Music Player Network
Previous Thread
Next Thread
Print Thread
Page 11 of 11 1 2 9 10 11
Joined: Aug 2000
Posts: 1,242
Platinum Member
Offline
Platinum Member
Joined: Aug 2000
Posts: 1,242
Quote:
Originally posted by Nika:
[QUOTE]Originally posted by Loco:
At 20 KHz it's actually around 380db.

OK, now you seem to be just making stuff up out of nowhere. Please show me the spec sheet that shows 380dB of dynamic range at 20KHz on a DSD system...
At roughly 6 dB per bit... 64 bit oversampling of 44100 yields around 384 dB. Asuming we lose one bit on filtering it's 192 dB which is still bigger than life.

Or show me the spec where it shows 120 dB.

Quote:

What you are describing is effectively neither PCM nor DSD, but it is close to both. It isn't DSD because you aren't showing "relative positions" with this scheme. It's not relative because you keep having "down" movements while the waveform's amplitude is increasing. Clearly it can't be "relative" if the data is moving in opposite direction to the waveform, can it?

This is referencing your 110110110... triangle waveform. If it's a triangle wave represented by "relative" position then why do you have "0's" in there during the increasing stage?

Nika.
Because after a while the expected step is lower than the actual step taken. Grab a milimetric piece of paper and draw sine wave. Sometimes the wave won't go to the next step but stays on the same (it's called quantization). Then it has to play catch up.

Just like a flat line. On DSD it's 101010101010 because the code has to go up or down. Once it goes up, the line is lower than expected. Next time, its bigger and then smaller again. That's why you lose one bit. The same bit you lose on PCM on DC when it's both and neither +0 and/or -0.

Time to stop the presses?


"There's no right, there's no wrong. There's only popular opinion" Jeffrey Goines
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
Loco,

So it's not relative as you were saying then, yes? Because if it were relative then whenever the waveform increased in amplitude the data would be a "1" and whenever it decreased in amplitude the data would be a "0." Clearly you are saying it is different than that if, while the amplitude increases, the sample values decrease.

This implies that the system is apparently more complicated than being "relative." Recognizing this, please now explain how you think it really works. And how is the data then turned from whatever you think it is to PCM? What kind of filters are used and how are they implemented?

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 Loco:
At roughly 6 dB per bit... 64 bit oversampling

Hold on, there. It's not "64 bit oversampling." It's a single bit system at "64 times oversampling." This means that it only has 1 bit of dynamic range - precisely 6dB of dynamic range.

Fortunately, however, since the dynamic range terminates in a white noise floor, the noise increases by a factor of 3dB per octave, so while the highest octave (750KHz to 1.4MHz) has a noisefloor of -9dB, the octave below that has a noisefloor of -12dB, etc. This means that the overall system, being a 1 bit system, only has 6dB of dynamic range, but by the time you get down to the audible range of DC to 20KHz you get a full 120dB of dynamic range.

of 44100 yields around 384 dB.

Which is complete malarky. Remember, it's a 1 bit system at 2.8224MHz. This means 6dB of dynamic range across the entire range of DC to 1.4112MHz, but the dynamic range improves as you get closer to DC because the noise floor has more energy at the high frequencies than the low frequencies.

Asuming we lose one bit on filtering it's 192 dB which is still bigger than life.

Which is fun to know, but not in any way related to what we're talking about.

Or show me the spec where it shows 120 dB.

Do a search on Google for "dsd 120dB" and it will pull up a littany of pages that refer to the 120dB dynamic range often attributed to DSD WITHIN THE AUDIBLE BAND. Nobody is saying it provides anywhere near 120dB of dynamic range across the entire range, or even up to 100KHz. Rather it is an approximate of the dynamic range in the audible band up to 20KHz, though I have yet to see a 1 bit converter actually accomplish this. Specs around 109dB are more common, though I'll run with the 120dB spec, recognizing that it's merely a limit to the accuracy and stability of the noiseshaping.

Try this page for info, and notice what they say, "residual noise at -120dB through the audible band.":

http://www.merging.com/2002/html/sacd.htm

There are many other pages you can view that will refer to this number of 120dB.

Because after a while the expected step is lower than the actual step taken. Grab a milimetric piece of paper and draw sine wave. Sometimes the wave won't go to the next step but stays on the same (it's called quantization). Then it has to play catch up.

So when the sine wave stays on the same step the "relative" value turns negative all of a sudden? Are you seeing the large, gaping holes in your "relative motion" theory yet? Have you found anything out there, anywhere that can substantiate this? Because you are still really just flat out incorrect about your notion on this one.

You're getting closer, though. But you'll realize when you finally figure it out that DSD is the same as 1 bit PCM that has been adequately noise-shaped.

Try this. Take a sine wave and add random noise to it of pretty high amplitude - maybe the noise you add is as high as the sine wave itself. Then convert it using 1 bit PCM. What does it look like?

Nika.

Joined: May 2003
Posts: 4,836
S
MP Hall of Fame Member
Offline
MP Hall of Fame Member
S
Joined: May 2003
Posts: 4,836
Nika

Is you book out yet. Does it cover SACD and all the different formats? Perhaps, you can post the table of contents. Thanks. \:\)

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
Johnny B,

The book does comprehensively cover how DSD is done. It is not out yet, however - expected Winter 2004.

Nika.

Joined: May 2003
Posts: 4,836
S
MP Hall of Fame Member
Offline
MP Hall of Fame Member
S
Joined: May 2003
Posts: 4,836
Thanks Nika.

I'll buy it because I'm truly tired of all the black magic. I hope it has some good solid and objective science in there.

Perhaps, just as important, I hope you have some example configs showing the folks some good setups.

And lastly, I also hope you have something like an "Executive Summary." Not everyone is a real techno-weeny, propellerhead as much as some us try. Some non-technical decisionmakers need something they can rely on when the make purchases.

What's the book's target price so far?

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 Johnny B:
Thanks Nika.

I'll buy it because I'm truly tired of all the black magic. I hope it has some good solid and objective science in there.


That's the goal.

What's the book's target price so far?

I haven't been in any discussions on that matter, yet, I'm afraid.

Nika.

Joined: May 2003
Posts: 4,836
S
MP Hall of Fame Member
Offline
MP Hall of Fame Member
S
Joined: May 2003
Posts: 4,836
Fair enough. Thanks.

Joined: Aug 2000
Posts: 1,242
Platinum Member
Offline
Platinum Member
Joined: Aug 2000
Posts: 1,242
Quote:
Originally posted by Nika:
So it's not relative as you were saying then, yes? Because if it were relative then whenever the waveform increased in amplitude the data would be a "1" and whenever it decreased in amplitude the data would be a "0." Clearly you are saying it is different than that if, while the amplitude increases, the sample values decrease.
that's why it is a Delta-Sigma and not a Delta. the sigma estimates where should be the next step. If the waveform doesn't go above it, then it just returns a zero instead of a one. Is that way too hard to understand? do I have to write a book?

Quote:
And how is the data then turned from whatever you think it is to PCM? What kind of filters are used and how are they implemented?
Jeeezzz... we are talking about DSD, which is NOT 1-bit PCM... but if you must know, a decimator takes chunks of 64 samples from the modulator and turn them into 24-bit words at 44100 words per second. Those 24-bit (or 8-byte if you will) words yield absolute values to DC with a limit where clips whenever the decimation goes beyond FFF.

Next...


"There's no right, there's no wrong. There's only popular opinion" Jeffrey Goines
Joined: Apr 2002
Posts: 242
D
Senior Member
Offline
Senior Member
D
Joined: Apr 2002
Posts: 242
Loco-

Funny, from the standpoint of a casual observer, it appears that Nika has answered most of your questions/challenges and backed his responses up as well...no response to those issues?

-Duardo

Joined: Aug 2000
Posts: 1,242
Platinum Member
Offline
Platinum Member
Joined: Aug 2000
Posts: 1,242
Quote:
Originally posted by Nika:
Hold on, there. It's not "64 bit oversampling." It's a single bit system at "64 times oversampling." This means that it only has 1 bit of dynamic range - precisely 6dB of dynamic range.
That is true for PCM systems, not for DSD systems... you made me write 848 ones and zeroes. I'm entitled to typos.

Quote:
Fortunately, however, since the dynamic range terminates in a white noise floor, the noise increases by a factor of 3dB per octave, so while the highest octave (750KHz to 1.4MHz) has a noisefloor of -9dB, the octave below that has a noisefloor of -12dB, etc. This means that the overall system, being a 1 bit system, only has 6dB of dynamic range, but by the time you get down to the audible range of DC to 20KHz you get a full 120dB of dynamic range.
That's 6 octaves from 20KHz to 1.4MHz. There's a gain of 18dB. You're missing like 100dB there, my man.

Quote:
Do a search on Google for "dsd 120dB" and it will pull up a littany of pages that refer to the 120dB dynamic range often attributed to DSD WITHIN THE AUDIBLE BAND. Nobody is saying it provides anywhere near 120dB of dynamic range across the entire range, or even up to 100KHz. Rather it is an approximate of the dynamic range in the audible band up to 20KHz, though I have yet to see a 1 bit converter actually accomplish this. Specs around 109dB are more common, though I'll run with the 120dB spec, recognizing that it's merely a limit to the accuracy and stability of the noiseshaping.

Try this page for info, and notice what they say, "residual noise at -120dB through the audible band.":

http://www.merging.com/2002/html/sacd.htm

It actually says "a dynamic range greater then 120dB" including typos.

And also:

"DSD can represent this with a frequency response from DC to 100 kHz. To cover the dynamic range of a good analog mixing console, the residual noise power was held at -120 dB through the audio band. This combination of frequency response and dynamic range is unmatched by any other recording system, digital or analog."

And that's bigger of what the analog devices can handle. Bigger than life.

Quote:
So when the sine wave stays on the same step the "relative" value turns negative all of a sudden? Are you seeing the large, gaping holes in your "relative motion" theory yet? Have you found anything out there, anywhere that can substantiate this? Because you are still really just flat out incorrect about your notion on this one.
See my previous post. Then you'll understand how that delta-sigma thing works for DSD. It's very simple. Forget about PCM. Just look at the problem from another point and you'll see how round the earth is.

Quote:
You're getting closer, though. But you'll realize when you finally figure it out that DSD is the same as 1 bit PCM that has been adequately noise-shaped.
OH...

MY...

GOD......

Put this in between your eyes:

DSD IS NOT 1-BIT PCM!!!!!!!

Quote:
Try this. Take a sine wave and add random noise to it of pretty high amplitude - maybe the noise you add is as high as the sine wave itself. Then convert it using 1 bit PCM. What does it look like?
Please post the JPEG. It will look as any 1-bit PCM file. Sqware waves. Regardless if it originally was sine, triangle, synth, whatever. gain, just + and -.

Most important is how it sounds. Through your filters (44.1, 96, 192, whatever). Like a highly distorted basic wave. Like what we have been hearing.

And WAY too different from DSD. Just put a SACD on your player. Better yet, use that Tascam DSD thing you seel there at Sweetwater.

And listen.


"There's no right, there's no wrong. There's only popular opinion" Jeffrey Goines
Joined: Mar 2002
Posts: 48
C
Senior Member
Offline
Senior Member
C
Joined: Mar 2002
Posts: 48
so, if i understand this correctly Loco, DSD is not what Nika says because....you say so? Good evidence.

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

Joined: Aug 2000
Posts: 1,242
Platinum Member
Offline
Platinum Member
Joined: Aug 2000
Posts: 1,242
Quote:
Originally posted by Nika:
I'm sorry for your misunderstanding on this one, but you seem very not prone to learning enough to discuss this issue with any credible sources or material. In the meantime I have tried to show you some sources to some information that would help you understand but you are only interested in picking through and finding information that you feel supports your untenable position. You are apparently not interested in reading for the sake of learning. You are interested in reading for the sake of furthering your misunderstanding on this issue. I don't know what to do to help, as you are most definitely not interested in any help on this.
Believe me. I've read. Researched. tested. Everything. That doesn't change the fact that the digital information you read from a DSD file gives you relative directions and the digital information you read from a PCM file gives you absolute directions. That is regardless of how it is encoded. Yes, we all know both systems share the same delta-sigma modulation front end but then PCM is achieved after another extra process.

Quote:
The term "modulator" comes from the notion that the 1's and 0's are being modulated by the input signal, and it happens to be the same as when "pulse code modulation" modulates the 1's and 0's in a code.
But then you're missing how you transfer from 1-bit relative data to 24-bit absolute values.

Quote:
Now if you continue to reject this notion as what is really happening then I really can't help. And I may never be able to convince you that the world is round either. And at a certain point I lose interest in doing so.
I know exactly what is happening. However, you still won't understand how different you read the data from DSD and PCM files. Even on a 1-bit PCM file which is still, absolute information. As absolute it is, it will sound different to DSD. And it does.


"There's no right, there's no wrong. There's only popular opinion" Jeffrey Goines
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 Loco:

Quote:
The term "modulator" comes from the notion that the 1's and 0's are being modulated by the input signal, and it happens to be the same as when "pulse code modulation" modulates the 1's and 0's in a code.
But then you're missing how you transfer from 1-bit relative data to 24-bit absolute values.
Loco,

What I have said is that the signal that comes off of a delta sigma modulator is 1 bit PCM code with noiseshaping in it.

What you are saying is that the signal that comes off of a delta sigma modulator is a different type of code that gives "relative values" as opposed to "absolute values."

I am curious about the question you pose above. How does a PCM converter take the "relative" values in the 1 bit "relative" code and turn it into the "absolute" values you are talking about? Please be EXTREMELY specific about this process and explain how it happens. Give us coefficient values and specific mathematical processes that can do this, please.

Because what I have said is that since the 1 bit signal IS PCM code it takes no process to do that. All you need to do is filter out the HF content and the sheer use of math on the signal will inherently create more quantization steps.

Try this very simple low pass filter, for example. Take a 1 bit data stream and take every sample and add to it the value of the sample before it. You will now no longer have just a 1 bit PCM data stream but will have a 2 bit data stream because your values are now 0, 1, and 2.

Or for an even steeper low pass filter try taking every sample in your 1 bit data stream and add it to the value of both of the samples immediately before it. Now you have 0, 1, 2, and 3 as your possible values.

The original data was 1 bit PCM data and each low pass filter you add inherently creates more quantization steps because any mathematical formula will create more data. Use the following formula and you'll get a very steep LP filter with many quantization steps possible, yielding 24 bit data, pending the coefficients used:

Y(n) = AX(n) + BX(n-1) + CX(n-2) + DX(n-3)... ?X(n-255)

Wherein Y is the output sample, X is the input sample, A...? are coefficients, and n is the current sample.

But you don't think that this is what is happening. You think that there has to be a system to first change the "relative" values to "absolute" values while you turn it into 24 bit data. Please, show us the specific process and explain the specific process used to do that.

BTW, the process I've described above is called "decimation" and is often done with a couple of stages of canonical integer coefficient (CIC) filters before a large, linear phone, many tap FIR filter is put in place for the final, more particular stages.

I am very curious about the process that yields 24 bit data based on the system you are describing. Again, please give us the coefficients and explain the math comprehensively, because it seems to conflict with "Theory and Application of Digital Signal Processing" by Rabiner and Gold. I recognize that my version of this book is from the 1980's, and much may have transpired in mathematical analysis of digital signal processing since then, but after your explanation I may need to ask Rabiner and Gold to rewrite their 1000 page textbook in accomodation of your new developments.

Nika.

Joined: Aug 2000
Posts: 1,242
Platinum Member
Offline
Platinum Member
Joined: Aug 2000
Posts: 1,242
Quote:
Originally posted by Nika:
Because what I have said is that since the 1 bit signal IS PCM code it takes no process to do that. All you need to do is filter out the HF content and the sheer use of math on the signal will inherently create more quantization steps.
Magic. How do you get a clean 100dB deep signal out of a noisy 6dB shallow signal? Try converting an 8-bit 96K signal into a 16-bit 48K signal and tell me if the original noise dissapeared. Now try it with a 4-bit 192K to 16-bit 48K and tell me that now you can hear the violinist breathe after it. No way Jose.

Quote:

Try this very simple low pass filter, for example. Take a 1 bit data stream and take every sample and add to it the value of the sample before it. You will now no longer have just a 1 bit PCM data stream but will have a 2 bit data stream because your values are now 0, 1, and 2.
Wrong. 0,1,2,3.

Quote:

Or for an even steeper low pass filter try taking every sample in your 1 bit data stream and add it to the value of both of the samples immediately before it. Now you have 0, 1, 2, and 3 as your possible values.
Wrong again. 0,1,2,3,4,5,6,7.

Quote:

The original data was 1 bit PCM data and each low pass filter you add inherently creates more quantization steps because any mathematical formula will create more data. Use the following formula and you'll get a very steep LP filter with many quantization steps possible, yielding 24 bit data, pending the coefficients used:

Y(n) = AX(n) + BX(n-1) + CX(n-2) + DX(n-3)... ?X(n-255)

Wherein Y is the output sample, X is the input sample, A...? are coefficients, and n is the current sample.
The coefficients. Where do you get those coefficients from?

Moresome. When you add values you're just totalizing (is that a word?) relative positions into absolute positions. You go from DSD to PCM. you are resolving the integral of a function. Two different beasts. Basic calculus.

However, if you go from PCM to PCM what you need is averaging and/or interpolation. Different formula.

Analogy to your model. You are walking down Greenwich meridian. Your GPS tells you that your friend is to your right (west). Maybe a couple fo feet away, maybe 2 miles away. Who knows? Next time says right again. Next one left (yes, your friend has a very fast UFO). Then you say you are taking his position (right, left) en every step you take so you can do some adding of those positions (east and west) and filtering so you can tell every mile you advance exactly how many feet away he is from you.

If you can do that, then you are gonna be so millonaire selling math-based GPS systems.

Quote:

BTW, the process I've described above is called "decimation" and is often done with a couple of stages of canonical integer coefficient (CIC) filters before a large, linear phone, many tap FIR filter is put in place for the final, more particular stages.
Have you actually read any of may previous posts? How many times I have mentioned the decimator?

Quote:

I am very curious about the process that yields 24 bit data based on the system you are describing. Again, please give us the coefficients and explain the math comprehensively
Well, you are the one who mentioned the coefficients. You start being comprehensive and then fall short. If you start with a formula, explain it thoroughly.

Also, the best way to quickly understand math models is by real life analogies. Give it a try and you'll see the big difference between both systems.


"There's no right, there's no wrong. There's only popular opinion" Jeffrey Goines
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 Loco:

[qb]
Quote:

Try this very simple low pass filter, for example. Take a 1 bit data stream and take every sample and add to it the value of the sample before it. You will now no longer have just a 1 bit PCM data stream but will have a 2 bit data stream because your values are now 0, 1, and 2.
Wrong. 0,1,2,3.
No, I said FIR, not IIR filters. Y(n) = X(n) + X(n-1) Not Y(n) = X(n) + Y(n-1)

Quote:
Quote:

Or for an even steeper low pass filter try taking every sample in your 1 bit data stream and add it to the value of both of the samples immediately before it. Now you have 0, 1, 2, and 3 as your possible values.
Wrong again. 0,1,2,3,4,5,6,7.
Here again. Y(n) = X(n) + X(n-1) + X(n-2) Not Y(n) = X(n) + Y(n-1) + Y(n-2)

Quote:
The coefficients. Where do you get those coefficients from?
http://www.cadenzarecording.com/papers

Quote:
Moresome. When you add values you're just totalizing (is that a word?) relative positions into absolute positions. You go from DSD to PCM. you are resolving the integral of a function. Two different beasts. Basic calculus.

However, if you go from PCM to PCM what you need is averaging and/or interpolation. Different formula.
Show me the formulas then.

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
dup

Joined: Aug 2000
Posts: 1,242
Platinum Member
Offline
Platinum Member
Joined: Aug 2000
Posts: 1,242
Quote:
Originally posted by Nika:
Show me the formulas then.
First prove your theory of PCM=DSD. If it's true I'll go and patent the GPS idea and then I'll show you the formulas. Show me a paper (not written by you) where it states that PCM=DSD and au contraire of all the documentation available, DSD is not relative information but absolute just like PCM.

I'll give a read to your papers on the link above. It's long, I'm moving this week, but I'll read it. Meanwhile, prove your PCM=DSD theory with math, papers and, most important, listening tests.


"There's no right, there's no wrong. There's only popular opinion" Jeffrey Goines
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
Loco,

I sent you something that might help. Let me know if you didn't get it.

Nika.

Joined: Aug 2000
Posts: 1,242
Platinum Member
Offline
Platinum Member
Joined: Aug 2000
Posts: 1,242
Quote:
Originally posted by Nika:
Loco,

I sent you something that might help. Let me know if you didn't get it.

Nika.
I was moving my apartment for the last week. No internet. I'll post back as soon as I read it all.


"There's no right, there's no wrong. There's only popular opinion" Jeffrey Goines
Page 11 of 11 1 2 9 10 11

Moderated by  gm 

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