Audio Myth - "DSD Provides a direct stream from A/D to D/A."

[Jud]

I've made contributions to many open source projects. Of greatest relevance to this thread is probably the DSD encoder I recently added to SoX. There's another thread on here with links and some graphs.

 

Link to the appropriate message in the other thread would be helpful to those interested.

[mansr]

Link to the appropriate message in the other thread would be helpful to those interested.

 

http://www.computeraudiophile.com/showthread.php?t=25556

[audiventory]

PCM is an archaic tricycle with training wheels dying format. It persists in audio music production because, with its minute market size, there isn't the necessary capital investment for the development of DSD production tools.

 

I remember, in my youth (end of 80's) I seen first digital reverberator scheme based on sigma-delta modulation. It was invented in 1954 and patented by Cutler.

 

PCM was invented in 1939.

 

Impossibly process DSD without multibit math into either computer or FPGA or other.

 

DSD and PCM are two orthogonal tools what can't replace each other, but perfect work together as solid system.

[Miska]

A simple SDM (i.e. no fancy features like lookahead) with an 8th order filter can achieve perfectly acceptable performance.

 

Well, I disagree on that...

 

A dedicated hardware implementation could in principle be constructed to run directly off the bitstream clock. Since the clock rates are anyhow quite low, it probably makes practical sense to use fewer computational blocks and run at something like 8x the bit clock.

 

Specialised hardware is almost always a few orders of magnitude more efficient at doing its one task than is a general purpose computer.

 

Maybe technically more efficient, but not necessarily more cost efficient. Designing chips for something like 14 nm process and manufacturing those is not cheap. And it becomes more problematic if it needs to be a mixed signal one, like DSP and D/A conversion on the same die. Especially if the analog side is supposed to have 100+ dB SNR with the DSP side less than a millimeter away...

 

If you think you can manufacture such a chip such way that the actual implementation is more cost effective than using current general purpose computers, by all means just do it.

 

I personally find it more cost efficient to use the computer I anyway already have (thus zero added cost) for performing all the DSP and put all the money in a discrete SDM-only DAC. This approach also allows easy software upgrades on the performance unlike chip implementations...

[Miska]

I've made contributions to many open source projects.

 

Me too, ironically in fact most of the development of my commercial software has been funded by my day jobs on open source software.

 

Unfortunately there are R&D costs for doing lot of the development. Asking some money for the license allows me to invest more on the R&D. So it's kind of a positive feedback loop.

[Miska]

If you take two DSD bitstreams and add them bit for bit, you end up with something where each sample can take on values [0,2]. Is this 2-bit DSD or 2-bit PCM? Does it matter?

 

There's a certain difference. Hint: the other one is bijection while the other is surjection.

[audiventory]

Miska, exactly, research and development is significant part of cost any software project.

 

What about technical support?

 

Who will support without proper funding?

 

I believe only in free projects, that developed in environment of commercial projects. These projects intended for development and/or promotion of commercial ones. I know many same good free projects not only in software.

 

No funding - no guaranted development and support.

[audiventory]

the other one is bijection while the other is surjection.

 

Sorry, what you want say?

[Jud]

Not to get too far off topic, but in at least some cases open source can be commercial (and successful). I'm not referring specifically or necessarily to audio processing.

[Miska]

Miska, research and development is significant part of cost any software project.

 

What about technical support?

 

Who will support without proper funding?

 

Most open source software companies base their business model on two principles:

1) Selling technical support

2) Selling development services for open source software components (Linux drivers, etc)

 

Plus then there are hardware companies for whom the development of software supports their hardware business.

 

This model works for components that have wide and diverse user base. It doesn't really work well for any niche market.

 

Anyway, this is off-topic for this thread, but subject worth it's own thread.

[mansr]

Maybe technically more efficient, but not necessarily more cost efficient.

 

Depends on the volume and the unit price. A custom chip is typically only cost effective at a rather large volume.

 

Designing chips for something like 14 nm process and manufacturing those is not cheap.

 

14 nm is cutting edge. For something as relatively uncomplicated as an SDM, the currently cheapest process should be plenty enough.

 

And it becomes more problematic if it needs to be a mixed signal one, like DSP and D/A conversion on the same die. Especially if the analog side is supposed to have 100+ dB SNR with the DSP side less than a millimeter away...

 

True.

 

If you think you can manufacture such a chip such way that the actual implementation is more cost effective than using current general purpose computers, by all means just do it.

 

I personally find it more cost efficient to use the computer I anyway already have (thus zero added cost) for performing all the DSP and put all the money in a discrete SDM-only DAC. This approach also allows easy software upgrades on the performance unlike chip implementations...

 

If a sufficiently powerful general-purpose computer is already available, using that is obviously cheaper. If you're aiming for a sub-$5 single-chip solution (which will naturally have compromises while still being good enough for some applications), not so much. The proper approach always depends on the demands of the target application.

[徐中銳]

It's Monday morning where I am, with little time, easiest just to quote myself quoting some of Channel Classics' Jared Sacks' views on... :

America's problem is actually at the dealer level. For years' date=' they didn't want to deal with SACD, and they didn't want to educate anyone coming in; that has continued to be the problem. So we need magazines and websites to educate the listener. Since what you never hear you do not miss, I organize regular listening sessions at my studio to let people hear what they are missing!

 

[img']http://www.nativedsd.com/media/uploads/Jared_Sacks_Shares_a_Story_at_DSD_Party.jpg[/img]

 

To me, DSD's superiority has to do with emotion, depth, and how the sound leaves the speaker. It's not a block anymore in the way it dissipates. When you listen to PCM, you can literally hear it as a block of sound coming out of the speaker. That doesn't happen with DSD. There's air around the sound. At the end of the day, we are talking about the air around the sound.

 

In our business, we have to do post-production, but not all the time. I always make a mix-down into stereo. The surround channels go directly to an A/D converter, so they don't go through a mixer, and I try to leave them like that. Then I make a master without going through post-production (without going through the sigma-delta converter again).

 

The moment I have to change levels or do some EQ, I have to go through the mixer, and that means going through the sigma-delta again, which lowers the quality. Of course, it's all high DSD, but you have to go into DXD if you do post-production, and there's really no way around it. This problem will be solved in the future. But we are talking about further research, which costs money, at a moment when there is not much to be made selling to recording companies.

 

When you listen to my raw data, and you compare it to the post-produced recording, there's a difference in the air around the instruments and the depth. There's a degradation of sound. It's slight, but it's there. It's unfortunate, but there's nothing we can do about it, because we have to go into the sigma-delta processor again. As with any other audio signal, if you have to keep on processing, it will change.

 

You may ask, given that, if there is a difference between the sound of 192 and DSD? You have to have a really good system, and it also depends on the repertoire, to hear the difference. I still do, especially because of the dynamic range. When I down-sample to 192, you can hear that it's PCM, absolutely.

For full Stereophile interview : DSD Present and Future

[audiventory]

徐中銳, I would divide DSD and SACD.

 

In my opinion, physical disks is obsolete now.

 

DXD, PCM, DSD, WDSD is only names of technologies. Meanful how it applied. How engineers can extract hidden opportunities or pass by troubles.

 

Multichannel systems, in state that it exists now, is intermediate variant only on path to real "wave field re-creators".

[Miska]

Depends on the volume and the unit price. A custom chip is typically only cost effective at a rather large volume.

 

Yes... I don't think audiophile market has such volumes. And once you have a custom chip you cannot change the algorithm in it anymore. So what if you realize a day after a new way to significantly improve the performance?

 

14 nm is cutting edge. For something as relatively uncomplicated as an SDM, the currently cheapest process should be plenty enough.

 

Good, while waiting for such to materialize I seek for different ways. I'm not holding my breath waiting.

 

HQPlayer does quite many billions of operations per second per channel when upsampling to DSD.

 

If a sufficiently powerful general-purpose computer is already available, using that is obviously cheaper. If you're aiming for a sub-$5 single-chip solution (which will naturally have compromises while still being good enough for some applications), not so much. The proper approach always depends on the demands of the target application.

 

I'm only interested on getting as good technical performance out of real world systems as possible.

 

IIRC, all the current SDM DAC chips are <$20 in volumes.

 

 

P.S. When I last checked what kind of FPGA it would take to implement what I currently run on a general purpose processor, the FPGA would have cost more than the CPU.

[Jud]

There's a certain difference. Hint: the other one is bijection while the other is surjection.

 

I am guessing the former is PCM and the latter is DSD. Or am I way off?

[Miska]

I am guessing the former is PCM and the latter is DSD. Or am I way off?

 

No, the other way around...

[Maldur]

For full Stereophile interview : DSD Present and Future

 

Thank You for link, this was very interesting reading and Mr.Mark Levinson comments was great .

[Norton]

Thank You for link, this was very interesting reading and Mr.Mark Levinson comments was great .

 

"The problem has always been that because I made a hybrid SACD, which I have to sell at normal CD price, my profit margin is minimal. By contrast, I can make a DSD download available in both stereo and multi-channel download versions for almost the same price."

 

Not sure I understand the whole point Jared is making here. IME a given recording is generally 2-3 times more expensive as a stereo DSD download on Native DSD than its "street price" as a physical SACD which may well include both stereo and MCH layers.

[徐中銳]

Thank You for link, this was very interesting reading and Mr.Mark Levinson comments was great .

 

You're most welcome We all want to share...

 

Yet, of Levinson, was he but again blowing his own « flugelhorn » (plus somewhat undermining DSD) ? For quoting fully :

I agree, an excellent interview. Hard to believe I didn't know about Jared and Channel Records until now. Thanks to Jason and Sterephile for introducing us.

 

There are a bizarre number of parallels here. He and I both are: American musicians playing brass intruments (he horn, me flugelhorn) who started small music-based companies; moved to Europe; work with our mates; have a passion for making quality recordings of classical music and reproducing them; have a connection with Switzerland and the EU; are early adopters of DSD, made some of the first SACD's, and worked with Sony and Philips in that regard; and are not afraid to go off the beaten path. I intend to learn more about Jared and Channel Records starting right now.

 

The only drawback to DSD is that there are no DSD mastering tools, so engineers have to do one of three things:

1. record in DSD and do no mastering.

 

2. record in DSD, convert to high resolution PCM, use high resolution PCM mastering tools, and convert to DSD for making SACD's.

 

3. record in DSD, convert to analog, master in analog, and convert to DSD.

 

Good results can be obtained with all the above, but real DSD tools would be great to have. Sadly, DSD does not lend itself to manipulation.

 

I recorded the SACD "Live Recordings at Red Rose Music" in 2000 with no processing at all, just pure DSD. When the musicians finished playing, that was it. This SACD was used to introduce SACD in the US.

 

Just recently, my Swiss company Daniel Hertz introduced Master Class, audio software for Mac. Master Class is intended to make PCM sound and feel more like analog and DSD. I have been using this technology in my own mastering for some time, but it was not in a form that could be sold. Master Class is a commercially available version, easy for non-technical people to use. I have found that with Master Class, PCM recordings can be more or less like DSD.

 

It is great to see that DSD is alive and finally being developed and appreciated by dedicated people like Jared and his Channel Records, and that Jason and Sterephile are bringing this to the public's attention with such well written articles.

And Norton, I'll look into finding the relevant write-up I had once read

[mkrzych]

You're most welcome We all want to share...

 

Yet' date=' of Levinson, was he but again blowing his own « flugelhorn » (plus somewhat undermining DSD) ? For quoting fully :

[/font']

And Norton, I'll look into finding the relevant write-up I had once read

 

For me Mark's short comment is kind of "AMEN" for DSD. Now, how many record labels doing pure DSD recording without mastering or converting to PCM domain could be a question - more precisely if they have enough content to shine globally on the music market. Now, what are the digital workstations able to record in pure DSD - Pyramix, but AFAIR it's DXD, so kinda of PCM right?

[Maldur]

Pyramix uses DXD only in editing points - fade-in/fade-out for example. How long is fade-in typically? Or fade-out...?

[tailspn]

A forth alternative Mark did not mention is session mixing and balancing in analog, then record in DSD. That's how the majority of Jared's catalog is recorded, as well a surprisingly large number of emerging boutique labels recording natively to DSD, and as most of the tape transcriptions to DSD are done today.

 

It's much easier, and less expensive, to roll into a location recording session, track each of the mics independently at full level, then fix it later in the mastering mix. And many times that's the only alternative. But labels and their producers and recording engineers striving for the most realistic recordings, as opposed to producing the most pleasing works of sonic art, find the venue and microphone selection and placement that produces a balanced session mix that then just requires editing.

 

Also not mentioned are the many recordings recorded, edited and mixed on Sonoma and SADiE workstations that do perform mastering functions (mixing, level adjustment/balancing, and effects processing), without converting to PCM /DXD). It really depends upon the producers objectives for the type of music to be released.

[tailspn]

Now, what are the digital workstations able to record in pure DSD - Pyramix, but AFAIR it's DXD, so kinda of PCM right?

 

Pyramix is a software system, like Pro Tools, that supports its own hardware subsystem. That was originally the Mykerinos series of I/O cards that supported external Analog to Digital and Digital to Analog Converters, and currently the Horus and Hapi A/D and D/A's networked system. It always supported native DSD recording, now up to 256fs/11.2MHz bit rates.

 

The Pyramix system, an outgrowth of the original Philips developed DSD recording/PCM post production system of the late 90's, uses DXD (352.8KHz 24bit PCM) as the processing format for most post production processes. This is in contrast with the Sonoma and SADiE Digital Audio Workstations that used the Sony Developed E-Chip, a processing engine and modulators allowing 8-bit DSD processing at the 2.82MHz original bit rate. Therefore none of the decimation filtering necessary for rate conversions.

 

There are several new entries from Tascam and Korg that all support DSD recording, but format unknown to me very limited editing and post processing functions.

 

Whether DSD or PCM, the objective as hardware technology gets faster is to pack more resolution and quantization noise separation into the "defined fixed" audio band of interest. It all sound much better than it used to.

[mkrzych]

Pyramix is a software system, like Pro Tools, that supports its own hardware subsystem. That was originally the Mykerinos series of I/O cards that supported external Analog to Digital and Digital to Analog Converters, and currently the Horus and Hapi A/D and D/A's networked system. It always supported native DSD recording, now up to 256fs/11.2MHz bit rates.

 

The Pyramix system, an outgrowth of the original Philips developed DSD recording/PCM post production system of the late 90's, uses DXD (352.8KHz 24bit PCM) as the processing format for most post production processes. This is in contrast with the Sonoma and SADiE Digital Audio Workstations that used the Sony Developed E-Chip, a processing engine and modulators allowing 8-bit DSD processing at the 2.82MHz original bit rate. Therefore none of the decimation filtering necessary for rate conversions.

 

There are several new entries from Tascam and Korg that all support DSD recording, but format unknown to me very limited editing and post processing functions.

 

Whether DSD or PCM, the objective as hardware technology gets faster is to pack more resolution and quantization noise separation into the "defined fixed" audio band of interest. It all sound much better than it used to.

 

That's fine, but editing must be done in another domain or no mastering at all right?

[Paul R]

P It all sound much better than it used to.

 

Yep! That's a statement I utterly agree with! :)