PCM 24/96 and 24/192 are better
So says John Siau, Director of Engineering at Benchmarks Media, one of the makers of the one of the best DSD and PCM dacs, interviewed by Mark Waldrep of AIX recordings, some of the best recordings available on DVD-Audio (I got my first recording from them something like 10 years ago) and now Blu-Ray and hirez downloads as well. I'm making this my top link because John Siau gives some of the best technical explanations I've seen anywhere, and I believe he is 100% correct in what he says here. Some of what other people I link to below I may agree with but leave me feeling a bit less than 100% certain.
John says Benchmark supports direct DSD processing basically because some people have those files, and if you do have a DSD file the direct processing of DSD is slightly (though just very slightly) better than conversion of DSD to PCM first. He stresses however that the DSD->PCM conversion is very benign, nonetheless.
But John says he does not endorse DSD, and believes that people should migrate to 24/96 and 24/192 because they are better in every way. He also points out while DSD makes an ok delivery format for high rez digital, somewhat better than 16 bit digital anyway, and it's all most people with most systems need, it makes a lousy professional recording format. And the main reason it makes a lousy recording format is that any kind of editing or mixing or even level adjustment requires conversion of DSD to PCM first, at least in some form, and then back to DSD if that is how you want to distribute the recording. And then the big loss is in the conversion of PCM back to DSD, that conversion is not so benign, he says. He gives way more technical detail than I do here, and also debunks the legendary "noise graph" Sony used to sell DSD, wherein the noise level of PCM formats is shown as a straight line higher than the noise level of DSD, whereas in reality the noise spectrum of 24/96 and 24/192 is below that of DSD (64x) everywhere, not just at supersonic frequencies (where the noise of DSD takes off like a rocket).
At least one DSD'er appears in the comments, arguing by authority of someone he knows who says DSD is better.
At least one DSD'er appears in the comments, arguing by authority of someone he knows who says DSD is better.
DSD = Never Twice the Same Digital Recording
Here's the best link I've found so far with a simple but visual comparison of PCM and DSD and the allegedly better impulse response of SACD which is actually worse. Basically, a delta sigma audio system like DSD generates a huge amount of noise at higher frequencies. The "Noise Shifting" that DSD uses to hide this fantastic amount of noise is a mathematical feedback loop which cancels out high frequencies generally (it can't actually discriminate between high frequency signal and noise). In the process of canceling out the high frequencies up to the Nyquist limit of 1/2 the sampling rate (1.4mHz for 2.8mHz DSD) this feedback loop moves the 1-bits around, and this creates noise in the time domain, also known as jitter. (I explore this further below in the section with Jitter in the heading.) This jitter can hide the fact that if you overlay enough DSD recordings of the same thing, such as a 10 kHz square wave, what you ultimately see looks very close to 88.2kHz 20 bit PCM. It's just that every time you make a DSD recording from a square wave, you get a different portion of the infamous (but actually harmless) ringing at the top of a square wave in PCM. So perhaps DSD should be called "Never Twice the Same Digital Recording" because the exact recording you get varies from one DSD recording to the next. (The playback, however, will be mathematically identical each time. So the randomness is no fun for the listener, who is actually being cheated out of getting the whole enchilada, no matter how many times he listens to it.) DSD also gives you nothing like the high frequency response you might expect from the sampling rate. While you might expect frequency response to 1.4 mHz from a 2.8 mHz sampling rate (not exactly DC to light, but getting up there), what you get is actually limited to about 40kHz, not any better than 88.2kHz PCM let alone 96kHz or 192kHz. Meanwhile, high resolution PCM formats potentially give you frequency response and low noise all the way to just below half the sampling rate. It's been proven that we are affected by such high frequencies, and/or the slew rates they embody, even when we can't hear them as tones. (Since this link above is available only through the Wayback website, and no recent comments, I wonder about it being 100% true, but you can see the corroboration below.)
Another Format War Now, just when PCM has gotten so good?
Nowadays we have wonderful PCM systems available, 24 bit 192 kHz recorded music is available just about everywhere, on DVD-Audio discs, Blu-Ray discs, and high rez downloads. And incredible 24/96 or 24/192 playback equipment is widely available at low cost, $100 and up. And PCM is suitable for endless digital signal processing (DSP) options, from digital preamps, to parametric and graphic equalizers, room correction systems, digital crossovers, vintage simulators, etc. Working with PCM inputs and outputs, these processors can be noiseless and distortion free. Note that it's basically impossible to do digital signal processing directly on DSD; it has to be converted to PCM or analog to do that. My own systems rely on digital equalizers and crossovers working with PCM at 24/96. It would be impossible to assemble the hybrid speaker systems I have without these digital processors. I also love the high resolution PCM recordings that I have obtained on DVD-Audio discs. As recently as last year I obtained the DVD-Audio anniversary edition of Lark's Tongue in Aspic from King Crimson, featuring 24/96 and 24/192 PCM masterings, and it has become one of my favorite recordings. This was released in 2012, long after many in the audiophile media declared DVD-Audio dead. In fact, many declared DVD-Audio dead before it was even introduced in 2002, and I can't count the number of wonderful DVD-Audio releases I have picked up since then.But just now that we have so many wonderful PCM options, Sony and a particular cadre of super high end equipment and software purveyors and studios are back to pushing the old 2.8 Mhz 1 bit DSD from 1996 as being somehow better than all PCM, while many of the most highly respected audio engineers and enthusiasts (including me) long ago thought 2.8 Mhz DSD was just barely better, if at all, than Redbook 16 bit 44.1 kHz digital, and very inefficient as well, as well as not being amenable to digital signal processing. If you ask me, this is a disaster for good recording, and adept home audio enthusiasts like me because another format war killing off high resolution PCM (now available on Blu Ray as well as DVD-Audio discs, as well as downloads), and wasting the time and effort of equipment manufacturers, is the last thing we need.
My Experience, and that of my friends
Before buying my first Sony SACD player, but just based on technical articles I had read (see below), I believed SACD actually would be inferior to Redbook digital, and was just a marketing scam by Sony to keep their ability to extract license fees intact after the expiration of CD patents, and keep music locked up in unrippable discs now that the public had realized CD's could easily be copied and uploaded into digital servers. I bought my first DVD-Audio player as soon as I could, a Toshiba 5700, having great hopes for high resolution digital on DVD-Audio. Only later did I grudgingly buy a Sony SACD player so that I could take advantage of the greater variety of claimed-to-be-high-resolution discs in their native format. Back in those days, I fed the analog output of my Sony SACD player straight into an all-analog preamplifier with a system having no digital processing. I continued to seek out the DVD-Audio versions of everything I could first, but had to accept that far more discs were available in SACD, as a result mainly of the possibility of Hybrid CD/SACD discs. I did feel after all that SACD was somewhat better than Redbook digital, just no where near 24/96 PCM. In all this time, only one recording on SACD has seemed to reach about the same heights as 24/96 PCM on DVD-Audio, and that is the SACD version of Santana's Supernatural. But I have found countless DVD-Audio discs (and such variants as the DVD-Video DAD's from Classic Records) that I though were stellar, far better than anything I could imagine hearing on CD. Yes I have been very unscientific because I got tired of doing endless listening tests long ago, and came to never trust the outcome of a small number of tests anyway. So why should you believe my listening tests? Of course you shouldn't, but many people seem to be interested in my observations and thinking anyway.
Meanwhile, many serious audiophile friends of mine (including one audio manufacturer, and another veteran audio engineer) have never bought into the notion that DSD/SACD was better than Redbook, indeed considering DSD/SACD to be inferior to Redbook CD digital. I must add that these friends ubiquitously used the analog outputs of the latest generations of SACD players available at the time, almost always made by Sony, into their all-analog preamplifiers. I must add this qualification because many DSD proponents are now claiming that "people who criticize DSD have never heard it properly" because in their imagination the detractors were so dumb as to take the PCM outputs of SACD players into their DAC's (or worse, use the Home Theatre setups on DSD, that requiring internal conversion to PCM) and that of course would not preserve the sanctity of DSD, as it would have to be translated to PCM before decoding.* But none of my friends have been that stupid, and none have bothered testing SACD players in home theater setups either (having DVD players dedicated to that task, if they even bother with Home Theater). They brought home the latest Sony SACD/CD players, plugged the analog outputs into their analog preamplifiers, and tried them out, using Redbook and SACD's, most often on the same exact SACD/CD players, though sometimes with different players and dacs attached to those other players, and concluded that the CD layer sounded better than the SACD layer (and being careful to find CD/SACD that were mastered identically, which was most often not true), or that the CD layer on the hybrid disc seemed dumbed down compared to regular CD's which could be purchased at the same time, and that those plain old CD's sounded better than anything on the SACD or hybrid SACD. And then, back went the expensive latest generation SACD player to the store or friend it came from. What actually bothers me is that some of these friends have never even tried hirez PCM on DVD-Audio. They just tried SACD because the herd was going there and they wanted to try it themselves. And there were so few DVD-Audio discs available, and almost never from local stores.
That has not been my experience, which I described above, though to be honest I haven't done much side-by-side testing of CD's and SACD's. I just play SACD's, on my Denon 5900 or Oppo BDP95, pretty much assuming SACD or DVD-Audio would be at least slightly better, and that has been my overall impression, from listening to all the discs in my collection, in other words casual testing if sometimes done through serious listening. And I've always used the analog outputs, and never external DAC's, which I feel are not worth the upgrade from the fine disc players I have, which I believe have better circuitry than most DAC's ever made, and don't suffer from jitter that may arise through external digital connections, especially ones that are not AES/EBU or better.
(*Though I suppose you could say that some particular DSD DAC available today is better than all the ones these friends of mine have heard, so much so that the (alleged) superiority of DSD from an SACD over an identical CD was not then apparent. Yes you can always make such arguments, ad nauseum. Frankly I don't think huge differences even exist among PCM DAC's since about 2000, or maybe even 1986, even from the DACs simply built into the better quality CD/SACD players which have typically used the best available DAC chips if not the most overbuilt associated circuitry, the technology and measured performance are only marginally different, and among DSD DACs since the SCD-1, which was a pretty much all-out effort by the inventors of SACD. I have no time or interest in testing each latest thing that has come out. I prefer more obviously fruitful and less expensive endeavors, such as fine tuning my digital crossovers, or actually listening to music. Or pointless but personally satisfying endeavors like posting to my blog site. However I'm fine with people playing their latest and greatest toys in their homes and audio society meetings and conventions for my amusement. Anyway, before I bother taking your argument about so-and-so being the best whatever, so much better it invalidates all previous tests on the same kind of thing, you'd also have to provide some sort of reason that makes sense to me, or if you have the unit in question right and hand, you can play it, and I'll give a suitably ambiguous observation, if you want one, or maybe even if you don't.)
The Technical Literature
Here's the first paper by Lipshitz and Vanderkooy blasting 1-bit Sigma Delta conversion. Lipschitz is one of the most well regarded audio engineers of all time. Another very knowledgable audio engineer, David Rich, summarized those ideas and others in a harshly critical review of DSD/SACD in 2001 that was (surprisingly) published in Stereophile. John Atkinson posted a one sentence follow up to Rich's ironclad denunciation (showing among other things that 1-bit DSD would generate idle tones and other bothersome forms of distortion) with the claim that DSD was not, in fact, an entirely one bit system, as he had apparently heard from Sony after informing them of the critique. However, I now wonder if that disclaimer by Atkinson was at least partly incorrect. Perhaps some DSD recorders use a multibit system, but it seems like that hasn't necessarily always been true, and virtually all DSD proponents are still talking today about DSD as if it is actually a 1-bit system. If DSD is a 1-bit system, then all the horrible things Rich, Lipschitz, and Vanderkooy talked about are indeed wrong with it. If DSD isn't a 1-bit system,then it requires decimation and interpolation filters just as PCM does, and there really isn't any magic "less processing" required for DSD, only a very inefficient means of encoding audio information so that more bits actually yield less information than Redbook PCM (see Arthur Salvatore's opinion discussed below--and note that while Salvatore clearly states he is not a digital expert, he is actually echoing some of the arguments made by the the serious engineer David Rich). DSD promoters seem to want to have it both ways.
After the paper by Lipschitz and Vanderkooy, some engineers employed by Philips responded in JAES, and Lipschitz and Vanderkoy respond again, and so on. This continued until about 2003, by which time Sony wasn't pushing SACD much anymore and had moved on to preparing the battle royale over high definition video formats, which they ultimately won with a frequently alleged to be inferior system, but by getting a majority of movie studios on their side. You can read about the intellectual battle over DSD in the Audio Engineering Society in the Wikipedia entry for Direct Stream Digital in the section entitled DSD vs. PCM.
While I'd recently become hopeful that DSD was a solution to the limited time resolution of PCM (see earlier posts), I now strongly believe that DSD is not the right solution. Not only does DSD generate large amounts of noise (which is shifted to higher frequencies…it does not "go away") in the amplitude domain…it generates noise in the time domain because of the noise-shifting. It doesn't actually have frequency response better than PCM, in fact it's substantially inferior to 24/96 in bandwidth.
A Blind Test
Listening tests at the River City Audio Society meeting for March 2014 were inconclusive for the group (compared with 16/44.1 digital) and very much for me as well. I had not a clue which was which. In contrast, a year before I clearly heard the difference between 16/44.1 and higher resolution digital formats like 24/96. In that test, I got every identification that I thought I got correct correct (all but one of the like-vs-like comparisons, and I don't count the mastering beauty contest last test). Unfortunately, the DSD vs PCM tests were likely all beauty contests anyway, with the sonic differences coming entirely from mastering or level differences. But even then, I would have expected the superior sonic resolution, if there were one, of DSD to shine every once and awhile. It didn't for me.
PCM continues to improve
It may be that whatever benefit DSD has could be duplicated better simply by having higher sampling rates, which are already available, such as 32bit/384khz PCM. PCM has had a history of continuous evolutionary progress. In DSD, one of the most highly regarded units is the very first one, the SCD-1 from Sony. Now there are 2x and even 4x DSD recorders available. But that's just catchup to the progress PCM has made since 2001, and at very high price levels
PCM also has many benefits, it makes possible wide ranging DSP such as speaker and room correction (which I use…manually tuned not automated), digital level control, and the like. For those recording music (which sometimes includes me) it is far easier to work with. To my knowledge there is only one (stratospherically expensive) DSD editing system. Virtually all DSD recordings are edited or processed either through PCM or analog. I think most of the recordings we listened to at RCAS were originally recorded in analog. There are now systems that can losslessly edit (without losing the DSD magic) DSD files by up sampling to an even higher resolution format known as DXD, which is essentially 8 bit PCM at 2.8Mhz. Such equipment is now stratospherically expensive. One wonders if the game being played now isn't intended to rub out all the small producers, leaving only Sony and a few majors who can afford the mega priced production equipment.
OTOH, perhaps the "randomness" of DSD, noise in both the analog and time domains, is a good thing. Life is random too. But I can't argue that seriously here (but I will take a stab at it later…I have a concept which I feel is far superior to DSD, using only PCM systems and analog.)
Jitter, you ask? DSD is made of Jitter
Well as I stated above, the "noise" in the time domain that "noise-shifting" creates (recall that noise shifting works by moving the 1-bits around to start filtering at about 10kHz) is also known as jitter, but somehow the J word is never mentioned among DSD enthusiasts. They don't have the numbers.
I have often noted that some of the best research published on jitter shows that it doesn't become audible until about 10nS or so, which is like 50 times the jitter levels in the better digital equipment, which is more like 200 pS or less. (I don't necessarily consider this the final word on the audibility of jitter, but it did represent the state of the art audio science in the mid 1990's.) The very best equipment can get down into double digits (actual clocks can get down to single digits or less, but we don't listen just to the clocks). So I think jitter may be overrated in importance, used more as a pseudo-explanation to cover up the fact that people don't really do controlled blind tests, but need to explain their gut feelings about why one CD player is better than another, and since the normal specs don't seem to say anything meaningful, audiophiles have latched onto jitter as the ultimate cause of all the bad things they think they hear, in unscientific listening tests, when in fact they don't really hear differences between digital playback gear, they just generally hear the lack of ultimate resolution in all digital sources, I think, maybe, and I am agreeing which Arthur Salvatore on this resolution thing (linked below) though this is completely unproven, it has always made intuitive sense to me that digital audio systems reduce noise at the expense of actual resolution. What I mean is that the difference between 270pF jitter (a dCS stack from 10 years ago measured about this well) and 225pF jitter (that's what a Sonos system measured--and I found it mind blowing that an inexpensive network player measured better than the dCS stack, when lowering jitter was one of the reasons for a dCS stack) probably doesn't mean much. But when you start getting into the nS of jitter, you definitely have cause to worry.
But what about DSD? Well I started looking back at Stereophile tests, ,and starting from the SCD-1 and other CD/SACD players, jitter is not measured for the SACD layer, only for the CD layer. John Atkinson gave the excuse that his Miller Research Jitter Test only comes on a very carefully made low jitter CD-R. So I just did a search for SACD and Jitter, and guess what, he did graph a sort-of jitter test for the Sony SCD-XA777ES player, one of the best SACD players ever made. This player had a very respectable 171pF jitter using the CD-R. Using the Sony 11.025kHz SACD tone test (this is a rather poor test of jitter, actually, because it's not an inter modulating signal as on the CD-R) it shows a "noise" floor much higher for the simple tone SACD playback than for the CD-R intermodulation. JA says the SACD playback noise floor was elevated 10dB. Now how can SACD have a higher noise floor with it's alleged 20 bit resolution compared with a mere CD?* Hehe, I can't help but believe what it's showing is the jitter, but because of the way SACD works it doesn't produce jitter distortion in spectral spikes, but rather an elevated broad spectrum across the entire audio band. JA doesn't give us any estimated jitter number for it, but obviously it's significantly higher than for the CD playback. Now 10kHz is just where the noise shifting is starting to kick in. Imagine how bad it would be at 20kHz, or 30kHz--which SACD players are generally considered as having response to. And the fact that this "noise" covers the entire audio band can't be good either.
(Note that at 10kHz, the background noise level for CD and SACD is identical. But in the presence of the 11kHz signal, the SACD noise floor rises 10dB higher. That's the awful sound of dynamic noise, noise that rises and falls with the signal. Such things sound horrible, like old fashioned noise gating systems. Even though SACD does not actually have noise gates, it seems like many noise reduction schemes, including the old Pulse Count Detectors from Kenwood, have a tendency to have dynamic noise which sounds as if there were noise gates.)
(Note that at 10kHz, the background noise level for CD and SACD is identical. But in the presence of the 11kHz signal, the SACD noise floor rises 10dB higher. That's the awful sound of dynamic noise, noise that rises and falls with the signal. Such things sound horrible, like old fashioned noise gating systems. Even though SACD does not actually have noise gates, it seems like many noise reduction schemes, including the old Pulse Count Detectors from Kenwood, have a tendency to have dynamic noise which sounds as if there were noise gates.)
Whatever it's sampling rate, or high frequency response, I believe SACD simply doesn't have time resolution at high frequencies. It smears them in order to have respectable measured noise levels. And the high frequencies are where we need that time resolution, because the high frequencies are where the detail exists (the tiny curves on an oscilloscope are the high frequencies), and I believe we are affected by that resolution or lack of it at frequencies higher than we can hear as tones.
(*Actually, though, at 10kHz, the "noise floor" of the XA777ES in CD mode is the same as it is in SACD mode, as is typical for CD vs SACD. That's where the two noise curves cross over, and the noise of SACD just takes off, despite the noise shifting. Still, the identical background noise level at 10kHz doesn't explain why the SACD measured 10dB higher spectral noise while playing an actual 10kHz signal. What would this mean in terms of Jitter? I don't know how it scales, but I'd think at least 3.3 times and maybe more than 10 times more. So at a minimum the SACD playback had 500pS jitter, or 2nS, or maybe even higher as the noise is spread across the entire spectrum. And if you really tried to force higher jitter with a combination of tones, for example, you might get well into 10 nS and above Interestingly, an Accuphase DP-85 measured over 4nS jitter through external input. Was it "upscaling" to delta sigma modulation?)
We really don't have enough data here, and my knowledge is lacking too, but even the lack of data suggests something very wrong. If SACD is better than CD, shouldn't it's jitter, a drum that audiophiles including Atkinson have been pounding for years, be less instead of more?
M.O.J Hawksford on the Meridian site says that "bitsream" (by which he means DSD and the like) is inherently more susceptible to jitter cause by intermodulation with the elevated high frequency noise floor. But they don't give any number either, or mention DSD by name.
We really don't have enough data here, and my knowledge is lacking too, but even the lack of data suggests something very wrong. If SACD is better than CD, shouldn't it's jitter, a drum that audiophiles including Atkinson have been pounding for years, be less instead of more?
M.O.J Hawksford on the Meridian site says that "bitsream" (by which he means DSD and the like) is inherently more susceptible to jitter cause by intermodulation with the elevated high frequency noise floor. But they don't give any number either, or mention DSD by name.
Opinions, opinions
Here's one of the seminal pro-DSD papers from 2001, essentially a response to the Lipschitz and Vanderkooy paper I linked above.
Here's an incredibly long blog at What's Best Forum debating DSD vs PCM, and it includes quite a few fans of DSD as well as detractors. More heat than light mostly, but I found the paper I linked at top therein.
Yet another blog regarding DSD vs PCM at WBF. This makes me wonder, many angels can dance on a single bit? If DSD is as good as it's fans say, where are the blind tests proving it? I'd be the first to say blind tests cannot tease out all that a person can hear, there have to be fairly big differences to hear differences in blind tests reliably enough. But that is precisely what the DSD fans are claiming, big differences reliably audible in a studio environment. Every conversion being audible to them. Somehow real audio scientists don't get involved with those studios. They probably wouldn't be liked much there.
Audiophile magazine reviewers are not going to say anything bad about DSD purveyors when they are large current and potential advertisers. But many reviewers who write for their own pleasure feel no such restriction. As well as manufacturers who don't use DSD anymore.
Linn Sondek is now calling DSD a good idea in 1999 but obsolete in 2013, followed by a 11 page discussion. Not as good as 24bit/192kHz, which can be played on Linn DS players since 2007. I have a number of Linn hybrid HDCD/SACD discs from the 2000's which sound reasonably good in either HDCD or SACD. Either way, I must use them on a universal disk player since my servers don't support HDCD or DSD. Haven't done enough testing to decide which layer I like better. My general feeling has been that properly decoded HDCD is about equivalent, or better, than SACD, and in fact I have more favorite sounding recordings on HDCD than SACD. HDCD played back without HDCD decoding is ok on Reference Recordings but mediocre on other labels, so if you haven't heard HDCD properly decoded, as it seems few audiophiles have, you haven't heard it, and this is incontrovertibly true.
Romy the Cat hates DSD with passion.
Arthur Salvatore (for whom I have very great respect, btw, more so than virtually any other audio reviewer, at fact value anyway--since I also respect John Atkinson but know he is very constrained by his commercial perch, so I read JA between the lines always and look at his measurements which I cherish, as well as Peter Aczel who has virtually opposite opinions on nearly everything--but see my comments on him below) hasn't much liked any form of digital, though he concedes it is superior to analog in many obvious ways, other than the lack of ultimate resolution which he believes is the most essential feature of good audio. He believes digital will ultimately catch up, even if it hasn't done that so far, with higher sampling rates and bits of resolution to meet human sonic requirements--I completly agree with this! He seems to greatly prefer CD to SACD/DSD, pointing out that in many ways CD's have more actual resolution. He points out that DSD encodes 64 different states at 44.1kHz, whereas Redbook CD encodes 65536 different states at 44.1kHz. Then he also points out that modern deltat sigma DAC's for PCM also degrade the inherent information content and therefore resolution similarly, but no where near as much as DSD does. BTW, he also recommends only disc players with or without external DACS. His one test of an (unnamed, at the manufacturer's request) expensive digital server proved very disappointing. I find that quite believable as I know digital servers use asynchronous and highly jittery connections like USB (gag!) and ethernet, as well as being internally jittery in many ways. I had not been thinking that way myself recently, but reading him it makes sense. BTW, the relative information content in DSD will actually increase at lower frequencies…so at what frequency are they the same? 43 Hz! Below 43 Hz, fwiw, DSD will have more information than Redbook CD played by a non-oversampling DAC. Unfortunately for DSD, most of the music is above, not below, 43 Hz.
A fixture on the PS Audio blog, Elk, finds more disadvantages than advantages to DSD. He says that the purist form of DSD is incredibly rare. That would be the recording that stays in DSD all the way. DSD can't be edited or processed because there is no headroom (the full bandwidth S/N is 6dB). So in the real world, conversions are done between DSD and PCM, and those conversions cost more than the alleged benefit of DSD. Further he describes conversion to analog for editing purposes as very lossy. He says the only huge advantage of DSD is the ease of implementing an acoustically transparent anti-piracy code. He also explains how modern PCM and DSD are very similar, and that modern DSD ADC's and DAC's are really 4 bit (Romy says the opposite, that DSD has stuck with or gone back to 1-bit, when 4 bit DSD was better, but I believe Elk more). He says that PCM has wider bandwidth and S/N, as well as being more easily edited (by the recording engineer or the consumer--I loved to read that, he's ok with us consumers having freedom to edit) and converted to other formats. However he chickens out from saying hirez PCM sounds better. He says decide for yourself. Speaking of which...
Cranky veteran audio journalist Peter Aczel points out that neither high rez PCM nor DSD has proven to be audibly superior to correctly produced Redbook CD digital. He says the clinching research was done by Meyer and Moran and published in the September 2007 issue of the Journal of the Audio Engineering Society. He says that often SACD's or DVD-Audio discs sound better only because they are edited better. He does in fact buy and review SACD's and DVD-Audio's for that reason.
I like Peter Aczel. I like reading what he says. I feel my disagreement is small, but he might not agree, and YMMV. As far as I know, there has been no scientific proof that high rez PCM or DSD is better than Redbook PCM. It sounds quite believable to me. However, scientific proof is not all that we go by. It depends on how you value type 1 and type 2 errors. If you only want to use a higher rez format if it has been proven better, otherwise you don't want to waste your money, then you should listen carefully to what he says. OTOH, if you think something just might be better, by a tiny amount, even though it might not have been, or might never be, proven to be better, and you want it, and you're willing to take the risk that it might not actually be better, and you have the money for it, then you shouldn't follow his advice. And it's interesting that he doesn't always seem to follow his own advice. He owns far more expensive amplifiers and disc players than have been proven to be necessary. Sometimes he has rationalized this, along similar lines, that he wants equipment sufficiently better than the minimum requirement that he doesn't even have to think about it. He tends to go for the stuff that's either measurably better or designed according to the most sound engineering principles, as he see's them (which is pretty much like most audio scientists and engineers). That's not far off from what I do.
But I think it's very refreshing to realize, nonetheless, to realize that hardly any of the hoopla that audiophiles make big deals about has really been proven to be better. The reason is that this gives me the liberty to do my own thing, and now worry about the people telling me I must do such and such, or I'm not with it, not a real audiophile, or whatever.
With regards to statistically valid blind testing, I think it's a very good idea. It's hard work, though, which should not be underestimated. It has to be done 100% correctly to be valid. And that requires precise level matching, use of absolutely identically mastered material, and pre-determined number of trials to reach statistical significance (typically, this is about 25 trials, though if you were sure to get every trial correctly, 10 would do). Interpretation of the results depends somewhat on expectation. If you believe a huge, obvious, and incontrovertible difference exists, and you get p value well above 0.05, you should at least adjust your expectations. OTOH, if you think the difference is actually quite hard to hear, and predict that you might not get it the first time, a p value even as high as 0.2 could be "suggestive" and warrant further experiments (note that all experiments must be reported, not only the ones with good results).
I think differences that have been shown to be important in such testing should be top priorities. Differences which lack such proof probably are smaller and less important, if they even exist at all. We do know that correct frequency response is quite important. We don't know that the omitting the interpolation filters, as DSD does (for the 1-bit DSD which may not actually exist now, if it ever did, as many I've quoted above say that actual DSD implementations are typically 4 bit nowadays, while the 1-bit mantras continue) is very important. Being able to omit the interpolations filters probably isn't very audibly important, or Sony probably would have proven that it was, in statistically valid blind testing. They had the means and motive long ago, and didn't. So it's probably not a big deal, if it's even a deal at all. And since the DSP devices I like to use require PCM, which with modern DAC's available to me require interpolation filters, I will continue not to omit interpolations filters, by not making my system into a pure DSD system. And I will not feel guilty about it. And I will not even feel I'm not a member of the What's Best crowd, simply on those grounds anyway.
Often audiophiles argue from authority on the one hand, (so-and-so says DSD is far better than PCM, and they use the whatsit system which everyone knows is the best) and/or then uselessly say you should listen for yourself (which is no substitute for scientific proof either, and the result of a single listening test is about as informative as a coin toss). The knowledge that certain things have NOT been proven means that you or I are free to disagree (except over whether things have been "proven") and not feel cowed by self-proclaimed experts having far more expensive equipment, etc., then we will ever do. The benefits of that far more expensive equipment has not been proven either!
Charlie Hanson has now made some tests available in which both the PCM and the DSD come from the identically same analog masters. I think those are worth getting.
I believe it is actually true as Arthur Salvatore argues that 16/44.1 has more actual information. DSD is a kind of lossy encoding. But remember that half of the information in a 16/44.1 recording is above 10kHz, and the other half below. DSD does an increasingly good job of encoding the information below 10kHz as you get lower in frequency, it merely does a lousy job of encoding the information above 10kHz. I'm not exactly sure how to compute this, though my computation of the point where DSD and 16/44.1 have the same state information is 43 Hz. That's the point at which they both directly encode 65536 states. But there must be more to it than that. Somehow the DSD is shown to have less background noise below 10kHz, and I don't quite how to quantify that, though it's clear the lower noise is a result of the "noise shifting", and it jitters the information below as well as above. So if I said above that the noise shifting begins at 10kHz, that is incorrect. Actually, the noise shifting is through and through.
People who can watch NTSC on actual NTSC television, or multiscanning crt's like my Sony XBR-960--a true classic still worth having, will see it as far sharper than NTSC remapped to 720p or 1080i, despite those HD formats having more resolution. And this added sharpness, mostly appreciated, is real sharpness, not the fake sharpening by peaking the horizontal amplifier that sharpness controls generally do. Real sharpness is always better. Meanwhile the 720p or 1080i presentations of 480i will appear smoother, but it isn't the smoothness of something which isn't artificially peaked, it's the smoothness of blurred edges, and that's what the delta sigma conversion of PCM does.
[this post may continue to be edited and expanded]
Audiophile magazine reviewers are not going to say anything bad about DSD purveyors when they are large current and potential advertisers. But many reviewers who write for their own pleasure feel no such restriction. As well as manufacturers who don't use DSD anymore.
Linn Sondek is now calling DSD a good idea in 1999 but obsolete in 2013, followed by a 11 page discussion. Not as good as 24bit/192kHz, which can be played on Linn DS players since 2007. I have a number of Linn hybrid HDCD/SACD discs from the 2000's which sound reasonably good in either HDCD or SACD. Either way, I must use them on a universal disk player since my servers don't support HDCD or DSD. Haven't done enough testing to decide which layer I like better. My general feeling has been that properly decoded HDCD is about equivalent, or better, than SACD, and in fact I have more favorite sounding recordings on HDCD than SACD. HDCD played back without HDCD decoding is ok on Reference Recordings but mediocre on other labels, so if you haven't heard HDCD properly decoded, as it seems few audiophiles have, you haven't heard it, and this is incontrovertibly true.
Romy the Cat hates DSD with passion.
Arthur Salvatore (for whom I have very great respect, btw, more so than virtually any other audio reviewer, at fact value anyway--since I also respect John Atkinson but know he is very constrained by his commercial perch, so I read JA between the lines always and look at his measurements which I cherish, as well as Peter Aczel who has virtually opposite opinions on nearly everything--but see my comments on him below) hasn't much liked any form of digital, though he concedes it is superior to analog in many obvious ways, other than the lack of ultimate resolution which he believes is the most essential feature of good audio. He believes digital will ultimately catch up, even if it hasn't done that so far, with higher sampling rates and bits of resolution to meet human sonic requirements--I completly agree with this! He seems to greatly prefer CD to SACD/DSD, pointing out that in many ways CD's have more actual resolution. He points out that DSD encodes 64 different states at 44.1kHz, whereas Redbook CD encodes 65536 different states at 44.1kHz. Then he also points out that modern deltat sigma DAC's for PCM also degrade the inherent information content and therefore resolution similarly, but no where near as much as DSD does. BTW, he also recommends only disc players with or without external DACS. His one test of an (unnamed, at the manufacturer's request) expensive digital server proved very disappointing. I find that quite believable as I know digital servers use asynchronous and highly jittery connections like USB (gag!) and ethernet, as well as being internally jittery in many ways. I had not been thinking that way myself recently, but reading him it makes sense. BTW, the relative information content in DSD will actually increase at lower frequencies…so at what frequency are they the same? 43 Hz! Below 43 Hz, fwiw, DSD will have more information than Redbook CD played by a non-oversampling DAC. Unfortunately for DSD, most of the music is above, not below, 43 Hz.
A fixture on the PS Audio blog, Elk, finds more disadvantages than advantages to DSD. He says that the purist form of DSD is incredibly rare. That would be the recording that stays in DSD all the way. DSD can't be edited or processed because there is no headroom (the full bandwidth S/N is 6dB). So in the real world, conversions are done between DSD and PCM, and those conversions cost more than the alleged benefit of DSD. Further he describes conversion to analog for editing purposes as very lossy. He says the only huge advantage of DSD is the ease of implementing an acoustically transparent anti-piracy code. He also explains how modern PCM and DSD are very similar, and that modern DSD ADC's and DAC's are really 4 bit (Romy says the opposite, that DSD has stuck with or gone back to 1-bit, when 4 bit DSD was better, but I believe Elk more). He says that PCM has wider bandwidth and S/N, as well as being more easily edited (by the recording engineer or the consumer--I loved to read that, he's ok with us consumers having freedom to edit) and converted to other formats. However he chickens out from saying hirez PCM sounds better. He says decide for yourself. Speaking of which...
Cranky veteran audio journalist Peter Aczel points out that neither high rez PCM nor DSD has proven to be audibly superior to correctly produced Redbook CD digital. He says the clinching research was done by Meyer and Moran and published in the September 2007 issue of the Journal of the Audio Engineering Society. He says that often SACD's or DVD-Audio discs sound better only because they are edited better. He does in fact buy and review SACD's and DVD-Audio's for that reason.
I like Peter Aczel. I like reading what he says. I feel my disagreement is small, but he might not agree, and YMMV. As far as I know, there has been no scientific proof that high rez PCM or DSD is better than Redbook PCM. It sounds quite believable to me. However, scientific proof is not all that we go by. It depends on how you value type 1 and type 2 errors. If you only want to use a higher rez format if it has been proven better, otherwise you don't want to waste your money, then you should listen carefully to what he says. OTOH, if you think something just might be better, by a tiny amount, even though it might not have been, or might never be, proven to be better, and you want it, and you're willing to take the risk that it might not actually be better, and you have the money for it, then you shouldn't follow his advice. And it's interesting that he doesn't always seem to follow his own advice. He owns far more expensive amplifiers and disc players than have been proven to be necessary. Sometimes he has rationalized this, along similar lines, that he wants equipment sufficiently better than the minimum requirement that he doesn't even have to think about it. He tends to go for the stuff that's either measurably better or designed according to the most sound engineering principles, as he see's them (which is pretty much like most audio scientists and engineers). That's not far off from what I do.
But I think it's very refreshing to realize, nonetheless, to realize that hardly any of the hoopla that audiophiles make big deals about has really been proven to be better. The reason is that this gives me the liberty to do my own thing, and now worry about the people telling me I must do such and such, or I'm not with it, not a real audiophile, or whatever.
With regards to statistically valid blind testing, I think it's a very good idea. It's hard work, though, which should not be underestimated. It has to be done 100% correctly to be valid. And that requires precise level matching, use of absolutely identically mastered material, and pre-determined number of trials to reach statistical significance (typically, this is about 25 trials, though if you were sure to get every trial correctly, 10 would do). Interpretation of the results depends somewhat on expectation. If you believe a huge, obvious, and incontrovertible difference exists, and you get p value well above 0.05, you should at least adjust your expectations. OTOH, if you think the difference is actually quite hard to hear, and predict that you might not get it the first time, a p value even as high as 0.2 could be "suggestive" and warrant further experiments (note that all experiments must be reported, not only the ones with good results).
I think differences that have been shown to be important in such testing should be top priorities. Differences which lack such proof probably are smaller and less important, if they even exist at all. We do know that correct frequency response is quite important. We don't know that the omitting the interpolation filters, as DSD does (for the 1-bit DSD which may not actually exist now, if it ever did, as many I've quoted above say that actual DSD implementations are typically 4 bit nowadays, while the 1-bit mantras continue) is very important. Being able to omit the interpolations filters probably isn't very audibly important, or Sony probably would have proven that it was, in statistically valid blind testing. They had the means and motive long ago, and didn't. So it's probably not a big deal, if it's even a deal at all. And since the DSP devices I like to use require PCM, which with modern DAC's available to me require interpolation filters, I will continue not to omit interpolations filters, by not making my system into a pure DSD system. And I will not feel guilty about it. And I will not even feel I'm not a member of the What's Best crowd, simply on those grounds anyway.
Often audiophiles argue from authority on the one hand, (so-and-so says DSD is far better than PCM, and they use the whatsit system which everyone knows is the best) and/or then uselessly say you should listen for yourself (which is no substitute for scientific proof either, and the result of a single listening test is about as informative as a coin toss). The knowledge that certain things have NOT been proven means that you or I are free to disagree (except over whether things have been "proven") and not feel cowed by self-proclaimed experts having far more expensive equipment, etc., then we will ever do. The benefits of that far more expensive equipment has not been proven either!
16bit 44.1kHz digital better sounding than DSD ?
After writing much of above and thinking, my best answer is, I don't know, but I am beginning to think so. For the last 8 years or so, I had been assuming that SACD was better than CD. I didn't actually believe that until I got my first SACD player. Then I let myself get taken in. I haven't done serious testing, and I suspect it would have to be very serious testing indeed. And very difficult to do, since the identical masterings are almost never available, even on the same disc. And to be "identical masterings" they would both have to come from some technology superior to both (live recording, certainly, or DXD?) or at least identical (analog tape, which may or may not be better, though I actually believe it is in the highest resolution available, say 1/2 or larger tracks at 30ips). If you simply take the DSD and convert to PCM, you are not hearing all that the PCM is capable of, because the initial translation or recording to DSD is one major limiting factor. Making a DSD recording throws away at least half of the information that would ultimately go to the PCM, even in 16/44.1 !!! Sony, which owned the masters for the pure and hybrid SACD's it made certainly had no incentive to do this. One of my friends strongly believed the CD layer on hybrid SACD's was inferior to previously existing CD's, and I now think he was probably right. Most likely Sony did in fact get the CD layer by conversion from the DSD, and that would not be "fair" to the CD layer.Charlie Hanson has now made some tests available in which both the PCM and the DSD come from the identically same analog masters. I think those are worth getting.
I believe it is actually true as Arthur Salvatore argues that 16/44.1 has more actual information. DSD is a kind of lossy encoding. But remember that half of the information in a 16/44.1 recording is above 10kHz, and the other half below. DSD does an increasingly good job of encoding the information below 10kHz as you get lower in frequency, it merely does a lousy job of encoding the information above 10kHz. I'm not exactly sure how to compute this, though my computation of the point where DSD and 16/44.1 have the same state information is 43 Hz. That's the point at which they both directly encode 65536 states. But there must be more to it than that. Somehow the DSD is shown to have less background noise below 10kHz, and I don't quite how to quantify that, though it's clear the lower noise is a result of the "noise shifting", and it jitters the information below as well as above. So if I said above that the noise shifting begins at 10kHz, that is incorrect. Actually, the noise shifting is through and through.
Still, DSD has more time Resolution
Thinking about the high sample rate…it still occurs to me that 2.8Mhz DSD has more time resolution. The very beginning of a sonic event is not lined up to the coarse grid of lower sampling rate PCM. This is a breath of the original sonic even that gets through pure DSD, if such a thing exists. OTOH, if you have a DSD converter based on 3 bits, say, then you have to divide the nominal sampling rate by that number to get the real time resolution. Any conversion, even conversion to 386kHz sampling rate DXD, will squash that resolution back down to that point. Then you've only again got a marginal increase in time resolution than DSD, and all the additional time distortion afterwards that DSD adds in the name of noise shaping. So while the very start of an acoustic even can ideally have more resolution, it gets smeared after that. Meanwhile, the regular time grid of PCM has a kind of coalescing and sharpening effect, as everything is combined to start at each sampling interval, whether it actually started a little before or less is lost. However, there again, with sigma delta PCM dacs, and even through it's kind of fake, the sharpening is removed, and there may even be a bit of apparent decoalescing of multiple events all having to start at the same instant, but that decoalescing is fake, the information really isn't there. Running PCM through a sigma delta DAC (preferably multibit, of course) will make it smoother, without any actual increase in resolution.People who can watch NTSC on actual NTSC television, or multiscanning crt's like my Sony XBR-960--a true classic still worth having, will see it as far sharper than NTSC remapped to 720p or 1080i, despite those HD formats having more resolution. And this added sharpness, mostly appreciated, is real sharpness, not the fake sharpening by peaking the horizontal amplifier that sharpness controls generally do. Real sharpness is always better. Meanwhile the 720p or 1080i presentations of 480i will appear smoother, but it isn't the smoothness of something which isn't artificially peaked, it's the smoothness of blurred edges, and that's what the delta sigma conversion of PCM does.
[this post may continue to be edited and expanded]
No comments:
Post a Comment