Jump to content
The Computer Audiophile

DSD Has Better Dynamic Rage = Misleading

Rate this topic

Recommended Posts

8 hours ago, tailspn said:

As formats, DSD has +6dB greater dynamic range than PCM.

 

DSD and PCM are digitally storable formats to record and transmit analog signals, mostly audio for our interest. 0dB is a definition in recording to signify a maximum level not to be exceeded. Below 0dB, both PCM and DSD are capable of the same signal level range, better known as dynamic range. Both can express an infinitely small signal, far below the practical minimal signal transferable by practical analog electronic circuits.

 

In the case of PCM, 0dB represents the maximum binary range that the the format can represent; there's no bits left. It's a 2's compliment binary word for each sample that's either all 1's, or all 0's at 0dB (full scale). It's 2's compliment to support both positive and negative values.

 

In DSD, there are no values represented. It's the density modulation of a bit clock, who's percent of modulation is proportional to the signal level, not a numerical value of a signal level (at an instant in time). 

 

All of that is background to the fact that from the lowest signal level deliverable to a  A/D converter to the maximum that produces 0dB, the same dynamic range is expressible in both a PCM or DSD format up to 0dB. The DSD format however has an additional +6dB of headroom above 0dB that is not achievable with PCM. That's because DSD 0dB is specified as 50% modulation, allowing an additional 6dB of signal level to be represented. 

DSD requires the extra headroom to prevent the sigma-delta modulators destabilising, so it's not actually available to encode the signal. Moreover, to compare dynamic range of formats, you need to look at both the high and low end. With PCM, the range between the highest and lowest level representable is trivial to calculate, and it increases with 6 dB for every added bit. Since the quantization noise in PCM is evenly distributed over all frequencies, doubling the sample rate also lowers the noise floor by 6 dB. With DSD it's not so simple. The noise level in the audio band depends primarily not on the digital format but on the quality of the noise shaping filter, and no matter how good it is, there will always be some residual noise as no filter has infinite stop band rejection.

 

In practice, both PCM and DSD are capable of representing the full signal range between the analogue noise floor and the maximum analogue level. The 6 dB headroom of DSD is an implementation detail and should be ignored for the purposes of this discussion.

Share this post


Link to post
Share on other sites
On 02.05.2017 at 6:06 AM, tailspn said:

As formats, DSD has +6dB greater dynamic range than PCM.

 

I reduced it to 0.4 dB. Though in the settings possibly add headroom value:

 

sigma-delta-modulator-settings.png

Share this post


Link to post
Share on other sites
18 minutes ago, audiventory said:

 

I reduced it to 0.4 dB. Though in the settings possibly add headroom value:

 

sigma-delta-modulator-settings.png

 

The reason?

 

In the posible settings, are those minus, like in -6 dB ?

 

Thanks,

 

Roch

Share this post


Link to post
Share on other sites
21 hours ago, elcorso said:

 

The reason?

 

In the posible settings, are those minus, like in -6 dB ?

 

Thanks,

 

Roch

 

Hi Roch,

 

-0.4 dB used for maximal amplitude using. It was accepted by number of test files (test signals and different musical ones). This headroom provide stability for known me audio stuff (that was converted since 3-rd generation of filters was released).

 

If used wide filter band mode (Settings > General > Filter mode "non-optimized wide resampling filter"), there is higher probability of break stability due overload. Especially for DSD 64.

Though I still don't got reports about break stability issue since wide mode was released.

 

If output file contains silence or constant oscillation, need to increase attenuation.

I suppose, 1 dB must cover all cases. For more fine tuning 0.1 dB may be used Main window > Levels > Output file volume slider (if it is available in used edition/configuration).

 

If you want to convert DSF files with "standard" headroom, you can set attenuator to -6 dB.

 

Best regards,

Yuri

 

P.S. Besides amplitude limitation, other efforts for modulator-stability improving are used in AuI ConverteR.

 

aui-converter-settings-wide-filter-mode.png

aui-manual-gain-control.png

Edited by audiventory

Share this post


Link to post
Share on other sites
7 hours ago, audiventory said:

 

Hi Roch,

 

-0.4 dB used for maximal amplitude using. It was accepted by number of test files (test signals and different musical ones). This headroom provide stability for known me audio stuff (that was converted since 3-rd generation of filters was released).

 

If used wide filter band mode (Settings > General > Filter mode "non-optimized wide resampling filter"), there is higher probability of break stability due overload. Especially for DSD 64.

Though I still don't got reports about break stability issue since wide mode was released.

 

If output file contains silence or constant oscillation, need to increase attenuation.

I suppose, 1 dB must cover all cases. For more fine tuning 0.1 dB may be used Main window > Levels > Output file volume slider (if it is available in used edition/configuration).

 

If you want to convert DSF files with "standard" headroom, you can set attenuator to -6 dB.

 

Best regards,

Yuri

 

P.S. Besides amplitude limitation, other efforts for modulator-stability improving are used in AuI ConverteR.

 

aui-converter-settings-wide-filter-mode.png

aui-manual-gain-control.png

 

 

Many thanks Yuri for the detailed explanation !

 

Best,

 

Roch

Share this post


Link to post
Share on other sites
On 4/25/2017 at 4:38 PM, The Computer Audiophile said:

I didn't mean to come across as not nice. Perhaps it was read that way. I just wanted to state the facts :~)

 

You were absolutely right, and none too hurtful!

 

Mastering(and to a big extent mixing) plays a far more significant role in how the finished product sounds than does format.

 

Comparing 8-Track Cartridge to DSD?  Then that poster of "DSD wins" might have a case! lol  But PCM(red book) vs DSD?  Splitting electrons.  ?

Share this post


Link to post
Share on other sites
On 25-4-2017 at 9:41 PM, The Computer Audiophile said:

Glad I took a screenshot. My comment has been removed.

 

 

Screen Shot 2017-04-25 at 2.40.58 PM.png

why would they do thatO.o

Share this post


Link to post
Share on other sites
On 5/2/2017 at 6:25 AM, mansr said:

In practice, both PCM and DSD are capable of representing the full signal range between the analogue noise floor and the maximum analogue level

This is the key point for me.  At best, in room during playback we might have 100 dB of DR.  And that is a very, very quiet room with a superb system.  So the original idea that 16 bits is enough (~96 dB) is mostly correct: of course it is nice to have a bit of headroom, and for recording, more bits are necessary for mixing without loss, making recording at 24 bits totally sensible.

Everyone who thinks about this should play a -90 dB signal in their system (with their volume control set to their normal "loud" critical listening level) and get an idea of how low in level -90 dB is.  Stereophile Test CDs have a -90 dB test tone which makes this easy.

 

I love DSD, and convert all PCM to DSD 128 or 256 during playback with Audirvana Plus (I feel my ESS based DACs sound better with DSD input).  But my love for DSD has nothing to do with DR... 

Share this post


Link to post
Share on other sites

Would love to hear more about the subjective or experienced side of DSD to compliment the technical comparisons in this thread.  In order to get an idea of what others are finding. 

 

I'm new to high res  music, having enjoyed Qobuz for about two months on their high rez subscription.  I have a Chord Qutest DAC that is DSD capable, so wanted to try some DSD content top evaluate if there are sonic differences that I can perceive.  I have downloaded a trial of Audirvana.  I then found via search engine that Blue Coast Music offers a free download of a well recorded track in multiple formats for evaluation, and I hope I am not breaking forum rules by including that link ( https://bluecoastmusic.com/free-downloads#.XC31CPx7mmk ).
 

I did not do blind testing, but I found it hard to tell much difference between the high rez FLAC file and the WAV file, but think that the WAV may have been a little more airy, that said I'd happily accept that this could be placebo as I know that WAV isn't compressed.  I would say categorically that that the two DSD formats did *NOT* sound better at all.

 

After playing with all the PCM upsampling in Audivarna I found in my system and with my ears that all degraded the sound quality with the exception of the setting 2X upsampling Izo filter (note not power of 2, but flat 2X).  Upsampling PCM to DSD that my DAC is most capable of receiving made the sound quality worse IMHO.  For straight DSD file playback I tried both the 1.0 and 1.1 DSD transfer setting, but not the ones converting DSD to PCM.

 

To expand on what I mean by stating "did *NOT* sound better" with DSD:

1. There was less width

2. There was less volume difference between the peaks and the lows

3. Oddly enough, my volume control on the pre-amp had less apparent effect (I did check that my Mac volume was at Max and that Audirvana was at 0db) 

 

My full main system details are in my profile. While far from the last word in audiophile I think the rig is OK, and IMHO musically revealing.

 

If any have time to conduct a similar subjective test then I'd be grateful to read about your findings.  Likewise any thoughts very welcome on whether this might be the limitations of the Audirvana player.  Further suggestions of other music tracks in multiple formats that could be used to test in a similar fashion, preferably samplers / free.

Share this post


Link to post
Share on other sites

@DavidFaik, The Qutest converts incoming DSD to PCM (although at a still high rate, I believe 705.6) so it is probably not the best converter to use for DSD vs. PCM comparisons.  Just something to consider...  DSD may sound better with a DAC which does not decimate (convert to lower sample rate) the incoming DSD.

Share this post


Link to post
Share on other sites

@barrows ... that's a bit of a shocker!  When I researched the Qutest I relied on Chord's features statement: "DSD support: Native playback supported. DSD64 (Single) to DSD512 (Octa-DSD)".  I cannot see how one can interpret the English word "native" to mean that the DAC down-samples DSD.  While I am HIGHLY impressed with the sonic qualities of the DAC and most likely would have bought it any way, I feel that if this is the case it is very underhand.  Would you mind sharing with me where you read this?

 

On your system would you how would you describe the DSD versus FLAC on the same track then?

Share this post


Link to post
Share on other sites

I never use any compressed formats here, all my files are either .dsf or .aiff.  I have two (DIY) DACs, one is based on the ESS 9038, and the other on a balanced version of Jussi's discrete DSC-1 approach.  I now oversample everything to DSD 256 in my system, but both my DACs have been specifically tailored to work best with high rate DSD, so comparisons are not really fair anyway.

I prefer this approach, but it is for my system.

 

I would not be dismayed by the Qutest's approach though, it is a very, very good DAC.  With the Qutest I would consider comparing sending native rates vs. oversampling to high rate PCM.  You can oversample in software to 705.6, for example, and compare that to sending native PCM rates and see which you prefer.  Every system is different, so do not expect your preferences to mate up with others.  I would leave DSD alone and just send it native to the Qutest and let the Qutest do what it does.

Share this post


Link to post
Share on other sites

@barrows Thanks, I'm very happy with the Qutest but that doesn't detract from my dissatisfaction at Chord playing loose with the truth and claiming native playback.  If the DAC is down sampling (more difficult than upsampling I'm told) then it is no wonder that PCM sounds far better than feeding DSD.

 

On Audirvana the "full throttle" upsampling sounds worse than the "only do 2X upsampling".  2X does sound darned good, mind. I will try some other players to see if they have better upsampling capabilities, but if not will probably buy the Audirvana license.  I find the interface a little basic, but could live with it.  

 

Share this post


Link to post
Share on other sites
1 hour ago, DavidFaik said:

@barrows Thanks, I'm very happy with the Qutest but that doesn't detract from my dissatisfaction at Chord playing loose with the truth and claiming native playback.

I do not really think this is the case.  When Chord claims "native" DSD playback I am sure they are referring to the fact that the Qutest can accept a native DSD stream on its input (as opposed to only accepting DoP).  What would you consider "native DSD"?  How about the ESS chip, which does not decimate incoming DSD, but does re-modulate it into a very high rate multi bit format (6 or 7 bits), similar to what dCS does as well.

Not many DACs convert DSD without some changes to it.  In fact, not many DACs are true NOS with PCM either...

Share this post


Link to post
Share on other sites

@barrows  thanks for the suggestions to try using the forced upsampling power of 2.   After a great deal of time I established that I had to make some quite extreme changes to the default "advanced settings" for Izoptope in Audirvana but THEN I was able to hit a complete sweet spot for the Chord Qutest. Now the upsampling power of 2 sounds better than the 2X, very wide sound stage remains (breadth of left to right beyond the speakers in case I got the audio vocab wrong), the "redbook" or 16 bit CD level files sound very much closer to high res 24 bit files and MP3s are much easier on the ear.  Keeps the Chord signature but just "better".  Especially with music where there is a lot going on at once.  I've yet to research more on what the Nyquest Cuttoff Freq means / does and so have left that as the default (from reading I think I could perhaps damage speakers if I screw that around?)  As I could find no one posting ideal Chord Qutest Audirvana settings online, I will add mine below in case it saves others the marathon fiddling, tweaking and listening that I did until the early hours (!)

 

WRT to Chord comments on "native DSD playback", that remains a complete fib in my eyes.  In a job interview if I state that I can "communicate with a native Dutch speaker" it means something totally different to saying "I can speak native Dutch".  DSD is marketed as a "higher / better" music format than PCM.  If the Qutest takes DSD and downsamples it to PCM that's not playing it "natively".   Had I not met a smart person like yourself on this forum, I could well have wasted days trying to get DSD playback to sound better than the default settings.  So THANK YOU VERY MUCH!!!, I am really most grateful.

 

Caveat to the above comment: I remain completely blown away by this Qutest DAC, one of the best audio purchases that I have ever made (along with the discontinued Chord Prime pre-amp that I scored 2nd hand).  Serving Qobuz high rez / 24 bit files to it was a revelation.  Very modest priced tweaks have then upped the DAC to levels that I think years ago may have been in the realms of exotica hifi costing the equivalent of a car: namely a €50 usb cable (viablue), some isolation feet (€20), a second hand sbooster linear PSU that was being used for Squeezebox (€70) and now a music player that will upscale 2 million taps (2X taps of a hugo m scaler) for €74.  Most things in audio are incremental and very few things in audio are really game changing.  This IMHO is truly the latter :-)))))

 

Best Chord Qutest Audirvana Audio Filter Settings (that I have found)

Converter: Izotope 64 bit

(note- didn't try the rest very long after reading the fora comments)

Forced upsampling: power of 2 only (second on drop down.  after advice I only played with this and the X2.  maybe the "max" option can also be configured well.  dsd options messed up SQ)

Advanced Parameters

Steepness 0db (slider totally left)

Filter Max Length 2 million (slider totally right.  my mac mini 8gb ram Intel i5 2012 does not break a sweat with this)

Cutoff Freq 1.00 x nyquest (I left this as default as I didn't understand it, so maybe changes improve further)

Anti-Aliasing 50 (slider totally right)

Pre-ringing 0 (slider totally left / min phase side)

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×