The sound quality of DSD?

[Saidera]

Multiple threads here deal with DSD, but I'm yet to see a bit more depth concentrated in one place. DSD For Beginners is boring and tells very little.

Based on listening to the sound quality of DSD native from Cirrus Logic chips like that in Sonata HD Pro I find:
• S/N ratio of DSD at high volume is bad. At high volumes, you can hear noise. Not good at all!
• There is no S/N problem with PCM volume, so it is not a problem with the DAC or analog circuit.
• DSD64/2.8Mhz has a 'dry' sound in the high range, probably because of residual noise. DSD128/5.6Mhz sound is beautiful.
• PCM and DSD have different tonality? It is said that DSD has a feeling of air, but it seems better to say that it uses airy sound as a decoration.
• There is no great difference between DSD and PCM in terms of sound quality such as attenuation length, depth of space, depth, and fineness of depiction.
So none of the results were as expected. Especially when the digital volume is reduced with DSD native, it becomes more apparent, but I think one of the sound quality characteristics of DSD is the effect of residual noise. But I understand that there are many people who like this sound. Especially in Japan, the direction of the sound that has been often heard in domestic audio has long been heard, and many music production engineers are willing to make such sounds. For female vocals, it's better to make audio sound like this. Perhaps?

On the frequency axis, DSD just looks like a disadvantage. DSD has a bad S/N and is disadvantageous compared to PCM.

But there is a DSD advantage. There is a square wave comparison of DSD and PCM in the link http://www.computeraudiophile.com/f...irect-stream-digital-18636-print/index11.html, but the DSD square wave response here shows a very quick response closer to NOS. The biggest advantage of DSD seems to be the fast time base response. This will not happen depending on the data converted from PCM, the intermediate processing, and the way of filtering, and it seems that the response is not much different from PCM like the image at the link destination above.

DXD is 24bit 352.8kHz. It's a lot like DSD.

Some reasons for the difference in sound between DSD and PCM:

I haven't made a strict comparison, so I can't make a hypothesis.

• Influence of high frequency noise due to noise shaping included in the DSD method itself
• Difference between DSD and PCM due to characteristics of digital filter applied inside DAC
• For DSD native data, the response on the time axis is as fast as NOS
• DSD affects the current/ power consumption and power supply noise due to the amount of computation inside the DAC. Higher sound quality = lighter processing with DSD the calculation load inside the DAC is lighter.
• The effect of deterioration due to jitter does not exist in DSD. Jitter countermeasures aren't necessary for DSD.

I think DSD's greatest advantage is the third point. In the case of properly filtered DSD data that does not go through PCM at all, the time axis response of the area that could not be reproduced by DXD is possible. When converting this to PCM, it is likely that the superiority of the time axis will be lost unless it is considerably high sampling data (ideal is 384 k or more?). Of course, when the data of the CD sound source is originally DSD, the information on the time axis has already been lost, so this advantage does not occur. The high-frequency noise of DSD is only added while the time base information is lost. Pure DSD is the only way.

Therefore, if you're not a time domain sensitive listener, you should not be so concerned about choosing between PCM or DSD. Converting files to DSD just to take advantage of the 'better sound' from the DAC or software is also conceivable. Convert all the sources you have into DSD? I can't recommend it. DSD does not increase the information on the time axis of PCM files.

Depending on the type of DAC, even if DSD is input, it may be converted to PCM internally, so in that case there is no point in sticking to DSD at all.

But http://www.computeraudiophile.com/f...ream-digital-volume-control-19955/index3.html
In short, it seems that it is converting DSD to DSD rated PCM and doing volume processing etc. Since the rate is not reduced by decimation and the original amount of information is maintained, this will not cause any deterioration in the data. It's converted to an ESS proprietary format that is neither PCM nor DSD. ESS may have less S/N deterioration from DSD volume control than Cirrus, perhaps due to better internal processing in the latter stages. Anyway, that doesn't serve as a reason to favor E1DA over Sonata Pro.

 

[PaulD]

Ok... For a start I would not believe ANYTHING on computeraudiophile.com.

There is a LOT of complete bullshit around DSD.

There are NO timing errors with PCM! It's a band limited signal!
See

Here are some bits from other posts I have made before:
--
DSD will not, cannot, improve the sound quality.

Here are 3 previous posts of mine on the topic:
https://www.audiosciencereview.com/...-rate-and-audible-frequency.9411/#post-247065
https://www.audiosciencereview.com/forum/index.php?threads/dsd-is-it-of-any-value.10682/#post-296499
https://www.audiosciencereview.com/...n-playing-native-dsd-files.11438/#post-326961

Other posts in those threads will also be of value to you. There will be more on DSD from others who are probably better qualified, I only pointed to a few.
--

A DSD stream cannot be modified at all, so you need a preamp if you want to remain in DSD land. If you want to modify it, then it must be converted to PCM, even for volume change (amplitude multiplication).

I would suggest that DSD is unnecessary and it's full of audiophile BS. High resolution PCM is at least as good, and without the ultrasonic noise-shaping issues, see https://archimago.blogspot.com/2013/04/musings-on-sacd-dsd-audio.html

DSD to PCM conversion is completely transparent, see https://archimago.blogspot.com/2015/04/analysis-dsd-decoders-2015-windows-mac.html So if you don't mind going to PCM then you can use a digital volume control.
--

I did do a blind test in a studio with the Korg MR-1000 DSD recorder when it came out, signal split after the mic preamps and sent to the Korg recorder and also to Avid Protools PCM converters at 24bit 96KHz. Level matched on playback, there was no audible difference between the DSD and PCM version.

Given the aggressive noise shaping required by DSD to make it work, and the fact that it cannot be edited or processed (even volume change is impossible) unless converted to PCM, I can see no value in DSD at all now. When DSD was developed PCM was stuck at 16bits and 48 KHz and perhaps it had an audible (or archival) advantage over that format of PCM, but today we routinely have 24bit 96KHz PCM which has none of the disadvantages of DSD, it has more resolution than we can hear and it can be edited and processed easily.

 

[Saidera]

There's simply too much out there on DSD that in the end, one has no idea why it exists and when to utilise it. For Primeseat streaming? Archiving and using Sony Sonoma SBM Direct on it (create best sounding CDs and 24/44.1)? DSD will perhaps become obsolete once 768 kHz becomes supported in mainstream DACs, as they sound similar? And why did DSD survive as a digital format?

Only recently has DSD once again become accessible for the poor listeners - Sonata HD Pro supports up to 11.2 MHz Natively at a cost of $25 to $40, taobao or aliexpress, equivalent to $60+ AUD. A bit more pricey - there is the popular DC02 which measures badly. Before these, VAIO DSD reigned from 2005-2011 using CXD9872 (to 2007) and ALC889DSD. But VAIO could record LINE IN or MIC IN using DSD! ALC1220 can do it via Realtek ASIO, so there's a reason to buy gaming mobos for recording in DSD. DSD = expensive bs. That image remains strong for us. We haven't fully tried it out, because it's been out of reach for so long.

Perhaps this 'inaccessibleness' of DSD - your casual computer or old phone can't play it directly! - has fed its success. Now of course, phones using Qualcomm audio codecs can play DSD 'directly', but even that seems to be through PCM according to rumors.

And I can't explain it well, but definitely some DACs work better with DSD depending on listener's preferences.

DSD is rare and thus people use it?

 

[firedog]

Ok... For a start I would not believe ANYTHING on computeraudiophile.com.

There is a LOT of complete bullshit around DSD.

There are NO timing errors with PCM! It's a band limited signal!
See

Here are some bits from other posts I have made before:
--
DSD will not, cannot, improve the sound quality.

Here are 3 previous posts of mine on the topic:
https://www.audiosciencereview.com/...-rate-and-audible-frequency.9411/#post-247065
https://www.audiosciencereview.com/forum/index.php?threads/dsd-is-it-of-any-value.10682/#post-296499
https://www.audiosciencereview.com/...n-playing-native-dsd-files.11438/#post-326961

Other posts in those threads will also be of value to you. There will be more on DSD from others who are probably better qualified, I only pointed to a few.
--

A DSD stream cannot be modified at all, so you need a preamp if you want to remain in DSD land. If you want to modify it, then it must be converted to PCM, even for volume change (amplitude multiplication).

I would suggest that DSD is unnecessary and it's full of audiophile BS. High resolution PCM is at least as good, and without the ultrasonic noise-shaping issues, see https://archimago.blogspot.com/2013/04/musings-on-sacd-dsd-audio.html

DSD to PCM conversion is completely transparent, see https://archimago.blogspot.com/2015/04/analysis-dsd-decoders-2015-windows-mac.html So if you don't mind going to PCM then you can use a digital volume control.
--

I did do a blind test in a studio with the Korg MR-1000 DSD recorder when it came out, signal split after the mic preamps and sent to the Korg recorder and also to Avid Protools PCM converters at 24bit 96KHz. Level matched on playback, there was no audible difference between the DSD and PCM version.

Given the aggressive noise shaping required by DSD to make it work, and the fact that it cannot be edited or processed (even volume change is impossible) unless converted to PCM, I can see no value in DSD at all now. When DSD was developed PCM was stuck at 16bits and 48 KHz and perhaps it had an audible (or archival) advantage over that format of PCM, but today we routinely have 24bit 96KHz PCM which has none of the disadvantages of DSD, it has more resolution than we can hear and it can be edited and processed easily.

DSD sounds different than PCM. I've also A/B compared the same tracks in both formats and I can distiguish them. A lot of people prefer the sound of DSD - even PCM converted to DSD. The simple fact that different modulators and filters are used for DSD playback means it sounds slightly different from PCM. It's a matter of taste, not that one is better or more acccurate than the other.

As far as the editing of it, etc., that's irrelevant for listening purposes.

A program like Roon can make changes to DSD by converting it to multibit-DSD and back to one bit. We can argue all day whether that is DSD or not. I'd say it is, as no decimation of the DSD stream takes place. I'm not going to argue that point as it's a pointless argument.

It's been shown with measurements by people like Miska that there are setups and DACs that produce their most accurate non-distorted output when fed DSD and even 2X or 4X DSD. Sorry if that doesn't fit your preconceived ideas and your linked pages from xiph. Sometimes real world facts inconveniently don't match your ideas that you want to crusade about. Your arguments would work better if you don't set up straw men to knock down.

 

[Saidera]

"DSD cannot be mastered". Not like that. The volume of 1-bit ΔΣ signal can be adjusted and mixed relatively easily. Actually, many DAPs adjust the volume of 1 bit ΔΣ of the headphone output in real time.

1-bit ΔΣ differs from PCM in that the sampling frequency is high. Trying to filter with FIR is time consuming because you have to process a large number of samples in addition to increasing the number of filter taps.

I've read Understanding Delta-Sigma Data Converters by Richard Schreier & Gabor C. Temes

I think there is only one textbook that helps you study ΔΣ modulation now. There are some points that are difficult to read. Although the theoretical conclusion is clearly stated, the process leading to the conclusion is omitted, so you have to think for yourself. Unless you are a university student in science and engineering who can consult with others around you, it will be difficult to read. The description in this book and the theory and experience of most listeners are inconsistent.

After hundreds of DSD v PCM threads over 20 years there's still no textbook statement to link the theory with what we hear. And there's a lot of DIY and different DACs when it comes to DSD. And my OP with 'time base response' isn't a popular explanation. And Sony's noiseshaping and filters too.

SBM Direct helps to avoid DSD v PCM by converting DSD to CD. But what exactly happens to make that CD sound different from the original DSD? I know it uses a 'single stage FIR Filter 32k taps and noiseshaping all processed at 64 bits' and it seems that 64k taps FIR makes sound even better.

Both Hidizs S8 and Sonata HD Pro have the DSD logo on their backs. Clearly it is the cheapest DSD DAC that ever existed till today, at just $25 USD! But it can't record in DSD.

What is high resolution?

"It is the data that can be expressed in the shape of a wave (in both the amplitude direction and the time axis direction), and it has six times as much information as a CD."

Some of my CDs I bought at the time when SBM became popular boast that they have 16 times as much information as conventional CDs.

Is this sound quality higher than high resolution?! I haven't touched on official material explaining in Japanese what Super Bit Mapping and similar technologies are in the first place. Is high resolution only 6/16 of SBM? Or is the amount of information increased 96 times by 16 times SBM × 6 times high resolution? The explanation was suspicious at that time.

Explanation written on CDs: "With the new technology, it is now possible to put 20-bit information of the master sound source on 16-bit CD. Since 20-bit audio information can be inserted 16 times more detailed than 16-bit, the amount of information is 16-fold." It's strange that 20bits can be stored and played in 16bits with lossless compression without changing the audio format of the CD or the player. Even if it becomes possible to put 20 bits in a 16-bit area without loss using data compression, the ratio of the amount of information is calculated in bit units, so isn't it (20/16) times more?

High resolution may be disadvantageous

Is there an auditory effect that can justify the amount of data increased by high resolution?

Increased capacity of recording media (flash ROM manufacturers are pleased with increased sales)

Data I/O speeds up (LSI manufacturers are happy to sell new models)

Power consumption of equipment also goes up (battery manufacturers are pleased to sell large capacity secondary batteries)

 

[BDWoody]

After hundreds of DSD v PCM threads over 20 years there's still no textbook statement to link the theory with what we hear.

First there would need to be more than anecdotal evidence. I haven't seen anything that actually clearly demonstrates this hearing distinction or preference. I saw one study that is often tortured to try to show the slightest of preference, once listeners receive some training, but that was certainly not conclusive.

What else is out there beyond the "I hear it too!" crowd?

 

[Saidera]

You're right. Those studies (ones I read were done in Japan) use the same DAC unit or seem inconclusive for us.

 

[bennetng]

DSD on some converters can be 15dB noisier than PCM below 20kHz. Would be interesting if someone's system has non-ideal gain staging.
https://www.audiosciencereview.com/...ds/new-dac-chipset-from-akm.14391/post-450135
https://www.audiosciencereview.com/...ds/new-dac-chipset-from-akm.14391/post-454955
Also, "multibit DSD" is an oxymoron, please use multibit delta-sigma instead.

 

[Buster Chestnut]

Just use the highest quality, best mastering you can find; either DSD or PCM given you have the equipment to choose.

I would steer away from blanket statements made in the OP.

 

[Archsam]

Don't forget SACD is the only surviving high rez format that comes in a physical disc format. I will say DVD-Audio / Blurry Audio are truly dead formats e.g. you can't really find new music sold in these format anymore.

SACDs are still kicking around in Asia. My music collection covers pop songs from Hong Kong and Japan, and whilst most of their latest releases are remastered albums from the last 10-20 years, that suits me just fine. Today I am firmly in the computer audio camp (I own a lot more DSF files than SACD discs) however I know plenty of folks who still enjoys popping a disc into a player rather than faffing with network playback.

Access to DSF files (either via downloads or ripping) are also relatively new, so I think the DSD vs PCM discussion needs to be framed in that context.

Personally I don't really mind whether a song is in DSD or PCM, I just want songs I like that are mastered well.

 

[Kal Rubinson]

DSD is rare and thus people use it?

DSD is not rare. Thus, people who want the music released in DSD must use it.

 

[MediumRare]

OP, IMO you are going down a rabbit hole of technical "explanations" for at best, infinitesimal differences. Evidence: @amirm ’s analyses of numerous redbook DAC results which are demonstrably transparent to humans. You *might* be able to measure a difference (140 dB of Jitter v. 150 dB) but that is orders of magnitude beyond human hearing. 100 microseconds of group delay? Think about that. Any audible differences are not due to the encoding Of DSD v PCM but the related electronics or mastering.

 

[AnalogSteph]

DSD is a pretty silly format if you think about it. Storing things as bitstream results in substantial data bloat, and on top of that it is impossible to dither a 1-bit bitstream enough for it to become fully transparent in the way PCM can be. You are basically embedding the characteristics of a (decent enough) 1-bit DAC into the audio.

There is no compelling technical reason for choosing DSD over hi-res PCM. None. I am sure the obscurity of the format was very welcome to label folks at a time where CDs were being copied and ripped left and right. Think of it as additional copy protection for SACD, just in case the need for special drives doesn't do.

If things had been done by the book, we would have had mainstream DVD-Audio instead.

 

[Buster Chestnut]

it is impossible to dither a 1-bit bitstream enough for it to become fully transparent in the way PCM can be.

Hmmm.... What is meant by “transparent”?

 

[raif71]

DSD is a pretty silly format if you think about it. Storing things as bitstream results in substantial data bloat, and on top of that it is impossible to dither a 1-bit bitstream enough for it to become fully transparent in the way PCM can be. You are basically embedding the characteristics of a (decent enough) 1-bit DAC into the audio.

There is no compelling technical reason for choosing DSD over hi-res PCM. None. I am sure the obscurity of the format was very welcome to label folks at a time where CDs were being copied and ripped left and right. Think of it as additional copy protection for SACD, just in case the need for special drives doesn't do.

If things had been done by the book, we would have had mainstream DVD-Audio instead.

I would say the simple popping of sacd and play instead of the dvd-audio/bluray-audio sifting through menus via a tv screen makes sacd a more enjoyable and practical choice for many but I agree with you that technical wise pcm may be more .... "practical" ?

 

[Head_Unit]

You are basically embedding the characteristics of a (decent enough) 1-bit DAC into the audio.

Well, yes. IIRC, Sony essentially stored the internal bitstream off whatever A/D converter they were using. That actually is more honest in a way than converting it to PCM and pretending it's pure. A few foibles do remain:
- The part about not being able to mix or otherwise process without conversion to PCM
- The dishonesty or at best disingenuity of their little graphic showing this pure flow which perhaps applies to I dunno 1.487% of recordings
- The fact that DSD or PCM is rather irrelevant when most of the source material is being scraped off ancient tape, with all those attendant problems
- The part about wondering why not use really good multibit converters...maybe they weren't as good in some specifications at the time? I don't know.

 

[Head_Unit]

A program like Roon can make changes to DSD by converting it to multibit-DSD and back to one bit. We can argue all day whether that is DSD or not. I'd say it is, as no decimation of the DSD stream takes place...It's been shown with measurements by people like Miska that there are setups and DACs that produce their most accurate non-distorted output when fed DSD and even 2X or 4X DSD.

...and then I once saw a demo at a surround sound conference showing that a small signal was unsteady and bobbled around in amplitude on DSD but not on PCM. Hmmm, all those demos *might* be artifacts of particular setups/chains? I was talking to John Eargle and he opined to me he felt 24/96 was quite transparent to the source. OTOH, a couple fellows from Telarc said they had set up parallel chains (mic feed, DSD, PCM) and felt the DSD was a bit closer to the mic feed. And yet another guy asked me "If DSD measures so bad, why does it sound so good?"-a great question.

?Could you explain more about "no decimation"? I'd like to hear why you consider the "wide" as still DSD. Not to argue, I've just always thought "well it's not one bit any longer so it's PCM" without any contrary ideas.

 

[firedog]

...and then I once saw a demo at a surround sound conference showing that a small signal was unsteady and bobbled around in amplitude on DSD but not on PCM. Hmmm, all those demos *might* be artifacts of particular setups/chains? I was talking to John Eargle and he opined to me he felt 24/96 was quite transparent to the source. OTOH, a couple fellows from Telarc said they had set up parallel chains (mic feed, DSD, PCM) and felt the DSD was a bit closer to the mic feed. And yet another guy asked me "If DSD measures so bad, why does it sound so good?"-a great question.

?Could you explain more about "no decimation"? I'd like to hear why you consider the "wide" as still DSD. Not to argue, I've just always thought "well it's not one bit any longer so it's PCM" without any contrary ideas.

Meaning the conversion is not full PCM representation (lower sample rate, wider samples) only multiplying/dividing the one bit and not a full conversion to PCM that produces a signal that looks like a PCM signal.

Here's part of Roon's explanation of how they DSP process DSD without converting it to PCM. (Like I said, I don't want to argue the point with those who insist the signal is no longer DSD. You can accept their POV or not. But IMHO, the tech people at Roon are at least as knowledgeable as some of the very opinionated posters here):

"We do indeed process DSD without performing a DSD->PCM conversion first....it "works like this:

1. Start with a DSD stream, and widen from 1 bit-per-sample to 64 bits-per-sample
2. Apply a low pass filter to remove the bulk of the inherent noise energy from the widened signal.
3. Apply processing steps to the wide intermediate format.
4. Send the signal through a sigma-delta-modulator to re-render the “wide” 64-bit stream into a 1-bit DSD stream.

The low pass filter (2) in this process might sound like the reconstruction filter we discussed above, but it is very different. It is much more lenient, less steep, and it only attenuates frequencies over 100kHz–and these already have a very poor SNR because of the inherent noise shaping in DSD, so we can be sure that no meaningful information existed there in the first place.

Without the filter, sound quality suffers significantly or the sigma delta modulator risks becoming unstable (i.e. starts outputting horrible sounds that ruin your ears and if you’re unlucky your gear too).

At step (3) the signal is structurally similar to a PCM signal–in that it is comprised of a series of multi-bit samples. However, it does not have content typical of PCM signals and it maintains the DSD sample rate. If you looked at a spectrogram of the intermediate format in (3), it would look just like DSD, except with the bulk of the noise above 100kHz severely attenuated by the low pass filter.

By maintaining the original sample rate through processing, the time-domain characteristics of DSD are maintained. By designing the filter to stay far away from musical content, the frequency-domain characteristics are maintained too.

Sometimes this form of processing, or this intermediate format is referred to as “DSD-Wide”. We didn’t use that term because some people have defined DSD-Wide as an 8 bit intermediate format (whereas we use 64 bits…a luxury of precision afforded to us by running on modern desktop-class CPUs) and I didn’t want to create confusion."

 

[Head_Unit]

widen from 1 bit-per-sample to 64 bits-per-sample

Uh, I think I see. Instead of PCM oversampling by adding more samples in time...they are oversampling in width? That doesn't sound right. But I kind of get it.

 

[LDKTA]

DSD = PCM

A highly oversampled, low bit/sample and noise shaped PCM. I wouldn't even bother giving it a second thought... There is a reason DSD is irrelevant everywhere aside from the likes of audiophiles that generally don't know any better.