DSD DAC

[Dusty Chalk]

I still think everything comes down to the mixing and mastering.

Welcome to team source material first ueber alles.

[Filburt]

I still think everything comes down to the mixing and mastering. For some reason, they took better care of them when the final product was going to be a DSD release than other formats. Good recordings sound awesome on any format.

 

[Note: In the interest of brevity and helpfulness, I sacrificed precision and detail. Also I just got bored trying to correct anything, and I noticed something nearly as soon as I posted it. So...whatever at this point; probably close enough]

 

The Scarlet Book standard (SACD) effectively forces 'better' behavior on the part of production engineers, while building in more margin for error. Anecdotally, I've heard that the peak meter on a lot of PCM recording and processing tools simply reports the instantaneous amplitude of incoming samples. Unfortunately, samples rarely fall along the maxima or minima of the incoming signal, which means analog peak is often in excess of the recorded peak value. Note that this does not mean that analog peak is unrecoverable; the familiar strictures of sampling apply. What it does mean is that the downstream playback device needs a certain amount of headroom to avoid clipping, and/or some means of rescaling the input. Historically, most consumer digital playback devices (e.g. CD players, 'DACs,' and so forth) made no such accommodations; meaning, you'll get clipping. One notable exception is that the PMD100 does rescale on the order of about 1dB, so devices using this filter do potentially have some headroom.*

 

The development of the SACD standard took a more systems-oriented approach and ultimately imposes a measure of conservatism in attempting to maximize peak and average amplitude. It restricts what can pass as "legal" data for the purposes of producing an SACD; that is, it imposes a concrete penalty (the SACD won't go to press) if one passes illegal data. Chiefly, it restricts peak modulation depth, as measured by an unweighted moving average of uh...I think it's 28 samples (I don't have the specification in front of me so I'll have to go by memory). A sort of industry 'rule of thumb' for DS modulators is that they are adequately linear up to about a 50% modulation depth, and 50% is what was ultimately chosen for 0dBFS. In practice, acceptable linearity tends to persist somewhere in excess of 50% MD for most modulators in commercial devices, while the SB restriction works out to (IIRC) something like 71% or about +3dBFS. So, some amount of headroom above 0dBFS can be built-in on the playback side if following the specification, though note that this comes at some cost to maximizing e.g. SNR specification at 0dBFS, so I suppose it isn't guaranteed that a manufacturer will opt to do so.** Secondly, a simple unweighted average in this case is going to integrate out of band noise and as a result will overestimate analog peak, with this effect tending to increase with the use of dynamic compression. Third, 0dBFS is still the canonical target peak, on the presumption that downstream headroom will not be built-in. If you combine these factors, what you get is some reduction in the likelihood of clipping on the playback side (though the degree to which this is the case will vary between devices), due to (1) the potential of violating the restrictions when following popular conventions in PCM recording, and (2) likelihood of having some measure of headroom built into the playback system. So, given these considerations, why did I express the prior opinion in the other post? Because sensible restrictions can be built into a PCM recording system (and safeguards in the playback system), too, and result in superior performance.

 

*Another example would be a 'NOS' R-2R DAC in cases where the reconstruction filter has adequate headroom. I don't recommend this solution, though - such DACs do not adequately filter aliases, which in turn intermodulate with the passband to produce in-band distortion and noise.

**Nor are manufacturers necessarily forthcoming about whether they opted to do so or not. Note that the ESS Sabre's modulator is linear beyond 71% MD and, as I recall, this particular converter does build in headroom although, so far as I understand, it does so for both DSD and PCM data.

Edited June 6, 2013 by Filburt

[blessingx]

LampizatOr DSD-only DAC http://feedly.com/k/18Wynva

http://www.lampizator.eu/newdac/lampizator/DSD_DAC.html

Edited June 8, 2013 by blessingx

[Hopstretch]

LampizatOr DSD-only DAC http://feedly.com/k/18Wynva

http://www.lampizator.eu/newdac/lampizator/DSD_DAC.html

 

Probably full of Albanian sperm.

[Hopstretch]

 

Antelope Audio, long respected in pro circles, showed two important products: the finally available Rubicon Atomic AD/DA preamp ($40,000), an all-in-one beauty that combines a 10M Rubidium atomic clock with a 384kHz converter, phono preamp, and headphone amplifier; and the due-this-fall Zodiac Platinum DSD-capable DAC/headphone amplifier ($4895) with optional Voltikus power supply ($995). Paired with ATC SCM100-AT active loudspeakers ($35,000/pair), the Rubicon produced supremely beautiful sound with exceptionally refined highs. And that was from a computer source equipped with a stock USB cable. Those who have experimented with aftermarket USB cables know how much more color and life the system would have produced had a better USB cable been in the chain.

Edited June 9, 2013 by Hopstretch

[spritzer]

Yes, distortion is a wonderful thing. 

 

There was a good USB comparison in the latest Hifi News... 

[Grahame]

Or you could attempt to measure the difference ...

http://archimago.blogspot.ca/2013/04/measurements-usb-cables-for-dacs.html

[Dusty Chalk]

Proving myself prescient yet again, it appears the Light Designs Da Vinci can play DXD files as per the Google syndication ad, but not per the product page proper.  WTF?

... but if you have the files there are commercial DXD DACs to play them. 
 
DXD is usually 352.8/24
DSD is usually 2822.4/1

Alright, maybe not so prescient, but honestly, I missed this post before.

 

PS  Emphasis mine.

[Filburt]

LampizatOr DSD-only DAC http://feedly.com/k/18Wynva

http://www.lampizator.eu/newdac/lampizator/DSD_DAC.html

 

Uh...you don't get "purity" by using a mechanically simpler conversion process in digital audio. You will probably get various types of errors (noise, distortion, flatness, linearity), which are not generally considered to contribute to "purity" of the reconstructed output. I'm not sure what is meant by converting "without silicon" in this case.

[Dusty Chalk]

DSD can be converted to analog directly, without a DAC.  That's probably what he's doing.

 

And according to the graph, there's plenty of room between the signal proper and the ultrasonic noise to filter it out.  I'd be interested in hearing the result.

 

The hard part is actually filtering out the DSD signal alone, without the metadata surrounding it as would probably exist in any digital transfer protocol, especially right off the disk or out of the file, and making sure it's decrypted, although there's probably a point where one can tap the signal somewhere along the chain.  I bet it's right there at the DAC, though, lollers.

[Nebby]

I was trying to remember where I had read about Lamizator before, but I found it. He was the one that did the wonky buffalo sabre32 dac this way:

 

[Filburt]

DSD can be converted to analog directly, without a DAC.  That's probably what he's doing.

 

And according to the graph, there's plenty of room between the signal proper and the ultrasonic noise to filter it out.  I'd be interested in hearing the result.

 

Which graph?

Edited June 10, 2013 by Filburt

[spritzer]

I was trying to remember where I had read about Lamizator before, but I found it. He was the one that did the wonky buffalo sabre32 dac this way:

 

 

Somebody sends this man money while not being under duress?   Holy fuck people are gullible... 

[Dusty Chalk]

Which graph?

This one:

 

 

From the interview with John Siau from Benchmark on the previous page, in which he claims it's impossible to filter out without destroying the purity of the signal (my paraphrase).

[complin]

What a quality piece of work, probably most of the workforce are not yet school age   

 

Somebody sends this man money while not being under duress?   Holy fuck people are gullible... 

[Filburt]

From the interview with John Siau from Benchmark on the previous page, in which he claims it's impossible to filter out without destroying the purity of the signal (my paraphrase).

 

Ah, OK. I definitely wouldn't describe that as 'plenty of room'. The floor is already at -100 by 30K. If using a digital filter, it's enough room that you can keep 20K out of the transition band and still have ample stopband attenuation by 26K. However, if you're using an analog lowpass, you won't get comparable attenuation without using a high-order filter that would assuredly breach the "purity" ethic.

 

In any case, I wouldn't put much stock (essentially none) in the claim that DSD can be converted with just a lowpass filter. If you physically attach an SACD to an analog lowpass filter, you aren't getting music on the other end. You still have to (1) obtain a code-equivalent voltage output (since the physical disc inconveniently fails to supply it outright), and (2) send that output to a lowpass filter. All digital audio, once you have the code-equivalent voltage output, can be "converted with just a lowpass filter" (hence the term "reconstruction filter"). Obtaining said output is precisely what one uses a DAC to do. As far as I am aware, real-world converters (multibit delta-sigma) run DSD through their modulator loop and DSP just like anything else, in order to gain anything like 'high resolution' performance, so Sony's block diagram is inapplicable even when one ignores the realities about how DSD is handled on the production side. I guess you could try some alternatives, although it would be difficult to avoid something functionally very similar, if not analogous, while still getting worthwhile performance. I certainly wouldn't describe a modern class D amplifier as architecturally minimalist, or analogous to what I think people interpret Sony's marketing claim to entail. Once factoring in the practical reality of taking in data and processing it, such an amplifier would include most (if not all) of the functional architecture of the aforementioned converters. The closest thing I can think of to Sony's idea would be a switch driving a filter, but this would essentially be the back-end of an uncontrolled switching amplifier; the performance is likely to be rather abysmal (and I'm not sure, in any relevant sense, 'pure').

Edited June 11, 2013 by Filburt

[spritzer]

The word pure does mean very different things to people.  Some claim it applies to simple amps even if the end performance of a single fet or a triode could be abysmal.  To some it is a short signal path, never mind if it is a chip amp with hundreds of components.  Then we this craziness...

[Dusty Chalk]

Filb -- But see, just because you're talking so much about it tells me you're thinking about it and the next thing you know you're going to try it because being an engineer that's what you do. Don't.

Okay that said, but if you do, why does the analog filter need to be high order? All it needs to do is match the increase in noise with the equivalent attenuation.

And I think 90% of the appeal to go DACless is the mental exercise/thought experiment/realization that DSD, once filtered, corresponds to the actual signal in the same way that digital amps drive speakers. Yes there's more to it than that, but that's probably why he won't ever put it to market.

Sent from mah phone-blet via Tapatalk

[Filburt]

If you find (most likely, experimentally) that having little (if any) attenuation of noise doesn't have excessively deliterious effects, I guess you could use something like a second-order filter simply to avoid passing noise at f > 100KHz. If you wanted attenuation by the time the noise floor starts to rise (~26KHz in that graph), or even by the end of that graph, you'd need a very high-order filter. A digital filter would make this operation trivial.

 

Second, my earlier point was that DSD isn't any more 'analog' than PCM, and it doesn't bring one substantively closer to 'DACless' in any practical sense, outside of marginal cases (that I'd still be tempted to call d/a conversion) that no one interested in audio fidelity would use. Also, if the 'digital amplifier' (i.e. class D) uses a multi-level modulator, it might run it through the modulator loop anyway as well (in addition to whatever else would be done in DSP). In Sony's idealized case (which is rarely the actual case) one at best is able to omit the modulator from the playback device, which systemically just means you've functionally divided the d/a process between devices. Calling this process 'DACless' is roughly as credible as my describing running decoded PCM (relative to the device in question) into a non-segmented ladder DAC as a 'DACless' process.

[Dusty Chalk]

If you find (most likely, experimentally) that having little (if any) attenuation of noise doesn't have excessively deliterious effects, I guess you could use something like a second-order filter simply to avoid passing noise at f > 100KHz. If you wanted attenuation by the time the noise floor starts to rise (~26KHz in that graph), or even by the end of that graph, you'd need a very high-order filter. A digital filter would make this operation trivial.

 

Second, my earlier point was that DSD isn't any more 'analog' than PCM, and it doesn't bring one substantively closer to 'DACless' in any practical sense, outside of marginal cases (that I'd still be tempted to call d/a conversion) that no one interested in audio fidelity would use. Also, if the 'digital amplifier' (i.e. class D) uses a multi-level modulator, it might run it through the modulator loop anyway as well (in addition to whatever else would be done in DSP). In Sony's idealized case (which is rarely the actual case) one at best is able to omit the modulator from the playback device, which systemically just means you've functionally divided the d/a process between devices. Calling this process 'DACless' is roughly as credible as my describing running decoded PCM (relative to the device in question) into a non-segmented ladder DAC as a 'DACless' process.

Okay, backwards -- okay yes, by comparing to a ladder DAC, I see that you get it.  I just wanted to make sure you weren't in your mind still putting a DAC in there somewhere, but the point is, a DSD point-to-point DAC is a lot simpler than a point-to-point ladder PCM DAC.  So keyword in your sentence there is "roughly".  You don't have to implement 2^n "addition" with the DSD DAC, so it's one less step, and to many, one particularly key less step.

 

A digital filter ... in DSD?  Someone (disassociated from Sony) needs to work up the maths for that.

 

And I'm perfectly alright with some mild signal attenuation at 26kHz.  I mean really, who cares?  Alright, the dog might care.

 

I am of the school that ultrasonic frequency matters, but I am definitely not of the school that ultrasonic frequencies matter much.  You can have attenuation at 26kHz with a 2nd order filter and I'd be perfectly happy.  Hey, let's go wild and make it with a tunable cutoff frequency.

[Filburt]

The point regarding filters isn't that you can hear ultrasonic noise; it's about avoiding audible intermodulation of noise with your passband. Ideally, the noise is uncorrelated but in practice cyclical errors are likely to show up in a 7th order 1-bit modulator as you can't employ dither. One would have to find out experimentally if this turned out to be audible. Digital filtering is already used in DSD playback; you can see that in datasheets for DACs that support DSD playback (and probably in the case of 'digital amplifiers' that support DSD playback, if they specify such things in their datasheets). However, the design of the filter may vary, particularly depending on how DSD is handled by the specific DAC. Aside from that issue, I disagree that DSD is relevantly 'simpler' except when working with very loose, largely undescriptive metaphors for the functional architecture of an audio system. That is, if one thinks that DSD permits a 'dacless' playback architecture, one has made too much of the notion of a DAC. DSD is not more of an 'analog' format than PCM, nor does it in some intrinsic sense somehow get you closer to a "pure" representation of 'the signal.' I somewhat obliquely suggested this earlier, but audio systems are probably best modeled as control systems. The 'signal' is a metaphor for an idealized reconstruction of the input; the sense in which the signal is in the format is relational to the system. Under Sony's functional diagram, all that is happening is the functional blocking is done slightly differently than CD Audio, such that the D/A conversion block is divided differently, where the front-end and back-end of a delta-sigma DAC is physically split. Sony's functional diagram isn't representative of real-world implementations of DSD playback, in any case, but this is essentially what that diagram represents.  Certain functional blocks of the total system are distributed in time and place, but they are all relatable in terms of a total system, and whether the reconstructed output of the playback device is 'pure' is only really intelligible within the confines of thinking about the system.  It makes little sense to describe a playback device, that is lacking ordinary features of the control system, as providing a more pure output simply on the basis of lacking such features. It may be a mechanically simpler device, but omitting certain processing (such as filtering, noise shaping/dither, and so forth) may effectively represent a failure in the control system and a less accurate or 'pure' reconstruction of the input. Therefore, you cannot tell that a DAC design contains superfluous parts, or is likely to adulterate 'the signal,' simply on the basis of its complexity and the extent of processing involved. You can only tell when taking into account the total system. People generally get around this caveat by inserting a tacit ceteris paribus clause - that is, all other things being equal; but that's the problem, they aren't. Design requirements are domain-specific, and simplicitly is not a particularly good universal indicium of fidelity in audio design. That doesn't mean that all complex designs are good (or that complexity itself is good); it means evaluation is domain-specific. In the specific case of DACs, simplicity often leads to a lower-fidelity reconstruction of the original recording (subject to production variables like post-processing). To the extent that DSD allows for 'simplicity,' it does in the sense that you could (in theory) use it as a control signal for what is essentially an open-loop switching amplifier, which is architecturally "simpler" in its most basic sense than an R-2R latter. However, as I noted before such a device would have abysmal performance that is not in any relevant sense 'pure' or architecturally equivalent to high-performance switching amplifiers designed for audio reproduction.

Edited June 12, 2013 by Filburt

[Dusty Chalk]

I stumbled upon a couple links of places that sell DSD files, not sure which have been posted before, so I'll just post all three again if anyone's interested.

 

DSDfile

Blue Coast

E-Onkyo (anyone read Japanese well enough to tell us exactly what to click?)

The point regarding filters isn't that you can hear ultrasonic noise; it's about avoiding audible intermodulation of noise with your passband. Ideally, the noise is uncorrelated but in practice cyclical errors are likely to show up ...One would have to find out experimentally if this turned out to be audible.

I bet you're exactly right.  I mean, all the DAC manufacturers (in this case DSD DAC manufacturers) pride themselves on their own way of filtering out the ultrasonic noise (amongst everything else), and that's an art.  I bet "find out experimentally" is exactly what he did, and the answer is, "it is audible", and that's why he'll never put it up for sale.

 

I had forgotten about that, thanks for reminding me.

 

And by the way, I completely understand and agree with your point about complexity -- there's a difference between complexity and necessary elements of a design.  As an unrelated example, I think most people don't understand how oversampling actually works, and that they usually introduce zeros in the data between the actual data and use filters to fill in the signal between the points.  (I know I used to.  :$  )

[Filburt]

Based on the datasheets it looks like, at least for the PCM1792, AD1955, and ES9018, DSD is run through both the DSP and modulator. That would make sense, considering the output stage is designed for a multibit input. They also use DEM, which can be used as an additional form of noise shaping. Interestingly, only ESS indicates a filter resembling the analog filter specification from Sony.

 

On the subject of complexity, I'd like to revisit the 'lowpass filter' claim, having now introduced the notion of audio as a system. The recording format in idealized terms is a means of feedforward control of the playback device by the recording device. However, this form of control requires that the playback device is adequately modeled. Sony's marketing takes advantage of the perception that lowpass filters are typically more homogenous in behavior across particular cases, and closer to ideal, as compared to DACs. In other words, they are claiming that DSD will, in practice, be a superior feedforward system compared to PCM. Sony's description, however, is incomplete. The full description is that, given an ideal voltage source, all you need is a lowpass. This claim, however, is true of any digital audio system; PCM, DSD, or otherwise. Likewise, the truncated claim is false - SACDs are not voltage sources, nor is a format in and of itself. In the case of DSD, the modeled device is an ideal switch; the noise shaping in DSD is modeled for that ideal. What is the closest thing we have, in practice? The same thing as in the case of PCM - a DAC

[Grahame]

And in other news,  some smart folks have thought outside the box, and combined DoP, FLAC, and existing software  to give you 

http://forums.slimdevices.com/showthread.php?98978-SB-Touch-amp-DSD-with-a-DoP-DAC!

via

http://www.computeraudiophile.com/f11-software/playing-dsd-any-players-including-squeezebox-touch-16346/

 

Exec summary 

DoP = DSD over PCM - play the DSD bits, (not converted as), but packaged as PCM - so you can use S/PDIF or toslink as the transport mechanism.

 

FLAC is a container with well a defined tagging ecosystem.

 

So tag + playback using software of choice - output as PCM - PCM input interpreted as DSD ...  looks fun stuff

 

I've played with the DSD->PCM conversion in foobar2K so I can play some DSD test tracks back on a PCM dac  - not that you'd want to , more as a proof of concept .

 

Lacking a DoP capable DAC , I can't test this out at the moment - but I like the idea.

 

So Benefits are existing PCM playback software / hardware (e.g. SB Touch) but will drive a DoP capable DAC ....

 

Would only work with bit perfect playback - so no replaygain , software volume / EQ - as the PCM data isn't really PCM Audio so can't be manipulated as such.

 

Gapless might work ...

 

Interesting stuff 

[Sherwood]

I picked up a Mytek 192 DAC a while back, and I've really been enjoying it with DSD files (.iso or .dsf).  I've got it set up on firewire AND usb, out of the same computer.  It's fun to switch inputs when using different programs.

 

System output is via firewire, and ASIO output is via USB.  For whatever reason (likely drivers), the USB sounds better via ASIO than the Firewire.  I truly expected the opposite.  I have no illusion that data cables make any difference if they're up to spec, but for those that do both the USB and the firewire cables are Audioquest Forest (because they look nice.)