To DSD or not to DSD?

[Hiro]

Full disclosure: I've known Mark Waldrep for years, have been to his studio a few times, and hang out with him now and then at various audio events. He has strong opinions and he has good reasons for them. And I own a Benchmark DAC2. I found Mark's interview with Benchmark's John Siau pretty interesting, even though it does have an anti-DSD slant.

 

It really does, doesn't it? Even the questions themselves seem to suggest an already negative answer.

 

I wonder if the true reason for not enabling 128x and 256x DSD playback on the Benchmark DAC wasn't really the fear that it could lead to even better DSD sound, with even more pronounced difference between DSD and 2fs ~ 4fs pcm formats.

[Hiro]

Somehow, even with all the high frequency noise, etc... That does not take away the fact that even more people who try DSD, prefer its sound to PCM and it justifies the fact that DSD has been getting more attention every time in the different forums and in the major publications. In the end, we believe that these users are trusting their ears more than technical numbers and that is all that matters.

 

Speaking of trusting one's ears...

 

Any Metric Halo LIO-8 users here ? ..... - Page 3

[bdiament]

Hi Hiro,

 

Barry, if you are using Metric Halo converters for Soundkeeper recordings then you're already using the format. They are 128x oversampling sigma delta A/D and D/A converters.

 

Not quite. In fact, not by a long shot. There is more to it than just the sigma delta aspect.

I find recordings made with and/or played back via the Metric Halo don't have that discomforting characteristic. On the contrary, unlike any DSD or SACD I've heard, the PCM results (to my ears) "get out of the way".

 

I understand you feel differently, as do many folks. Enjoy your music any way you enjoy it.

 

Best regards,

Barry

Soundkeeper Recordings

Barry Diament Audio

[Hiro]

Just that all your modern "PCM" ADCs are SDM converters that internally convert SDM to PCM (with crappy algorithms) and all your modern "PCM" DACs are SDM converters that internally convert PCM to SDM (again with crappy algorithms). That's the problem DSD is trying to avoid by having SDM end-to-end...

 

Downsample something like 128fs SDM to 2fs or 4fs PCM and they are happy. Play to them a native 128fs SDM recording, and tell them it's DSD, and the problems start

[hardline]

Fellow forum members allow me to share my experience last month when I had the pleasure of attending the NY Audio Show where I experienced first hand first rate audiophile gear. Proprietors pushed all types of wares, along with their commensurate technologies, onto the mass of audiophiles who showed up with eager and open ears. This Audio Show, to me, clearly demonstrated the power and sensibility of the "computer audiophile" who can at several mere clicks of an input device (be it mouse or keyboard) could summon music among several thousand (if not tens of thousand) tracks for instant gratification. I became acquainted with the Mytek and Auralic Vega DACs - Channel D’s - puremusic programs.

 

One big take away from this show is that we, all of us - be it principals of audio-related hardware-software concerns, owners of music labels, or consumer/audiophiles- have similar stakes in this DSD debate. The concerns can include financial, sonic and the overall aesthetic ones - the goals of aural satisfaction.

 

We are fortunate to have the energy and passion of leaders and principals such as Dr. Waldrep (who I believe also chaired some events at the NY show), Miska, and Barry Diament (to name a few) participate in this discussion. So when there is an audience and a sizable forum (thanks to Mr. Computeraudiophile) we should engage in discourse - not so much to relish victory - but to get to the heart of the matter and advance the cause of furthering the audiophiles’ goal of enjoying music.

 

Now (back to some business) in relation to this issue of investments (stakes), I would like to bring up an undercurrent expressed by, it so happens, two designer-principals of DSD-enabled Dacs: John Siau and Charles Hansen . Mr. Siau contends that some of Sony’s marketing materials are not so forthright (see interview link in first post) while Mr. Hansen goes a little bit further and makes the claim (see concurrent thread http://www.computeraudiophile.com/f6-dac-digital-analog-conversion/ayre-wants-%241-5k-dsded-qb-9-a-15650/index6.html#post226219 please refer to this concurrent thread for some of Mr. Hansen’s most illuminating posts and reply-posts concerning DSD and PCM which we probably do NOT want to repeat nor rehash here) that Sony has been downright deceptive (and may be worse).

 

As an aside it is interesting that both commented on the DSD option as just that an “option”; DSD apparently is the “new wave” and this would be a nod towards the format’s popularity and a hedge against obsolescence. I believe in the sincerity and integrity of these two individuals. The equipment associated with these two individuals have been highly praised for form and function. Although I have not listened to their DSD enabled models, I have no reason to doubt that these would also be among the highly rated. If I am misrepresenting their positions please let me be corrected. On the other hand I’ve had so many bitter experiences with the megaloptic behemoth called Sony that I feel the need to follow The Who’s directive and “Don’t get fooled again."

[Hiro]

Now (back to some business) in relation to this issue of investments (stakes), I would like to bring up an undercurrent expressed by, it so happens, two designer-principals of DSD-enabled Dacs: John Siau and Charles Hansen . Mr. Siau contends that some of Sony’s marketing materials are not so forthright (see interview link in first post) while Mr. Hansen goes a little bit further and makes the claim (see concurrent thread http://www.computeraudiophile.com/f6-dac-digital-analog-conversion/ayre-wants-%241-5k-dsded-qb-9-a-15650/index6.html#post226219 please refer to this concurrent thread for some of Mr. Hansen’s most illuminating posts and reply-posts concerning DSD and PCM which we probably do NOT want to repeat nor rehash here) that Sony has been downright deceptive (and may be worse).

 

You're talking as if all the information presented in DVD-Audio marketing brochures was true, starting from the 144dB performance of 24bit DACs (THD of the best TI ladder type 24bit converter PCM1704 measured at 17bit), to the supposed sonic bliss of 192kHz (see John Siau's opinion on the marketing claim here ).

[Jud]

It really does, doesn't it? Even the questions themselves seem to suggest an already negative answer.

 

I wonder if the true reason for not enabling 128x and 256x DSD playback on the Benchmark DAC wasn't really the fear that it could lead to even better DSD sound, with even more pronounced difference between DSD and 2fs ~ 4fs pcm formats.

 

So Benchmark would purposely sabotage the opportunity for best performance of their own DAC because of an irrational dislike for the format that would enable it?

 

Come on. No reason to try to impugn someone's motives with wild conspiracy theories just because they don't share your views.

[Hiro]

So Benchmark would purposely sabotage the opportunity for best performance of their own DAC because of an irrational dislike for the format that would enable it?

 

Come on. No reason to try to impugn someone's motives with wild conspiracy theories just because they don't share your views.

 

Jud, when Benchmark enables 128x and 256x DSD on DAC2 they will dispel my doubts... I'm waiting to be proven wrong on this one.

 

PS Take note that people who listen to DSD 128x on their USB DACs report that it sounds better than 64x, yet the feature wasn't implemented by Benchmark. Five/Four Recording Engineer Robert Friedrich was quoted as saying "Taking DSD from 2.8 MHz to 11.2 MHz (256x) doesn't just step it up to the next level, it catapults it!". If they are not scared of DSD, why not enable all three sampling rates, like the exaSound has done it?

[Miska]

This is because you cannot dither perfectly 1 bit SD. Whatever you do, you will have distortions;

Just read this: http://sjeng.org/ftp/SACD.pdf

 

I'm bored to argue with this, it's just wrong. SD is not dithered in parallel bits period. And you are obviously mixing bits and levels...

 

(and to be exact, noise-shaped PCM is not dithered in LSB either, only boring inefficient flat dithers like TPDF)

[Jud]

Jud, when Benchmark enables 128x and 256x DSD on DAC2 they will dispel my doubts... I'm waiting to be proven wrong on this one.

 

PS Take note that people who listen to DSD 128x on their USB DACs report that it sounds better than 64x, yet the feature wasn't implemented by Benchmark. Five/Four Recording Engineer Robert Friedrich was quoted as saying "Taking DSD from 2.8 MHz to 11.2 MHz (256x) doesn't just step it up to the next level, it catapults it!". If they are not scared of DSD, why not enable all three sampling rates, like the exaSound has done it?

 

Yes, I'm sure every DAC builder that allows max 192k PCM input does so for fear their products will sound way too good at 384.

[Hiro]

Yes, I'm sure every DAC builder that allows max 192k PCM input does so for fear their products will sound way too good at 384.

 

They are doing it because they either believe there's no sonic benefit in it or their DAC chips don't support this sampling rate. We already know that the ESS Sabre chip Benchmark is using in DAC2 to do the D/A job does all three DSD sampling rates, so it can't be the compatibility issue. Now, does Benchmark believe that there's no sonic benefit in going above 64fs DSD (namely to 128fs and 256fs DSD)?

[Miska]

Just read this: http://sjeng.org/ftp/SACD.pdf

 

Funny thing is, the distortions presented in the paper are way below -120 dB which is lower than 90+% real world DAC devices manage for their analog stages...

 

It is so lovely when those guys are so disconnected from reality of real world.

 

Plus for the section (3) "further comments" is completely broken in multiple ways. It is in fact hilarious by for example becoming self-recursive.

 

Anyway, I'm fine with either format. My software can output ninth-order noise shaped 32-bit PCM at 1.5 MHz or seventh-order 1-bit SDM at 24.576 MHz. There are no real world DACs that can take former (?), but there are for handling the latter one...

[Jud]

They are doing it because they either believe there's no sonic benefit in it or their DAC chips don't support this sampling rate. We already know that the ESS Sabre chip Benchmark is using in DAC2 to do the D/A job does all three DSD sampling rates, so it can't be the compatibility issue. Now, does Benchmark believe that there's no sonic benefit in going above 64fs DSD (namely to 128fs and 256fs DSD)?

 

Or that there isn't a large enough market to make it cost-effective for them to do so.

[Hiro]

Or that there isn't a large enough market to make it cost-effective for them to do so.

 

Every DSD64 recording recorded to date can be upsampled to DSD128 or DSD256 with the use of DSD-capable software player like the Signalyst, so the potential market for this is as big as the current SACD/DSD64 market. Or even greater, as there are first audiophile labels like Five/Four, Opus3, MA Recordings, Blue Coast that are starting to offer 64fs+ DSD content.

[Jud]

Every DSD64 recording recorded to date can be upsampled to DSD128 or DSD256 with the use of DSD-capable software player like the Signalyst, so the potential market for this is as big as the current SACD/DSD64 market. Or even greater, as there are first audiophile labels like Five/Four, Opus3, MA Recordings, Blue Coast that are starting to offer 64fs+ DSD content.

 

The point is not so much whether you are convinced, but whether enough other people can be relied on to feel the same way that it becomes worthwhile for Benchmark to risk their smallish company on the possibility.

[luka]

That's only for the DAC side. And it shows up as non-linear distortion behavior. Main reason for them doing it is simplification of analog reconstruction filter and that they don't have enough processing power inside DAC to have a proper 1-bit SDM. And that really shows up. PCM4201, PCM4202, PCM4204 and PCM1804 are true 1-bit ADCs. For ADC they can afford to do 1-bit because they don't need to have a digital modulator. But although those ADCs can output 192/24 PCM too, you get better result by capturing the 6.1 MHz DSD stream and then converting it to PCM in a computer using good algorithm.

 

Not sure I agree with your reasoning here Miska. A 1-bit SDM is not difficult to do in silicon.

 

As I understand it DAC manufacturers were initially attracted to 1-bit SDM because it meant they could cut costs without sacrificing performance - standard CMOS process vs laser trimmed BiCMOS. However, they soon discovered that 1-bit SDM comes with it's own limitations. The main problem with 1-bit SDM, or any 1-bit format such as PWM, is jitter sensitivity. Any variation in edge timing results in an error that spans the entire output range. Timing jitter is akin to frequency modulation so the jitter signature modulates the output of the SDM which leads to shaped quantisation noise being mixed down into the audio band. It therefore becomes very difficult to achieve a noise floor that approaches that of the best multi-bit converters. The attraction of 1-bit converters, however, is that they are at least theoretically capable of perfect linearity. Unfortunately, this is impossible to achieve in practice due to the imperfect performance of real world transistors.

 

Thankfully, DAC design has progressed significantly since the early days of 1-bit sigma delta. Multi-bit SDM DACs have low jitter sensitivity and lower levels of quantisation noise but they no longer have the inherent linearity of 1-bit systems. However, thanks to some clever processing it is possible to transform the distortion generated by multi-level linearity errors into noise which can then be shaped in much the same way as quantisation noise is shaped in a SDM.

 

ADC designers have begun moving to multi bit too (e.g. PCM4222), again because it reduces the noise floor.

 

In 0 - 20 kHz band both look nice, DSD usually giving a bit better measured THD and IMD figures. Ultrasonic range they definitely look different, PCM has the music content reflected there (every second with inverse spectrum), DSD has just hiss.

 

Most 1-bit modulators can usually only be used up -6dB or so, so maybe the reduced distortion is due to reduced analog signal levels. The distortion of most modern DACs only gets above the noise floor in the upper few dBs. Regarding ultra-sonic output, for a PCM input DAC this really depends on how they get to the modulator rate. Some DACs have filtered upsampling beyond 8fs which gives improved performance over a basic sample and hold. Also, a DSD spectrum will have multiple images too, but both DSD and PCM noise/images can be dealt with easily with an appropriate analog filter.

 

I'm especially interested what comes out of the DAC, digital domain presentation is only half of the picture.

 

Me too!!!

[Miska]

This discussion is becoming hard to follow due to mix up of multi-level and multi-bit SDM.

 

Not sure I agree with your reasoning here Miska. A 1-bit SDM is not difficult to do in silicon.

 

I didn't say it was difficult, but making a good one would require much higher master clock frequency than currently used in DAC chips. Making it proper would also mean that the DAC chip will need at least large heat sink.

 

So far, all multi-level (>2) modulators I've seen in DAC chips are also really poor, but nature of multi-level reduces side effects of this.

 

The main problem with 1-bit SDM, or any 1-bit format such as PWM, is jitter sensitivity. Any variation in edge timing results in an error that spans the entire output range.

 

It is still much less jitter sensitive than multi-bit ladder PCM DACs running at 4x/8x/16x rates. Sure, you'll need a good clock. Difference between two-level SDM and typical multi-level SDM is anyway just 14 dB.

 

The attraction of 1-bit converters, however, is that they are at least theoretically capable of perfect linearity. Unfortunately, this is impossible to achieve in practice due to the imperfect performance of real world transistors.

 

Perfect anything is of course impossible in real world, it is still much less problem than non-linearity of real world two's complement converters. DSD is two-level 1-bit only in transport. Nothing says how many bits you use for conversion while still keeping it two-level.

 

Multi-bit SDM DACs have low jitter sensitivity and lower levels of quantisation noise but they no longer have the inherent linearity of 1-bit systems. However, thanks to some clever processing it is possible to transform the distortion generated by multi-level linearity errors into noise which can then be shaped in much the same way as quantisation noise is shaped in a SDM.

 

If we now don't again mix levels and bits, yes, and you can still play native DSD through this. Of course you can have two-level SDM with any number of bits you desire.

 

ADC designers have begun moving to multi bit too (e.g. PCM4222), again because it reduces the noise floor.

 

...and increases non-linearity. But it's because customers seek for lower noise floor figures while being happy to trade linearity and distortion. Good side of that chip is that you can get the multi-bit SDM data output, instead of poor PCM. That's why I use it in some cases, although my primary converter uses PCM4202 at DSD128. I can then convert either one to what ever PCM or DSD in software.

 

Most 1-bit modulators can usually only be used up -6dB or so, so maybe the reduced distortion is due to reduced analog signal levels.

 

Almost all DACs compensate for the DSD spec of max 50% modulation depth and keep output levels matching PCM. So that's not a problem. In my measurements output level of fundamental frequency has been the same between PCM and DSD.

 

Regarding ultra-sonic output, for a PCM input DAC this really depends on how they get to the modulator rate. Some DACs have filtered upsampling beyond 8fs which gives improved performance over a basic sample and hold.

 

Already the 8x filters are pathetic, and almost all chips employ just 8x + SAH. They just don't have enough clock cycles to do better. (ESS is better)

 

Also, a DSD spectrum will have multiple images too, but both DSD and PCM noise/images can be dealt with easily with an appropriate analog filter.

 

DSD64 images repeat every 2.8 MHz, not every 352.8 kHz which makes quite big difference when it comes to simple analog filters.

 

Another DAC... Not really much of a difference between noise output. But DSD128 gives much less time domain distortions, plus a bit lower THD and especially IMD in audio band.

 

PCM sweep 192/24:

 

DSD128 sweep:

 

(source sweep file is 352.8kHz PCM, so DSD output span is limited by that)

[Miska]

So far, all multi-level (>2) modulators I've seen in DAC chips are also really poor, but nature of multi-level reduces side effects of this.

 

OK, not all, ESS is pretty good. Probably due to just bundling number of simple modulators together to get "multi-bit".

 

Cirrus Logic (CS4398) is second best on my list...

[PeterSt]

As for 24bit PCM, the technology was also developed as a superior codec to 16bit PCM. Interestingly enough, when you look at the actual performance of the best TI ladder type 24bit PCM dacs (BB PCM1704) you'll see that, for example, their THD is at 17bit level.

 

Actually this is interesting (and hopefully not too much offtopic);

 

 

 

This is a -3dBFS 1KHz (16 bit upsampled to 705600) signal, 120dBFS attenuated (so total -123dBFS), taken from the Phasure NOS1 outputs (8x 1704). With the experience of looking at 20 bit converters it is clear that this resolves way over 20 bits (20 bits is 120dB) because otherwise you'd see squares instead of a still clear sine as you see it here. Attenuate 1 bit more and the signal disappears in the noise. So notice : it is the noise which prevents resolving more than 21 bits (the picture effcitively shows 21 bits of resolvement).

But also : the dynamic range of the 1704U-K is 120dB, which is 20 bits.

 

It is noteworthy that the TI specs IMO are not 17 bits for "THD" but 16 bits or a half less. Still, I measure your exact 0.00076% (which is 17 bits, unless I miscalculate one bit) and sometimes 0.00043% (not exactly half of 0.00076%) somewhat depending on the purity of the test signal. So I actually wonder, Hiro, where you get this data from (but maybe you were able to decypher the Japanese extended datasheet, unless I forgot what's in the normal English one).

But also notice that this is the net output of the NOS1 and not any lab measurement directly on the chips with "impossible" PSU etc. That I can obtain these figures will be because of the chip arrangement (differential + parallelled).

 

In addition I am almost sure that the noise you see is "my" noise, and not that of the chips. This noise (hence the dynamic range in the datasheet) surely is there, but only when the output is more than -20dBFS. If we then do the math (this needs some more extensive outlay but I hope you'll get it it), the DR still is that 17 bits because beyond -20dBFS the noise rises linearly (and wonder oh wonder, noise line is at -120dB at FS - it is at -145 or so in above picture).

Lastly notice that pratice depicts that attenuation less than 20dBFS (so -10dBFS etc.) is hardly used because "we all" do not use preamps for the NOS1, thus digital attenuation is used. This by itself implies 3-4 bits loss and better than 21 bits is not practically achievable in that case.

If I am able to tear down the inherent noise of some 6-7uV RMS to something like 2uV RMS it would resolve 21 bits after 24 dBFS attenuation.

 

If you again look at the picture, you see that my mentioned 6-7uV of RMS noise does not work out like that for peak noise. This is more close to 30uV. This now is devistating to THD of course. To make this clear better, see this picture of -96dBFS attenuation (-3dBFS sigbnal, thus attenuated signal is at -99dBFS :

 

 

where you can see with some interpretation of the vertical scale that the p-p noise is still the same - riding on the signal. No difference with -0dBFS (actually worse as I explained).

 

Lastly, I know exactly how TI tests their chips (for the K grade etc.), and you have to trust me that more can be squeezed out of the 1704 than TI thinks herself. But as said, it needs the proper arrangement and some more (like inherent noise is as aspect of it). And, as we will know, the chips are speced for -0dBFS and -20dBFS and it could be interesting to notice that the NOS1 does a sheer 9dB better on -20dBFS (-21dBFS actually) than the specs (something like 0.0038% IIRC vs the 0.0080% from TI, again IIRC).

 

Maybe the key point of this post (and why I posted) in response to the quote above) is that no matter I can't attenuate one bit more because the signal disappears in the noise, the NOS1 hence 1704 resolves -what I estimate- 23 bits or otherwise you'd see a square (I can post a picture of this if you like). That this is quite noisy is one thing, but when it would be that square THD would be quite worse (around 6dB as I measure it).

 

This is just practice and no stupid theory only, or a lab thing.

Peter

[Hiro]

The point is not so much whether you are convinced, but whether enough other people can be relied on to feel the same way that it becomes worthwhile for Benchmark to risk their smallish company on the possibility.

 

You missed the point of my post. Other people already hear the benefit in using DSD128 and DSD256, at least when making native DSD recordings and anolog to DSD transfers, so there's no need for convincing anyone. What's more, all current DSD64 recordings can be upsampled to 128x and 256x DSD, so if anyone (Benchmark included) is worried about the noise in the inaudible band, that's a solution for them.

[Hiro]

It is noteworthy that the TI specs IMO are not 17 bits for "THD" but 16 bits or a half less.

 

Oh, so it's even "better". Thanks for the info.

 

I bet that many people bought into the marketing claim that all DVD-Audio players deliver 144dB performance.

[Miska]

If you again look at the picture, you see that my mentioned 6-7uV of RMS noise does not work out like that for peak noise. This is more close to 30uV.

 

7 µV means -103 dBV (17 bits) and 30 µV is -90 dBV (15 bits).

[Hiro]

I have advanced a definition of HD-Audio as a recording that meets or exceeds the ability of human hearing...put into specifications, that means at least 96 kHz/24-bits. This rules out analog tape, vinyl

 

If only our hearing abilities could be fully described by two or so parameters...

[PeterSt]

With my previous post as a kind of lead-in, I too seem to have an opinion on DSD sourced music. ;-)

But please notice : I am not saying this at all because I produce a sheer PCM DAC; as an audiophool myself I just hunt for the best music reproduction.

 

I have been focusing on Hires SACD rips lately, and what I notice is that they sound more spatious than Hires DVD-A (once I am able to find properly done DVD-A which is rare anyway). It is easy to like "DSD" better for that matter.

But please notice : I listen through my own NOS1 and of course -thus- the SACD rips were converted to PCM.

 

So, DSD based music sure sounds different and this is already so with Redbook material taken from it (SACD).

Assumed the rips have been done well (PS3 etc.) it is my opinion that SACD receives a clear flavor. All highs sound the same throughout albums. And if there's one thing I strive for, it is NOT letting sound all the same. This is in my software (bit perfect/"lossless") all the time and this is in the DAC which should be 100% neutral. And 100% neutral means that nothing will receive a flavor hence sound similar. Also notice that the NOS1 does nothing to the sound, since it doesn't contain anything else than "simple" D/A conversion. No filters - nothing. And PCM (no SDM).

 

Of course it is hard to compare apples to apples, escpecially since the SACD needs to be converted to PCM (which would be necessary for each non-capable direct DSD DAC). But since the flavor is there everywhere while this is not the case at all for DVD-A or normal Redbook, it may tell something;

 

Apart from the flavor, I think the SACD rips are lifeless. Sound is nice and nicely spatious, maybe never disturbing, but there's not much in it. No power. Nothing of which I say (album to album) WOW. Never. Now :

 

What makes it so difficult to judge, where it for all of us to have an opinion, is that the filtering I apply (which is done in software) does not ring even one sample. This, while any normal SDM based DAC does this. So, SACD already rings more than my playback from normal Redbook. This is of vast importance, because with that normal filtering Redbook sounds lifeless (and way distorted) to me just the same - and way more than the SACD rips.

 

Sidenote : with a non-ringing filter - which is perfect in the time domain - the frequency domain receives distortion especially in the higher frequencies. To me (and/or my gear) this is not audible at all but should be audible in some way. Btw notice that the pictures in my previous post are from that same non-ringing filter.

What's done with DSD (which rings minimal) a totally obvious "distortion" is audible, though indirectly through that flavor applied. Remember though, after conversion to PCM.

 

It has been said by others before (and in this thread), but to me it is clear that the high frequency noise influences the sound in a similar way all over.

Also again look at my second picture from my previous post, which clearly shows that (HF) noise rides ON the signal. This is maybe not clear to most but it just is so. IOW, no matter how much down noise or other spuria are, they just influence the signal (hear) hard. This, and this is my personal opinion based upon a couple of years trying to squeeze out the best SQ from Redbook, is exactly what I perceive from SACD : a decorrelated HF noise which inluences all exactly the same. This is different from correlated noise which would imply anomalies.

 

my2c

Peter

[Hiro]

Apart from the flavor, I think the SACD rips are lifeless. Sound is nice and nicely spatious, maybe never disturbing, but there's not much in it. No power. Nothing of which I say (album to album) WOW. Never.

 

That's a common experience among folks converting DSD to PCM.

 

See the Absolute Sound exaSound DAC review

 

"...To evaluate the e20’s performance with DSD files, I selected Channel Classics’ album Super Artists on Super Audio Sampler Vol.5 and downloaded both the 192/24 FLAC version (2406.9MB) and the original DSD (DFF format, 2956.4MB) version. I listened first to several FLAC files, beginning with Dejan Lazic playing the “Allegro” from Scarlatti’s Sonata in C major. I remember thinking how clean and detailed this recording of a piano sounded. Then, hoping I would be able to detect the difference, I cued up the DSD version of the same recording..."

 

 

"...Holy flaming cow! The DSD version of the recording made the FLAC version sound flat and mechanical. With DSD, it sounded like a different piano! The DSD version had the relaxed sound I associate with analog playback. In comparison, the FLAC piano sounded bleached and harmonically threadbare; and for some reason, the DSD version of the piece also sounded distinctly more dynamic. Reach-out-and-touch-it textures further increased the impression that, with the DSD recording, I was listening to a real piano. With the FLAC file, the piano notes just splatted into existence, while the DSD file’s piano notes sounded spookily real, beginning with the hammer hitting a string, the note launching into space, and then decaying off into silence as the note ended..."

 

or the Positive Feedback review of the Resonessence Invicta DAC

 

"Pure Music from the MacBook Pro sounded pretty good as well, and offered the option of real-time conversion to 88.2/24 and 176.4/24 if I was curious to hear how it (DSD recording) sounded in PCM. Uh, not as good. Converting DSD to PCM definitely alters the character, and oddly, the imaging of the mix, not something I'd expect."