Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Jordanosaur]

Hello everybody -

I work for a semi-major audio/video equipment manufacturer who happens to market a few models of universal/SACD players.  A frequent question that I receive in tech support concerns the encoding/printing process to the physical SACD disc.  I know that the single-bit DSD encoding process can acheive up to 2.8 MHz,  and is claimed to be a superior conversion process over PCM due to lack of filtering and decimation involved during conversion.  Here's my question(s):  When the SACD media is printed, is the same one-bit DSD information printed onto the disc, or is there a downsampling process (not lossy compression) involved that allows that data to fit onto the physical media?

The units that I am familiar with do contain a DSD decoding chip, but I am curious to know if this will be faithfully decoding DSD information, as opposed to decoding a lower resolution digital stream.  

Other than that, any thoughts on the quality of the SACD?  I have only listened to a few in our demo room, and although I don't prefer multichannel mixes, the 2ch mixes do sound pretty good.  The common "audiophile" opinion is that SACD mixes promote less "fatigue" to the listener as opposed to the standard PCM format - I don't neccessarily agree with this ideal, as I have no serious problems with regular CD's.

Anybody?

Jordan Schultz
Marantz America

     

[Nika Aldrich]

Jordan,

Oh, wow.  Big questions.  Let me see if I can help.

Jordanosaur wrote on Fri, 01 October 2004 21:46

I know that the single-bit DSD encoding process can acheive up to 2.8 MHz,

This is a little bit of a misnomer.  The SACD spec requires that the data be sampled using only 1 bit, and that it do this 2.8224 million times per second.  Many processes can "achieve" 2.8224MS/s, and DSD can actually "achieve" higher than that, if you will.  The SACD spec defines what goes on an SACD disk, and that spec has determined that the material that qualifies as SACD is 1 bit, 2.8224MHz.  This coincides with 64x the speed of sampling on a traditional audio compact disk.

Quote:

and is claimed to be a superior conversion process over PCM due to lack of filtering and decimation involved during conversion.

Yes, there are people who claim that.

Quote:

Here's my question(s):  When the SACD media is printed, is the same one-bit DSD information printed onto the disc, or is there a downsampling process (not lossy compression) involved that allows that data to fit onto the physical media?

No, it is the 1 bit material.  You ask if it is downsampled so it can "fit" onto the physical media.  Well lets look at the numbers:

16 bit, 44.1KS/s means that every 44,100th of a second 16 bits are used.  This means that 705600 bits are needed per second.

At 24 bit, 96KS/s (as DVDs are capable of) we use 24 bits every 96,000th of a second.  This means that we use 2,304,000 bits per second.

At 24 bit, 192KS/s (as DVDs are also capable of) we use 16 bits every 192,000th of a second.  This means that we use 4,608,000 bits per second.

For DSD we use 1 bit 2,822,400 times per second.  This means we need 2,822,400 bits per second.  Notice that that is LESS than 16 bit, 192KS/s requires.

Also remember that DVD disks (which is what SACD actually uses) is capable of much greater than this anyway.  A DVD disk can spool off more than three times that amount of information per second.  A DVD disk can spool off 9,600,000 bits per second, and this is approximately what is used with DVD video disks you buy, or with full surround material in uncompressed PCM format (24 bit 96KS/s in the front three channels and lower rates in the rear and sub).

So to answer your question, no downsampling needs to occur in order to "fit" the SACD material on the disk.  The disk can actually hold about three times the data that SACD requires.

Further, DSD manufacturers WOULDN'T do this downsampling, because that would negate specifically the supposed benefits they claim above - that no filtering or downsampling is required!  

Quote:

The units that I am familiar with do contain a DSD decoding chip, but I am curious to know if this will be faithfully decoding DSD information, as opposed to decoding a lower resolution digital stream.

We need to know more about this "decoding" chip.  My guess is that this is because when the material is put on SACD disks it is "scrambled," if you will, for the sake of error correction and copy protection.  In other words, the data doesn't go on the disk as sample 1 goes first, then sample 2, then sample 3, etc.  I am not sure exactly how it is done, but I have to imagine they take the first 64 samples as a chunk of data and then scramble that data in various ways before putting it on the disk - complete with a header for the "block."  This is essentially what is done with CDs as well.  Sample 1 does not go first, etc.

Quote:

Other than that, any thoughts on the quality of the SACD?

Well, now, there's a big can of worms.    

Quote:

The common "audiophile" opinion is that SACD mixes promote less "fatigue" to the listener as opposed to the standard PCM format

Right.  uh-huh.  You betcha.

Nika.

[Barry Threw]

Nika,

I remember it being explained to me, in a way that made sense, that DSD had inherant high end problems compared to regular PCM.  I can't recall the exact details, but I think it was based on the noise shaping filters used to push the quantization noise out of the audible range.  If you could clairify this argument for me it would be greatly appriciated.

Thanks for you time,

[locosoundman]

I think that in theory, 96k/24 bit PCM has less total noise energy throughout the entire spectrum than noise shaped 1-bit convertors.  To me it makes more sense to use 96/24 PCM, because the audiophile argument is that the high-frequency information that we cannot hear still affects our perception of the sound (which is no doubt true and makes sense even in an acoustic argument).  Since noise-shaping shunts the bulk of the noise into this realm of the sound spectrum, it follows that this high level of noise would mask any sound going on up there anyway.

24 bit PCM has a flat noise floor that is below the threshhold of hearing throughout the reproduced sound spectrum.  Thus using a high sample rate (like 96kHz) would reproduce those frequencies without covering or modifying them with noise.

Of course, with really high sampling rates, we could get into the argument that clock jitter makes more of a difference in audio quality than the coding scheme.

Best,
Rob

[Nika Aldrich]

Barry Threw wrote on Sat, 02 October 2004 07:17

Nika, I remember it being explained to me, in a way that made sense, that DSD had inherant high end problems compared to regular PCM.  I can't recall the exact details, but I think it was based on the noise shaping filters used to push the quantization noise out of the audible range.  If you could clairify this argument for me it would be greatly appriciated. Thanks for you time,

Barry,

I'm going to have to think on this and see if I can find a simple way to explain it.

Nika.

[danlavry]

“So to answer your question, no downsampling needs to occur in order to "fit" the SACD material on the disk. The disk can actually hold about three times the data that SACD requires.
Further, DSD manufacturers WOULDN'T do this downsampling, because that would negate specifically the supposed benefits they claim above - that no filtering or downsampling is required!

Nika.”

The claim for DSD is the fact that with 1 bit you get perfect linearity, at least in theory. I never heard them say that that “no filtering or downsampling is required!”.
In fact, filtering is required. You would not want to feed your speaker a 2.8Mhz data stream of 1’s and 0’s!

Feeding DSD to an analog filter is a very difficult thing to do. The filter is relatively easy, but the data at that point is not just a digital train of 1’s and 0’s. At that point it is analog train of 1,s and 0’s so the voltage levels need be very steady and low noise, and the time durations must be such that all 1’s have the same contributions. Given that rise time (going from 0 to 1) and fall time (from 1 to zero) are not equal, the weight of a 1’s in say a pattern of 001010 is not equal to 001100…. That is a big problem. There are some solutions, but it is not easy.

Early attempts by IC makers (Crystal) to allow for external filtering of a DSD train were so difficult that they all decided to have the filter inside the IC - a switch capacitor filter is the common (only?) way it is done.

Also, many of the DA IC’s do decimate the data by some factor to ease the problems.
One way or the other, when you look at the whole process, there is filtering involved, and at times decimation.

BR
Dan Lavry

[Nika Aldrich]

danlavry wrote on Sat, 02 October 2004 21:26

The claim for DSD is the fact that with 1 bit you get perfect linearity, at least in theory. I never heard them say that that ?no filtering or downsampling is required!?. In fact, filtering is required. You would not want to feed your speaker a 2.8Mhz data stream of 1?s and 0?s!

Dan,

Yes, good call.  I should have said "digital filtering" or "decimation" or "downsampling."  As you stated above, though, even those features are implemented on some playback systems, despite the supposed benefits of the system.  Also, I would hardly claim that a system that has built in distortion has "perfect linearity," but then again, that was their claim, not yours.  I know that you and I are on the same side of this equation.

Cheers!
Nika.

[Brandon Schexnayder]

Barry Threw wrote on Sat, 02 October 2004 07:17

I remember it being explained to me, in a way that made sense, that DSD had inherant high end problems compared to regular PCM.  I can't recall the exact details, but I think it was based on the noise shaping filters used to push the quantization noise out of the audible range.

I don't know if this is in a similar direction to your question; however, most of what I have read has essentially stated that 1-bit converion in and of itself is an extremely noisy quantization process, thus much more noise-shaping has to occur.  Unlike a 44.1 kHz recording that has to do noise-shaping through an EQ curve that diminishes the noise at highly audible frequencies like 1k or 2k, DSD has a high enough bandwidth to just throw all that noise up completely above 20k.  In addition, if I understand it correctly, the noise-shaping "curve" is more of a linear increase in power starting at 20k.

Thus most opponents of DSD have been arguing over a sort of high frequency "ringing."  Don't know how someone would be able to hear that at 20k, but that is possibly one of the biggest arguments I have been seeing against DSD.

[Loco]

Schexnayder wrote on Mon, 15 November 2004 03:45

Thus most opponents of DSD have been arguing over a sort of high frequency "ringing."  Don't know how someone would be able to hear that at 20k, but that is possibly one of the biggest arguments I have been seeing against DSD.

I have this SACD disc thas someone gave me at a surround conference a while ago with some sweep tones. So I went to the consumer side on a Denon multiplayer into an Onkyo receiver to Polk LSi speakers which claim to be capable of 26KHz. Measured the tone and indeed the system went to 28 KHz at -3db  from a SACD source which is way beyond of what you really need. I saw no ringing on the spectrum analyzer, though.

As for the bitrate usage of SACD being lower than a DVD-A.... SACD carries a single layer of DVD format with both stereo and multichannel DSD material, and a layer of CD material. It's a compromise that is good enough for consumers as my test proved.

[Nika Aldrich]

Schexnayder wrote on Mon, 15 November 2004 03:45

I don't know if this is in a similar direction to your question; however, most of what I have read has essentially stated that 1-bit converion in and of itself is an extremely noisy quantization process, thus much more noise-shaping has to occur.

It is the same amount of noise-shaping that was used from around 1990 to 2000, while single bit delta sigma modulators were the basis for all A/D and D/A conversion in our industry.  The DSD format simply took an off-the-shelf A/D converter at the time and used the code straight off the modulator, prior to downsampling to 44.1kS/s.  It is more noise-shaping than is required with today's technology, however.

Quote:

Unlike a 44.1 kHz recording that has to do noise-shaping through an EQ curve that diminishes the noise at highly audible frequencies like 1k or 2k, DSD has a high enough bandwidth to just throw all that noise up completely above 20k.  In addition, if I understand it correctly, the noise-shaping "curve" is more of a linear increase in power starting at 20k.

It's actually not very linear, but this is inconsequential.

Quote:

Thus most opponents of DSD have been arguing over a sort of high frequency "ringing."  Don't know how someone would be able to hear that at 20k, but that is possibly one of the biggest arguments I have been seeing against DSD.

I think we should call it "warbling" because "ringing" is a different sensation that is often brought up these days with regards to filter behavior.  The warbling or chirping is only manifested in certain conditions (that obviously wouldn't be met on a "test disk" released by Sony!).  It can be manifested with very low levels of DC, for example.

Nika

[Bob Olhsson]

To answer his question, yes, a lossless data algorithm is employed although I understand only for multichannel SACDs.

[Brandon Schexnayder]

Not to take too much stock in a Sony produced paper regarding DSD, but I was wondering about the accuracy of the noise-shaping illustratin in  "Super Audio CD: A Technical Proposal" (http://www.sel.sony.com/SEL/consumer/dsd/dsd.pdf).  This is obviously an extremely basic explanation of DSD and its properties, so is there a better illustratin or explanation around that shows where the SDM actually puts the noise (an explanation that someone like me, a not so electronic engineering attuned individual, could understand)?

thanks

[danlavry]

Schexnayder wrote on Mon, 15 November 2004 22:08

Not to take too much stock in a Sony produced paper regarding DSD, but I was wondering about the accuracy of the noise-shaping illustratin in  "Super Audio CD: A Technical Proposal" (http://www.sel.sony.com/SEL/consumer/dsd/dsd.pdf).  This is obviously an extremely basic explanation of DSD and its properties, so is there a better illustratin or explanation around that shows where the SDM actually puts the noise (an explanation that someone like me, a not so electronic engineering attuned individual, could understand)? thanks

The plot on page 7 of the article is a simplification. I do not believe Sony or anyone ever pinned down a specific curve (a specific noise shaping filter, or even a filter order). There is a compromise between the dynamic range and the bandwidth (the usable noise free range). Most DSD gear provides about 0-22KHz of audio, and the noise start rising at about 22KHz. One could go for say "30KHz audio", but the dynamic range would suffer. You can view the process as yielding 2 regions, one for the audio, the other for the noise. Say the audio is 0-22KHz, then the frequency range left for the noise is 22KHz-1.4112MHz.

Gerzon proved that the best you can do is to shift the noise (you can not reduce or eliminate it). The "area under the curve" is fixed. It is analogous to digging a hole in the ground. The dirt removed from the hole needs to go "somewhere". So digging deep or digging wide yields the same amount of dirt...

So why not dig both deep and wide? Why not pile up more noise at the high frequencies? There is is one more issue constraining what can be done. The system needs to be stable (no oscillations) and one can not overdo the noise shaping beyond a certain point.

At the end of the day, most DSD shaping is done by a 5 pole filter in the feedback loop (maybe an occasional 7 pole filter). I believe it is at the point of diminishing returns, unless one is ready to add bits or increase the oversampling ration from 64fs to 128fs.

Regards
Dan Lavry

[chap]

I don't know about the industry or consumer embrace of
SACD.  I know this.  Last year a friend of mine remastered the
Dylan catalog.  On each disc is a regular cd, an SACD and a surround mix..
We attended a Sony sponsored premiere of the recordings.  The regular cd mix sounded really good, the SACD mix sounded fantastic.
A very focused and solid center, great width and depth of field
and amazing presence without eq.  I was amazed.
Then, the surround mixes.  I didn't get that at all.  It's Dylan!
Some mixes had the stereo return behind us and some were a guy  with an acoustic guitar, singing songs.  Why?  He doesn't move around, the snare drum is never behind us.
I left with a good impression of SACD and the idea of a folksinger being mixed in surround, as being an absurd excercise.  He's over there, wait!, he's over here, wow, he's in 2 places.  What about the songs?  It was tricks for tricks sake.
Remember those stereo demo records with a guy walking from one side to the other?  I think he was looking for his seat at the Dylan show.
chap