DSD Rendering vs Mixdown Test

[CastLog]

Hey everyone,
Maybe a quick one to sort out:

System:
Laptop
Pyramix Native v 12.0.5 Build 1633
w/ SACD / DSD support

I'm trying to set up / test out workflow for DSD mastering in Pyramix. Source material in test project is stereo DXD, project is DXD, too. I have added sacd track markers for one cd track.

a) When using Rendering
"Generate Master" > "SACD Edited Master" (DSDIFF) > "DSD64" > "DSD Rendering"
the outcome is a file with peak -54.9 dB and no waveform modulation or audible audio whatsoever.

b) When using MIXDOWN
"Generate Master" > "SACD Edited Master" (DSDIFF) > "DSD64" > "Mixdown"
the outcome is a file with 1 second of inaudible noise, 2 seconds of complete silence and then the audio from the source file


QUESTIONS:
a) any ideas as to why the rendering function doesn't work?
b) does the noise at the beginning of the mixdown file have anything to do with sacd specs? the 2 sec of silence certainly looks familiar (red book)...

help & thoughts appreciated...
Thank you!

[mpdonahue]

a) When using Rendering
"Generate Master" > "SACD Edited Master" (DSDIFF) > "DSD64" > "DSD Rendering"
the outcome is a file with peak -54.9 dB and no waveform modulation or audible audio whatsoever.

The render only works with DSD media. It bypasses the mixer and re-modulates the DSD audio only where there are gain changes or crossfades.

b) When using MIXDOWN
"Generate Master" > "SACD Edited Master" (DSDIFF) > "DSD64" > "Mixdown"
the outcome is a file with 1 second of inaudible noise, 2 seconds of complete silence and then the audio from the source file

If you are using DXD media, the best workflow is to create a DXD MTFF PCM digital Release>DXD>32 bit float. Then open that file in Album Publisher and generate the DSD delivery media. You can generate DSD media in any of the DSD rates.
All the best,
Mark

[CastLog]

Mark, thank you for taking the time to sort this out. That's some first-rate advice right there.
And Album Publisher means a huge step in efficiency, too. Thanks, great!

What still seems difficult from an newbies perspective is dBSACD.
The source material in my test project is peak at -1.2 dBFS
When processed to DSDIFF through Album Publisher the readings show 0.0 dBSACD peak value.

Judging from the Pyramix workflow guide for DSD I would suspect that 0.0 dBSACD is -6 dBFS. So either there's some background processing applied or I'm missing something in my understanding of the dBSACD scale.

Also, I've not found a bit of information for the meco SDM new to v12, Obviously SDM algorithms matter in a way similar to SRC conversion types?

Any thoughts appreciated. Thank you!

[mpdonahue]

Mark, thank you for taking the time to sort this out. That's some first-rate advice right there.
And Album Publisher means a huge step in efficiency, too. Thanks, great!

What still seems difficult from an newbies perspective is dBSACD.
The source material in my test project is peak at -1.2 dBFS
When processed to DSDIFF through Album Publisher the readings show 0.0 dBSACD peak value.

The change in level can happen for several different reasons. Inter-sample peaks is one and if you modulated to DSD64fs the added HF noise can also cause the level to rise.

Judging from the Pyramix workflow guide for DSD I would suspect that 0.0 dBSACD is -6 dBFS. So either there's some background processing applied or I'm missing something in my understanding of the dBSACD scale.

No, DSD has headroom above 0dBFS. Scarlet book gives +3.1 dBDSD as the acceptable maximum level, but it is possible to have transients go above that without noticeable distortion.

Also, I've not found a bit of information for the meco SDM new to v12, Obviously SDM algorithms matter in a way similar to SRC conversion types?
Any thoughts appreciated. Thank you!

MECO modulator is designed to go between different DSD rates (DSD256 to DSD64 for example). For DXD to DSD conversions, you should use the SDM D Modulator. Trellis was designed to reduce high frequency noise buildup when re-modulating from DSD64 to DSD64.

All the best,
Mark

[CastLog]

No, DSD has headroom above 0dBFS. Scarlet book gives +3.1 dBDSD as the acceptable maximum level, but it is possible to have transients go above that without noticeable distortion.

That's good to hear. There is an option to apply pre-SDM gain in Album Publisher when adding the DSDIFF Edited Master as Output Format.
But judging from your response there is no need to apply any gain change for SACD when coming from a properly mastered CD level, as long as SACD peak values remain below/equal +3.1 dBDSD?

Thank you!

[Graemme]

0 dBfs is full scale and equals +6dbSACD. Max. level for SACD is +3.1dbSACD or -2.9dBfs

However, in practice things are a little different:

The Pyramix DSD real-time metering is not sufficiently accurate enough to consistently report signals over about +1.1dBSACD. In other words, you're never really sure what level you've actually got, right where it matters most in the 'over-zero' range. Steady-state test tones don't reveal this but programme material does.

When first discovered, I had several discussions with a couple of the Philips R&D engineers about maximum levels and they (Erwin Janssen and Dirk Reefman) came up with a basic rule of thumb: Allow the programme to peak at +1dBSACD. They suggested that levels above this, even transients, could cause audible artifacts. I found, in subsequent tests of recordings with levels closer to +3.1dbSACD, artifacts that lasted longer than the event that caused them, as displayed in the ReNovator app's spectrogram. The good news was that ReNovator could effectively remove these 'overs' and also the 'scar-tissue' with no audible penalty.

Somewhere, there is an AES paper (Reefman, Janssen and maybe others) that addresses the topic of max. levels in Pyramix DSD: my copy is buried in a storage container, currently inaccessible until this pandemic tones down a bit...

Mark do you have this paper or at least remember it better than I do?

Cheers,

Graemme

[CastLog]

Thank you, Graemme for jumping in the conversation and providing further context. I'm sure the paper would be an interesting read, an initial search through AES E-library came up short but I'll keep trying.

The Pyramix DSD real-time metering is not sufficiently accurate enough to consistently report signals over about +1.1dBSACD. In other words, you're never really sure what level you've actually got, right where it matters most in the 'over-zero' range. Steady-state test tones don't reveal this but programme material does.

Does that also apply to the peak value shown during Album Publishing Encoding?
If so, how do you validate peak level of a SACD master?

And - if I may throw in another towel - concerning DST compression:
From the Pyramix workflow guide on DSD (which has helped me a lot as an entry point, so thank you for your contribution there, too) I can read that DST would not be necessary for a stereo release - even on a hybrid disc? Since DST being losless - is it nevertheless used in like 99.9 % of cases? And if not, who makes the decision to compress or not? Label / Manufacturing plant... Maybe there are other factors to consider with DST?

Thanks again a lot for all your help & stay healthy...

[Graemme]

Does that also apply to the peak value shown during Album Publishing Encoding?
If so, how do you validate peak level of a SACD master?

These SACD verification tools are offline processes, so use different (and accurate) methods to determine 'overs' and other errors in the DSD files. They don't have the 'real-time' burden. The Sony Sonoma hardware had an entire ASIC-chip dedicated to metering (I think it was called the 'TACO-Chip' or something similar?) Philips' cutting-master playback hardware also devoted quite a bit of silicon real-estate to metering, I recall.

And - if I may throw in another towel - concerning DST compression:
From the Pyramix workflow guide on DSD (which has helped me a lot as an entry point, so thank you for your contribution there, too) I can read that DST would not be necessary for a stereo release - even on a hybrid disc? Since DST being lossless - is it nevertheless used in like 99.9 % of cases? And if not, who makes the decision to compress or not? Label / Manufacturing plant... Maybe there are other factors to consider with DST?

No need for DST unless disc-space is an issue. As to who makes that decision? For some, it's a company policy (Well-known audiophile company says no-way!), for some it's left to the authoring person (I never use it unless disc space is an issue.) Finally, some leave it to the factory (which is basically leaving it to the authoring person who happens to sit in a factory ;->)

When I say 'disc-space issue,' I'm referring to the additional presence of a multi-channel programme, it's length and the crest factor of said programme. The latter affects DST-encoding efficiency.

Having done extensive, empirical listening tests with DST-encoded and non-encoded DSD files: I can't tell a difference.


Thanks for your kind words; I hope you and everyone reading this is weathering the storm and staying connected in some way. Being a student of Peter A. Levine's work has certainly helped.

Kind Regards,

Graemme