PCM vs DSD. Why DSD and its variants?

[garylkoh]

Even at 20-bit, that's only -120dB.

I was reading the article where its talking about CD

Bruce, since you're doing most of HDTrack's DSD conversions, you might want to get David to correct the paper.

 

[FrantzM]

Bruce


Still I don't see the advantages of DSD over PCM. In the best of cases it involves a transformation to analog then back to DSD if mastering is involved. Perhaps a direct to DSD scheme would be great but do they exist? What does the future hold for DSD? Has it advanced the SOTA? While I can seeus going toward even higher sampling rates on PCM do you see DSD as a format that will stand the test of times ... PCM seems to be quite poised to be that entrenched as it is.

@Mike

i agree that in the short term trusting your ears can be less than reliable. however, over time, when you can listen in many states of mind and awareness the truth will 'out'. your body does not lie to you while your mind is fickle. music is one of those things to be enjoyed and savored, not dissected and proved.

I can enjoy music from a clock radio. I am currently enjoying a lot of music from a system inferior to what I had before. I don't need proofs on music .. I need proofs when a mode or medium of reproduction is claimed to be superior. I will accept preferences, no proofs needed I do require it when superiority is claimed. Such imply a common standard by which the superior item can be compared to.
We audiophiles do take a break from the music from time to time (or so I hope since many simply come to a point where the music is "unlistenable" if from digital, not you I hasten to say) and we do focus on its reproduction. I want my system to perform its best and would like the industry that serves it to progress. Not go sideways listening (pun intended) to the mermaid songs of Big Corporations marketers.. I can't see what it brings and neither could the vast majority of audiophiles who didn't abandon their CD and flock toward DSD/SACD, the way they are now toward computer-based music systems. I do continue to see some of the High_rez sites claiming that their source was recorded in DSD and would like to understand why this is such a big deal .. The (multiple) conversions are not transparent, worse can't be transparent .. SO why bother? When confronted with that question, the answer is:" one needs to listen" .. I have and find especially some Hi-Rez spectacular. I like some of the HRx even on my Benchmark which seems to do some trick on HRx .. Mastering ? could be but done in PCM anyway in most instances ...

So for the present the question remains : Why DSD..and its variants?

 

[Bruce B]

Bruce, since you're doing most of HDTrack's DSD conversions, you might want to get David to correct the paper.

Believe me... David reads every one of my posts!

I'll drop a line to Malcolm to see what he says.

 

[Bruce B]

Just changed:

https://www.hdtracks.com/files/DSD_to_LPCM.pdf

But I still don't see how 16-bit noise can be at -144dB?

 

[weiss2496]

Just changed:

https://www.hdtracks.com/files/DSD_to_LPCM.pdf

But I still don't see how 16-bit noise can be at -144dB?

The -96dB figure for 16 bit is the total noise power in relation to the signal. The FFT does not give the accumulated noise, but rather all the separate frequency bins. Often a source of confusion....

Daniel

 

[FrantzM]

Hi

Aside from SACD are there consumer DSD sotware available , the Playback Desin 5 cliams the ability to decode DSD?

 

[RBFC]

Hi

Aside from SACD are there consumer DSD sotware available , the Playback Desin 5 cliams the ability to decode DSD?

The Playback Designs MPD-3 and MPS-3 both handle DSD. The "D" model also is a disc transport if I'm correct.

Lee

 

[Bruce B]

The Playback Designs MPD-3 and MPS-3 both handle DSD. The "D" model also is a disc transport if I'm correct.

Lee

The "D" model is the DAC and the "S" has the transport.

I believe Pure and iTunes will both pass a DSD signal via USB.

 

[chrisr]

DSD playback

FrantzM, Signalyst HQPlayer is PC software that plays back native DSD (dsf and dff) without converting to PCM. Also plays PCM up to 32/768. DSD playback is via ASIO.

 

[microstrip]

(...) This might have been true 25 years ago.. (...)

Frantz,
Should we remember this advertisement of Linn Products Ltd in the back cover of the June 1984 issue of Hifi News and Record Review?

 

[amirm]

FrantzM, Signalyst HQPlayer is PC software that plays back native DSD (dsf and dff) without converting to PCM. Also plays PCM up to 32/768. DSD playback is via ASIO.

At the risk of stating the obvious, that only works if the audio hardware has a DSD DAC. Otherwise, conversion to PCM is mandatory.

 

[DonH50]

FYI, and to echo Daniel, the spur (noise) floor (spurious-free dynamic range, SFDR) for an ideal converter is roughly 9N, or 144 dB for a 16-bit DAC (or ADC). The math is ugly, and the number does not hold well at low resolution, but for a sine wave and reasonable number of bits (say maybe 6 or more; I do not recall the curve off the top of my head) it is pretty accurate. When you add up all the energy across the band, as mentioned earlier, you get the familiar 6N+1.8 dB for SNR.

 

[fas42]

FYI, and to echo Daniel, the spur (noise) floor (spurious-free dynamic range, SFDR) for an ideal converter is roughly 9N, or 144 dB for a 16-bit DAC (or ADC). The math is ugly, and the number does not hold well at low resolution, but for a sine wave and reasonable number of bits (say maybe 6 or more; I do not recall the curve off the top of my head) it is pretty accurate. When you add up all the energy across the band, as mentioned earlier, you get the familiar 6N+1.8 dB for SNR.

Gee, Don, I don't get that one. Had a quick look around and figures for SFDR are around 90-100dB for 16 bit devices that I've seen; which makes sense. 144dB out of 16 bits sounds like fancy hand waving to me ...

Frank

 

[DonH50]

Textbook stuff, Frank, I am speaking of ideal converters... I think there are pictures in my sampling tutorials that will make it clear.

SFDR = spurious-free dynamic range = 9N dB = distance from signal to highest peak.

For an ideal converter there is no distortion, only the quantization noise floor. When you add up all the noise across the signal band, by RSS'ing, and compare to the signal power, then the familiar SNR = 6N+1.8 dB equation results.

Think of it this way: the SNR is the total integrated noise across the converter's bandwidth. Having a total of 6N means any individual spur is much lower, theoretically 9N; when you add up all the individual spurs, you get 6N.

I actually have papers proving the 9N number, but they make my eyes glaze... Bessel functions and some high-level math is required. I will say I have worked on converters that are pretty close to the 9N ideal, but it was a lot of work!

In practice, SFDR is often close to SNR for a lot of converters, though not for all (look through more specs and you'll see do much better than others). The reasons have to do with design trades, especially among noise, bandwidth, and distortion. More bandwidth and greater bias currents usually improve THD at the cost of noise. How that trade space is optimized is (very) application-dependent. 100 dB SFDR is pretty impressive for any converter, and probably near the practical noise floor for a real (and broadband) device.

HTH - Don

p.s. Look here: http://www.whatsbestforum.com/showthread.php?1209-Sampling-101

 

[Orb]

Just changed:

https://www.hdtracks.com/files/DSD_to_LPCM.pdf

But I still don't see how 16-bit noise can be at -144dB?

Just to add to what has been said before in a subtly different way.
It is worth noting there is a difference between what your thinking of I think being dynamic range and the article regarding noise at -144dB (which has been explained above by Don and Daniel.

The figure most are used to regarding 96dB is from the expression:
20 * log10(65638 / 1 ) = 96.3 (I think lol)
This is the linear amplitude, or dynamic range in dB for 16-bit, for 24-bit it is 144db.

I would have to hunt them out but I am sure there are several products with noise floor close to 120dba, more ideal for hi-res though.

Just my take on it anyway.
Cheers
Orb

 

[DonH50]

The logarithm leaves out part of the analysis, Orb... I have a couple or three derivations of quantization noise but that is outside the scope of this thread (I am pretty sure, plus I am too lazy to walk two flights up to find them! ). It can be found numerous places, though perhaps I should add a techie thread on that as well. Unlike SFDR, the SNR analysis is pretty straight-forward.

 

[fas42]

Got a handle on it now, Don. As you say, an individual spur or one specific harmonic is down by the 9N figure at the most, and that's very relevant for communication technology, but the actual noise is an enormous swath of spurs which add up to the 6N. As far as your ear if concerned that that one spur is so far down is pretty meaningless because it still can hear all the multitudes of other spurs all adding up, so it's only relevant for communications that you get this sort of performance with devices.

Frank

 

[Orb]

The logarithm leaves out part of the analysis, Orb... I have a couple or three derivations of quantization noise but that is outside the scope of this thread (I am pretty sure, plus I am too lazy to walk two flights up to find them! ). It can be found numerous places, though perhaps I should add a techie thread on that as well. Unlike SFDR, the SNR analysis is pretty straight-forward.

Agreed but the focus is on the practical aspect as Bruce is focusing on and quering the 144dB figure, which is beyond the boundary of 16-bit dynamic range.
but it needs to be shown that dynamic range/bit depth/resoultion/Snr/etc are different to sfdr, and IMO more applicable to what Bruce and others focus on..
To tell practical use of a CD or DAC, to me it is the S/n related against the dynamic range of the technology, in case of traditional CD that brings us to the 96dB boundary and needed S/n with dBa better than that.
If you feel SFDR is applicable in a practical way in regards to 16-bit or 24-bit audio and I appreciate I may be missing this it definitely is worth specifically expanding upon it in this thread.
I should had clarified for others the formula I gave was relating dBFS and is for going from a linear amplitude value (+ and - 32,768 values for 16-bit) to decibels, but I appreciate the maths you go into with the 6.02 (and how this is derived) and your own good thread goes into more detail.
Thanks
Orb

 

[Orb]

Got a handle on it now, Don. As you say, an individual spur or one specific harmonic is down by the 9N figure at the most, and that's very relevant for communication technology, but the actual noise is an enormous swath of spurs which add up to the 6N. As far as your ear if concerned that that one spur is so far down is pretty meaningless because it still can hear all the multitudes of other spurs all adding up, so it's only relevant for communications that you get this sort of performance with devices.

Frank

And which you can see with the measurement graphs by both John Atkinson and Paul Miller.
Cheers
Orb

 

[DonH50]

Nearly there, Frank... It is not "at most" 9N, some are lower than that. The math says for a single-tone input the maximum (highest) spur is 9N dB down, so most are actually a little lower. The rest you have right.

Orb, somebody with a better psycho-acoustics background will have to address SNR vs. SFDR, but I will comment that we can pull out signals below the noise floor, and that includes distortion so would argue SFDR can in fact be important in audio as in other applications.