Direct Stream Digital and the Power of Imagination

[Hiro]

If your system is to include any kind of mixing, EQ, or other signal processing, you pretty much have to go the PCM route. Fortunately, a conversion to PCM and back doesn't cause any audible degradation. Besides, many high-end DAC chips (ESS, AKM) process DSD digitally into whatever internal representation they use anyway.

 

If they do process DSD, its internal representation is still delta-sigma modulated signal.

 

AKM, among others, offer a more direct path for DSD (bypassing decimation, and two steps of internal oversampling) in their SDM DACs...

[Paul R]

Seriously... you are stating this on an audiophile forum?

 

Where the only criteria for "better" or "best" is absolutely and only SQ. No matter how much "processing" power it takes in the form of HQP?

 

Are you really stating that?

 

Or are you really trying to get out of a corner you have boxed yourself into... without starting another flame war.

 

At least man up.

 

I make a mistake, I man up at the very least... proof as in my apology a couple of posts in this very thread.

 

No, I didn't make a mistake and I didn't box myself into a corner.

 

You must have missed the last sentence in there.

 

If you choose to use SQ - how so ever you choose to measure that - as the absolute and only criteria for "better" or "best" - more power to you chum. I do not think you will go far wrong with that.

 

So why are you trying to pick a fight with or "educate" people here over something you say you don't care about?

[Paul R]

That's not how I'd describe it.

 

If a sampled signal is made continuous by a simple hold mechanism (producing the familiar stair-step curve), the resulting spectrum consists of the original baseband (up to Fs/2, aka Nyquist frequency) followed by images around Fs and its multiples. Applying a sharp lowpass filter at Fs/2 removes the images, leaving only the original signal. This is all pure maths.

 

A PCM DAC could be constructed as a resistor ladder with its inputs updated on every sample tick and its output feeding directly into an analogue lowpass filter. The trouble with this is that the unwanted images start immediately at Fs/2, so the filter must be very steep in order to remove the images without affecting the baseband. As we know, constructing such a filter is all but impossible.

 

If the sampled signal is first digitally resampled to a higher rate, the images are shifted to correspondingly higher frequencies. For CD audio upsampled 8x, the lowest image frequency is at roughly 330 kHz (8 * 44100 - 22050 Hz). This allows for a much simpler (lower order) analogue filter due to the wider transition band.

 

A DSD DAC works fundamentally in the same way: the digital values are converted to a voltage, and the resulting analogue signal is lowpass filtered. The important difference is that there are only two possible values (zero and one) which greatly simplifies the conversion from digital value to analogue voltage. Instead of a high-precision resistor ladder, all we need is a simple on/off switch. The spectral characteristics are also different. Instead of repeating images (well, there are images, but in the MHz range), there is a gradually rising noise floor which again allows a reasonable analogue lowpass filter.

 

You know, I think we honestly said the same thing... save for you went into more technical detail on the audio engineering than I did. (As well you should, that is your speciality, is it not?)

 

So look at it this way - nobody is saying that PCM is evil. In fact, very high sample rate PCM sounds about the same as DSD128 or DSD256 to me, even when up sampling or transcoding from a red book data file ripped from a CD.

 

To me, this suggests that, at least with our current DACs, the less processing the DAC does the better, even when there is a whole lot more actual data from the computer to transfer and process.

 

Expanding that, the "pure DSD" music I have almost always sounds better to my ears than the red book equivalents, even when the red book file was created from the high rate DSD "master."

 

And for "pure DSD" vs. red book recordings, as George and others have pointed out, the mastering and sonics of the original recording may definitively override any format considerations.

 

At which point, I suggest that up-sampling or transcoding the red book file to very high rate PCM or DSD may further improve the sound.

 

YMMV.

[master]

So why are you trying to pick a fight with or "educate" people here over something you say you don't care about?

 

I'm not trying to pick a fight with you and certainly not trying to educate anyone.

 

My apologies if you felt that way.

 

My only contention is against dealing in absolutes.

 

For instance we may both be listening to the same speaker in the same setup, but we both will look for different things with regard to speaker performance and as a result our rating/evaluation of the speaker will vary.

 

That holds true for everything in music - all of the equipment, the DACs, and even the formats.

 

Nobody can state anything for sure without really knowing what the customer/end user is listening for or even wanting from the said equipment or format.

 

In my situation, based on what I'm looking for, I've found RBCD with a NOS DAC delivers the best. Based on what I'm looking for and it need not be the same for everyone... heck even my wife differs in this and looks for something entirely different when evaluating a speaker's performance.

 

The same holds true for you too... in your situation and based on your needs DSD offers the best results and I'm fine with that too.

 

However, we need to see and acknowledge there is a different shoe of a different size for every foot. Having an issue with everyone who does not see DSD as God's greatest gift to music is the real "silliness"...

[Tony Lauck]

If your system is to include any kind of mixing, EQ, or other signal processing, you pretty much have to go the PCM route. Fortunately, a conversion to PCM and back doesn't cause any audible degradation. Besides, many high-end DAC chips (ESS, AKM) process DSD digitally into whatever internal representation they use anyway.

 

A conversion from DSD64 to DXD (352.8/32fp) and back to DSD64 can most definitely degrade the sound quality, as I have heard both versions of the same recording and the differences were obvious. This was two generations of noise shaping: analog to DSD64 and then DXD to DSD64. And this was no low rate (88/24) PCM either. There are two plausible explanations for why transparent conversions between these formats are difficult. First, DSD64 noise is not easy to separate from the music, and the necessary filtering causes degradation. Second, using noise shaping on a previously noise-shaped signal can create audible artifacts. (This can be obvious when remastering from original noise shaped 44/16 material, so this can be a purely "PCM" issue as well.)

[Hiro]

Not if a Sonoma DAW is employed in post processing, and not for long in the future for at least one other DAW supplier.

 

An important and most welcome development. I trust that it's going to support the current DSD128+ rates...

[mansr]

A conversion from DSD64 to DXD (352.8/32fp) and back to DSD64 can most definitely degrade the sound quality, as I have heard both versions of the same recording and the differences were obvious. This was two generations of noise shaping: analog to DSD64 and then DXD to DSD64. And this was no low rate (88/24) PCM either. There are two plausible explanations for why transparent conversions between these formats are difficult. First, DSD64 noise is not easy to separate from the music, and the necessary filtering causes degradation. Second, using noise shaping on a previously noise-shaped signal can create audible artifacts. (This can be obvious when remastering from original noise shaped 44/16 material, so this can be a purely "PCM" issue as well.)

 

This is why modern ADCs and DACs internally use multi-bit formats at DSD128 rate or higher.

[Hiro]

A conversion from DSD64 to DXD (352.8/32fp) and back to DSD64 can most definitely degrade the sound quality, as I have heard both versions of the same recording and the differences were obvious. This was two generations of noise shaping: analog to DSD64 and then DXD to DSD64. And this was no low rate (88/24) PCM either. There are two plausible explanations for why transparent conversions between these formats are difficult. First, DSD64 noise is not easy to separate from the music, and the necessary filtering causes degradation. Second, using noise shaping on a previously noise-shaped signal can create audible artifacts. (This can be obvious when remastering from original noise shaped 44/16 material, so this can be a purely "PCM" issue as well.)

 

Fortunately, DSD-wide DAWs don't require decimation, and the conversion to low rate PCM and back can be avoided.

[mansr]

Fortunately, DSD-wide DAWs don't require decimation, and the conversion to low rate PCM and back can be avoided.

Yes, but there's only so much DSP you can do in that format.

[Hiro]

Yes, but there's only so much DSP you can do in that format.

 

IIRC, Miska said that pretty much everything is possible in terms of SDM-wide DSP for DSD. So, let's wait and see what the new DSD DAW will be capable of...

[mansr]

IIRC, Miska said that pretty much everything is possible in terms of SDM-wide DSP for DSD. So, let's wait and see what the new DSD DAW will be capable of...

Even if that's true, I doubt there are many standard plugins that support it. Most music production will still be done in PCM, and I'm perfectly fine with that.

[Hiro]

Even if that's true, I doubt there are many standard plugins that support it. Most music production will still be done in PCM, and I'm perfectly fine with that.

 

But that could change...

 

SDM-wide DAWs certainly bring us closer to end-to-end SDM...

[tailspn]

Try figuring out the DSD examples here:

 

https://en.m.wikipedia.org/wiki/Pulse-density_modulation

 

 

Hi Carl,

 

That's really a great reference! Try leaving out the confusing referring to PWM, and all the math. Just consider the text from the beginning through "Examples" Ignore the rest.

 

If you'll excuse/forgive me for using my Mr. Rogers sweater for an explanation attempt:

 

For purposes of understanding PDM (and all its other names), forget the concept of "bits". That's what throws everyone off, for they're attempting to use their prior knowledge and understanding of "digital". It's like an airplane pilot trying to learn the flying of a helicopter; it's much easier if you didn't know how to fly a plane.

 

Instead of populations of bits, think potatoes. They're equivalent, for in PDM, neither represent quantitative values, only elements in the measure of population. Better yet, think fence pickets, for then a visual corollary exists.

 

An analog signal is a continuously varying force (in our case measureable), like sound pressure, or voltage from a microphone output. The simplest form of PCM does measure those quantities on a periodic basis (sampling rate), to some resolution (digital word bit depth), and expresses for audio as we know it in a agreed upon digital word format (2's compliment binary). It's a nice (antiquated) system designed to provide a medium of digitally processable data which to eventually shape and deliver recorded music.

 

PDM (DSD) has no such objective. It merely, through a complex modulation scheme, transforms continuously changing analog voltage levels to continuously changing populations of potatoes/fence pickets/bits along a timeline. The maximum allowable number of elements (I like fence pickets) represents the maximum allowable positive PERCENT of MODULATION, and the maximum absence of elements represents the max minimum PERCENT of MODULATION. There's no actual (digital) values being represented, simply percents of modulation along a timeline represented by the DENSITY of the things we're using as elements representing proportionally the signal (force) being modulated. If you use fence pickets, you get the picture in your reference under "Examples".

 

It really is that simple. To make it useful for us music lovers, the rule it follows by the modulator designer's definition is that alternating (dare I say 1's and 0's?) pickets/potatoes represents zero signal. All 1's (Max density) represents maximum positive signal, and all 0's (lack of 1's) represents maximum negative signal. Also, 50% modulation represents 0dB for us people that watch VU meters. There are no voltage values (ala PCM) at this point, just percentages of signal amplitude represented proportionally by the density (over time) of the elements we choose to feed our modulator (how bout clock bits?).

 

To further make this SD modulator output useful, the "bits" spilling out of it are made at voltage levels that a computer bus and storage media recognizes, making the bit stream storable and retrievable. Package that up on CD frames packets, deliver it to a music lover with a simple (not really) integrator DAC, and you've got music!

 

If however, you'd like to shape that music, you'll need to convert that simple potato stream(s) into something a digital computer can recognize, and process. Now you've reached my Peter's Principle.

 

Hope this is useful.

 

Tom

[tailspn]

Even if that's true, I doubt there are many standard plugins that support it. Most music production will still be done in PCM, and I'm perfectly fine with that.

 

Actually, the same plug-ins operate on DSD, once it's converted into a processable form. DXD, or some other PCM sample rate conversion is what's used presently, because it's cheap to convert DSD to PCM. But other encoding formats which did not require decimation to be processable would allow all current plug-ins to operate.

[Nikhil]

For me the matter was settled when I heard the same performance simultaneously recorded in PCM and DSD. No amount of discussion is going to convince anyone more than just listening to these recordings.

 

In the end what you really hear is how well your setup can handle the two formats.

DSD is very sensitive to the entire chain starting from the computer transport.

PCM tends to 'start' from the DAC.

 

For those interested here are the files:

 

PCM Files:

https://www.dropbox.com/s/w9rc24vy33ifqd5/PCM.zip?dl=0

 

DSD128 Files

https://www.dropbox.com/s/8lrzn2qbem...SD128.zip?dl=0

 

DSD256 File

https://www.dropbox.com/s/d0dsg832b1...SD256.zip?dl=0

 

For those interested in the provenance of the files,

 

Orfeu Negro" by Japanese Trio, sax.: Y. Tomakomachi, piano: M. Ishizuka & Percussion: S. Ishikawa. Recording engineer is Mr. H. Takada of Victor Entertainment, Chairman of Technical Committee of Japan Association of Professional" Recording Studios (JAPRS).

" ... files were prepared by capturing an identical analog output signal of API console of CR-506 studio of NHK(Nippon Hoso Kyokai: Japan Broadcasting Corporation) with six DAWs, Avid ProTools for PCM 48kHz, 96kHz, 192 kHz and Merging Pyramix/Horus for 384kHz/32bit,DSD 5.6MHz, DSD 11.2MHz simultaneously."

 

... purpose of the release of the sources is to provide "a kind of reference standard sources" for professional engineers to recognize actual differences on sound characteristic depending on formats.

 

 

[esldude]

For me the matter was settled when I heard the same performance simultaneously recorded in PCM and DSD. No amount of discussion is going to convince anyone more than just listening to these recordings.

 

In the end what you really hear is how well your setup can handle the two formats.

DSD is very sensitive to the entire chain starting from the computer transport.

PCM tends to 'start' from the DAC.

 

For those interested here are the files:

 

PCM Files:

https://www.dropbox.com/s/w9rc24vy33ifqd5/PCM.zip?dl=0

 

DSD128 Files

https://www.dropbox.com/s/8lrzn2qbem...SD128.zip?dl=0

 

DSD256 File

https://www.dropbox.com/s/d0dsg832b1...SD256.zip?dl=0

 

For those interested in the provenance of the files,

 

Orfeu Negro" by Japanese Trio, sax.: Y. Tomakomachi, piano: M. Ishizuka & Percussion: S. Ishikawa. Recording engineer is Mr. H. Takada of Victor Entertainment, Chairman of Technical Committee of Japan Association of Professional" Recording Studios (JAPRS).

" ... files were prepared by capturing an identical analog output signal of API console of CR-506 studio of NHK(Nippon Hoso Kyokai: Japan Broadcasting Corporation) with six DAWs, Avid ProTools for PCM 48kHz, 96kHz, 192 kHz and Merging Pyramix/Horus for 384kHz/32bit,DSD 5.6MHz, DSD 11.2MHz simultaneously."

 

... purpose of the release of the sources is to provide "a kind of reference standard sources" for professional engineers to recognize actual differences on sound characteristic depending on formats.

 

 

Do we know if these are precisely level matched? Or do you know if the level is precisely the same with your playback setup? The latter being the key issue which varies from system to system.

[mansr]

Actually, the same plug-ins operate on DSD, once it's converted into a processable form. DXD, or some other PCM sample rate conversion is what's used presently, because it's cheap to convert DSD to PCM. But other encoding formats which did not require decimation to be processable would allow all current plug-ins to operate.

I'm guessing here, but I doubt many (any) plugins currently operate on 8-bit DSD-wide (if you have an actual spec for that format, please show me). There's also very little incentive for the developers to support it for two reasons: 1) the vast majority of users don't mind PCM, and 2) those few who want to avoid PCM also shun any kind of processing. In other words, the only users who might buy such tools don't want them.

[Hiro]

I'm guessing here, but I doubt many (any) plugins currently operate on 8-bit DSD-wide (if you have an actual spec for that format, please show me). There's also very little incentive for the developers to support it for two reasons: 1) the vast majority of users don't mind PCM, and 2) those few who want to avoid PCM also shun any kind of processing. In other words, the only users who might buy such tools don't want them.

 

Many people upsample PCM to DSD so they "don't shun any kind of processing," and DSD processed without any downsampling/decimation at MHz rates, is superior to the first option, as such files can be played in NOS mode on DSD DACs.

[mansr]

Many people upsample PCM to DSD so they "don't shun any kind of processing,"

 

I'm talking about producers like Cookie Marenco and others who rather use analogue tape than let their music be tarnished by filthy PCM.

[Hiro]

I'm talking about producers like Cookie Marenco and others who rather use analogue tape than let their music be tarnished by filthy PCM.

 

Well, DSD-wide DAWs will be an alternative for them, so that they don't need to downsample their 256FS SDM masters to 8fS PCM or such.

[plissken]

Do we know if these are precisely level matched? Or do you know if the level is precisely the same with your playback setup? The latter being the key issue which varies from system to system.

 

In the thread he linked to the DSD track has higher volume output.

[esldude]

In the thread he linked to the DSD track has higher volume output.

Thanks for the info plissken.

 

Not uncommon. No surprise he preferred DSD.

[tailspn]

I'm guessing here, but I doubt many (any) plugins currently operate on 8-bit DSD-wide (if you have an actual spec for that format, please show me). There's also very little incentive for the developers to support it for two reasons: 1) the vast majority of users don't mind PCM, and 2) those few who want to avoid PCM also shun any kind of processing. In other words, the only users who might buy such tools don't want them.

 

I didn't say anything about DSD-Wide, I said a processable DSD form. There's numerous ways of achieving that.

 

I disagree that there's little incentive to develop a non PCM DAW. I'm only concerned there's a production market left to buy it, with the ongoing collapse of the acoustic music recording industry as we know it.

[master]

In the thread he linked to the DSD track has higher volume output.

 

DSD pulling a Nordost fast one...

 

Just kidding folks... lets get back to enjoying the music.

[Tony Lauck]

In the thread he linked to the DSD track has higher volume output.

 

Please explain what it means for a DSD track and a PCM track to be "precisely level matched" My DAC uses a different signal path for DSD and PCM and I am not aware that there is a consistent standard for how to match levels. Therefore, other DACs (or conversion players such as HQPlayer) might have different gain from each other.

 

If I were doing a demonstration, I might be able to deal with this, if I created both files from a common (e.g. analog) source and began with a test tone, or if I trusted the file creator and he provided a test tone. I could also equalize the levels on my own DAC by capturing the analog output and comparing it, but this would be a difficult task for anyone who didn't have two separate computer audio systems.

 

Someone carrying on a demonstration would have an ethical obligation to equalize the levels at his speakers, but my observation is that most demonstrations are implicit or explicit sales pitches for some product or dogma.