Music Player Network
Previous Thread
Next Thread
Print Thread
Page 5 of 6 1 2 3 4 5 6
Joined: Apr 2001
Posts: 157
R
Senior Member
Offline
Senior Member
R
Joined: Apr 2001
Posts: 157
George,

If we keep pushing for ever larger amounts of information content in our digital signals, I would guess that the conversion process will start moving faster in the direction of higher sampling rates. As David suggested above, the information content of a signal is the product of the length of the code words (bit depth) and sampling rate.

With today's technology, we're much better at building super fast clocks than we are at resolving very small voltages. DSD's ~3MHz sampling rate has a long way to go before it bumps up against the ~100GHz bandwidth offered by high-end oscilloscopes. This isn't exotic equipment in the larger (non audio) world. The ADC's of these beasts are 12-bits wide and can resolve single microvolts using off-the-shelf instrument grade preamps. Today's best audio converters are skimming the same voltage range and there's not much farther to push since Johnson Noise Land lives just a few dB down from there (at least at room temperature). In other words, there's lots of land to explore bandwidth wise, and not much more amplitude wise.

So if the industry keeps pushing for more bits-per-second, we will all see higher conversion rates. But is that important for production? I recall from last year's monster 96k thread, that we really don't have all the kinks sorted out in real-world converters--at least not to the point where we can take advantage of all those bits flying by really fast.

-Dennis

As an aside ...

Indeed you can view the triumph of digital computing over analog computing as a victory of the precision clock-makers over precision voltage measurers. Digital has two levels of interest but takes time (you need a fast clock). Analog computers work almost instantaneously, but in order to get high-quality answers you need to measure really precisely. Well who won?

Joined: Feb 2002
Posts: 2
D
Member
Offline
Member
D
Joined: Feb 2002
Posts: 2
I'd made the observation that the choice between 24-bit, 96ksps and 4-bit, 576ksps was an irrelevancy from an information-theoretic standpoint.

Quote:
Originally posted by George Massenburg:

...or, as Ed Meitner suggests, you can do single-bit at 256FS.


George, I think that's the one thing we don't want to do! Using a one-bit as an intermediate storage format to which we return after each processing step is a really, really bad idea. That's because to get back there, we have to go through another single-bit modulator, and those are inherently and irretrievably flawed (c.f. Lipshitz and Vanderkooy). If the marketplace dictates that we have to put our audio through a one-bit modulator in order to deliver it to the consumer, well, so be it. But let's do that once, and only once, at the end of the production chain.

Quote:
GM again:

In fact, it may matter how you get there. What if there were sample-rate related artifacts (as distinct from the bandwidth limitations, including filtering problems, that have everyone's attention) in PCM? Then, only a higher sample rate is going to improve things? And, if that's true, what's the 'next frontier' in sample rates? 96kHz doesn't seem like a big step. Maybe 192 (which sounds pretty good to me)?


You may be correct. Some people think that PCM (as implemented) doesn't start to compete sonically with DSD until 192ksps. If that's so, it's not because of "channel capacity", but because of what we're putting into the channel. Whether it's the time response of the filters, a crappy quantized-coefficient implementation of them, or aliasing artifacts inadvertantly modulated down to baseband, something is messing up the conversion process and the semiconductor manufacturers are too focused on price-point to fix it.

Sony's solution was to avoid having to decide what's the best decimation filter at record time. (Unfortunately, they've effectively left the decision in the hands of the folks who make $200 playback equipment.) I'm rather inclined to agree with them, at least in the following way: If I can't buy a 96ksps PCM converter with a decimation filter that I can trust, then maybe I should use a 192ksps converter, followed by a software down- (or up-) conversion algorithm whose provenance and performance is known to me. In fact, I can do all my intermediate production at 192ksps if I decide that sounds any better (and I don't have to mortgage my house to buy the tools). Finally, if it takes a one-bit modulator to get my work to consumers I can still do that at the end.

Rereading your comment, I realize that you what you meant by "sample-rate related artifacts" was not problems in the decimation filter, but subsequent processing artifacts which alias into the audio band. I agree that oversampling will reduce these. But how much oversampling is actually needed? If we process our audio with something as egregious as a squaring circuit, it will only double the bandwidth. So if we perform the processing at 192ksps, being careful to filter to 96kHz bandwidth before each processing step, we should be in good shape.

Best regards,

David L. Rick
Seventh String Recording

Joined: Mar 2001
Posts: 2,938
N
MP Hall of Fame Member
Offline
MP Hall of Fame Member
N
Joined: Mar 2001
Posts: 2,938
I just wanted to throw in that I'm agreeing with one of the opinions above.

There is nothing wrong with SACD as a delivery format, and I believe that there are probably advantages: it will be much more practical for manufacturers of home hi-fi equipment to design that equipment so that they can meet their demands for affordability while providing better sound quality. SACD is a fairly practical way of allowing the mastering studio to do the D/A filtering, standardizing this one aspect of the playback system. At least we engineers will be able to ensure the quality of the D/A filter, while there are still many areas where we will have no control.

I am still not in agreement that DSD converters can be made to sound better than PCM converters can be made to sound, and as this is the case, I still support working at lower rates, conserving harddrive space, and not simply subsribing to higher rates of everything for unsubstantiated everything. Therefore, I do not see any reason to be in support of recording DSD, and processing at DSD is practically impossible without first converting to PCM of some form anyway. But delilvering at DSD is something I haven't seen a valid reason to avoid, and I do see benefits of supporting.

Nika.

Joined: Apr 2001
Posts: 413
E
Senior Member
Offline
Senior Member
E
Joined: Apr 2001
Posts: 413
Gentlemen,
I respectfully suggest that Mike Bishops posts be re read and that he be prompted for more input.
Several pages back I think his posts got shouted down a bit.

As a former Telarc artist I can clearly state that I have absolutely nothing positive to say about the label. I hated being there.
BUT to their credit, Telarc relentlessly pursues Audio Excellence.
They do this without being a multinational conglomerate that also manufactures gear (ie. sony)
Telarc has always been on the very bleeding edge and it is no surprise to see them out there now. Somehow Telarc spends studip amounts of money on the highest calibre gear and still stay in business!
And again, I clearly state that I was treated like shit as an artist on there roster. My respect for their quest for audio excellence remains in spite.
Mike mastered my first two records. Not for a minute did I sense the slightest air of audio superiority (or bullshit) about him.
He is all about SOUND.
He was always available to solve tech issues (and I had a "first batch" 02R that needed much help!) and helped me get the most out of the gear that I had. Not once did he put down my studio gear. (02R-DA88's at that time, far below Telarcs usual stuff)
Is it possible to be a cutting edge audio elitest and a humble, practical, down to earth engineer?!
Yes.
I have never seen evidence of Mike having an agenda of any sort. His picture is seldom in MIX mag standing next to the latest Digital Mixer and Phil Ramone.
His car was like all of ours (late late late model - he aint getting rich endorsing anything).
SO,
If HE hears a difference, well shit then, I have to believe that there is a great deal to it.
There are only a very few people whose opinion would carry that much weight with me.
Mr. Bishop is one of them

Also,
Crafty, Please post some bio info on yourself and perhaps you real name. Almost every one in here puts there entire reputation on the line when posting. Please be kind enough to do the same.

as a side note:
I am sure that when digidesign comes out with PT DSD - the upgrade path will suck my bank account dry, as the one to HD just did!

Joined: Mar 2000
Posts: 930
M
Gold Member
Offline
Gold Member
M
Joined: Mar 2000
Posts: 930
Ed-

Thanks for your post.

I also hear nothing but good things about Michael Bishop (know nothing of Telarc and how they treat their artists) and was sorry to see him 'shouted down' as well.

Hoping to see him and other DSD users post here with not only their scientific arguments, but also their personal opinions.

ML

Joined: Oct 2001
Posts: 3,974
Rog Offline
MP Hall of Fame Member
Offline
MP Hall of Fame Member
Joined: Oct 2001
Posts: 3,974


"That's what the internet is for. Slandering others anonymously." - Banky Edwards.
Joined: Oct 2001
Posts: 63
P
Senior Member
Offline
Senior Member
P
Joined: Oct 2001
Posts: 63
3 Ás is 333 kHz ! Far away outside the audible range.

All signals looks perfectly clean to me : same energy (surface under the curve), but more or less lowpassed in frequency.

Joined: Oct 2001
Posts: 63
P
Senior Member
Offline
Senior Member
P
Joined: Oct 2001
Posts: 63
PS : not the same energy, of course (silly me), the highs are removed.

Joined: Jul 2000
Posts: 238
M
Senior Member
Offline
Senior Member
M
Joined: Jul 2000
Posts: 238
Quote:
Originally posted by Rog:
http://www.merging.com/2002/images/dsdresponseneon.gif

Any thoughts?


Notice the extreme difference in ringing around the main impulse between the various pcm and DSD samples. I had never before seen such an analysis and comparison of the various formats, but this very well illustrates the difference that I ***hear*** between them. In all versions of pcm recordings - at any sample rate - I hear "clatter" around the detail of the signal. It was one of those things I got used to in pcm over the past 22 years, but now find really grating after living with DSD the past 4 years. That "clatter" is also how I identify that a track is of pcm origin in comparison to a DSD track. Once you know the "clatter" is there it's easy to find on a repeatable basis, as in blind testing - and you don't have to have "golden ears" to find the artifacts.

Thanks for posting the link, Rog!

With Best Regards,
Michael Bishop
Telarc International


Best Regards;
Michael Bishop
Telarc International
http://www.telarc.com
SACD, DSD & DVD-A Editing and Mastering available now at:
Wired 4 Music
http://www.wired4music.com
also: DSD, SACD, DVD, surround recording, and pro-audio information & links
Joined: Mar 2001
Posts: 2,938
N
MP Hall of Fame Member
Offline
MP Hall of Fame Member
N
Joined: Mar 2001
Posts: 2,938
I am actually confused by the chart. How is it possible that a 333kHz impulse could get into a 44.1kS/s PCM system in ANY capacity?

Anyone? What am I missing. I must be missing something, here...

Thanx!
Nika.

Joined: Dec 2001
Posts: 135
D
Senior Member
Offline
Senior Member
D
Joined: Dec 2001
Posts: 135
Quote:
Originally posted by Nika:


Dennis,

Don't forget that the D/A's of a DSD system still have filters in them.

Nika.


I think it might be possible to move these filters all the way to the speaker driver, given enough time and R&D monies.

Thinking ahead,
Dan

Joined: Oct 2000
Posts: 4,223
C
MP Hall of Fame Member
OP Offline
MP Hall of Fame Member
C
Joined: Oct 2000
Posts: 4,223
Quote:
Notice the extreme difference in ringing around the main impulse between the various pcm and DSD samples. I had never before seen such an analysis and comparison of the various formats, but this very well illustrates the difference that I ***hear*** between them. In all versions of pcm recordings - at any sample rate - I hear "clatter" around the detail of the signal. It was one of those things I got used to in pcm over the past 22 years, but now find really grating after living with DSD the past 4 years. That "clatter" is also how I identify that a track is of pcm origin in comparison to a DSD track. Once you know the "clatter" is there it's easy to find on a repeatable basis, as in blind testing - and you don't have to have "golden ears" to find the artifacts.
Thanks for posting the link, Rog!
With Best Regards,
Michael Bishop
Telarc International


Michael,

Do you hear the same amount of "clatter" at 10K or 20K as the amount you hear at 333K? What about the 20hz to 1K range? Are there illustrations of the amount of "clatter" in that range floating around that we could look at?

Thanks!

Eric \:\)


Eric Vincent (ASCAP)
http://www.curvedominant.com
Joined: Dec 2001
Posts: 135
D
Senior Member
Offline
Senior Member
D
Joined: Dec 2001
Posts: 135
Quote:
Originally posted by td:
Curve - maybe a switch to Sanka would help and maybe you should take up David's Pepsi challenge in & around Philly instead of flogging every person that uses DSD and tries to state some thoughts about it ......... I mean they're out there on the edge of technology, trying something new. Breath in ~ breath out. Ahhhh.

Tony

[ 02-01-2002: Message edited by: td ]


Speaking of Pepsi...what do they use for sweetening up there in Philly. When I moved from Chicago to Miami, I was a bit disappointed that my daytime drink of choice (Pepsi in a can) wasn't the same. It seems that down in Florida there is a greater concentration of sugar to corn syrup AND a greater concentration of syrup to the carbonated water - I suppose this is because they expect you to have to use ice, and sugar is more plentiful down here than corn.

Perhaps a bit too sensitive,
toast

Joined: Oct 2001
Posts: 63
P
Senior Member
Offline
Senior Member
P
Joined: Oct 2001
Posts: 63
There might be some clatter in PCM that is not in DSD, but it's not what we can see in this picture.

Look at what happens to a pulse when it's mathematically lowpassed : http://pageperso.aol.fr/Lyonpio2001/pictures/lowpass/lowpass.htm

In the picture we're discussing about, PCM versions of the pulse are perfectly lowpassed versions of the analog one, with no distortion at all exept the removal of ultrasounds.

Therefore they should sound exactly the same to any human, exept maybe some asthmatic children, that are said to be able to hear up to 30 kHz. For them, the 48 kHz version will sound different from the others.
That, is valid only if the samples are actually played as we see them displayed on the picture, of course. That must be impossible since the air itself doesn't transmit the sound well, if at all, at such frequencies as 333 kHz !

For me, this picture is aimed at fooling people into believing that DSD is superior to PCM. What MAY actually be, independently of this picture, but this advertising is just deceiving of the masses.

Joined: Oct 2000
Posts: 4,223
C
MP Hall of Fame Member
OP Offline
MP Hall of Fame Member
C
Joined: Oct 2000
Posts: 4,223
Quote:
posted by digitaltoast:
Speaking of Pepsi...what do they use for sweetening up there in Philly.


I generally don't consume refined sugar or anything that contains it, so I couldn't tell you.

Rog,

Where did you find that jpeg you posted?

Any idea who created that demonstration?

E \:\)


Eric Vincent (ASCAP)
http://www.curvedominant.com
Joined: Oct 2001
Posts: 63
P
Senior Member
Offline
Senior Member
P
Joined: Oct 2001
Posts: 63
Quote:
Originally posted by Nika:
How is it possible that a 333kHz impulse could get into a 44.1kS/s PCM system in ANY capacity?


It depends on the DAC circuitry itself, i.e. how it measures the voltage. If it measures the electric charge accumulated by the current, then the same final accumulated charge in a 1/44100 of a second can be produced by bursts of current as short as you want (3 Ás) inside the 1/44100 s window.
After that, the oscillations come either from the playback filters (DAC), either from the display used.
Cool Edit actually displays pure digital pulses exactly like that, with an oscillating line between the samples : http://pageperso.aol.fr/Lyonpio2001/pictures/lowpass/cool.gif
The squares are the samples themselves. This digital file presents one sample at full scale.
The curve displayed is the regular display of Cool Edit, that draws the perfect mathematical curve fitting the points. Another proof that the oscillating pulses are not distorded.

Yet another proof : The Fourier transform of a Dirac distribution is a constant function. To lowpass it is to cut it so as it becomes a "step" function, and the inverse Fourier transform of a step function has a Cardinal Cosinus shape (Cos X/X).

Joined: Oct 2000
Posts: 4,223
C
MP Hall of Fame Member
OP Offline
MP Hall of Fame Member
C
Joined: Oct 2000
Posts: 4,223
Quote:
posted by Pio2001:
There might be some clatter in PCM that is not in DSD


If there was "clatter" in the source PCM, of course you wouldn't hear it in DSD, because all of the 3rd harmonic distortion caused by the cyclic non-random HF noise from the Delta Sigma Modulator would smother it...right?

Reference Stan Lipshitz:
Quote:
At the Audio Engineering Symposium in September, 2000, Stan Lipshitz showed that a 1-bit system is inherently non-perfectible, and he is right. You cannot properly dither without overloading the converter, because the correct dither (white noise with a triangular pdf and a width of 2 LSBs) covers the full range of the converter! Thus, as soon as you put in any input, the converter overloads. Of course the manufacturers don't put in 2 LSBs of dither, but then the converter is not correctly dithered and suffers from both distortion and noise modulation (and idle tones). This is why the state-of-the-art converters do not use 1-bit internally but rather a low number of bits, such as 3. The trend for professional audio equipment is 8-bit, 64 times oversampling, noise-shaped.

Mike Bishop? Glasser? Is this why you think DSD "sounds" like analog? Yeah, it's "less fatigueing..." on a bad mix, perhaps. I could buy a BBE Aural Exciter off of Ebay for cheap if I want to mix harmonic distortion into my tracks, or buy an "Analog Simulator" plugin - but at least I can disable a plugin. DSD-generated distortion cannot be disabled, as far as I can see.

Milk and cookies have kept The Curve up late thinking about this. I just don't see how it's a "cleaner" path as Phillips is suggesting. There are a lot of folks more knowledgable than myself who participate on this board, and this thread has been up for awhile now, but still, no one has yet explained how you dither a 1-bit signal, and where the dithering noise goes.

What am I missing? Thanks in advance.

Eric \:\)

(Edited for spelling errors)

[ 02-17-2002: Message edited by: Curve Dominant ]


Eric Vincent (ASCAP)
http://www.curvedominant.com
Joined: Oct 2000
Posts: 4,223
C
MP Hall of Fame Member
OP Offline
MP Hall of Fame Member
C
Joined: Oct 2000
Posts: 4,223
Quote:
delilvering at DSD is something I haven't seen a valid reason to avoid, and I do see benefits of supporting.
Nika.


Nika,

Read my previous post, and then re-iterate your view.

Thanks in advance.

Eric \:\)


Eric Vincent (ASCAP)
http://www.curvedominant.com
Joined: Mar 2001
Posts: 2,938
N
MP Hall of Fame Member
Offline
MP Hall of Fame Member
N
Joined: Mar 2001
Posts: 2,938
Quote:
Originally posted by Curve Dominant:


Nika,

Read my previous post, and then re-iterate your view.

Thanks in advance.

Eric \:\)


Eric,

OK, no problem. Let's review how a D/A works for a moment:

A signal at 44.1kS/s has absolutely no data in it beyond 22.05kS/s, as we know from Nyquist. But all it is is a series of points with no data between them. Depending on how that data is turned into analog signals additional information CAN get into the system.

For instance, if a D/A uses "sample and hold" then any particular value will be held until the next value comes along. This takes our series of dots and turns them into a bunch of miniature stair-steps, like square waves. This, obviously, is not desireable. What we want is a smooth line connecting all of the dots in sinusoidal fashion.

What happens is we "oversample" the material, creating a bunch of dots between all of the dots, and using filters to insure that they conform to the waveform that we actually want there. If we use an "8x oversampling" D/A it will add 8 additional dots between each pair of dots in our timeline, adding a filter that is a brickwall at 22.05kHz so that the resultant stairsteps now contain no material that is less than 529.2kHz (3rd harmonic of 176kHz, which is 8x 22.05kHz). The bottom line is that we still have stair-steps, but they are much smaller steps, and the minimum frequency over 22.05kHz that is contained in their material is up at around .5MHz. At this point the signal is actually turned into a 1 bit datastream, akin to DSD. Then it is converted to analog using "sample and hold". Then an analog low pass filter is put in at just below .5MHz that essentially "rounds" this small series of stair steps into the sinusoidal waveforms that we were looking for. This entire process is done in the D/A converter, and the quality of these filters is one of the differences between "adequate" and "great" converters. Quality is important because the digital filter that is used for the oversampling process can suffer from errors that will negatively affect your audio. This process is just as imperfect as the A/D conversion process.

The problem here is that home hi-fi equipment generally has very poor filters in it - sometimes none at all. This means that the way you hear the material in your studio does not translate as well to home hi-fi equipment.

Let's discuss the paradigm of tracking and mixing and recording PCM, and then turning it into DSD for delivery. The process for turning PCM into DSD is essentially the same process as oversampling in a D/A. The material is oversampled at 64x with a filter put in and the result is then "encoded" in a 1 bit datastream. This process would obviously happen after any dithering that would applied, and after all other processing and mastering. Basically the mastering engineer would run your 44.1kS/s datastream into a little black box that would oversample and filter it into DSD, just like your D/A's do. Except that the mastering engineer has the tools to do this process at a level that we can't afford in the studios and most of our listening audience can't afford either. Then your material is put on an SACD disk for duplication and distribution.

We're not saying here that our listening audience does not have the ability to purchase D/A's that can perform to the level that a mastering engineer's can. We're not saying that SACD will inherently sound better than we can make PCM sound, because it's really the same thing, SACD as a delivery format just being one step beyond PCM. Instead what we're conceeding here is that by taking the filters and the oversampling technology out of the consumers hands and putting them into the mastering engineer's hands we have helped to standardize what our consumers are listening to. This way:

Record, edit, mix, master -> storage (delivery format) -> {consumer playback} oversampling D/A with filtering turning it into a 1 bit DSD datastream -> conversion to analog -> to amplification -> to speakers

turns into:

Record, edit, mix, master, oversampling D/A with filtering turning it into a 1 bit DSD datastream -> storage (delivery format) -> {consumer playback} conversion to analog -> to amplification -> to speakers

Now if we could only standardize what speakers our consumers listen to!?

So again, SACD is not inherently going to make the music sound better, but by using it as a delivery format it helps to remove one potential for sound degradation from the hi-fi componentry, further standardizing the way that our consumers hear what we intended for them to hear.

Does this help?

Nika.

Joined: Oct 2000
Posts: 4,223
C
MP Hall of Fame Member
OP Offline
MP Hall of Fame Member
C
Joined: Oct 2000
Posts: 4,223
Quote:
posted by Nika:
The process for turning PCM into DSD is essentially the same process as oversampling in a D/A. The material is oversampled at 64x with a filter put in and the result is then "encoded" in a 1 bit datastream. This process would obviously happen after any dithering that would applied, and after all other processing and mastering. Basically the mastering engineer would run your 44.1kS/s datastream into a little black box that would oversample and filter it into DSD, just like your D/A's do. Except that the mastering engineer has the tools to do this process at a level that we can't afford in the studios and most of our listening audience can't afford either. Then your material is put on an SACD disk for duplication and distribution.


Yes, but don't you still end up with all of the flawed audio inherent in putting it through a 1-bit Delta Sigma Modulator? That's one of the points I seem to be stuck on. Isn't the 3rd harmonic distortion from statistical noise modulation inherent from the DSM still there? And what about all of the HF noise generated, because we have a signal that has full scale peak to peak noise up to 1.4Mhz, correct? Won't that burn some tweeters if we don't filter it - which brings us to: what type of filters? Analog? And won't that roll-off audio as well...that can't possibly compete with DVD-A standards.

Quote:
posted by Nika:
what we're conceeding here is that by taking the filters and the oversampling technology out of the consumers hands and putting them into the mastering engineer's hands we have helped to standardize what our consumers are listening to.


This is something else I don't get: What type of D/A filter? If it's 1-bit data going through a FIR filter: doesn't that just turn it back into multi-bit PCM, but with all of the flawed audio from the DSM in tact?

The only logical reason I can see for Sony/Phillips to opt for 1-bit, is because they can manufacture cheaper consumer gear because there's less information that needs to be processed at the end.

I still don't see how DSD is not PCM, except that it has to really bend over backwards and jump through hoops in order to maintain integrity as a 1-bit data stream. Which it can't anyway if it needs to be processed - "DSD Wide" is multi-bit PCM as far as I can see.

Thanks, Nika, for you patience. Sorry if I sound like a retard sometimes.

E \:\)


Eric Vincent (ASCAP)
http://www.curvedominant.com
Joined: Oct 2000
Posts: 4,223
C
MP Hall of Fame Member
OP Offline
MP Hall of Fame Member
C
Joined: Oct 2000
Posts: 4,223
Nika,

I apologize if there seems to be a delayed reaction in my posts, but as you know I'm a little new to this field, and sometimes I have to think for awile to make sure I have my facts straight (and of course I'm still not sure but anyway...).

You posted:
Quote:
What happens is we "oversample" the material, creating a bunch of dots between all of the dots, and using filters to insure that they conform to the waveform that we actually want there. If we use an "8x oversampling" D/A it will add 8 additional dots between each pair of dots in our timeline, adding a filter that is a brickwall at 22.05kHz so that the resultant stairsteps now contain no material that is less than 529.2kHz (3rd harmonic of 176kHz, which is 8x 22.05kHz). The bottom line is that we still have stair-steps, but they are much smaller steps, and the minimum frequency over 22.05kHz that is contained in their material is up at around .5MHz. At this point the signal is actually turned into a 1 bit datastream, akin to DSD. Then it is converted to analog using "sample and hold". Then an analog low pass filter is put in at just below .5MHz that essentially "rounds" this small series of stair steps into the sinusoidal waveforms that we were looking for. This entire process is done in the D/A converter, and the quality of these filters is one of the differences between "adequate" and "great" converters. Quality is important because the digital filter that is used for the oversampling process can suffer from errors that will negatively affect your audio. This process is just as imperfect as the A/D conversion process.


This seems to contradict what you've posted on your legendary "96K..." thread, as well as on the DUC. As I understand (and based somewhat on your guidance), the so-called "staircase" waveform you referred to does not exist at the output of the DAC, because the reconstruction filter removes all frequencies above the passband (20K). Isn't the fundamental frequency of the staircase at the sample rate, which is 44.1?

Also, one of the things I seem to remember you stressing constantly on the "96K..." thread, was that it was impossible to create information that has not passed via the input. This can't be what you're referring to when you say, "...we 'oversample' the material, creating a bunch of dots between all of the dots..." because it would imply that you are creating new information. Am I simply mis-interpreting your analogy?

And I don't understand how the signal could end up on a 1-bit DAC, which with its 1-bit signal path restriction must be vastly inferior to a 24-bit PCM DAC.

Can you clarify? Thanks!

E \:\)


Eric Vincent (ASCAP)
http://www.curvedominant.com
Joined: Oct 2000
Posts: 4,223
C
MP Hall of Fame Member
OP Offline
MP Hall of Fame Member
C
Joined: Oct 2000
Posts: 4,223
Quote:
posted by synesthesia:
in a conversation with a Tascram (yes, tascam of all people) rep a couple of months back, he had mentioned a conference in western canada where a mathematician proved to sony a 3bit format would be more efficient, and yield higher fidelity. i'm not sure how this relates to two's compliment...


It doesn't. It relates to dither.

Again, Stan Lipshitz, Professor of Applied Mathematics at Waterloo University:
Quote:
A 1-bit system is inherently non-perfectible...You cannot properly dither without overloading the converter, because the correct dither (white noise with a triangular pdf and a width of 2 LSBs) covers the full range of the converter! Thus, as soon as you put in any input, the converter overloads. Of course the manufacturers don't put in 2 LSBs of dither, but then the converter is not correctly dithered and suffers from both distortion and noise modulation (and idle tones). This is why the state-of-the-art converters do not use 1-bit internally but rather a low number of bits, such as 3.


Eric \:\)


Eric Vincent (ASCAP)
http://www.curvedominant.com
Joined: Oct 2000
Posts: 4,223
C
MP Hall of Fame Member
OP Offline
MP Hall of Fame Member
C
Joined: Oct 2000
Posts: 4,223
Quote:
posted by Mike Bishop:
Transferring 16-bit 44.1 pcm recordings to DSD does not get you anything more than what you started with.


That is correct, yet deceitful: it will get you less than what you started with. 1-bit conversion of a 16/44.1 pcm recording will produce artifacts inherent of the limitations of 1-bit conversion. The 1-bit data path will introduce errors in the signal that cannot be removed, by any means, ever.

Mike, unless you can show me hard math that contradicts this, you stand corrected. Show me the math.

Eric \:\)


Eric Vincent (ASCAP)
http://www.curvedominant.com
Joined: Mar 2001
Posts: 2,938
N
MP Hall of Fame Member
Offline
MP Hall of Fame Member
N
Joined: Mar 2001
Posts: 2,938
[QUOTE]Originally posted by Curve Dominant:
Yes, but don't you still end up with all of the flawed audio inherent in putting it through a 1-bit Delta Sigma Modulator? That's one of the points I seem to be stuck on. Isn't the 3rd harmonic distortion from statistical noise modulation inherent from the DSM still there? And what about all of the HF noise generated, because we have a signal that has full scale peak to peak noise up to 1.4Mhz, correct? Won't that burn some tweeters if we don't filter it - which brings us to: what type of filters? Analog? And won't that roll-off audio as well...that can't possibly compete with DVD-A standards.

The point here is to put it through exactly the same signal path and exactly the same D/A conversion as PCM does once PCM is upsampled and reduced to a DSD type of data stream.

You have to remember that a D/A converter has (2) filters in it that function together to perform all of the filtering. One is digital. The other is analog. The digital filter is added during upsampling and reduction from PCM to DSD. The analog filter is added after the D/A conversion. both, however, are said to happen "in the D/A converter", and both are an integral part of the D/A conversion process.

The idea behind DSD as a delivery format is to take half of this filtering process out of the consumer's hands. While in the 96k thread we discussed the theoretical aspects of whether or not it was possible to design a perfect filter, in here we're discussing the real world applications, that we can't leave it up to hi-fi companies to design perfect filters, so we might be better off removing as much of that from their control as we can.

This is something else I don't get: What type of D/A filter? If it's 1-bit data going through a FIR filter: doesn't that just turn it back into multi-bit PCM, but with all of the flawed audio from the DSM in tact?

I don't think that this is how it works. It'd be nice if Mr. Frindle was around. We take PCM data and filter/upsample it in one fell swoop, then modulating it to a one bit datastream? So it's not 1 bit going through an FIR. It's 24 bits going through an FIR.

I still don't see how DSD is not PCM, except that it has to really bend over backwards and jump through hoops in order to maintain integrity as a 1-bit data stream. Which it can't anyway if it needs to be processed - "DSD Wide" is multi-bit PCM as far as I can see.

Well it kind of really IS just a really fast PCM data stream where we leave off a bunch of the bits - kind of.

Notice that I'm not addressing doing ANY processing in DSD. All I'm addressing is taking a 24 bit signal and upsampling and filtering it to the same rate as DSD, then turning it into DSD 1 bit and putting that on a disk for delivery. Then all the consumer's gear has to do is have a very gentle analog filter to roll off all of the high frequency stuff that we haven't gotten rid of yet - basically the last filter in a typical PCM D/A, but in this case it will be the only filter, because the first half of that filtering process was done at the mastering house.

Nika.

Joined: Mar 2001
Posts: 2,938
N
MP Hall of Fame Member
Offline
MP Hall of Fame Member
N
Joined: Mar 2001
Posts: 2,938
Quote:
Originally posted by Curve Dominant:


That is correct, yet deceitful: it will get you less than what you started with. 1-bit conversion of a 16/44.1 pcm recording will produce artifacts inherent of the limitations of 1-bit conversion. The 1-bit data path will introduce errors in the signal that cannot be removed, by any means, ever.

Mike, unless you can show me hard math that contradicts this, you stand corrected. Show me the math.

Eric \:\)


Eric,

Allright, here's the math:

The process that the data goes through to become a 1 bit datastream is exactly the same process that the data goes through to become converted to analog from normal PCM anyway, namely: upsampling and filtering, then turned into a 1 bit signal. That's the same process as any PCM converter you have on your shelf.

So if there's going to be inherent loss because the upsampling/filtering algorithms aren't perfect, it happens to both your PCM datastream at the conversion AND EQUALLY to the PCM datastream that gets converted to DSD, only to be further converted to analog. This is because DSD is the halfway step of the conversion process.

Now, in the land of theoreticals, those filters can be made to be perfect, right? But in a practical world we may have issues with that. This is just a way of taking SOME of the conversion process out of the hands of the end users and putting it into the hands of mastering engineers who have VERY GOOD, or rather VIRTUALLY TRANSPARENT upsampling algorithms.

Nika.

Joined: Mar 2001
Posts: 2,938
N
MP Hall of Fame Member
Offline
MP Hall of Fame Member
N
Joined: Mar 2001
Posts: 2,938
[QUOTE]Originally posted by Curve Dominant:

This seems to contradict what you've posted on your legendary "96K..." thread, as well as on the DUC. As I understand (and based somewhat on your guidance), the so-called "staircase" waveform you referred to does not exist at the output of the DAC, because the reconstruction filter removes all frequencies above the passband (20K). Isn't the fundamental frequency of the staircase at the sample rate, which is 44.1?


Yes, but the "reconstruction filter" is actually two filters - the digital and the analog working in combination with each other - one before conversion to analog and one after. That's all one big "reconstruction filter". I have attempted in this thread to dig into more of how that "filter" or "filters" work, describing each step of that filtering process and what the signal is at each step along the way.

Also, one of the things I seem to remember you stressing constantly on the "96K..." thread, was that it was impossible to create information that has not passed via the input. This can't be what you're referring to when you say, "...we 'oversample' the material, creating a bunch of dots between all of the dots..." because it would imply that you are creating new information. Am I simply mis-interpreting your analogy?

Hmm. No. We're not "creating" that information. The two dots rather "inherently imply" a waveform passing through them. We are simply further defining where that waveform passes by adding additional dots in it's path. That's different than "creating new information". Do you follow?

And I don't understand how the signal could end up on a 1-bit DAC, which with its 1-bit signal path restriction must be vastly inferior to a 24-bit PCM DAC.

Not at all. The 1 bit DAC is actually a part of the 24 bit PCM DAC.

I hope this helps. I'm going to bed now.

Nika.

Joined: Jan 2002
Posts: 4,649
MP Hall of Fame Member
Offline
MP Hall of Fame Member
Joined: Jan 2002
Posts: 4,649
I follow the gist of what you're saying, Nika, and I do think SACD as a delivery format sounds appealing in this regard. I have a profound distrust for super-cheap digital. So much better to have a simple cheap analog only for the "consumer" piece. Even a very expensive reproduction system would be hard pressed to have the quality of D/A conversion a good mastering house has.
Ted


A WOP BOP A LU BOP, A LOP BAM BOOM!

"There is nothing I regret so much as my good behavior. What demon possessed me that I behaved so well?" -Henry David Thoreau
Joined: Oct 2001
Posts: 3,974
Rog Offline
MP Hall of Fame Member
Offline
MP Hall of Fame Member
Joined: Oct 2001
Posts: 3,974
Quote:
Originally posted by Curve Dominant:


I generally don't consume refined sugar or anything that contains it, so I couldn't tell you.

Rog,

Where did you find that jpeg you posted?

Any idea who created that demonstration?

E \:\)


Sorry Eric, I forget - it may have been from the Nuendo forum. Maybe you could find out by mailing someone at Merging?


"That's what the internet is for. Slandering others anonymously." - Banky Edwards.
Joined: Oct 2000
Posts: 4,223
C
MP Hall of Fame Member
OP Offline
MP Hall of Fame Member
C
Joined: Oct 2000
Posts: 4,223
Quote:
posted by Nika:
The 1 bit DAC is actually a part of the 24 bit PCM DAC.


What??!!!

The 1-bit DAC is a part of WHOSE 24-bit PCM DAC exactly? Crystal? AKM? Burr Brown?

It is possible that there are some old, badly sounding, inefficient convertors floating around out there that used the 1-bit format. Most manufacturers are steering clear of 1-bit these days, as I understand, because of its inherent flaws.

C'mon, Nika, I thought you were the "go-to" guy on this stuff. Don't almost all of the latest generation of high quality convertors use multibit internal delta/sigma modulation techniques? As I understand, some very high quality ones actually use oversampled 24bit direct conversion.

Nika, at any rate, please let us know which convertor manufacturers are still using 1-bit DSM components, so we can all steer clear of that manufacturer and its convertors.

E \:\)


Eric Vincent (ASCAP)
http://www.curvedominant.com
Joined: Mar 2001
Posts: 2,938
N
MP Hall of Fame Member
Offline
MP Hall of Fame Member
N
Joined: Mar 2001
Posts: 2,938
Eric,

From my understanding (which is admittedly a little thin at this point in the conversation) there is an inherent benefit in multibit at the same sample rate, but the one bit DSD is at a much faster rate than multibit oversampled PCM, to the point that I think that the drawbacks don't compare.

8x oversampled, but multibit vs. 64x oversampled but one bit. Is there an advantage? I don't know. I was led to believe that they could be effectively the same.

Nika.

Page 5 of 6 1 2 3 4 5 6

Moderated by  gm 

Link Copied to Clipboard
Powered by UBB.threads™ PHP Forum Software 7.7.5