The Ear - DSD under Fire

[Doak]

As just one example, Opus3 Records analog master tapes transferred to DSD128 (DSDfile.com) are GREAT sounding - wonderful in fact. Not a thing "nonsensical" about them. That's for certain.

[tailspn]

Digitally delivered recordings are made through a process of converting one or more analog levels (continuously changing voltage level(s) without discontinuities or quantization) to a digitally storable/deliverable medium (the entire process aka: A/D conversion). The end result may be native, or they may be converted to digital audio workstation processable formats for sweetening. In all cases, the analog signal(s) to digital conversion is initially through Sigma-Delta modulation. So the entire concept of PCM recording is bogus; all recordings originate from a form of Pulse Density Modulation, and in the case of a "PCM recording", converted from analog signal > PDM > PCM).

 

Separately, Bruno's argument is one of delivery format. He believes if the original recorded PDM (DSD) format is to be digitally processed through PCM conversion, then it should be made available in that PCM format also, for ultimate quality. But please recall, there's still lots of Sonoma and SADiE DAW's out there producing recordings without "Philips" (DXD) PCM conversions.

[spdif-usb]

I think you are slightly taking that sentence out of the context its meant in...

 

 

Its not just that he "considers" DSD a lossy format, it is by definition lossy if any conversion takes place as there is no mathematical function which is the exact reverse. As soon as you convert DSD to something which can be manipulated in editing - any mixing, eq or level adjustment (e.g. anything more than just splicing) - you cannot convert it back to DSD exactly.

 

Thats why its ideal as an archival format for analogue sources; but for original "full digital" recordings PCM is a better choice.

 

Ironically I think (and I may be wrong here) Barry Diamante's Soundkeeper recordings would be an ideal test of PCM vs DSD as (iirc) he records simply with a pair of microphones and does no post recording production.

 

Eloise

The term 'lossy format' just means that the digital data it contains is not bit perfect. But then, neither is PCM. This is because both modern PCM and modern DSD are storing data that has gone through various complex (parallel/cascaded) DSP filter stages that must exist within the ADC unit. Granted, there are (some rare) exceptions to this. But a case can be made that omitting the digital filters yields worse performance in both formats as opposed to not omitting them. So that's why these exceptions can be considered not to actually matter in the practical sense. (Even, if we can live without anything more than just "Philips" style splicing). The bottom line... everyone else but Classic Records can't water their plants in the desert.

[Audio_ELF]

The term 'lossy format' just means

Sorry but the term lossy has a specific meaning to mathematicians and therefore engineers. A lossy equation is something which does not have a function which can reverse it.

 

All the filtering which goes on inside a ADC may also lossy. However thats not (I think) what Bruce was talking about.

 

Once outside the ADC, it is possible to apply EQ to a PCM signal and be able to reverse that process exactly (iirc) where as a DSD signal would need first converting to PCM (or multibit SDM) then the EQ applies and then converted back to DSD. Once that signal was converted back to DSD you cannot reverse the process exactly (again afaik).

[ted_b]

The term 'lossy format' just means that the digital data it contains is not bit perfect.

? Not my definition, nor those in the industry. Eloise's defiintion is spot on for those that use the term in the industry. It is most often used for downsampled formats like MP3 (where music information is lost and never mathematically recovered), but also used when formats change (say PCM-to-DSD or vice versa). Furthermore, most would not use the term lossy when describing taking a 24/96 recording and upsampling it to 24/192; you are certainly not bit-perfect but you are also not typically describing it as lossy (in the industry sense).

 

The bottom line... everyone else but Classic Records can't water their plants in the desert.

 

Classic Records is a reissue label specializing in audiophile-grade vinyl and DAD/HDAD reissues. Not sure how they play a role here. You might have meant Channel Classics. If so, you are welcome.

[Hiro]

The bottom line... everyone else but Classic Records can't water their plants in the desert.

 

As Eloise mentioned earlier, even Barry Diamant bypasses digital mixing for his recordings...

[spdif-usb]

Sorry but the term lossy has a specific meaning to mathematicians and therefore engineers. A lossy equation is something which does not have a function which can reverse it.

 

All the filtering which goes on inside a ADC may also lossy. However thats not (I think) what Bruce was talking about.

 

Once outside the ADC, it is possible to apply EQ to a PCM signal and be able to reverse that process exactly (iirc) where as a DSD signal would need first converting to PCM (or multibit SDM) then the EQ applies and then converted back to DSD. Once that signal was converted back to DSD you cannot reverse the process exactly (again afaik).

He (Putzeys) probably was referring IMO to both the editing (bar "Philips" style splicing) and the digital filters of the ADC itself. Since modern ADCs that do not apply digital filtering of any sort perform worse than ones that do (as well as are rare), in the practical sense you can't really consider DSD a 'lossless format'. Using it as a 'true' lossless format necessitates the use of a 'true' 1-bit modulator. This is because DSD can't be used to store anything other than a 1-bit datastream. It means that you can't use a multi-bit modulator (MBM), or multi-level modulator if you want to capture to DSD, without irreversably mangling them by incorporating a 1-bit re-modulation DSP method prior to capturing to DSD, the bits that are the output of said MBM. However, using an MBM as opposed to a 'true' 1-bit modulator improves performance in a well designed ADC. So again, in practice, the point would still be moot anyway in the first place.

[Hiro]

He (Putzeys) probably was referring IMO to both the editing (bar "Philips" style splicing) and the digital filters of the ADC itself. Since modern ADCs that do not apply digital filtering of any sort perform worse than ones that do (as well as are rare), in the practical sense you can't really consider DSD a 'lossless format'. Using it as a 'true' lossless format necessitates the use of a 'true' 1-bit modulator.

 

Miska's DSC1 DAC delivers bit-perfect DSD conversion with 33-level Delta Sigma Modulator.

[Hiro]

In all cases, the analog signal(s) to digital conversion is initially through Sigma-Delta modulation. So the entire concept of PCM recording is bogus; all recordings originate from a form of Pulse Density Modulation, and in the case of a "PCM recording", converted from analog signal > PDM > PCM).

 

True. It's very misleading to refer to PCM as a recording format these days. In most cases it's simply a delivery format.

[Hiro]

Furthermore, most would not use the term lossy when describing taking a 24/96 recording and upsampling it to 24/192; you are certainly not bit-perfect but you are also not typically describing it as lossy (in the industry sense).

 

The 7kHz square wave tests Miska performed with 24/96 vs DSD 128+ showed that DSD was actually closer to the original signal.

[spdif-usb]

? Not my definition, nor those in the industry. Eloise's defiintion is spot on for those that use the term in the industry. It is most often used for downsampled formats like MP3 (where music information is lost and never mathematically recovered), but also used when formats change (say PCM-to-DSD or vice versa). Furthermore, most would not use the term lossy when describing taking a 24/96 recording and upsampling it to 24/192; you are certainly not bit-perfect but you are also not typically describing it as lossy (in the industry sense).

 

 

 

Classic Records is a reissue label specializing in audiophile-grade vinyl and DAD/HDAD reissues. Not sure how they play a role here. You might have meant Channel Classics. If so, you are welcome.

My point was that, in theory DSD is lossless because it is still technically possible to adhere strictly to 1-bit quantizers, but in practice hardly anyone in the industry does that anymore these days due to quantizers that incorporate MBM inherently being superior to that. I.e., does it matter to you the fact that the DSD format itself is lossless when in actual reality the digital data stored therein has almost certainly gone through DSP filters, even BEFORE it actually left the ADC that was used to create said data, anyway in the first place?

 

The Classic Records remark was toungue-in-cheek as vinyl to me is superior to DSD in that the former makes the music genres I like most sound way better to me than the latter, generally speaking. Have you ever tried listening to prog rock on SACD?

[Audio_ELF]

I'm not sure arguments about "what happens inside the ADC" is very interesting to a recording engineer. No one (except the designer of the ADC) can control that.

 

What he/she can control is the work flow after there. Given a choice between DSD and PCM out of the ADC can be better made by an engineer if they understand what compromises are required.

[spdif-usb]

Miska's DSC1 DAC delivers bit-perfect DSD conversion with 33-level Delta Sigma Modulator.

DSD cannot store the output of a 33-level Delta Sigma modulator bit-perfectly except if this modulator itself already digitally processes the bits in an irreversible manner before outputting them. If the modulator itself is lossy so it outputs lossy data as a result from lossy modulators being lossy, does the fact that the lossy data gets stored in a lossless format magically turn the lossy datastream into a lossless one? You know, most modern ADCs that output PCM data are using lossy DSP so the data itself is lossy. But if you really believe that a modern ADC that outputs DSD and that doesn't use lossy DSP can perform more accurately than any of the ones that do use. Then such is your belief.

[Hiro]

DSD cannot store the output of a 33-level Delta Sigma modulator bit-perfectly except if this modulator itself already digitally processes the bits in an irreversible manner before outputting them. If the modulator itself is lossy so it outputs lossy data as a result from lossy modulators being lossy, does the fact that the lossy data gets stored in a lossless format magically turn the lossy datastream into a lossless one? You know, most modern ADCs that output PCM data are using lossy DSP so the data itself is lossy. But if you really believe that a modern ADC that outputs DSD and that doesn't use lossy DSP can perform more accurately than any of the ones that do use. Then such is your belief.

 

You would need to ask Miska about the details of his DSC1 DAC, but from what I remember, he definitely described its operation as bit-perfect NOS conversion of DSD sources up to 512fs.

[mansr]

You would need to ask Miska about the details of his DSC1 DAC, but from what I remember, he definitely described its operation as bit-perfect NOS conversion of DSD sources up to 512fs.

 

How do you define a bit-perfect DAC? Since the output is not bits, what is it perfect in relation to?

[Hiro]

My point was that, in theory DSD is lossless because it is still technically possible to adhere strictly to 1-bit quantizers, but in practice hardly anyone in the industry does that anymore these days due to quantizers that incorporate MBM inherently being superior to that. I.e., does it matter to you the fact that the DSD format itself is lossless when in actual reality the digital data stored therein has almost certainly gone through DSP filters, even BEFORE it actually left the ADC that was used to create said data, anyway in the first place?

 

PCM undergoes far more conversions than DSD when you look at the whole chain. And with DSD 128 ADCs like Emm LAbs ADC8, Tascam DA-3000, Sony Pro DSD 128 A/D, Korg MR-2000s, there's no DSP involved whatsoever when outputting DSD signal.

[Hiro]

How do you define a bit-perfect DAC? Since the output is not bits, what is it perfect in relation to?

 

You can find the DSC1's design principles described at the Signlayst website.

[徐中銳]

Of special interest is an excerpt of Bruno Putzkeys, Chief Designer of Philips Digital Systems Labs

 

Let's quickly consider what that actually signifies, particularly his start in (studying/developing) DSD. From Putzheys' LinkedIn profile :

Chief Engineer Class D Audio

Philips

July 1995 – May 2005 (9 years 11 months)

 

Develop new class D circuits, assist implementation in consumer audio products.

 

And with Wikipedia listing Putzkeys as born in 1973...

 

Of Kennedy, his write-up is but Hansen's easy mop-up ?

[Hiro]

Separately, Bruno's argument is one of delivery format. He believes if the original recorded PDM (DSD) format is to be digitally processed through PCM conversion, then it should be made available in that PCM format also, for ultimate quality. But please recall, there's still lots of Sonoma and SADiE DAW's out there producing recordings without "Philips" (DXD) PCM conversions.

 

That's fair enough, I guess, but even if you make it available as "PCM" it will be converted to SDM/PDM/PWM anyway in 99.9% of cases during D/A conversion.

[sandyk]

Well, to me, and many others (like Neil Young, Michael Fremer, Cookie Marenco, etc) archiving analog masters to CD resolution 16/44.1kHz is nonsensical.

 

Agreed, but that is another completely different thing. They could be archived to high res. LPCM as Barry does .Seeing that so many love their LPCM to DSD conversions, it could be interesting to convert Barry's 24/192 files to the best DSD format based purely on the more relaxed playback filtering , and compare the results ?

[Hiro]

Let's quickly consider what that actually signifies' date=' particularly his start in (studying/developing) DSD. From [b']Putzheys' LinkedIn profile[/b] :

Chief Engineer Class D Audio

Philips

July 1995 – May 2005 (9 years 11 months)

 

Develop new class D circuits, assist implementation in consumer audio products.

 

To my knowledge Philips never developed native DSD editing tools, which were the product of Sony R&D. Eventually in 2013 Philips decided to exit consumer audio market, and focus on what they knew best, i.e. lightbulbs.

 

"Philips, which has become primarily a maker of medical equipment and lighting products, has sold the audio, video, multimedia and accessories activities to the Japanese consumer electronics company for the almost token sum of €150 million ($201.8 million) in cash and a brand-license fee."

 

Philips Exits Consumer Electronics - The Source - WSJ

[Hiro]

Agreed, but that is another completely different thing. They could be archived to high res. LPCM as Barry does .Seeing that so many love their LPCM to DSD conversions, it could be interesting to convert Barry's 24/192 files to the best DSD format based purely on the more relaxed playback filtering , and compare the results ?

 

We were discussing production workflows that do not involve PCM...

[tailspn]

To my knowledge Philips never developed native DSD editing tools, which were the product of Sony R&D.

 

Yes, both the Sonoma hardware and software to operate them (Sonoma and Sonoma Mixer) as a DAW were Sony developed. The Philips DAW solution, now Merging's Pyramix, was always DXD PCM based.

[spdif-usb]

PCM undergoes far more conversions than DSD when you look at the whole chain. And with DSD 128 ADCs like Emm LAbs ADC8, Tascam DA-3000, Sony Pro DSD 128 A/D, Korg MR-2000s, there's no DSP involved whatsoever when outputting DSD signal.

It can easily be shown that increasing the number of conversions in, and complexity of the modulator can actually help to improve rather than deteriorate accuracy for both PCM and DSD if the right design criteria are met. The insertion of properly chosen digital filters inside the multi-bit Sigma Delta modulator's architecture can be a far more excellent design choice than many would suspect, and the necessity of using fewer processing steps to achieve more accuracy is just another typical example of those long overdue myths that keep on popping back up in pretty much every DSD related topic.

[Hiro]

It can easily be shown that increasing the number of conversions in, and complexity of the modulator can actually help to improve rather than deteriorate accuracy for both PCM and DSD if the right design criteria are met. The insertion of properly chosen digital filters inside the multi-bit Sigma Delta modulator's architecture can be a far more excellent design choice than many would expect, and the necessity of using fewer processing steps to achieve more accuracy is just another typical example of those long overdue myths that keep on popping back up in pretty much every DSD related topic.

 

That's fine by me, if you want to introduce as many conversion steps between formats and use as many upsampling and oversampling filters as possible. But I'll still take a direct output from the modulator as the closest representation of the actual modulator. And for DSD playback, a multi-bit SDM DAC such as DSC1, for instance, is still considered 999.9 Pure when handling DSD.