Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Hilton]

Hi All,

Since I now have a new DAC capable of up to 32bit 352.8 and DSD64/128 and DXD I've been trying to understand a bit more about DSD and the whole DSD vs PCM and think I have learnt something very interesting.  

I'd like to discuss my thoughts with people smarter than me with more experience with DSD to understand technically what "should sound best", using MC for playback to a PCM, DSD DoP and DXD capable DAC.

From what I understand, the filters are the secret to the quality of the sound, its got very little to do with the format.
When I say filters, that means many things.....

The filters and algorithms used in ADCs and DACs, the filters which are used in MC21 and then the filtering that may or may not be present in the amplifier.
And of course the filters/algorithms being used in DSD and PCM recording and playback.
I believe (as a newb in this topic) the biggest audible difference these chains of filters make is usually heard in the form of phase shift and coherence.

So with that said, DSD in theory should introduce less phase shift and timing errors and be more lifelike because of less ADC DAC conversions because nearly all PCM ADCs and DACs incorporate a SDM (sigma-delta modulation) step (same technology as DSD).

There are people out there with far more knowledge and experience of these things that say to play native DSD or DXD to get the minimum number of conversions and filters which is ultimately what changes the phase and timing of the music.  (going through too many conversions)

These phase and timing errors are more prominent when you know what to listen for.  If I understand correctly, the phase and timing is what gives sound it's directional quality and size or space quality.  Ever noticed how even with one ear on the pillow, you can usually pin point that pesky mosquito? That because our ears are highly designed to hear phase shift and timing variations, and its this change in the sound that we are hearing with different formats, frequencies and filters.

I'm probably way off base with my infantile knowledge of the subject, but thought it would be an interesting discussion. Especially since the recent addition of the TPDF dither and its apparent very subtle effect on sound quality.

So what say you boffins?  
Should I now upsample everything to DSDx2 (128) since my shiny new active headphones support it?
People have been saying for a while that even redbook CDs when upsampled and re-filtered and dithered sound better at 24/176.4... ?!!!? (less oversampling done at higher sample rates which leads to less rounding errors in the oversampling steps (or something like that?)

[Mark_Chat]

I would ignore peoples ears, including your own, with regards to upsampling sounding better.

It may sound "nicer" or more appealing, but it is most absolutely and definitely a less accurate representation of what was actually recorded.

Upsampling will not recreate lost detail nor provide new detail that was never there in the first place.
It will provide extrapolation and interpolation fill-in data and dither to smooth it all over.

[Mark_Chat]

To answer your main point, if you have a quality DAC plus connectors that will output and input these formats then listen to it in it's native format.

The DAC should have optimised filters for each, so trust them.

It doesn't seem sensible to convert to a different format in the player and then again in the DAC.

Do one D-A conversion and leave it at that for the PCM sources, including DXD

Bitstream DSD in whatever data rate the file is in and leave it in DSD rather than DOP to minimise conversions, which are each approximations. (as long as your DAC is playing DSD "direct" without an intermediary conversion to PCM (i.e. just using sigma-delta conversion plus noise shaping plus a 50kHz low-pass filter))

[ferday]

there is evidence that upsampling to DSD from PCM changes the signal, and some may find the effect pleasing (even order harmonics)

http://archimago.blogspot.ca/2013/09/measurements-pcm-to-dsd-upsampling.html

[dtc]

The filters are clearly a significant part of the process and one reason DACs sound different. This is particularly true for DSD. That is why most comparisons of PCM and DSD are flawed. What you are really listening to is the particular DAC's implementation of PCM or DSD conversion to analog, including the filtering, rather than any inherit difference in the formats.  Even DAC's that implement both, use different filters for the two different formats. And some DSD DACs convert DSD to PCM internally, without telling you. So, the real question comes down to how your particular DAC implements PCM and DSD and how that sounds to you. I know that sounds trite, but it really is true in this case. Two people can listen to the exact same comparison and hear different things. That is the wonder, and the frustration, of human hearing.

One reason to go to higher resolutions is in order to minimize the effects on the filters. At 16/44 or 2.8 you need a pretty steep filter to keep noise out of the audio range. The higher the resolution, the less steep the filter needs to be and the less filter anomalies enter into the normal audio range. With DSD 5.6MHz, DACs can move down from 5th or 7th order for 2.8 MHz to 5th or even 3rd order filters, which can mean significant difference in filter anomalies. Again, which filter to use is up to the DAC designer.

People often scoff at the idea of these higher resolutions because people cannot hear above 20 KHz and most cannot hear that. However,  the higher resolutions do move noise higher up into the spectrum which can clean up the normal audio band. In addition, there is certainly evidence that the brain, which is ultimately the processor for sound, does recognize subtle differences due to frequencies higher than 20 KHz, even though we cannot hear them as a sustained tone.

I had the chance to talk to a well know speaker designer recently and I asked him what he thought about having speakers that went above 20 KHz. He talked in depth about one of his most recent designs. He originally put a 20 KHz tweeter in it and it just did not sound right to him. So he put in a 40 KHz tweeter (they design their own tweeters) and it sounded much better. He ascribed the difference to the harmonics that were being generated by the two designs, similar to the observations in the article that ferday linked to.

Some DAC have the option to choose between multiple filters. I sometimes think it would be interesting to have the DAC do no filtering and have a "filter box" in line between the DAC and the pre-amp that would allow the user to dial up different filters and listen to the difference. Even order, odd order, steep, gradual, even different types of filter algorithms. I think it would show that the filters can have a significant effect on the final sound. It would be particular interesting for DSD. No practical, but something to think about.

In the end, it depends on the DAC implementation and how it sounds to you. Trite, but as true for DACs as for any other component.

[blgentry]

To answer your main point, if you have a quality DAC plus connectors that will output and input these formats then listen to it in it's native format.

The DAC should have optimised filters for each, so trust them.

*Should*.  But is the conversion and filter in the DAC or in software better?  The real question (to me) comes down to this:

For your particular DAC, which input sample rate produces the most transparent (and/or) pleasing sonic output?  Does your DAC upsample everything to 192 kHz?  If so, does it's upsampling process from 44.1 to 192 sound best?  Or does upsamping from 44.1 to 192 *externally* in software, sound better?

As I understand it a HUGE part of the buzz over Schiit's Yggdrasil, Gungnir (MB), and new Bifrost (MB) is due to the super technically advanced upsampling algorithm.  That and the fact that it's not Sigma Delta, but rater "multi-bit".

It doesn't seem sensible to convert to a different format in the player and then again in the DAC.

Agreed.  One conversion if at all possible.

Bitstream DSD in whatever data rate the file is in and leave it in DSD rather than DOP to minimise conversions, which are each approximations.

As I understand it, DoP is simply a container format.  It packs the DSD bitstream into PCM frames for transport.  At the other end, the PCM frames are unpacked and reassembled into DSD which is identical to the DSD that was fed into the DoP process.  It is "bit perfect" as I understand it.  No conversion.  I'm sure someone will correct me if I'm wrong. 

This is all in the realm of "you have to try all the combinations with your hardware".  Just like other posters have said.  Because the DSD implementation in your (for example) Sony headphones might be great.  Or it might be crap and PCM fed to it might sound better.

I think you would be hard pressed to find a DAC that does both DSD and PCM that sounded best with everything converted to DSD.  It's possible, but it seems unlikely.

Brian.

[kstuart]

I have the aforementioned Bifrost Multibit DAC and it does a really excellent job of imaging, instrumental timbre, detail and room ambience.

  If I understand correctly, the phase and timing is what gives sound it's directional quality and size or space quality.

This is exactly what Schiit is trying to preserve with multibit and their closed form digital filter.

===

Also, note that there are no DSD based workstation software for processing and mastering, so almost all SACDs were made by converting DSD to PCM, doing processing, and converting back to DSD.  So, in most DSD and SACD music, you have the all the weak points of DSD and the weak points of PCM.

[Hilton]

Thanks guys this is very interesting conversation!

I wonder why Sony didn't advertise that the MDR-1ADAC does DXD "Extreme PCM" LOL...

Here's a very interesting article on DSD mixing and DXD that talks about the more relaxed filtering that you can use at the extreme sample rates and the benefits to minimising digital pre and post ringing which I've seen demonstrated in some traces of digital signals.

http://www.realhd-audio.com/?p=2974

I find it all very interesting and fun to learn something new so thanks and please keep up the constructive conversation.

Im interested in the multi-bit DACs, I researched a little on them yesterday and understand that it reduces the number of anti-aliasing filters and oversampling required which in theory should sound better.

PS. Interestingly I just discovered the DAC also does 32bit 384Khz too as well as 32bit 352.8 (DXD)
Must be some interesting DAC Sony have hidden away in there and they refuse to tell anyone what it is.

[Hilton]

Interesting. MC on the Pi2 will output 32bit 384k to the headphones full bit depth. Does anyone know if ALSA would be downsampling? It appears not to be.  This is pretty darn cool!

Pi2-DXD1 by Hilton, on Flickr

[dtc]

Im interested in the multi-bit DACs, I researched a little on them yesterday and understand that it reduces the number of anti-aliasing filters and oversampling required which in theory should sound better.

You might add these companies to your reading :

Look at the Lampizator DSD DACs. Lukasz Fikus hand builds them in Poland. He has somewhat of a cult following. He has handcrafted his DSD filters and gets rave reviews from his followers. These are tube based, hand wired units. He does minimal processing in his DSD implementation.

At the upper price range, you should look at dCS. They use an unique ring architecture and have upscaled to DSD for a long time. Very pricey, but the technology is very interesting.

EMM Labs also has interesting products, based on Ed Meitner's designs. He is one of the fathers of DSD.

All of these are expensive DACs, but it is interesting to look at what some of the leaders are doing without the use of off the shelf chips.

[blgentry]

Im interested in the multi-bit DACs, I researched a little on them yesterday and understand that it reduces the number of anti-aliasing filters and oversampling required which in theory should sound better.

As I understand it, the Schiit MB DACs are unique in that their filter does things other filters can't do.  Specifically, it retains all of the original samples delivered to it.  Other filters throw away some (or most) of the original samples and create new ones.  The Schiit "mega combo burrito filter" keeps every single sample that it receives *and* adds more (for upsampling).  I'll admit to being unclear on the exact details, despite having read quite a bit of discussion about it.

Mike Moffat (the designer) says that this filter is more time correct than almost any other filter and thus sounds as people describe it:  More spatial queues, more low level details, more ability to pin point the locations of instruments. 

I have NOT HEARD ONE.  But I'd like to. 

Brian.

[Hendrik]

One reason to go to higher resolutions is in order to minimize the effects on the filters. At 16/44 or 2.8 you need a pretty steep filter to keep noise out of the audio range. The higher the resolution, the less steep the filter needs to be and the less filter anomalies enter into the normal audio range. With DSD 5.6MHz, DACs can move down from 5th or 7th order for 2.8 MHz to 5th or even 3rd order filters, which can mean significant difference in filter anomalies. Again, which filter to use is up to the DAC designer.

People often scoff at the idea of these higher resolutions because people cannot hear above 20 KHz and most cannot hear that. However,  the higher resolutions do move noise higher up into the spectrum which can clean up the normal audio band. In addition, there is certainly evidence that the brain, which is ultimately the processor for sound, does recognize subtle differences due to frequencies higher than 20 KHz, even though we cannot hear them as a sustained tone.

Its important to note here that any good modern DAC would oversample internally before filtering and D->A conversion just to avoid these limitations and not depend on the input signal being high enough sample rate.

[kstuart]

Its important to note here that any good modern DAC would oversample internally before filtering and D->A conversion just to avoid these limitations and not depend on the input signal being high enough sample rate.

There is also some semantics going on.

Schiit calls their custom oversampling a "filter", whereas most consumer level articles reserve the use of the word "filter" for the analog low-pass filter.

BTW, on 16/44.1 material, some JRiver users have compared the Schiit DAC oversampling to MC's real-time upsampling to 176.4 or 192 and claim that the sense of space is much better with the internal oversampling.

[dtc]

Its important to note here that any good modern DAC would oversample internally before filtering and D->A conversion just to avoid these limitations and not depend on the input signal being high enough sample rate.

I agree that most modern DACs upsample internally, although there certainly are people who prefer NOS DACs. I should have pointed out that out. Using hi rez files means that oversampling in the DAC does not need to done or has to produce fewer new data points. That is why one might want to use  24/96 or24/192  or 2x/4x DSD files. I am not sure how many DSD DACs oversample 2.8MHz to higher DSD rates. Certainly most original SACD players did not.

[kstuart]

I agree that most modern DACs upsample internally, although there certainly are people who prefer NOS DACs. I should have pointed out that out. Using hi rez files means that oversampling in the DAC does not need to done or has to produce fewer new data points. That is why one might want to use  24/96 or24/192  or 2x/4x DSD files. I am not sure how many DSD DACs oversample 2.8MHz to higher DSD rates. Certainly most original SACD players did not.

(DSD playback is a level more complex than that, and it is too detailed to deal with here.)

The overall problem is that you want to listen to a particular recording.  Sometimes it is only available in 16/44.1, sometimes there is a good 24/192, and many different variations.  Usually, there is more difference between different releases, than between the formats.  (We now know in hindsight, a majority of the time, the word "remastered" means that the sound quality is worse, it is just a marketing term designed to lighten our wallets.)

So, you want your DAC/Music_software combination to be able to play all variations, and its only an academic issue as to which format is better.  The record labels choose which format is used by the great mastering engineer, not us.

[theoctavist]

then there is the whole other issue of DSD being a waste of time. It is only useful for mastering, say direct from Analogue to SACD.   Anyother time (like when using a DSD recorder for example, or mixing in a daw) all benefits of the technology are lost, due to the HF in the source signal hitting a second delta sigma modulator.


same with 24/192.

Ideal is around 24bits/60kHz . anything above that is not only a waste of time, but is potentially destructive.

[kstuart]

You say "same with 24/192" but provide no reason.

There is no similar problem with 24/192.

In fact, recording engineer Paul Stubblebine commented in another Forum (the "Barry" he refers to is mastering engineer Barry Diament, and Keith Johnson is also a recording engineer):

"Barry has mentioned that he hears a qualitative difference between the 2X rates (88 and 96) and the 4X rates (176 and 192) and I hear it pretty much the way he describes it. As we go up from 16 to 24 bits, and as we go up from the 1X rates to the 2X rates, I hear a number of specific improvements. When we get to the 4X rates done well (and here I agree again with Barry--easier said than done) it's more of a feeling that we have turned a corner and we are almost dealing with a musical experience rather than a facsimile of a musical experience. And I'll confirm that Keith Johnson has said something similar in several conversations."

These are guys who are listening critically all day every day.  They are paid specifically because they can hear and notice small differences.

[dtc]

You say "same with 24/192" but provide no reason.

There is no similar problem with 24/192.

In fact, recording engineer Paul Stubblebine commented in another Forum (the "Barry" he refers to is mastering engineer Barry Diament, and Keith Johnson is also a recording engineer):

"Barry has mentioned that he hears a qualitative difference between the 2X rates (88 and 96) and the 4X rates (176 and 192) and I hear it pretty much the way he describes it. As we go up from 16 to 24 bits, and as we go up from the 1X rates to the 2X rates, I hear a number of specific improvements. When we get to the 4X rates done well (and here I agree again with Barry--easier said than done) it's more of a feeling that we have turned a corner and we are almost dealing with a musical experience rather than a facsimile of a musical experience. And I'll confirm that Keith Johnson has said something similar in several conversations."

These are guys who are listening critically all day every day.  They are paid specifically because they can hear and notice small differences.

There are many people who believe that the same is true for higher order DSD and for DSD versus PCM.  Not everyone will hear the differences. You need the right mastering and production, the right equipment and listening conditions, and the right ears. Obviously, not everyone hears the differences. But there is a segment of the professional and consumer population who believe they hear a real difference.  It is hard to simply dismiss them.

[theoctavist]

You say "same with 24/192" but provide no reason.

There is no similar problem with 24/192.

In fact, recording engineer Paul Stubblebine commented in another Forum (the "Barry" he refers to is mastering engineer Barry Diament, and Keith Johnson is also a recording engineer):

"Barry has mentioned that he hears a qualitative difference between the 2X rates (88 and 96) and the 4X rates (176 and 192) and I hear it pretty much the way he describes it. As we go up from 16 to 24 bits, and as we go up from the 1X rates to the 2X rates, I hear a number of specific improvements. When we get to the 4X rates done well (and here I agree again with Barry--easier said than done) it's more of a feeling that we have turned a corner and we are almost dealing with a musical experience rather than a facsimile of a musical experience. And I'll confirm that Keith Johnson has said something similar in several conversations."

These are guys who are listening critically all day every day.  They are paid specifically because they can hear and notice small differences.

the nyquist theorem lays it all out ... but here.

https://xiph.org/~xiphmont/demo/neil-young.html

http://www.lavryengineering.com/pdfs/lavry-white-paper-the_optimal_sample_rate_for_quality_audio.pdf

http://drewdaniels.com/audible.pdf

[theoctavist]

the nyquist theorem lays it all out ... but here.

https://xiph.org/~xiphmont/demo/neil-young.html

http://www.lavryengineering.com/pdfs/lavry-white-paper-the_optimal_sample_rate_for_quality_audio.pdf

http://drewdaniels.com/audible.pdf

the human ear-brain system is horrifically prone to bias.  as such, "professionals" or not,  one simply cannot usurp the laws of physics/sound propagation. Although some true believers think it so.

"f we built up our analog gear and our loudspeakers so that the Least Significant Bit (the 24th one) equaled the analog noise floor (-100 dBV) and the threshold of hearing (0 dB SPL), why, then we’d hear this hi-rez benefit for sure. Personally, I think it would be fabulous!

Our normal calibrated listening level would then be, of course, something like –60 dBFS for 84 dB SPL, and our loudspeakers would require water cooling and our power amps would have to be capable of generating perhaps a mega Watt of power per channel – they’d probably need water cooling as well, maybe even liquid hydrogen."

[theoctavist]

http://www.moultonlabs.com/more/24_bits_can_you_hear/

[Starchild1]

I have the aforementioned Bifrost Multibit DAC and it does a really excellent job of imaging, instrumental timbre, detail and room ambience.

This is exactly what Schiit is trying to preserve with multibit and their closed form digital filter.

===

Also, note that there are no DSD based workstation software for processing and mastering, so almost all SACDs were made by converting DSD to PCM, doing processing, and converting back to DSD.  So, in most DSD and SACD music, you have the all the weak points of DSD and the weak points of PCM.

+1 I have a Gungnir multibit dac and can attest to yourt observations.  I'm rather suprised to see such an audiophile oriented thread appear in this forum.  There's a lot of discussion by professionals in the field about the pluses and minuses of DSD, Delta-Sigma vs R2R Ladder Dacs and the 24 bit recordings (Meridian's MQA encode/decode process is based on the fact that there's very little data beyon 18-20 bits).  Very fascinating reading.  Schiit Audio hosts a forum on Headfi that's moderated by Jason Stoddard (one of its founders along with Mike Moffat).  You should check it out.

[theoctavist]

http://old.hfm-detmold.de/eti/projekte/diplomarbeiten/dsdvspcm/aes_paper_6086.pdf

[kstuart]

The xiph paper has been rebutted dozens of times in dozens of forums.

He takes oversimplifications for lay people (such as "bit depth is signal-to-noise") and then acts as if they are actual technical descriptions of digital audio.

He also uses the same sort of irrational assertions that you and most other objectivists use:

" the human ear-brain system is horrifically prone to bias "

Wow, how scientific.  Is "horrifically"  greater than 10 ?   Is it as much as 100 ?

The whole "bias" claim is just unscientific voodoo.  Here is the objectivist definition of bias:

Bias -

A mysterious magical power in the brain that can improve the sound quality of any musical experience.

[JimH]

That was fun.