[StudioSound]

There has been a lot of interest in DSD recently, with a lot of the higher-end DACs adding support for it.
But is DSD actually any better than PCM?
 
Even from papers which are pro-DSD, you have graphs like this:

So DSD has lower noise below 20kHz, and significantly higher noise above 20kHz.
 
But one of the arguments I hear about DSD or "HD" audio in general, is that ultrasonic sound can be "experienced" even if it cannot be heard.
 
In that case, with PCM having a flat noise floor of -144dB to the nyquist limit (half the sample rate - so 24/96 can represent signals up to 48kHz) doesn't PCM fare better?
 
What audio gear actually has a noise floor lower than -144dB, and furthermore, when is a -144dB noise floor ever going to be audible anyway? To hear that noise floor, isn't the signal going to be so loud that it will destroy your hearing?
 
Another pro-DSD argument I hear is that DSD doesn't require the anti-aliasing filter that PCM does - but we can clearly see that DSD needs a lot of filtering above 20kHz due to that ultrasonic noise.
 
Doesn't that also work in PCM's favor? With 24/96 PCM, you can use a gentle roll-off, rather than the brickwall filter required for redbook audio. (16/44.1)
 
The "Dirac pulse"

 
Wow, PCM looks terrible here!

But it's my understanding that all a Dirac pulse is entirely simulated, and all it actually does is show the filters in use.
 
What you see here is not going to be a part of an actual audio signal, and while you can see a difference in this graphic, there would not be an audible difference.
 
 
And finally - you can't edit DSD. If there has been any editing performed during the mastering phase, the audio will have been converted to PCM (probably DXD - 24/352.8) to do the processing, and then converted back to DSD... so you lose any "advantages" DSD may or may not theoretically have, and have just added ultrasonic noise compared to straight PCM mastering.
 
 
Is all this recent DSD hype just another thing manufacturers have come up with to sell you new hardware or "better" recordings?

 

[jcx]

plenty already mentioned here in other hi res threads - check out Meridian, Xiph white papers, some video's
 
 
consider human hearing threshold in quiet - after 10s of minutes of accommodation in a anechoic chamber
 
then look for really good studio sound proofed room - or architectural acoustics firms goals for decent home theater room noise floors
 
or how headphones give acoustic noise from mechanical microphonics from pulse, muscle microtremor, cable dragging on clothes with head motion...
 
in short - a very optimistic noise floor for practical audio listening is nearer the -100 dB line on your graph
 
 
then the DSD plot isn't adequately filtered - few audio amps like 10s of mV of AM band noise coming straight into their inputs - much higher order analog filtering is needed for DSD
in fact Sony had to revise the original 100 kHz DSD analog output filter recommendation down to 50 kHz when embarrassed by some audiophile amps melting down with early SACD at audio shows
 
 
and higher than RedBook CD 16/44.1 isn't clearly established as needed in the literature - the still very few conflicting studies with questionable methods haven't been accepted yet as "proof" that more is needed (ie can be distinguished AB/X under controlled conditions) in home music listening

 

[StudioSound]

Quote:

and higher than RedBook CD 16/44.1 isn't clearly established as needed in the literature - the still very few conflicting studies with questionable methods haven't been accepted yet as "proof" that more is needed (ie can be distinguished AB/X under controlled conditions) in home music listening

 
Oh, I agree. I was just curious about the resurgence of DSD when PCM seems like it is the better format.
 
I found this article today, which is very interesting: http://www.realhd-audio.com/?p=74

 

[jcx]

actually that article still assumes 24 bits -  but the 24 bit depth really isn't justified as a hi res consumer distribution format either - even if you suspect 16/44.1 isn't quite adequate for some small % several sigma out tail of the population
 
by upping sample rate to any of 88/96/176/192 you increase the frequency space for dither - then 16 bits with noise shaped dither can deliver your pretty extreme"requirement" of 120 dB in the audio band