Does DSD sound better than PCM?

[dc655321]

Can't think of a shorter "sound" at the moment, nor how to make one.

A wee bit of Python?
10 seconds of impulses at 100 Hz:

import numpy as np
import soundfile as sf

data = np.zeros(44100)
data[::441] = np.sqrt(2)/2

track = np.tile(data, 10)
stereo = np.column_stack((track, track))
sf.write('100hz_3dB_impulse.flac', stereo, 44100, subtype='PCM_16')

buzzzzzzz.
It will never make the "recommended listening" list.[/code]

 

[Graph Feppar]

Now I'd like to know, what about DSD seems like a good idea?

DSD is good idea,just not for consumer.Its smart way to turn the lack of knowledge among audiophile community into profit.Its good for marketing,just imagine how many times someone bought something becose it does DSD,not jumping on DSD scam bandwagon decrease profit except in DACs aimed at professionals and studio use becose most pro people know its BS.

 

[RayDunzl]

A wee bit of Python?

I'm thinking sound in the air, not in a file...

 

[dc655321]

I'm thinking sound in the air, not in a file...

Ah... Carry on...

 

[graz_lag]

Being honest and frank you need to learn a little about how digital audio works. Higher sample rates extend frequency response. They don't increase dynamic range.

There is also a very high probability CD mastering is not the same as on other releases. Upsampling can't add details to a file.

I have not used the 2 terms in an interchangeable way, I know they are not exactly the same thing ... Come on ...
My knowledge of the English language is not as good as your, and I am very sorry for that, I should have probably written "signal to noise ratio" or make it another way, "dynamic range" ... Do your like more this way ?

Logically, oversampling algorithms can't add information if that isn't there in the original file, that's also clear to me ...
However, IMHO, it is possible that a software-oversampled file (for example at the ripping stage from a disc) may sound better than playing back the original 16/44.1 source, if the DAC converts all inputs to 192kHz and the software over-sampler introduces less distortion than the one implemented in the DAC's hardware ...

 

[Thomas savage]

I found that oversampling can thin out the sound vs the native 16/44.1 , certainly could with the esoteric CDP that I used to use.. maybe that’s upsampling ,, I don’t know either way it sounded thin lol

 

[Graph Feppar]

Higher sample rates extend frequency response. They don't increase dynamic range.

Actually,higher sample rates do increase dynamic range.Doubling samplerate increase dynamic range by 3db,so it works,its just ridiculously inefficient way to do it,increasing bit depth is far better.

You can make 1bit sound format have same dynamic range as 16bit without any dither or noiseshaping,but the samplerate must be 47352014438400 samples per second.

 

[graz_lag]

Yes,I can recommend that xiph video too.I busts the time resolution = sample lenght myth.

I give you helpful tip on how to judge if some filter is "smearing" your music,if it audibly degrades the transients.Filters smear sounds in certain frequency band depending on how is their frequency and phase response at that point.

If the filter phase response is flat and frequency response is flat,then absolutely no smearing or ringing is happening at that frequency.Thats why those brickwall digital linear phase FIR sync filters and analog brickwall bessel filters are the best choice and any of the different "flavor" filters can only degrade the sound since they arent as flat in passband. ...

Hey, you're still omitting the spaces after the commas and points ...

 

[folzag]

DSD is good idea,just not for consumer...

Right. DSD wasn't created for us consumers, it was created by labels, for labels. Executives saw it "solving" two problems:
* New format to incite fresh demand to resell back library. Again.
* Bloat file sizes as impediment to online piracy. (Not a factor today, but 20 years ago when it came out...)

 

[Graph Feppar]

Hey, you're still omitting the spaces after the commas and points ...

I have been doing this all my life, this thing is so deeply rooted inside my brain I cant just stop it

But I did put atleast one space after comma today so its small progress, I deserve cookie for effort.

 

[graz_lag]

Right. DSD wasn't created for us consumers, it was created by labels, for labels. Executives saw it "solving" two problems:
* New format to incite fresh demand to resell back library. Again.
* Bloat file sizes as impediment to online piracy. (Not a factor today, but 20 years ago when it came out...)

Sorry mate if I correct you a bit ... In fact it was created by the ancient Greeks for their mythological and hero creatures ... under the eye of Zeus ... (Sony Audio mega-corporation of audio sub-companies ...)

 

[graz_lag]

I have been doing this all my life, this thing is so deeply rooted inside my brain I cant just stop it

But I did put atleast one space after comma today so its small progress, I deserve cookie for effort.

OK, go for a cookie this time ...

 

[Wombat]

I'd be curious to see any evidence of 'better transients / dynamics' in DSD compared to PCM.

Delayed reply.


Archimago looked at impulse response of converted DSD files. He sums up his findings in the Conclusion:

Filtering, Impulse Response, Frequency Effects


His look at DSD to PCM conversion quality:

http://archimago.blogspot.com/2015/04/analysis-dsd-decoders-2015-windows-mac.html

 

[dc655321]

Actually,higher sample rates do increase dynamic range.Doubling samplerate increase dynamic range by 3db

Can you explain how that might work exactly?

 

[andreasmaaan]

Delayed reply.


Archimago looked at impulse response of converted DSD files. He sums up his findings in the Conclusion:

Archimago DSD Impulse


His look at DSD to PCM conversion quality:

Conversion quality

Links don't seem to point to where you intended them to?

 

[Wombat]

Links don't seem to point to where you intended them to?

Fixed.

 

[Graph Feppar]

Can you explain how that might work exactly?

The total amount of quantization noise remains same if you double samplerate but its spread over larger bandwidth. That means the band of our interest which is up to 20 KHz will have less quantization noise spread over it.

 

[Blumlein 88]

Can you explain how that might work exactly?

It would be true of quantization noise. If all of our quantization noise is distributed over a 20 khz bandwidth, and we increase the bandwidth via higher sampling rates to 40 khz bandwidth, then the quantization noise is the same amount spread over twice the bandwidth which means the first 20 khz has 3 db less quantization noise.

Now with dither things can get more interesting. Shaped dither can put more of the noise at higher frequencies and less at lower frequencies. With some dither you might lower the noise floor in our more sensitive areas by more than 10 db.

Delta sigma DACs take advantage of this. They run at say 5 bits which doesn't sound like enough. Yet they run at very high sample rates, and with noise shaping they can give us perhaps 120 db of dynamic range by moving noise up way past where we care and lowering dramatically the audible band noise.

Well, Graph Feppar beat me to it by a minute.

 

[Graph Feppar]

One thing I am not sure about, I hope someone smarter than me will chine in, I know this dynamic range increase works for pulse density modulation ane pulse width modulation but I am not sure if it works with pulse code modulation.

The PCM will just ouput million ones or zeroes in row without dither,thats unlike PDM and PWM that are more suitable for this one bit oversampling thing, or am I wrong? Does 1bit oversampled PCM work even without dither?

 

[Wombat]

His poor grammar makes reading it an unnecessary strain.

Harrumph!

As repeatedly demonstrated on this forum there are those whose lesser grasp of English does not prevent them from conveying information better than others who are native English speakers.

Kal See Grammar perhaps.
(Frasier pun intended)