective matter, from musical taste and gear involved that you like.

 

I don't want to state nothing and will not defend nothing. But sometimes write here about my believings (or findings?), and I like to know about others. But I never will be influenced for ABX test even if they would be made by Einstein in Harvard University Dept. of Fine Arts (if such dept. exist).

 

I have friends who loves solid stated and I love tubes.

 

BTW, I also have some good music player friends, and they care more for the execution (when listen to reproduced music), than for the SQ.

 

Roch

 

[dallasjustice]

There is little application to the real world with this study. In the real world, most recordings are mastered in either pcm or dsd, not both. If originally done in DSD, my experience is that even the best mastering gear can't make the PCM conversion sound exactly the same.

 

I compared Ray Charles Genius Loves Company, "Fever". There's a lot of cool reverb and cymbal decay in this song. I have both the DSD rip and the conversion to 24/88.2 Bruce Brown did for HDtracks. There's no question the DSD is superior. Anyone could hear the difference.

 

[dallasjustice]

My understanding from the paper is that they used a "Y adapter" to split the mic feed into 2 separate signals so as to contemporaneously capture said mic feed in PCM and DSD. There was no conversion done. This is the right way to do it.

 

Upon further review, maybe I misunderstood what you are saying. My apologies if I did.

 

[elcorso]

BTW, I was listening to this track in native DSD, great!

 

I have some mastered in analog, then converted to PCM, and also to DSD. DSD win (to my ears)

 

Cheers!

 

Roch

 

[Miska]

My understanding from the paper is that they used a "Y adapter" to split the mic feed into 2 separate signals so as to contemporaneously capture said mic feed in PCM and DSD. There was no conversion done. This is the right way to do it.

 

No, what I meant is what is the native format for the converter's internal processing.

 

You cannot have the same converter hardware for both PCM and DSD and avoid any processing of either data. Now, if those dCS models were multibit-SDM, was there for example 176.4/24 intermediate format for DSD before remodulation to multibit-SDM? Or was the 176.4/24 PCM converted to DSD? Given dCS, I guess I can leave the case of native multibit PCM converters out.

 

As an example, my Sony SACD-player converts PCM to DSD when CD-layer/CD is being listened. Because it has only DSD-converters in the hardware.

 

 

[zenmastering]

"Would have been even better if they had tested at higher sample rates rather than down-converting everything to 44.1 khz."

 

They did not downsample the audio to 44K1. They used 'bit-splitting' to spread the 2ch DSD info across 6 audio tracks in the workstation. It was stored as 'data' and then when played back, re-constructed by the DAC back into a DSD stream. Done properly, this is entirely lossless and was used in many early DSD recordings, as it allowed DSD converters to store audio to PCM workstations.

 

Graemme

 

[ted_b]

It supports my view (not mine uniquely of course) that direct native hirex recordings, whether 24/176k+ or DSD, are wonderful! Furthermore, it reinforces my goal of listening to DSD in DSD, and PCM in PCM (both ultimately converted to analog, of course). And moreover, it reinforces my feeling that old analog recordings transferred to DSD are better than old analog recordings transferred to PCM...??WHY? because DSD is DSD. There is no low hirez (24/44 or 24/48) version of DSD that might require upsampling to get to true DSD. However, in the PCM world, too many analog tapes were transferred to PCM at a significantly lower resolution than 24/176k...significantly lower! Many, likely a majority, were done at 24/44 or 24/48k. (Note: it might be interesting to do the same experiment with ever-decreasing resolution to find out where the statistical significance creeps in where native DSD is king over native PCM).

 

Finally, although I chose to do DSD-to-PCM conversions at 24/176.4k also (prior to getting a Mytek DSD capability) I am NOT convinced the same logic is in place. I am NOT convinced that 24/176.4k is a great conversion resolution just because of this experiment. What if the DSD-to-PCM conversion process is theoretically or inherently a mess, a round peg into a square hole? What then? Couldn't the choice of higher resolutions than, say, 24/88k be theoretically detrimental? Like the early days of hidef tv when the tv dramas and sitcoms showed makeup smudges and fake propped backgrounds cuz they weren't specifically built for that resolution? When 720p actually conveyed a better reality than 1080p due to the above (no longer true, but early HD shows were kinda funny that way. I have a friend who is a newscaster, and he says they do way more detail on their makeup now that people can see their pores.)

 

Anyway, this is a great subject, and renews my excitement for pushing for native hirez recordings, and solidifies 24/176k as being a minimum goal for native PCM. Conversions? I'm still not sold on what's best.

 

[esldude]

Thanks for the explanation of this Graemme.

 

My hurried reading of the linked document caused me to mistake their procedure for what was really done.

 

[Miska]

I don't know if any of the listeners had been trained to find detail differences in the sound (not in presentation!) or knew what to actually look for. For example couple of Navy passive sonar operators would have been good selection (I used to train those in the past).

 

For example same goes for MP3 comparison. Once someone is familiar with certain sonic fingerprints it is easy to detect the difference.