I am afraid you are fixated on one or two figures in that paper and you're missing the forest for the trees.
This NRZ (zero-order hold) is not a problem in reality. There is no such thing as the DAC you talk about because it doesn't exist. One, there isn't a linearity problem. Second, go find me a real DAC (not ICs misused in a NON-OS converter) that shows any attenuation from the ZOH. I will be waiting.
The figures right below the ones you are stuck on explain how oversampling makes this a non-issue. Guess what every commercial converter does? It helps to know just a little bit about the current state of the art before you try to reinvent the wheel based on a partial understanding of one paper.
Last edited:
Which audibility claim did I make that you seek a supporting evidence of?
When did you remove me from your ignore list again? Or was this image photoshopped?
Very good! You are leading the chart this year. 😀
You just took the lead of the year from Hans. Congrats!
Bill,
You essentially announce when the cookie is there. So if they randomly check it won't be there.
You essentially announce when the cookie is there. So if they randomly check it won't be there.
Is that like Richards called Dick?
Speaking of JJ he agrees on the timing stuff I was talking about saying essentially the same thing here, in fact he was moved to call the claim that 16/44.1 cant preserve sub-usec timing a crock of s***. If you don't listen to anything I say maybe this will help.
A warning. There is a new web article going aroud that is a crock of schmidt. | Page 4 | Audio Science Review (ASR) Forum
Last edited:
I'm home now. What I see is 2 sides arguing about sample rate. I have no dog in this race yet hear a difference on upsampled 16/44k to 24/96k, but as Matt put it, "How do I know it is better?".
When I mentioned measurement like a null test, the other side said "Why don't you do it?". Clear sign of disinterest for me to gain proper understanding nor settle the matter.
Your side supported Lavry's premise on sampling, provides theoretical plots of ideal DAC behavior yet show no measured performance on available devices on the market. Lavry also stated "Data found in compact disk recording and similar formats is often oversampled prior to digital to analog conversion, to simplify anti imaging filtering." But when I mentioned preference on upsampled conversion you chose condescending tone to reply. Seems clear enough that you have little or no interest for me to gain better understanding and see the matter as resolved either.
I have no choice but agree with Matt.
Last edited:
No,
It is like calling Peter, Dick.
I haven't given you grief on the time resolution.
As usual, we don't communicate on both issues.
BTY he used to go by "Sparky" until his first job where there were other "Sparkys" and "Jim's."
I am fixated on nothing, so you have nothing to be afraid of. I certainly am not missing the forest for the trees.
NRZ, until I see proof as opposed to nothing but assumptions, I remain questioning. If you choose to accept a premise without proof, that is fine. It may be fully linear, It may not. As I said, it does not matter to me for casual listening.
Attenuation from the zoh/N4z? That is not my statement, it is Lavry's. I believe 4.1dB was the number he cited.
As to over sampling, I already stated that Lavry showed 2x with images at 88Khz, requiring only a 4 pole, and stated that the 1/2 NRZ produced the exact same spectra.
You said "guess what every commercial converter does"? Instead of spending time putting a sentence together that is only designed to badger me, why not actually state what "every commercial converter" does.
Trust me, if you put it in print, I will understand it.
But the best part? If I don't understand it because you are so much more advanced than me, I will ask you questions. I have no problem with that.
So please, get off your high horse, I am not the enemy.
Talk and act civil please..
Jn
Sorry that was for Mr. M and jn.
I thought Peter's were Pete or Pecker or something like that. James M. Williams (RIP) was Jim to all his friends.
Scott,
I would like you to elaborate here.
As I have never intentionally tried to build a strawman, I would like to know why you would say that given my history.
Also, I keep re-reading my post, and cannot figure out how it could even be perceived as such.
(I assume you are being honest, so I would like to know what in the post you perceived as "strawmanish" (I assume that is a noun??).
If your intent was dishonest, nevermind the question.
Thanks, john
If you were just having fun, feel free to publicly humiliate me as I would certainly deserve it for misreading you..
Last edited:
Ah, my name in vain again.. (Well, moniker anyway)
So JJ said that.. Ok, did he provide any factual evidence of it we mortals can peruse?
Don't give me that sinx/x ideal equation stuff. Did he show by actual test on real hardware, that NRZ flipped into a brickwall maintained all temporal accuracy and preserved superposition?
Again, I said actual test, actual hardware?
If so, please provide a link, I would love to learn.
As I stated previously, these questions are not my pedestal, not my hobby horse, not my windmill.
They are questions that arose from my reading of Lavry...nothing more, nothing less..
Edit. Checked the link..great, more simu, looks exactly like my excel data. And a statement that it comes outta the filter fine,,,I have a pic somewhere..
Sigh.. Actual measurements, what happened to those? Is the world nothing but simulations and the assumption that they are always right?
I never forget GIGO...or Murphy. Or actual proof.
Jn
Last edited:
I suspect Jim Williams wasn't worried about being the new guy and having a confusing name. I don't recall ever calling him James.
Did you understand why calling a Peter, Dick was not a nice thing to do?
Well, my friends call me John..
Ok, if I had friends, they would call me John.
That is, if anybody actually called me...
Hello...hello... Is anybody out there???
Jn
Last edited:
Could you please point out where my reply was not civil? I certainly didn't tell you to get off your high horse. If I were trying to be impolite, you would know it. Maybe there is a tone of frustration because we are rehashing solved problems, as Hans and syn08 already mentioned.
The reason I said I thought you were missing the forest for the trees is because there is no need to be concerned about the NRZ (ZOH) in practice. What Lavry said is true and it's well known, but the problem has been designed out of audio DACs more than 20 years ago, if not 30. You don't need to get stuck in the minutiae wondering if adding the sinc function is linear or not (why would it not be?). All you have to do is look at the measurements of any decent DAC on the market. They are all ruler flat and don't show the in-band attenuation. Why? Because, excepting some pathological designs, they all use an oversampling digital filter before conversion. The vast majority use a factor of 8x with 44.1 kHz input material, so the data is fed to the converter or sigma-delta modulator at 352.8 kHz. That pushes everything very far out, to where it's inconsequential. It may be even less of an issue for sigma delta DACs where the output runs at an even higher multiple of that rate. Even if you didn't oversample, you can just implement the inverse response in the digital filter. Some parts do implement this.
The AD1955 is an almost 18 year old DAC chip, but is representative and has a good datasheet. You can find frequency response graphs starting on page 7.
https://www.analog.com/media/en/technical-documentation/data-sheets/AD1955.pdf
Equalizing Techniques Flatten DAC Frequency Response - Maxim
First of all, you did not tell me to get off my high horse, I told you.
And in rereading your post, I was completely out of line, my apologies.
By "adding the sinc" function, you mean the DC extension of the sample by hold, NRZ?
Don't forget, the Lavry paper was presented as the paper to end all myths, but raised more questions than answered..hence my questions.
Can I assume 8x oversampling is waveform prediction, a least squares data fit?
Cheers, jn
Sims what sims? Current CPU's can run all this processing native real time if desired, don't need no stinkin' DSP. What about the hardware would make the basic digital computations fail?
I'm confused with your distinction between CPU processing and DSP. Aren't they the same?
The link you provide showed the exact same sampling differences I did in excel. No new info there. Then he said "filter gets rid of all the differences". Um, that is not proof.
As Chris pointed out, hardware at 8x really kills the filter concern, unlike the Lavry paper which was supposed to finish off the myths but didn't.
Since new hardware really wipes out any NRZ concerns, why was the Lavry paper even cited?? I guess we'll never know.
Jn
Last edited:
No, it's ok, I could stand to be less snarky. 🙂
Yes, I meant the NRZ.
In terms of the 8x oversampling, typically it's done by zero stuffing followed by a brickwall LPF.
You were not snarky, period. My apologies.
By zero stuffing, do you mean NRZ with 1/8th width hold? If so, that was exactly my question regarding 1/3rd or 1/4 hold to get the images out there. It was a question begging an answer.
Actually, I was hoping for a wide window polynomial fit, or least squares, that would have been far more sexy than thin pulses..😉
At 8x, I would almost think an rc roll off would work...
Jn
Usually a polyphase IIR filter, can be made very close to a linear phase FIR filter.
Yeah, I suppose it is equivalent to that, it's just not how I ever conceptualized it. I see now what you were getting at.
- Status
- Not open for further replies.