Dsd - how far does one go?

This is audiophile land after all. There will there be a push for higher and higher rates which, just so coincidently, cost more and more to purchase. Personally, I won't purchase any higher files than 24/192 and double DSD. More than enough for me.

For the past couple of years I've said 128fs or DoubleDSD is all you need. I still stand by that. Even though I have QuadDSD capability in my Horus, just about all of my transfers are at Double. And to me, I still can't hear the difference between Quad and Double. If there is a difference, then the DAC must be handling the files differently!

Audiophile Bill said:

Hi all,

Just browsing the pages of my monthly subscription to Hifi News and noticed that CES saw the launch of the EMM labs DA2. Now said Dac is hexadeca DSD capable (16x) - I asked myself immediately I wonder when my GG will be able to do that after a new USB interface is installed. I then thought imagine the processing power required to upsample quad dsd or PCM to that.

Anyway - what level of dsd is the maximum that offers any theoretical benefits or can you just keep going?

Cheers,

Bill

Click to expand...

Clearly a numbers game, horsepower race, megapixel count, or whatever you wish to call it. It makes for good marketing.

Bruce B said:

For the past couple of years I've said 128fs or DoubleDSD is all you need. I still stand by that. Even though I have QuadDSD capability in my Horus, just about all of my transfers are at Double. And to me, I still can't hear the difference between Quad and Double. If there is a difference, then the DAC must be handling the files differently!

Click to expand...

I agree with transfers I have heard. it's very hard to hear any difference with 2x verses quad with transfers from tape, vinyl, PCM or single dsd. 2xdsd is the sweet spot of transfers.

my limited exposure (about 20 so far) to native quad recordings tells me it's something more. but I only have one recording where it's native 2x verses native quad so in my mind it's still a question. it's possible that native quad is worth doing.

but I would likely defer to your much more informed perspective if someone is looking for truth right now.

The added bandwidth required for quad probably increases the noise floor and hurts linearity, though could also provide more natural dither... I don't really need GHz bandwidth front ends for my MHz AM radio... Reminds me of the THD numbers game years ago when claims kept getting lower and lower; marketing pushed them there despite the obvious other problems that resulted. Always gets me how so many who claim nothing but the sound and their ears matter get caught up in the numbers game, be it SNR, THD, W/ch, clock (sampling) rates, or whatever... Of course the other side is just as bad about neglecting a body of evidence without trying to see if something is rally perception bis or a real effect worthy of deeper study.

Audiophile Bill said:

Hi all,



Anyway - what level of dsd is the maximum that offers any theoretical benefits or can you just keep going?

Click to expand...

Soon you will need a 500GB drive for one file, and somebody will say storage is cheap.

DonH50 said:

The added bandwidth required for quad probably increases the noise floor and hurts linearity, though could also provide more natural dither... I don't really need GHz bandwidth front ends for my MHz AM radio... Reminds me of the THD numbers game years ago when claims kept getting lower and lower; marketing pushed them there despite the obvious other problems that resulted. Always gets me how so many who claim nothing but the sound and their ears matter get caught up in the numbers game, be it SNR, THD, W/ch, clock (sampling) rates, or whatever... Of course the other side is just as bad about neglecting a body of evidence without trying to see if something is rally perception bis or a real effect worthy of deeper study.

Click to expand...

The way DSD works would it not be true that increasing to a higher rate spreads the noise over a wider bandwidth which means the bandwidth in the audible range would be less of the total bandwidth and therefore a lower level noise floor in the audible range with quad vs double rate DSD?

Disclaimer: I am not all that familiar with DSD though I have a good understanding of sampling theory (and application -- made a career of it) and various data converter architectures. I am analog guy, nothing is black and white, all shades of grey (and other colors since we are talking about noise), so I could be completely out to lunch. We have WBF experts who know way more about audio data conversion than I.

It's complicated, thus I qualified my answer a bit...

Quantization noise would be spread over a wider bandwidth and if filtered then in-band noise would be reduced. However, the analog circuits must have higher bandwidth to deal with the higher sampling rate and all that it entails (such as switching transients and such), and that higher bandwidth potentially translates to higher noise which is passed on to the rest of the system. How much additional noise depends upon how the actual DSD signal is processed (filtered) and the final circuit implementation. My assumption is that higher rates lead to wider output bandwidth (otherwise why bother?) If the final bandwidth is the same, then any intermediate analog stages must have higher bandwidth, but if it filtered ("decimated") entirely in the digital domain so analog bandwidth is the same then your argument is valid. If not, which is my understanding, and output (analog) bandwidth increases with increasing DSD rates, then more noise is output and over a broader band (note higher currents needed to achieve higher bandwidth lead to greater noise at all frequencies). How the noise is handled and its impact depends upon the bandwidth of the rest of the system.

Higher sampling rates mean circuits have to settle faster to maintain high linearity, ands that means more noise (and power). A data converter's performance (primarily distortion, which includes both harmonic and nonharmonic spurs) is often better when the sampling rate is lowered a little from its maximum value and internal (and external) nodes have more time to settle. There are also inevitably switching glitches from transistors that are worse as speeds go up, though technically you could use a completely passive filter to process a DSD signal. Of course you then have to worry about ringing the filter could introduce...

Hopefully all that babble helps a little. - Don

Good explanation Don. I will add that I too am in the dark with respect to what noise shaping is done in multiples of DSD. The extra bandwidth it provides allows that to be different but since noise shaping is not part of a format, i.e. the choice of the person creating said content and equipment used, it is hard to make blanket statements. I went searching for measured spectrums. Found this one in an CA forum:



This shows that as Don mentioned, the noise floor in the audible band is maintained and not lowered using double DSD. It does push out the ultrasonics to higher frequencies which a simpler filter could get rid.

Here is another good link and this image that demonstrates what happens beyond the limit in the above graph: https://www.promates.com/music-store/hd-audio#prettyPhoto



So perhaps any preference for double DSD stems from elimination of ultrasonic noise that could be causing intermodulation distortion in DSD and/or causing amplifier oscillation, etc.

Audiophile Bill said:

Haha - you are right. Each double dsd lp I rip at the moment is 5GB

Click to expand...

Come talk to me when you start shooting 4K video!!

This article -- http://www.whatsbestforum.com/showthread.php?2487-DACs-102-Delta-Sigma-DACs -- may offer a bit of insight about noise shaping. As Amir said, the filter implementation for DSD is up to the manufacturer. I have not really been tracking it, but they seem to be using fairly low order filters, like 4 to 8, but hopefully somebody with more knowledge can correct me on that.

DonH50 said:

This article -- http://www.whatsbestforum.com/showthread.php?2487-DACs-102-Delta-Sigma-DACs -- may offer a bit of insight about noise shaping. As Amir said, the filter implementation for DSD is up to the manufacturer. I have not really been tracking it, but they seem to be using fairly low order filters, like 4 to 8, but hopefully somebody with more knowledge can correct me on that.

Click to expand...

No expert on DSD myself. I was thinking the extra bandwidth would be used as often is the case to lower noise floor in the audio band. Looking at some measures like the one Amir posted above it looks like that is not the case. Looks like they extend the flat noise floor up to 40-50 khz and then noise rises. So your explanation would be the correct one DonH50.

Your other comments about settling time etc. are spot on of course. So DSD64 or maybe 128, what are we really getting at higher rates? I know bigger and more is better sometimes even when it isn't.

Bruce B said:

For the past couple of years I've said 128fs or DoubleDSD is all you need. I still stand by that. Even though I have QuadDSD capability in my Horus, just about all of my transfers are at Double. And to me, I still can't hear the difference between Quad and Double. If there is a difference, then the DAC must be handling the files differently!

Click to expand...

Resampling PCM to DSD with HQplayer, with all the DAC's I've tried that use the Sabre chips, I can't tell DSD 128 from 256. But when I try DAC's that have a directDSD bypass option to bypass the internal SDM/SRC/multibit conversion section of the DAC chip, I can easily tell the difference with quad DSD.

Even with the Merging Horus sourced files processed with Pyramix from 2L (originally recorded in DxD) I can hear improvement on the quad DSD versions.

They have recently added MQA to the list. So excellent reference to compare all the formats, and test the new Hqplayer MQA filter out.

http://www.2l.no/hires/

Some info on their recording/mastering techniques:


http://www.merging.com/news/use-cases/morten-linderg-2l-norway

Audiophile Bill said:

I bet you can't tell under blind conditions Blizz. If you can tell, the difference must be minuscule. What do you notice specifically? I listened to the quad and double dsd files that Bruce sent through my GG and could not tell under blind conditions. I was using pipeline SDM mode.

Click to expand...

Did you use the directDSD mode? Not pipeline SDM, directSDM. Things take a noticeable step up. Enough to easily tell. For example I had it set to 128 upsampling one time and forgot about it. I was listening and it just didn't sound right. I was wondering what was wrong. Then I noticed that it was only set to 128 rather than 256. So I changed it to 256, and everything was good again.

I have sold my superstream image to a couple GG owners. What 1 guy told me is using HQplayer to upsample PCM to DSD is almost too smooth. This is because the HQplayer algorithms are already smooth sounding, and when combined with the tube output stage, and Dueland bypass caps the GG uses to add warmth and smoothness, it's too much.

The tube output stage isn't required to polish turds when Hqplayer is used is what I've gathered from it. However with native DSD recordings, if directSDM is enabled, you aren't using any of the Hqplayer SRC/SDM. You are using whatever the mastering house used with their DAW. So the end results will be based on that.

Audiophile Bill said:

Ah ok - I did not know there was 2 options. I will check tomorrow which one but I clicked the one that blocks you from changing the sample rate I think?

Click to expand...

PipelineSDM is just to more efficiently utilize multiple cores in the processor when resampling PCM to DSD. The directSDM is the important one. It's under the DSDiff/DFF settings. If that box isn't checked, all the audio is being upsampled and processed through the same filter algorithms, as well as going through the multibit conversion process. So yes very hard to tell apart in this case.