DSD?

[John Phillips]

...SACD tries to avoid this as much as possible by measures like keeping the peak signal levels encoded well below the max range nominally possible. But you can't beat the maths...

FYI, there was a rather entertaining presentation from ESS (SABRE DACs) at a RMAF which shows complex/chaotic state variable behaviour and noise levels depending on signal input in one-bit sigma-delta modulators. It reveals that ESS has patented secret sauce to minimize the audible impact (sorry that this version of the online video may not work in your IT but the slide deck is available from ESS).

I remember reading the Lipschitz paper you mentioned earlier and deciding (maybe based on not enough understanding) that the right cure was a multi-bit SDM at the core of a DAC with enough margin by design to avoid overload problems. However I never got round to trying out that hypothesis.

 

[alanbeeb]

SACD is still very much alive and kicking and plenty new classical recordings are being released on SACD by a number of labels. Quite a few of them are actually made as hires PCM recordings then being converted to DSD. Which might be a bit pointless.

In general, they sound very much better than redbook CD. But 24/96 or 24/192 PCM does too - so my preference nowadays is to download 24/96 flacs when I'm buying new classical releases.

I've ripped my SACDs, obtained over last 20 years, to .DSF files on my server, which my DAC can convert natively. The DSF files use approx same space on disk as 24/192 flac of the same material.

 

[mansr]

In principle, yes, you can 'convert' DSD by using a carefully chosen analogue low-pass filter. This then tries to mimic the behaviour of a digital DSD 'demodulator'. However the official demodulators have complicated feedback taps, etc, to try and get optimum recovery of the DSD, try to avoid signal-pattern-dependent idler patterns at audible frequencies, and minimise the ulstrasonic 'hash' that can upset later kit.

Doing all that is quite hard - indeed essentially mathematically impossible to do perfectly once you go above about 4th order. Too many plates to juggle.

SACD tries to avoid this as much as possible by measures like keeping the peak signal levels encoded well below the max range nominally possible. But you can't beat the maths...

You seem to be slightly confused. Converting a 1-bit (DSD) signal to analogue or PCM requires only a standard low-pass filter. The complexity is all in the modulator which, as you say, is prone to instability.

 

[Jim Audiomisc]

FYI, there was a rather entertaining presentation from ESS (SABRE DACs) at a RMAF which shows complex/chaotic state variable behaviour and noise levels depending on signal input in one-bit sigma-delta modulators. It reveals that ESS has patented secret sauce to minimize the audible impact (sorry that this version of the online video may not work in your IT but the slide deck is available from ESS).

I remember reading the Lipschitz paper you mentioned earlier and deciding (maybe based on not enough understanding) that the right cure was a multi-bit SDM at the core of a DAC with enough margin by design to avoid overload problems. However I never got round to trying out that hypothesis.

In essence you need more than one bit per sample when recording to ensure you can avoid problems. You can then fully dither and take steps to avoid instabilities, etc, at least in principle. But there is no way to ensure you do so with 1 bit per sample.

 

[Jim Audiomisc]

It depends on what is assumed by words like "requires". The demodulators can also employ multistage 'feedback', and may do so to try and optimise recovery of the wanted audio whilst minimising the HF hash output. Yes, you can use simple demodulators just as you can use a simple analogue LPF that avoid those risks, but then it means you can't do as well in cases where the problems don't crop up.

The big distinction is that idler 'orbits' and failures to dither sufficiently when modulating live recordings are 'unrecoverable' as they are burned into the results. Whereas a demodulation system only damages replay using that system. So you can flip a switch and try again with a different demodulator (filter) if you get a problem.

Low-bit DACs fed LPCM also potentially have the same problems. But given more than one bit/sample you can design them in a form you can mathematically assess to show are 'clean'. Just that this gets harder as the orger goes up. For higher orders it all depends on the design. 3rd or higher with memory allows for 'space filling' problems as the orbits can now fail to intersect. In a way analogue can then be worse as the result could in some situations be chaotic (in the mathematical sense) due to the analogue stage noise making things worse, not better!

 

[mansr]

The demodulators can also employ multistage 'feedback', and may do so to try and optimise recovery of the wanted audio whilst minimising the HF hash output.

Do you have a reference or example of this? I have never seen anything like that mentioned in related literature.

 

[davidsrsb]

FYI, there was a rather entertaining presentation from ESS (SABRE DACs) at a RMAF which shows complex/chaotic state variable behaviour and noise levels depending on signal input in one-bit sigma-delta modulators

Memory lane, I used the Ferranti PCM1/2 set in my McMichael days. It did have some odd artifacts that I did not understand at the time

 

[marshanp]

I don't pretend to understand the technicalities, but...

I have recently been ripping SACDs using a Sony Bluray player. My DAC, a TEAC UD-503, is DSD-capable but doesn't seem willing to play .dsf files sent from my Auralic Aries Mini. I have therefore converted the SACD rips to 24/88 FLAC files.

On the Channel Classics SACD of Rachmaninov's Second Symphony (recorded in Budapest and conducted by Ivan Fischer) the filler track is the famous Vocalise. There is audible traffic noise on it - at the beginning, and again at 1:17 and 1:52. I never noticed this when playing the SACD on my Marantz KI Pearl Lite, but it is quite audible on the FLAC file played through the TEAC DAC. It is mercifully absent from the Symphony's famous Adagio.

Perhaps the TEAC is a better DAC than the one in the Marantz... but there is, it would seem, no magically "musical" quality to DSD in itself. SACDs sound pretty good as a rule; but that is because recordings originally made for SACD release are made with particular care. A signal of very high quality sounds just as good via PCM, maybe better if the PCM reproduction chain is superior.

 

[Jim Audiomisc]

Do you have a reference or example of this? I have never seen anything like that mentioned in related literature.

Nothing to hand I'm afraid as I did my look at this decades ago. I may be able to find something if it turns up at some point, though. But a key distinction point is if you use a 1bit system of over 3rd order that has feedback taps.

I looked at this as a result of also looking at chaotic and semi-chaotic systems of other kinds - i.e. not used in audio. Mainly because I was interested in steganoraphy in signalling for some defence-related work. Given my interest in audio that led me on to DSD out of curiosity. It is also a potential pitfall for low-bit processing of LPCM, but easier to avoid because you can use the advantages of > 1 bit as makes sense.

Added: This is as close to something, to hand:

http://jcgl.orpheusweb.co.uk/history/sa_1990-2/ChaosAndNoise.html

It then got extended when I worked on the Combat ID systems as per

http://jcgl.orpheusweb.co.uk/history/sa_1995-7/AVeryBusyTime.html

 

[Jim Audiomisc]

So far as I'm concerned, both DSD/SACD and LPCM can sound fine - or lousy. But this is all down to how well a recording was made, etc, etc. Not the technical difference between DSD and LPCM.

 

[DimitryZ]

It's from my old audio site, around 1998 or so(*). There was a caption:

Avid audiophools cheer for SACD as they turn in their redbook CDs.



A DSD DAC is a very rudimentary thing, no more that a few switches driven from the raw DSD data stream, feeding into a switched-capacitor filter (IIRC).

I think it was Allen Wright who first, in the early 2000s?, realised that there is no fundamental difference between the post-DAC DSD current or voltage pulses as found in a regular player, and the pre-DAC raw bit stream. So he tapped into the latter and exported that via one of his own buffer/amp designs.

http://www.vacuumstate.com/index.dna



(* That same page explained in some detail my concept for a PCM-based recording and distribution format based on 96k or 192k sampling with low-order AA and AI filters matched to the natural spectrum of music. Sounds familiar?)

Not much new under the sun, I guess.

 

[mansr]

Nothing to hand I'm afraid as I did my look at this decades ago. I may be able to find something if it turns up at some point, though. But a key distinction point is if you use a 1bit system of over 3rd order that has feedback taps.

Yes, that's all true for the encoder/modulator. I just don't see how it applies at the playback end. For a digital conversion, you'd typically use a FIR filter, same as for any sample rate reduction, which has no stability issues.

 

[Nagraboy]

For many years, pretty much all SACD players internally converted the DSD data into some form of multi-bit PCM, so there was actually no way to listen to the sound of DSD converted directly into analogue.

Marantz now have made a few disc players (SA10, SA-KI Ruby, SA12SE) that convert all incoming data to DSD256 and then use a simple low pass filter instead of a DAC. I’m not sure what goes on inside the AKM and ESS DAC chips of today as I lost interest in this sort of thing some years ago, but I’ve read that the AKM ones do not convert the DSD to PCM.

 

[allthingsanalog]

I’ve downloaded quite a few hires files and tbh, unless it's down the useless remasters, in many occasions I prefer the original 44.1k 16 bit versions.