Does DSD sound better than PCM?

[John Deas]

There's a lot of limitations mixing ITB actually. Mixing and editing may not be a big deal, but when it comes to processors/plug-ins, many of them claim to have emulated the machines, very few actually did

Enough is not enough, all they can do, is to capture the original signal/a master as close as possible to minimise loses, use DAW as transport and work it out from there

The very best engineers mix an orchestra live in analogue, capture in DSD then call it a day.

I realise music production and music creation, acoustic and "pop" recording are very different kind of mindset

Very difficult to record an orchestra because of all the mike feeds and not have to do post production mixing, I know the LSO record in DSD then mix in Pyramix back to DSD which I would say is not ideal but they still sound excellent. DSD is best suited to projects like small acoustic recordings and mixed through analog en route to capture or possibly not even that if performers and miking is all set up carefully. Recording projects requiring lots of mixing, plug-ins processing and well... I don't think Nine Inch Nails should bother with DSD

 

[John Deas]

Yes ok, we'll end up round in circles if we keep going

My point is that PCM doesn't have a "sound" as it is transparent, i.e. noise and all forms of distortion are well below audible thresholds (assuming sufficient bitrate, sample rate, and filtering).

If she thinks PCM sounds inherently different from DSD (as opposed to how particular masters of recordings that may happen to be in either PCM or DSD format might sound) she is mistaken, just as she seems to be about WAV and FLAC.

I've explained clearly the reasoning behind why Cookie Marenco records in the way She does, what She hears when She listens to music I couldn't possibly tell you as I'm not Cookie Marenco. What my ears think is She produces incredible recordings that I'm very happy to buy.

 

[andreasmaaan]

What my ears think is She produces incredible recordings that I'm very happy to buy.

I won't argue with that. All I'm saying is that if She (capital "S" for her Superhuman hearing abilities) thinks there is an audible difference between PCM and DSD then she is mistaken

 

[John Deas]

I won't argue with that. All I'm saying is that if She (capital "S" for her Superhuman hearing abilities) thinks there is an audible difference between PCM and DSD then she is mistaken

Yeah, apologies I can never remember if you need to use capitals for She/He or not.

Mistaken.

So this massively experienced, respected and conscientious recording engineer is unable to realise that PCM and DSD sound exactly the same and has stupidly soldiered on with a format that is time consuming and testy for the last 9-10 years for no worthwhile reason whatsoever? I suggest you contact Her/her without further delay to tell Her/her this shocking bad news.
I'm done here

 

[andreasmaaan]

Yeah, apologies I can never remember if you need to use capitals for She/He or not.
Mistaken?
so this massively experienced and respected conscientious recording engineer hasn't realised that PCM and DSD are exactly the same and sound exactly the same and has soldiered on with a format that is time consuming, controversial and expensive to try to recoup the work needed when all she needs to do is just record and mix everything in PCM - I better tell her/HER then!!!

Ah I’m sorry about the She joke, didn’t realise you weren’t a native English speaker as your writing is so perfect. I do believe she is mistaken in this case, but have no argument with her technical abilities as an engineer

 

[Ron Party]

https://www.audiosciencereview.com/...vement-or-marketing-tool.130/page-5#post-2719

 

[pkane]

Yeah, apologies I can never remember if you need to use capitals for She/He or not.
Mistaken?
so this massively experienced and respected conscientious recording engineer hasn't realised that PCM and DSD are exactly the same and sound exactly the same and has soldiered on with a format that is time consuming, controversial and expensive to try to recoup the work needed when all she needs to do is just record and mix everything in PCM - I better tell her/HER then!!!

Experience is great, but is no guarantee of infallibility. Show me that FLAC and WAV files sound different in an objective way, and I'll agree. Show me that DSD is better than PCM in an objective, repeatable way, and I'll agree. As I (and others) said already, DSD isn't better from a theoretical point of view. There can certainly be differences in DAC and ADC implementations that could cause audible differences when processing two formats. But a blanket statement that DSD is superior, in my view is unfounded regardless of what Cookie says or does. And from a practical standpoint, DSD is much less convenient.

 

[Pluto]

What percentage of recording engineers prefer to work in DSD?

Relatively few for the simple reason that it's a pain in the butt. Unless you have, to hand, the highly specialized kit capable of working entirely within the DSD space, there is little point. The market for DSD is absolutely minute and the people who want to buy this stuff are, frankly, the obsessives for whom it's never good enough.

Something that has to be remembered is that DSD acquired its momentum at the heyday of SACD, the early noughties, having been developed in the last few years of the nineties. At that time, 96kHz converters were, largely, a curiosity unlikely to be accurate beyond 18 bits if you were lucky. So anything DSD-based was, at that point in time, a real step forward. Moving on 15 years, even cheap A to D converters can easily out-perform whatever was SOTA in 2001 and we can manage sample rates that were unheard of 20 years ago.

In short, DSD had its place and time. But things have since moved on.

 

[Pluto]

So this massively experienced, respected and conscientious recording engineer is unable to realise that PCM and DSD sound exactly the same and has stupidly soldiered on....

Maybe, maybe not. But I expect the fact that it's what I would call “a nice little earner” counts for something

 

[dc655321]

So this massively experienced, respected and conscientious recording engineer is unable to realise that PCM and DSD sound exactly the same and has stupidly soldiered on with a format that is time consuming and testy for the last 9-10 years for no worthwhile reason whatsoever?

Product differentiation is a helluva drug...

 

[captain paranoia]

She also thinks wave and flac sound different.

Oh. Well, that gives us a woo-factor reference, at least...

 

[andymok]

Speaking of technical advantage, I genuinely wish to hear any comments regarding info given below

(source: https://www.gearslutz.com/board/rem...nd-location-recording/30990-dsd-wdsd-dxd.html )
(I've been avoiding to quote as ppl would take pieces out of context not by reading the whole discussion, but I really wish to hear more. If you wish to know where I quote from, go back and see p.30)

Hi. My name is Peter Scheelke. I am one of the founders of DAD, Digital Audio Denmark. We are the manufacturer of theAX24 AD/DA converter and the Merging Sphynx 2 converter.
In the discussion about DSD/DXD, Sadie/Sonoma/PYRAMIX I can help clarify some of the advantages/disadvantages in the different audio formats.

DSD 64fs (the format used for SACD) is a 1 bit format with a sample rate of 2.8224 MHz (64 x 44.1 kHz). The main advantage of this DSD format is the as close to perfect as we have ever seen impulse response, which is just a little more than 100% measured with a 3us pulse, plus the very wide frequency band. The disadvantage of DSD compared to digital PCM is the noise performance and the fact that DSD can not be edited since it is only 1 bit.

In our AX24/Sphynx 2 we have been able to keep the noise down until 23 kHz (<-120 dBfs measured 20-20 kHz), however the energy from the noise is still part of the signal, so after 23 kHz the noise becomes increasingly significant.

Since DSD can not be edited as a one bit signal, it is necessary to convert to some kind of multi bit for editing. When converting back to DSD the modulator noise will be added again, and after just a few editing back and forth the audio will no longer comply with the noise performance specification stated in the Scarlet Book (specifications for SACD’s). DSD 64fs is used as recording format in Sonoma, Sadie and PYRAMIX DSD systems. (PYRAMIX can also record in DXD).

There is also DSD 128fs, also 1 bit, but with a sample frequency of 5.6448 MHz (128 x 44.1 KHz). Since it is a 1 bit signal, quantization noise is equal to DSD64, but since the frequency band is twice as wide, the noise is only half that of DSD 64fs within a given frequency band. With the same type of noise shaping as DSD 64 the noise will start rising at the double frequency. In our implementation the noise starts rising from about 46 kHz.

PCM 44.1-192 kHz/24 bit: Since all frequencies, at more than half the sample rate, will be mirrored around half the sample rate (aliasing), all PCM formats needs an anti aliasing filter. The normal anti aliasing filter is the 0.45/0.55 filter which starts at 45% of the sampling rate and has full attenuation at 55% of the sampling rate.

A major disadvantage in the normal 0.45/0.55 anti aliasing filter is that the filter is only attenuating 10-12 dB at half the sample rate (Nyquist), so frequencies between 50% and 55% of the sample rate will get mirrored around half the sample rate and will create new frequencies without any harmonic relationship to the audio. Another disadvantage is that some of the energy from the audio is lost in pre/post ringing; a stronger anti aliasing filter will create more pre/post ringing than a less intense filter. Since some of the energy is lost, the anti aliasing filter attenuates the impulse response.

Due to bandwidth a steep anti aliasing filter at 44.1 and 48 kHz sampling rate can be justified, however at higher sampling rates (96kHz, 192kHz) it would be better to use a less steep filter. All anti aliasing filters cause delay in the A/D converter which is about 0.8 ms at 44.1 kHz sampling rate with a 0.45/0.55 filter.

DXD is something new, a 352.8 KHz/24 bit sampling rate in the A/D converter intended for 32 bit floating point DAWs. Until now DXD is supported by DAD, by Merging in their PYRAMIX DAW, by AMS Neve in their new DFC PS/1 console, and in the Saracon SRC from Daniel Weiss, but also other manufactures are one their way with DXD equipment.
In our implementation we have chosen to add only very soft anti aliasing filter for DXD and for an even higher sample rate we offer, 384 kHz; since there is very little audio above half of the sample rate. Our implementation of DXD has a great impulse response (88% with a 3us pulse) and a significant better out of band noise performance compared to DSD64fs.
We think it’s a big step forward, offering far better impulse response that previous PCM, avoiding the filtering problems of previous PCM and better out of band noise performance than DSD. We think this is the future of digital audio.

Merry Christmas from Denmark to all audio lovers at Gearslutz!

Peter

It is much better to record directly in DXD format. The principle of the DAD AD converter is to sample in 5-bit at 128 times the sample frequency. When sampling is based on 44.1 kHz the sample rate of our delta/Sigma front end is 5.6448 MHz. This gives a much better and accurate signal, than if it was only derived from a 1-bit 64 og 128 fs signal as DSD.
If the media is SACD SRC to DSD is only needed once after all editing has been done.

For DXD our filter start very gentle at 125 KHz. At 200 KHz the filter attenuates about 12 db.

The simple answer to the question about 1 bit processing is that the delta/Sigma modulated 1 bit signal is giving the information if the sample is one higher or lower than the previous. Further more to process at only one bit does not give any room for calculations, since any digital processing will have +/-1bit accuracy. Therefore the delta Sigma signal has to be decimated to so a called multi-bit signal. DXD is indeed such a signal, but other formats are used by other vendors.

The problem is not so much to convert from DSD to any multi bit format nor the Sony e-chip (8bits 64fs). The signal deteriation appears when the multi bit signal has to be re modulated back to DSD. Then the characteristic noise of the 1-bit delta/Sigmamodulator again is “infected” on the signal making the out of band noise worth than it originally was on the DSD signal. When the source signal has a higher resolution (like 5bit) only the last modulation noise will affect the signal.

Best regards,

Peter

In the figures below the relationship between sample rate, impulse responses, noise performances and frequency response from the A/D converter in our AX24 can be seen.

It is clear that DSD has the very best impulse response and band wide but significant more “out of band” noise than PCM and DXD. The impulse response of DXD is still good and the noise performance is better than DSD64fs.

Everybody who has ever compared DSD64fs, DSD128fs, DXD and 384 KHz to analog would probably agree, that all these formats are sounding much more transparent than PCM between 44.1 and 192 KHz. The main advantage of high sampling rate is however not the wider frequency band itself since we can not hear frequencies above 20 KHz.
High sampling rates is mainly about timing due to improved impulse response.

The front-end of our AX24 DSD/DXD converter is a 5-bit delta-Sigma modulator running at 128fs. (5.6448 MHz). A multi bit A/D front-end is much more efficient than a 1 bit front-end.

I our implementation I would say that DXD/384 KHz is the most transparent format. Very close to DXD comes DSD128fs and then we have DSD64fs. There is a quite a jump between DSD64fs and 192 KHz.

If the media is SACD and editing is not needed (except for cross fades) it is probably the best solution to record in Pure DSD, since any sample rate conversion has an impact on the sound.

If editing is needed I would recommend DXD/384 KHz. If the media is SACD you will only need to convert to DSD once, if the media is something else, you will have the benefit of a higher resolution in the production phase. Also EQ, dynamic and reverb are working much better at higher sampling rates.

Best regards,

Peter

 

[Pluto]

I genuinely wish to hear any comments regarding info given below

I'm sure his comments are largely correct. His company makes great, though expensive, products. But the simple fact is that DSD and its derivative formats are frozen in the era 1995-2005 and many proponents simply forget just how good modern (and relatively low cost) PCM converters can be. Once you are sampling at 96kHz (with a properly designed converter) all your filtering anxieties simply disappear into the noise!

Perhaps, maybe, possibly there are a few people in this world with such abnormally superhuman hearing (and the necessary cognitive ability to support it) that they really can tell the difference.

But then again, perhaps not.

 

[M00ndancer]

Perhaps, maybe, possibly there are a few people in this world with such abnormally superhuman hearing (and the necessary cognitive ability to support it) that they really can tell the difference.

And even if there were, they would only listen to the difference and not the music.

 

[mansr]

Speaking of technical advantage, I genuinely wish to hear any comments regarding info given below

I have a few comments.

The main advantage of this DSD format is the as close to perfect as we have ever seen impulse response, which is just a little more than 100% measured with a 3us pulse, plus the very wide frequency band.

This obsession with the visual appearance of the impulse response is more than a little misguided. Also, what is 100% compared to what?

A major disadvantage in the normal 0.45/0.55 anti aliasing filter is that the filter is only attenuating 10-12 dB at half the sample rate (Nyquist), so frequencies between 50% and 55% of the sample rate will get mirrored around half the sample rate and will create new frequencies without any harmonic relationship to the audio.

While technically correct, the disadvantage isn't all that major. The aliasing only affects frequencies in the 0.45-0.5 fs range. With a sampling rate of 88.2 kHz or above, this is well above what any human can possibly hear, and there's hardly any content at those frequencies anyway. At 44.1 kHz sample rate, the aliasing might become audible in some circumstances. Even 48 kHz is enough to make it pretty much a non-issue.

Another disadvantage is that some of the energy from the audio is lost in pre/post ringing; a stronger anti aliasing filter will create more pre/post ringing than a less intense filter. Since some of the energy is lost, the anti aliasing filter attenuates the impulse response.

This is incorrect. The lost energy is that contained in the removed, inaudible frequencies. The peak of the impulse response being lowered a bit is not a defect. Also, "ringing" is a misnomer.

All anti aliasing filters cause delay in the A/D converter which is about 0.8 ms at 44.1 kHz sampling rate with a 0.45/0.55 filter.

True, but so what? Latency only matters for live effects processors, and there several milliseconds is still acceptable.

High sampling rates is mainly about timing due to improved impulse response.

This is incorrect. Timing accuracy does not depend at all on sampling rate.

 

[andreasmaaan]

Speaking of technical advantage, I genuinely wish to hear any comments regarding info given below

(source: https://www.gearslutz.com/board/rem...nd-location-recording/30990-dsd-wdsd-dxd.html )
(I've been avoiding to quote as ppl would take pieces out of context not by reading the whole discussion, but I really wish to hear more. If you wish to know where I quote from, go back and see p.30)

The ringing contained in properly filtered PCM is outside the audio band, just as is the rising noise in DSD. In each case, we have a problem that is visible in measurements but inaudible to humans (unless a poor decision has been made at some point in the chain).

 

[John Deas]

Speaking of technical advantage, I genuinely wish to hear any comments regarding info given below

(source: https://www.gearslutz.com/board/rem...nd-location-recording/30990-dsd-wdsd-dxd.html )
(I've been avoiding to quote as ppl would take pieces out of context not by reading the whole discussion, but I really wish to hear more. If you wish to know where I quote from, go back and see p.30)

It's an interesting read but all from around fourteen years ago now. There are people recording in DXD notably Morton Linberg 2L and Music Liaison and they make great records. Don't know of anyone recording in DXD just for the purpose of converting and releasing it in DSD though (pointless) certainly plenty of labels (NativeDSD.com) using it to mix DSD through Pyramix.

 

[Kal Rubinson]

There are people recording in DXD notably Morton Linberg 2L and Music Liaison and they make great records. Don't know of anyone recording in DXD just for the purpose of converting and releasing it in DSD though (pointless) certainly plenty of labels (NativeDSD.com) using it to mix DSD through Pyramix.

Well, that's what Morten and others did (and do) for their SACD releases.

 

[John Deas]

Well, that's what Morten and others did (and do) for their SACD releases.

Sorry I don't understand the point you're making? Yes Morten Linberg records in DXD (he prefers it to DSD - please don't start...) into Pyramix and creates all kinds of versions because people want all kinds of versions of his recordings from the DXD multichannel original to Blu-ray 192/24, 96/24 down to CD. The point I was making was I am not aware of anyone recording in DXD just to convert and sell it in DSD. Many people record in DSD and mix through Pyramix back to DSD for the simplicity, how much this effects the original capture I've no idea but the consensus I've had from many correspondence is it's not ideal. Don't shoot the messenger

 

[John Deas]

Product differentiation is a helluva drug...

Trying to do something as well as you can is a better one.