Review and Measurements of Audio-gd NFB28.28 DAC and Headphone Amp

[Bluespower]

DS DACs function in the megahertz range which is high enough to perform this operation correctly with any signal within the audio band.

However, why don’t you propose a specific test and see if someone here with adequate instrumentation is willing to perform it?

Perhaps something like the SMPTE test but with the higher frequency tone at 1/128th power (-42dB)?

Yes exactly

 

[Bluespower]

@andreasmaaan @amirm this leads to the conclusion that old 16 bit r2r dac has 16 bit resolution whatever frequency whereas dsd has 7bit (130 steps) un'dithered resolution at 20khz for 2,8 mhz and 8 bit resolution (260 steps) at 20 khz for 5,6 mhz. So what is the resolution of new akm ess dacs at 20khz? Maybe 7 bit +5 bit from the multibit? No test show that dsd loose resolution whith high frequencies but the theory proove it. It means that some tests are missing. That is why i'would'like to think'on test with muti tone with high dynamic differencies.
At 5khz that leads to 9bit resolution for dsd @2,8mhz and 10 bit resolution for 5,6mhz.
This'illustrate the advantge of r2r over delta sigma dac. R2r has more precision on bits and dynamic while ds is better in thd and frequency.

 

[RayDunzl]

What exactly is shown on these graphs? Speakers response driven with 2 different DACs, or ..?

For the referenced post, the excitation was the Measurement Sweep Tone (log sweep) in REW, sent through two DACs, a Benchmark DAC2 and a Pro-ject DAC Box Fl S non-oversampling filterless jobbie, intp the air via Martinlogan reQuest speakers (with DRC applied), and sensed with a UMIK-1 at the listening position, with the various traces displayed using the REW Overlay window.

 

[RayDunzl]

Oh, I know he used a mic, probably a good calibrated one. I just want to check with him if he made multiple measurements and averaged them or some other method to ensure accuracy.

In the case of the sweep tones, no averaging. Nonetheless, the traces of the two DACs overlaid nearly perfectly.

 

[RayDunzl]

All i want is a proof test that show that i m wrong.

Since you are not accepting as "proof" any explanations that others regard as "proof", and are apparently not willing or able to perform and post the results of experiments to your own specification, well, ???

@andreasmaaan with modern dac chips you can still have let's say 16 bit resolution at 20khz? Cause with dsd wich is dure delta sigma you loose resolution as the frequency grows.

You've stopped making sense. To me.

Have fun arguing/rejecting/inventing.

 

[RayDunzl]

I keep seeing the idea that a DAC is "stressed" somehow producing different selections of samples to recreate in the analog domain.

A sine is not "stressful" but some other combination of sample values is "stressful".

Is there any basis for this conjecture?

@amirm

 

[dc655321]

@RayDunzl there is certainly a lack of comprehension here, but it is not on your part...

 

[Krunok]

In the case of the sweep tones, no averaging. Nonetheless, the traces of the two DACs overlaid nearly perfectly.

Of course they did, but the question is if this small difference is variation between measurements? Can you make few measurements with the same DAC so we can check that?

 

[andreasmaaan]

For 2,8 mhz dsd to render a 20khz tone you'will have only 140 steps (2800/20)that means you have only 7 bit resolution at 20khz. Of course dither can improve but artificially. So maybe on modern dac chips the true resolution for high frequency is less than 16 bit.

Not sure I follow. Do you have a source that explains your argument here in more detail?

 

[Krunok]

*distortion measurement is dominated by ambient room-noise at the tested SPL

After 300 Hz (or so) that is probably true, but up to that I believe speakers make more distortion than noise floor.

I got very similar distortion graph with my speakers but only after 300Hz it coincides with the measured noise floor, which is logical to me as there's no much in-room noise so low.

 

[Bluespower]

Since you are not accepting as "proof" any explanations that others regard as "proof", and are apparently not willing or able to perform and post the results of experiments to your own specification, well, ???




You've stopped making sense. To me.

Have fun arguing/rejecting/inventing.

Well your tests are proof i was speaking for test of audioscience review for new dacs. I mean there should be a test like your test. I ve nothing to test i just wonder why there are never test with complex tones that contains high and low'level harmonics. It's a proof that delta sigma dacs loose precision on high frequency so why not proving that the new hybrid multibit delta sigma dac have solved this'problem ?

 

[Bluespower]

Not sure I follow. Do you have a source that explains your argument here in more detail?

Dsd is 1 step possible at 2.8 mhz that means if you calculate that you ve got only 140 steps possible at 20khz.(2800/20) It,s mathematic. It's well know that with ds dac you'loose precision on'level comparing to the r2r multibit. I read somewhere that ess has make some tricks to improve precision iver classic ds by having very fast frequency and a pistol stuff. But pure dsd has less than 16 bit resolution that's real. Pcm 24 bit 96khz has much more information.
I found this link that explain dsd problem on precision.
https://medium.com/@jolon/thoughts-on-dsd-3fc9aaff8548
Read the part The problem with DSD

 

[andreasmaaan]

Dsd is 1 step possible at 2.8 mhz that means if you calculate that you ve got only 140 steps possible at 20khz.(2800/20) It,s mathematic. It's well know that with ds dac you'loose precision on'level comparing to the r2r multibit. I read somewhere that ess has make some tricks to improve precision iver classic ds by having very fast frequency and a pistol stuff. But pure dsd has less than 16 bit resolution that's real. Pcm 24 bit 96khz has much more information.
I found this link that explain dsd problem on precision.
https://medium.com/@jolon/thoughts-on-dsd-3fc9aaff8548
Read the part The problem with DSD

I read that section of the article and although I don’t agree with a lot of it (another story), there’s nothing about steps and how these relate to bit depth.

Do you have a source for that, pls?

Or could you explain how you got from 140 “steps” to 7 bits?

 

[Krunok]

I read that section of the article and although I don’t agree with a lot of it (another story), there’s nothing about steps and how these relate to bit depth.

Do you have a source for that, pls?

Or could you explain how you got from 140 “steps” to 7 bits?

2 to the 7th power is 128, so let's round it to 140?

 

[Bluespower]

I read that section of the article and although I don’t agree with a lot of it (another story), there’s nothing about steps and how these relate to bit depth.

Do you have a source for that, pls?

Or could you explain how you got from 140 “steps” to 7 bits?

7 bit means 2^7 values possibles.
2*2*2*2*2*2*2 = 128 that 7means bits can make 127 different levels. 8 bits can make 256 values so 140 levels possible is equivalent between 7 bit and 8bit.

 

[Krunok]

7 bit means 2^7 values possibles.
2*2*2*2*2*2*2 = 128 that 7means bits can make 127 different levels. 8 bits can make 256 values so 140 levels possible is equivalent between 7 bit and 8bit.

Because 256-127=140?

 

[Bluespower]

Because 256-127=140?

You re funny 16 bit resolutions means 65536 values possible. With one bit dac to render good a 20khz tone with 65536 levels possible you'would need a frequency of 20000 * 65536=1310 mhz. I know there are some tricks like dither that can compensate. Maybe i'm wrong but i don't see failure in my math. @andreasmaaan @RayDunzl

 

[Krunok]

You re funny 16 bit resolutions means 65536 values possible. With one bit dac to render good a 20khz tone with 65536 levels possible you'would need a frequency of 20000 * 65536=1310 mhz. I know there are some tricks like dither that can compensate. Maybe i'm wrong but i don't see failure in my math. @andreasmaaan @RayDunzl

Oh, I actually think there are lot of failures in your math, but let's go one step at a time. Can you please elaborate how exactly did you get 140 steps?

After that we can discuss your new 1310 MHz DSD standard. I guess we shoud call it DSD8192?

 

[amirm]

I keep seeing the idea that a DAC is "stressed" somehow producing different selections of samples to recreate in the analog domain.

A sine is not "stressful" but some other combination of sample values is "stressful".

Is there any basis for this conjecture?

@amirm

My DACs are so stressed I give them one day of a week holiday. They are asking for more but that is all I can do.....

 

[Bluespower]

Oh, I actually think there are lot of failures in your math, but let's go one step at a time. Can you please elaborate how exactly did you get 140 steps?

After that we can discuss your new 1310 MHz DSD standard. I guess we shoud call it DSD8192?

Well one step at a frequency of 2.8 mhz that means one step takes 1/2,8mhz s to come (t=1/f).
So one step take 0,35us to come.
So the 140 steps wilo take 50us to grow up'the level to the max. Because dsd is one bit after the other to grow the level of the signal.
50us means a frequency of 20khz. F=1/t
If you make more steps the time to reach the 141th step wil be more than 50us so it s impossible.
If you make a low level it Ill be rounded to one of the 140 steps.
For exemple for a low level the 50th step will take 17us to'come that means it s less than 50us so it s ok for 20khz.


In dsd for 20khz you have only 140 steps.
In pcm 16 bit you have 65536 steps.

If just calculate 2,8Mhz/20000khz it gives you directly the number 140.
Is it logic?

It seems very strange i agree but that is'how'i'understand dsd.

Dsd one bit at 2.8 mhz is One bit 1 value (on or off) after the other every 0,35 us.

Pcm 16 bit at 44'1 khz is 16 bit 65536 values every 22 us.