darrenyeats
pfm Member
There were several opinions on the thread, of which that was one.
PCM vs DSD, can you hear the difference?
- Thread starter kristoffer
- Start date
onlyconnect
pfm Member
Yes, though oddly it chimes with my own subjective impression of DSD "sound" if it has a sound, which technically it should not if it is meant to be accurate. And I think DSD fans claim PCM is "harsh" which could be the same characteristic in reverse. Personally though I don't associate any particular sonic characteristic with PCM unless it is the over-compressed loud mess often found on CDs, which is not the fault of PCM. I would also not be surprised if all these differences are imaginary.
Tim
Igloo Audio
Registered User
We just want to hear the microphones!
Either DSD 'direct to disk' or DXD 24/384 IMVHO
Peter
onlyconnect
pfm Member
Why? Nobody is saying that a DSD recording cannot sound very very good. It could still be flawed.
Tim
While DSD (or high resolution PCM/DXD) may be misused by marketing to re-sell existing music, it may very well have audible benefits.
This document - http://www.merging.com/uploads/assets//Merging_pdfs/dxd_Resolution_v3.5.pdf - is linked from a Wikipedia article on DXD. It's mainly focused on SACD mastering, possibly with some marketing BS, but if the measurements and graphs on the second page can be trusted, DSD (or PCM with very high sampling rate) has an excellent impulse response, even when compared to 96k or 192k. There's this theory that the time domain may be more important than the frequency domain and if it's at least partially true, impulse response (and pre/post ringing) matters.
You can take a look what antialiasing on 44.1k does to the pulse response on ie. old known http://src.infinitewave.ca/ - select the Impulse test result and compare "SoX 14.4 High Quality" to "SoX 14.4 High Quality (Aliasing Enabled)". Note that the more relaxed requirements on the filter steepness reduce the ringing effects. Switch to Sweep and note that the aliased signal is still above 20KHz.
Now imagine that you have up to 192/2 KHz available. It kind of makes me think that the charts in the PDF linked above might be in fact correct.
The SRC comparison chart probably used the default 95% bandwidth setting (passband up to ~21KHz), but it got me thinking - if people generally don't hear much above ie. 18KHz, perhaps the filter could be even more relaxed, resulting in even better impulse response (at the cost of added distortion above 18KHz). I could do that.
This isn't necessarily practical for generic usage, but it could be useful for testing whether humans can distinguish between these slight changes in ringing and whether the result subjectively sounds better.
I can literally play with samples using sox, incl. downsampling without any decimation filter if I wanted. If you're interested, I could prepare a few 44.1Khz samples with various settings for you to audibly test, though I would need a good quality 96/192KHz original (not just remastered upsampled 44.1KHz) with good indicators (sharp strings, vocals, distorted guitar, ...) and a license/permit to modify and publish the downsampled samples.
The test would - by no means - be the same as recording in DSD/DXD, but it could shed some light on whether / how much impulse response matters.
I'm also no experienced digital audio mastermind, but I like to play with stuff.
edit: some images:
white noise-based frequency spectrograms (no sweep generator here): original, normal VHQ, steep filter (99% passband), 85% passband, antialias, 85% passband, alias
manually created single sample, resampled, opened in Audacity, screenshot cropped: normal VHQ, steep filter, 85% passband, alias
This document - http://www.merging.com/uploads/assets//Merging_pdfs/dxd_Resolution_v3.5.pdf - is linked from a Wikipedia article on DXD. It's mainly focused on SACD mastering, possibly with some marketing BS, but if the measurements and graphs on the second page can be trusted, DSD (or PCM with very high sampling rate) has an excellent impulse response, even when compared to 96k or 192k. There's this theory that the time domain may be more important than the frequency domain and if it's at least partially true, impulse response (and pre/post ringing) matters.
You can take a look what antialiasing on 44.1k does to the pulse response on ie. old known http://src.infinitewave.ca/ - select the Impulse test result and compare "SoX 14.4 High Quality" to "SoX 14.4 High Quality (Aliasing Enabled)". Note that the more relaxed requirements on the filter steepness reduce the ringing effects. Switch to Sweep and note that the aliased signal is still above 20KHz.
Now imagine that you have up to 192/2 KHz available. It kind of makes me think that the charts in the PDF linked above might be in fact correct.
The SRC comparison chart probably used the default 95% bandwidth setting (passband up to ~21KHz), but it got me thinking - if people generally don't hear much above ie. 18KHz, perhaps the filter could be even more relaxed, resulting in even better impulse response (at the cost of added distortion above 18KHz). I could do that.
This isn't necessarily practical for generic usage, but it could be useful for testing whether humans can distinguish between these slight changes in ringing and whether the result subjectively sounds better.
I can literally play with samples using sox, incl. downsampling without any decimation filter if I wanted. If you're interested, I could prepare a few 44.1Khz samples with various settings for you to audibly test, though I would need a good quality 96/192KHz original (not just remastered upsampled 44.1KHz) with good indicators (sharp strings, vocals, distorted guitar, ...) and a license/permit to modify and publish the downsampled samples.
The test would - by no means - be the same as recording in DSD/DXD, but it could shed some light on whether / how much impulse response matters.
I'm also no experienced digital audio mastermind, but I like to play with stuff.
edit: some images:
white noise-based frequency spectrograms (no sweep generator here): original, normal VHQ, steep filter (99% passband), 85% passband, antialias, 85% passband, alias
manually created single sample, resampled, opened in Audacity, screenshot cropped
: normal VHQ, steep filter, 85% passband, alias
JohnW
pfm member
The DSD impulse response in the paper confirm what we heard during our listening tests...
DSD's 2.8MHz "sampling Rate" gives DSD superior transient precision which it trades in exchange for High Frequency Dynamic range performance (PCM has a Dynamic range advantage at higher Frequencies, but DSD Dynamic range output performs even 24bit PCM at lower frequencies where you have more musical information – so I’d be inclined to say that this is also an advantage in DSD's favour).
As I’ve repeated in many posts that as I get older and I’d like to believe wiser I’ve come to appreciate that transient (Time) precision is far more important then frequency domain performance for HiFi reproduction.
DSD's 2.8MHz "sampling Rate" gives DSD superior transient precision which it trades in exchange for High Frequency Dynamic range performance (PCM has a Dynamic range advantage at higher Frequencies, but DSD Dynamic range output performs even 24bit PCM at lower frequencies where you have more musical information – so I’d be inclined to say that this is also an advantage in DSD's favour).
As I’ve repeated in many posts that as I get older and I’d like to believe wiser I’ve come to appreciate that transient (Time) precision is far more important then frequency domain performance for HiFi reproduction.
That depends on what you define as "musical information" - PCM at high sample rates (DXD - 8.4672 Mbit/s, "three times that of DSD64") with the right filters could also have some advantages (worth testing at least). It's apparently "much better than 192KHz" according to some sources.
Although DSD128 still seems rather superior if all you need to do is record/playback without any mastering (ie. vinyl digitization).
merlin
Avatar changed - Town names deemed offensive.
Tim,
My feeling is that any recording can be flawed but that the digital conversion that is applied to it is, these days, unlikely to add to that.
The only process that seems to make any really important difference is the mastering IMHO.
darrenyeats
pfm Member
JohnW,
Back to your delay line test!
Further to my comments about "A/D optimised for PCM" I perused the Gearslutz thread. This morning something popped into my head. Your flow diagram shows the final stage as always a DSD DAC. This means the PCM path has a second delta-sigma stage ... The GS thread explains although there is a computational method to convert between DSD and PCM, going from PCM back to DSD always adds a new layer of modulator noise. That's why DSD editing suites, once they convert something to PCM for edits, keep it PCM for as long as possible. Each return journey is harmful.
So this is me returning to the optimising argument but at the D/A end ... If your flow chart showed a PCM DAC versus a DSD DAC (instead of both paths leading to a DSD DAC) perhaps the result would have been different?
Thinking about it, a more optimal PCM path would have been just multi-bit delta-sigma A/D and PCM D/A.
Back to your delay line test!
Further to my comments about "A/D optimised for PCM" I perused the Gearslutz thread. This morning something popped into my head. Your flow diagram shows the final stage as always a DSD DAC. This means the PCM path has a second delta-sigma stage ... The GS thread explains although there is a computational method to convert between DSD and PCM, going from PCM back to DSD always adds a new layer of modulator noise. That's why DSD editing suites, once they convert something to PCM for edits, keep it PCM for as long as possible. Each return journey is harmful.
So this is me returning to the optimising argument but at the D/A end ... If your flow chart showed a PCM DAC versus a DSD DAC (instead of both paths leading to a DSD DAC) perhaps the result would have been different?
Thinking about it, a more optimal PCM path would have been just multi-bit delta-sigma A/D and PCM D/A.
onlyconnect
pfm Member
OK, so let's *assume* for a moment that DSD has significant benefits over PCM but *only* if:
1. It is a straight to DSD live recording
2. It has not been through any PCM encoding eg DXD
I'd imagine that the majority of SACDs out there do not conform, but if I could find one, then if I have a player that can do unconverted DSD (which I believe I have), I should hear a significant different if I interpose a PCM conversion or use the bass management which also converts to PCM, right?
Can you suggest a suitable SACD so I can test?
Or can someone post some samples that would demonstrate this if I can hold of a suitable DSD DAC?
Tim
1. It is a straight to DSD live recording
2. It has not been through any PCM encoding eg DXD
I'd imagine that the majority of SACDs out there do not conform, but if I could find one, then if I have a player that can do unconverted DSD (which I believe I have), I should hear a significant different if I interpose a PCM conversion or use the bass management which also converts to PCM, right?
Can you suggest a suitable SACD so I can test?
Or can someone post some samples that would demonstrate this if I can hold of a suitable DSD DAC?
Tim
onlyconnect
pfm Member
Here is another massive thread on the subject:
http://www.whatsbestforum.com/showthread.php?9598-DSD-comparison-to-PCM
Lots of familiar names though. It is as if there is one big thread on the internet where the same 25 people who care have a never-ending debate.
Tim
http://www.whatsbestforum.com/showthread.php?9598-DSD-comparison-to-PCM
Lots of familiar names though. It is as if there is one big thread on the internet where the same 25 people who care have a never-ending debate.
Tim
onlyconnect
pfm Member
tomek
pfm Member
Really "pure" DSD recordings are very rare. Is there any list of them ?
I can't listen to Allison Kraus or whatever their name is ....
It will be interesting to compare pcm and pcm converted to DSD stream.
Is there any advantage to convert pcm to dsd "on the fly" when playing and feeding with DSD stream one good DSD dac ?
In 1984 Linn owner was not able to recognise if his LP12 was played directly, or with 16bit (!) ADC-DAC conversion. It was stone age of digital era and both adc and dac where pcm.
http://www.bostonaudiosociety.org/bas_speaker/abx_testing2.htm
I gave up my lp collection and turntable more than 20 years ago. I have some digitalized lps in my streaming system and when i am playing them to a friend who is absolutely analogue terrorist he is shocked and says - it sounds like LP.
OTH i believe the people, who had a chance to compare pcm and dsd in perfect conditions. Maybe there is a difference.
I also now that a good truntable can sound very good (or bad with bad recordings), but a master tape is even better, and the question is what is closer to original - lp or digital copy of the master tape ?
I suppose that Sony & co want earn more money with the same recording again, which where already sold as lp, lp 200g, cassette, CD, CD remastered, CD 24 bit remastered, 40th years edition remastered, mp3, 2448, 2496, 24192, maybe again lp 200g and now DSD and in future as DSD X MHz, DSD XX MHz and and and ...
I can't listen to Allison Kraus or whatever their name is ....
It will be interesting to compare pcm and pcm converted to DSD stream.
Is there any advantage to convert pcm to dsd "on the fly" when playing and feeding with DSD stream one good DSD dac ?
In 1984 Linn owner was not able to recognise if his LP12 was played directly, or with 16bit (!) ADC-DAC conversion. It was stone age of digital era and both adc and dac where pcm.
http://www.bostonaudiosociety.org/bas_speaker/abx_testing2.htm
I gave up my lp collection and turntable more than 20 years ago. I have some digitalized lps in my streaming system and when i am playing them to a friend who is absolutely analogue terrorist he is shocked and says - it sounds like LP.
OTH i believe the people, who had a chance to compare pcm and dsd in perfect conditions. Maybe there is a difference.
I also now that a good truntable can sound very good (or bad with bad recordings), but a master tape is even better, and the question is what is closer to original - lp or digital copy of the master tape ?
I suppose that Sony & co want earn more money with the same recording again, which where already sold as lp, lp 200g, cassette, CD, CD remastered, CD 24 bit remastered, 40th years edition remastered, mp3, 2448, 2496, 24192, maybe again lp 200g and now DSD and in future as DSD X MHz, DSD XX MHz and and and ...
onlyconnect
pfm Member
Of course they do; and it is the older generation that is buying, which is why whenever we get a new format (eg Blu-ray audio) the same old warhorses appear, Dire Straits, Rolling Stones etc.
Tim
merlin
Avatar changed - Town names deemed offensive.
No but most of Tony Faulkner's LSO live SACD's were recorded straight to DSD as were many Telarc releases.
It's Alison Krauss and Union Station. It's a shame. Might not be trendy like Pat Barber et all but musically it's far more wholesome IMHO.