Pure DSD

[Paul]

Figured we might as well dedicate a new forum area strictly to DSD. If you don’t mind, let’s keep this generic to the format, the music and SACDs, downloads, DoP, etc. I am sure it will morph into DirectStream stuff, but if we’re able to keep it clean this may become a valuable resource for DSD.

So what is DSD? Well, as our computer scientist, genius, DirectStream designer Ted Smith would tell you, it’s any stream of bits that when you low pass filter it, you get music.

Typically, a more common definition is DSD is a marketing term coined by Sony and Philips when they launched the SACD. Perhaps more accurately, DSD (in its common usage) is PDM or Pulse Density Modulation. Even more typically, DSD (PDM) is referred to as 1-bit data, although that is not a requirement to qualify as PDM.

Some advantages of DSD:

• Simplicity. DSD is simple to convert to analog: just low pass filter it.
• Linearity. DSD is inherently linear: it’s hard to build a PCM DAC that always takes the same sized step in the output for any possible unit increment of the PCM value because of component matching challenges. DSD doesn’t need this level of component matching.
• Soft clipping. Like magnetic tape,DSD soft clips when overdriven: signals which exceed the nominal full scale value (by less than, say 3-4dB) only get slightly compressed if at all. With PCM the consequences are flat tops which induce extra energy at the squared off edges, or worse, wrap around which is very audible.

There's plenty of controversy surrounding the advantages and disadvantages of DSD vs. PCM. PS Audio's support and affinity for DSD should be reasonably obvious, as our DirectStream product is nothin' BUT DSD.

I hope we can engage in a spirited knowledge filled discussion on this forum. Thanks.

[alrainbow]

Great topic. As I think most here who have heard a good analog system know dsd is where we get that same relaxed feeling when hearing our music. I look forward to hearing this sound reproduction. Quality from the direct stream .

Al

[Elk]

DSD (PDM) v. PCM

DSD is roughly equivalent to 88/20-bit PCM, so both are high resolution. Both provide excellent sound.

The disadvantage to DSD (Sony’s name for PDM - pulse density modulation) is that it cannot be edited and, thus, must be converted to PCM for editing. Thus, it is very rare to obtain a “pure” DSD recording that has not spent a good share of its life as PCM. (Query: if it is turned into PCM at some point, has any benefit of being DSD at one time been permanently lost?)

(As an aside, converting a DSD/PDM stream to PCM and back again is a lossy process. We like to think it is just math, but it is not: PDM data has no direct equivalent to PCM data. Thus, the conversion requires interpolation. The inaccuracies are tiny, but present.)

On to the next concept: All digital data begins and ends as PDM, whether it is PCM or PDM/DSD in the consumer’s possession as a stored music file. Weird, huh?

PCM ADCs work by first converting analog into PDM, as a single bit data stream. This signal than passes through a decimation filter (love this term) to turn the signal into multi-bit PCM.

At the other end, PCM is transcoded into PDM prior to decoding. In both cases, this is because it is a lot easier to build a quality PDM DAC or ADC. As a result, almost all are PDM based. (There are a few exceptions.)

One would think this would give the advantage to PDM/DSD as long as it is recorded, never edited, and then played on a PDM DAC. Not only are such recordings rare, modern PDM/DSD ADCs and DACs are almost always multi-bit - not pure 1-bit PDM. As a result, any theoretical difference/advantage of PDM is at least partially lost as the pure PCM/DSD signal will be decimated to 4-bit (I think this is the most typical) making it more like PCM as PDM is played back. (Kudos to the DS for avoiding multi-bit decimation).

In addition to being impossible to edit, PDM also contains the disadvantage of carrying a lot of noise on the signal. The noise begins at roughly 30kHz and rapidly increases so that by 40kHz the noise is louder than any high frequency signal that may be present in the original sound. The low pass filters necessarily applied to the signal begin significantly filtering out the signal at roughly 35kHz. Does this matter? I do not know. Does it matter if a PCM file is 96/24 and thus able to produce up to a 48kHz waveform?

Both high resolution PCM and PDM allow for gentle reconstruction/low pass filters. PCM employs them to avoid aliasing artifacts to reach the playback system; DSD/PDM to keep high frequency noise from doing the same. This strikes me as a draw.

While there is some truth that 1-bit PDM/DSD is more analog like in its digital representation of the signal, this does not recognize that PDM DACs are multi-bit. Even if they were 1 bit, a lot more processing takes place even in 1-bit playback than PDM/DSD aficionados like to admit. (The DS appears to avoid this. )

Bottom line:

The advantage of PDM/DSD is that, in its purist form with true 1-bit ADCs and DACs, it is a less processed signal. Such equipment does not exist in the current world of audio recording and playback (or is incredibly rare). The big disadvantage to PDM is that it must be turned into PCM to be edited in any way, making it very hard to work with. It can only be transcoded to PCM or FLAC (etc.) process.

PCM has both a wider bandwidth and greater S/N. It is easy to edit for both the recording engineer and the consumer, and it can readily be transcoded to other formats (FLAC, MP3, etc.)

A huge advantage for purveyors of PDM/DSD/SACD: it is easy to encode an acoustically transparent anti-piracy code in the data stream.

Whether one sounds better or is up to the listener.

[alrainbow]

As much as understand some of the information above. But my ears tell me dsd sounds like analog , now this may make no sense to some . But if those same people heard a high quality analog mad next the same music in dsd it’s obvious dsd is closer . PCM at it’s best is great but it does lack the relaxation of the music . Did you ever read where a reviewer quotes the music presentation was relaxed . Well it is a true statement . PCM In general sounds digital as opposed to dsd , now sacd,s although they sometimes sound good never get close to a quality dsd 128 Downloads from good sites. An example is opus 3 now this music is supposedly directly recorded from analog masters to dsd . There js no question to my ears this is way above most if not all sacd,s . So as I do understand the concepts and all,of it’s gains and losses in the end we must judge with our ears on a really high quality system. As lesser systems just muddy up conclusions just like the above paragraphs do.

Al

[Elk]

I intend my post to be agnostic as to whether PCM or PDM/DSD is “better.” The end quality of the sound depends on the implementation of the chosen technology. There appears to be no one “better” solution.

And, amusingly, there are few pure DSD ADC-to-DAC recording to playback streams that allow us to even experience DSD which has not been either PCM or multi-bit PDM.

The more interesting question is whether our recordings - however sourced, edited and processed - ultimately sound better with final digital to analog conversion as 1-bit PDM, mutli-bit PDM or PCM. The DSD is an excellent example of the first; the MSB, of the last.

Paul did a nice job summarizing the reasons why final conversion as 1-bit PDM has advantages. My post is intended to further this aspect of the topic.

[alrainbow]

Thanks elk.

in preparing for the direct stream arrival I have been playing almost all dsd tracks. I have also ordered a new music server win server 2012 as well. I did a comparison of the mytek playing native direct dsd 128 and the msb on DOP . Although as good as the mytek sounds and I am sure the new direct stream will beat it by an easy to hear margin. I was surprised how much better the msb sounded . I did this with headphones and the office rig with speakers. Dsd is a marvel to hear there is no doubt it just sounds easier to listen too . I look forward to the direct stream in this mix. I am hoping the server 2012 will bring things to a new level as well.

Elk do you have any recommendations for great sounding dsd downloads I can offer two I am aware of one is Elvis stereo 57 essential Elvis . It s a dsd 64 download the other is Anything opus 3. But the one released from last year no 3 dsd 128 is amazing .

Al

[Elk]

alrainbow said Dsd is a marvel to hear there is no doubt it just sounds easier to listen too .

Subjective impressions are always interesting.

I do not find either PCM or DSD inherently more “relaxed” than the other. For me, the of sound depends primarily on the quality of the recording, and then on the particular playback equipment. I have heard both dreadful and excellent examples of both DSD/SACD and PCM.

[tedsmith]

Howdy

There are a lot of good papers on DSD theory in the AES Library - for those that aren’t members of the AES and don’t want to pay for a paper that may or may not be “worth reading” I’ve found one “illegal” copy of one of the good papers: Derk Reefman’s “Why Direct Stream Digital (DSD) is the Best Choice as a Digital Audio Format”: http://tech.juaneda.com/en/articles/dsd.pdf

Anyway many of the “problems” or statements by Elk are not true for all DSD, they are arguably true for single rate DSD - this is like condemning PCM for the faults of Redbook.

One should remember that you can’t get excellent quality results by editing Redbook in 16/44.1 - you need to use at least wider samples for any sample by sample processing (e.g. mixing) and you need a higher bit rate for any processing where contiguous samples interact (e.g. filtering).

Double rate DSD allows amelioration of many of the issues cited. For example the DSD noise hump in the DS doesn’t go to the -40dB that Elk states, it stays less than -80dB or -90dB. Also the hump doesn’t peak around 30-50kHz - in the DS it doesn’t grow above -120dB until approx. 60kHz. This leaves the effective bandwidth well over that of 88.2kHz PCM (closer to that of 176.4 or 192kHz)

It’s not that you loose all of the “DSD goodness” when you go to PCM - it’s that low rate PCM doesn’t cut it. Keeping the rate at double rate DSD or above allows any math or filtering you want. When you are “done” you “dither” back to one bit. I put dither in quotes because that’s what you’d do in PCM - dither back to the archive sample width. The equivalent in noise shaped single bit based processing is to requantize (run it thru a sigma delta modulator) back to one bit for archiving. With double rate DSD or higher this causes less damage to the signal over the audio band than converting a PCM signal back to 24 bits (dithering or not). Yes, the noise grows (as it does with dithering or truncating the width in PCM) but the number of operations required to get an audible change is huge. Similarly as I mentioned the noise hump can be as low as -80dB FS - for that to grow to significance also takes a huge number of operations.

Another approach to doing DSD processing is to do your processing in the analog domain and then convert back to double rate DSD - quite a while ago I remember being surprised that some of the CDs I liked best had a DAD SPARS code (digital recording, analog mixing, digital delivery.) Going from double rate DSD to analog and then back to double rate DSD can be a very transparent process - arguably more transparent than any process that includes going to lower rate PCM (e.g. 384kHz PCM)

-Ted

[tedsmith]

Howdy

On the other hand DSD is mathematically intractable. Perhaps I should say instead that we just don’t have any good mathematical models for DSD. The tractable models (linearizing the process with assumed random dither) somewhat work but don’t describe the more troublesome behaviors of sigma delta modulation well.

As it turns out, most of the introductions to sigma delta you might see online are first order modulators - one delay, or a first order loop filter, or one integrator, etc. A first order modulator is at least stable - kick it in the head and it will eventually settle down. As soon as you get to more practical sigma delta modulators they aren’t unconditionally stable (SACDs typically use fifth order modulators). The most practical thing to do when designing a high order modulator is to run gobs of simulations - try “all” of the expected inputs. You should of course also do the best job you can analyzing the linearized model of the modulator as well, but neither simulations nor mathematical modeling (with current models) prove stability.

SACD defines the full scale input as 50% modulation. What does this mean? Well if the output of the modulator were all 1s or all 0s you’d be at 100% modulation and at one “power rail” or the other. 50% modulation means that your “full scale output” is at 50% of the power rails. The good effects of this are that it’s reasonably easy to build a modulator that’s stable up to, say, 75% or 80% modulation so if full scale is at 50% we have 3 or 4dB of headroom. I.e. “soft clipping” - if the input temporarily spikes up to about 3 or 4dB over the full scale input the modulator is still stable - it won’t start chirping, oscillating, etc.

Obviously throwing away 1/2 of your available voltage range isn’t something chip makers want to do. So they take heroic efforts to get the unconditionally stable modulation level up to, say, 90 or 95%. This lets them get about 5 or 6dB more S/N ratio and a higher output level both of which are desirable. The downside is that they don’t have as much headroom for musical peaks. Since we’re all used to CDs (and PCM in general) having no headroom, this isn’t totally evil. Having a high modulation level in a modulator which only has PCM inputs a good trade off since you know PCM’s maximum levels anyway.

I both don’t want to alarm people and I don’t want to sweep under the rug some of DSD’s problems.

As I mentioned recently to Paul, just because something is impossible, it doesn’t mean it doesn’t work Parts of DSD are impossible in the sense that we can’t describe them clearly or analyze them with our current models. But that’s true for many processes in nature that are chaotic. Just because we can’t predict the future doesn’t mean it’s not going to happen. Just because we can’t predict the rest of the bits in the output of a sigma delta modulator doesn’t mean we won’t like the sound when we low pass filter them. We have an existence proof that DSD can be done well, many of us hear the results daily.

-Ted

[alrainbow]

BUT can we measure why it sounds better than pcm? and do you agree it sounds more like analog . the relaxed sound I am speaking of ?

al

[tedsmith]

alrainbow said BUT can we measure why it sounds better than pcm? and do you agree it sounds more like analog . the relaxed sound I am speaking of ?

al

We’d have to have a model of the differences that the ear/brain is sensitive to to measure them. Comparing two PCM formats is relatively easy - only sample width, sample rate, and channel count/arrangement can matter. Even then do we infer that 192kHz sounds better than 96kHz? Or that stereo sounds better or worse than MC?

[Elk]

Ted Smith said Anyway many of the "problems" or statements by Elk are not true for all DSD, they are arguably true for single rate DSD - this is like condemning PCM for the faults of Redbook.

Double rate DSD allows amelioration of many of the issues cited. For example the DSD noise hump in the DS doesn't go to the -40dB that Elk states, it stays less than -80dB or -90dB. Also the hump doesn't peak around 30-50kHz - in the DS it doesn't grow above -120dB until approx. 60kHz. This leaves the effective bandwidth well over that of 88.2kHz PCM (closer to that of 176.4 or 192kHz)

It's not that you loose all of the "DSD goodness" when you go to PCM - it's that low rate PCM doesn't cut it.

All good points. I hope all realize I am merely raising issues for discussion and not advocating one format over the other. I am agnostic on this point.

Single rate DSD has greater limitations, just like like 44/16 PCM. SACDs are single rate and most DSD files are single rate and sourced from single rate (or PCM for that matter). Thus the issues raised.

I agree DSD is not necessarily “damaged” when it is transcoded to PCM (I love the phrase “DSD goodness”), but many DSD proponents claim DSD is somehow superior in all respects. Theirs is an argument of faith that somehow DSD is pure; PCM, polluted. I find this argument specious as almost all DSD people have heard has spent a good deal of time as PCM. Additionally, almost all DACs decode all digital streams they receive as four to six bit DSD. That is, we are typically listening to something which is neither PCM or DSD, but sort of a hybrid.

For me the vastly more interesting question - and one which I have raised on this forum long before the DS was announced - is whether there is something special or “better” when decoding single bit DSD to analog v. decoding PCM. If there is magic, I argued, this is where it should reveal itself.

I wish I could claim prescience with respect to the DS’ entry as a partial answer to this question.

[tedsmith]

BTW Thanks Elk, it’s nice to talk about this without people acting like religious extremists on either side. Sites (and people) who can do that are few and far between.

[Elk]

Ted Smith said

On the other hand DSD is mathematically intractable.

Another wonderful turn of phrase.

And a fascinating post. I was unaware of the definition of DSD full scale and its practical impact. Extremely interesting.

My inner nerd is giggling.

[alrainbow]

But neither one of you has answered the question does dsd 128 sound better or more natural to your ears. I have found that if I take a good dsd 128 and convert to 24/192 it looses something . So my understanding would lead me to think there is something to dsd being better in some ways. Now if the change is totally due to the conversion software being used and not just the down sample and conversion Then I would be wrong. Either way dsd 128 recordings I have shown me it sounds better . Ted when you say lower jitter makes your toes start to tap I agree but what does this mean . To me it means it’s closer to what makes our ears happy and just what is this. Recordings I have heard or own that were from analog tape from the 50 or 60,s and converted to dsd sound above almost all dsd 64 and most PCM .

Does anyone here agree or disagree not arguing but asking ? All of this is subjective anyway .

Al

[tedsmith]

alrainbow said But neither one of you has answered the question does dsd 128 sound better or more natural to your ears. I have found that if I take a good dsd 128 and convert to 24/192 it looses something . So my understanding would lead me to think there is something to dsd being better in some ways. Now if the change is totally due to the conversion software being used and not just the down sample and conversion Then I would be wrong. Either way dsd 128 recordings I have shown me it sounds better . Ted when you say lower jitter makes your toes start to tap I agree but what does this mean . To me it means it's closer to what makes our ears happy and just what is this. Recordings I have heard or own that were from analog tape from the 50 or 60,s and converted to dsd sound above almost all dsd 64 and most PCM .

Does anyone here agree or disagree not arguing but asking ? All of this is subjective anyway .

Al

Personally I’ve been listening to everything upsampled to DSD128 for so long that it’s hard for me to say what DSD64 looses, before I built my DAC I was using the Meitner (EMM Labs) DAC6e which also upsamples to DSD128.

To my ears, yes, the material I’ve heard in unadulterated DSD64 or DSD128 sounds better than that which has been converted to 24/192, but I must say that I haven’t heard a high end PCM DAC built by a true PCM believer with material of known providence for more than a decade. I do trust the opinions of my friends like Gus Skinas who work with this stuff every day. They definitely prefer DSD128 over DSD64 and both over PCM @ 96kHz or 192kHz.

I expect that it takes a high sample rate to maintain the waveform of impulses (like most percussion) well enough that it sounds natural to the ear/brain. And the sample rate needed seems to be over 300 or 400kHz.

Most of the listening tests that people do are flawed just because it’s often quite hard to know the provenience of the material being listened to. I have thousands of SACDs and tho many of them started out life as 24/96 (or even 24/50) I do enjoy most of them.

[alrainbow]

Ted your honesty is really refreshing to read. And weather you agree or disagree with me I always learn from you. I am a bit sorry for forcing an answer it’s just there is some many ways to say nothing . And you have actually said something that so many refuse to say . Now given that the direct stream does up sample and makes you happy wow I am even more happy I ordered it. As always thanks for the reply ted .

al

[oddeophile]

alrainbow said Ted your honesty is really refreshing to read. And weather you agree or disagree with me I always learn from you. I am a bit sorry for forcing an answer it's just there is some many ways to say nothing . And you have actually said something that so many refuse to say . Now given that the direct stream does up sample and makes you happy wow I am even more happy I ordered it. As always thanks for the reply ted .

al

Al, your going to be surprised I think. I am taken aback with what I have been hearing with mine. I use PWT exclusively so that needs to be taken into account when reading my thoughts. 95% of my collection is either pure RBCD or down sampled from DSD mastering.

[alrainbow]

Thanks , have you tried to use the dop USB and not down sample from dsd. Is there any improvement . Also have you tried the PWT in dop ISO disks for playback ?

al

[Elk]

alrainbow said But neither one of you has answered the question does dsd 128 sound better or more natural to your ears. . . . Does anyone here agree or disagree not arguing but asking ? All of this is subjective anyway .

I do not have a definitive opinion. Both high resolution PCM and DSD sound wonderful if done right.