E1DA Cosmos ADC

[IVX]

C9 and C10 aren't connected(footprints shorted by 0ohm 0402), and ADC will not be happy to get DC higher than +/-.5V. Even 3VDC (9039S) is safe but affects the distortion level. I recommend adding the Cosmos Scaler to the chain to cut DC bias. I moved C7 and C8 to the ES9822 input directly to keep the op amp unbiased for maximal symmetry = minimal H2. Also, the serial resistor between op amp and ADC chip is eliminated to achieve a cleaner noise floor. For example, 1k input at -.5dfs gives all harmonics under -150db or around.

 

[morillon]

C9 and C10 aren't connected(footprints shorted by 0ohm 0402), and ADC will not be happy to get DC higher than +/-.5V. Even 3VDC (9039S) is safe but affects the distortion level. I recommend adding the Cosmos Scaler to the chain to cut DC bias.

And if we were to consider installing these capacitors at the input..., what values would you recommend? How much space is available? Are there any measurements available to assess the impact? Thank you

 

[IVX]

And if we were to consider installing these capacitors at the input..., what values would you recommend? How much space is available? Are there any measurements available to assess the impact? Thank you

I have suitable Nichicon (kzh?) 330-470u/10V 6*11mm. To keep H2 as low as possible, need to select a pair of such caps with close capacitance.

 

[morillon]

Thanks
Some sweeps THD, THD-N vs frequency with and without caps?
(has "standard" line levels like 4.4 VRMs etc.)
;-)

 

[capslock]

no composite amp in the first stage?

 

[IVX]

Thanks
Some sweeps THD, THD-N vs frequency with and without caps?
(has "standard" line levels like 4.4 VRMs etc.)
;-)

Input caps increase the THD at low frequencies and with an unbalanced source(as found by ADC users on the diyaudio.com forum). This is the second reason why I have moved the caps to the ADC chip input, which always operates in differential mode. So, do you wanna see the THD difference with an unbalanced source with/without the caps, right? Or do you need to see the difference in H2 with/without the caps from a balanced source?

 

[IVX]

no composite amp in the first stage?

No composite.

 

[morillon]

Input caps increase the THD at low frequencies and with an unbalanced source(as found by ADC users on the diyaudio.com forum). This is the second reason why I have moved the caps to the ADC chip input, which always operates in differential mode. So, do you wanna see the THD difference with an unbalanced source with/without the caps, right? Or do you need to see the difference in H2 with/without the caps from a balanced source?

Unbalanced..
"2,2vrms"?
;-)

 

[IVX]

Funny, but I found no differece with/withot the caps. Maybe I didn't reproduce the same conditions as on diyaudio.com.

 

[morillon]

But how many harmonics are measured in a THD?
And clearly, we shouldn't present, limit , ourselves to h2...
A measurement at 192kHz/24kHz, BD 96kHz... down to at least h5/6?
;-)

 

[danadam]

I prepared the print under the sensitivity switch(top 4 are same as ADCiso, the second group of 4 sensitivities are low-noise). IMO, it is clear, but if you have some idea about how to make it more intuitive, welcome.

"1/7.8.1 K" seems like a typo (but maybe I don't understand something)

 

[IVX]

"1/7.8.1 K" seems like a typo (but maybe I don't understand something)

right, my fall..

 

[IVX]

But how many harmonics are measured in a THD?
And clearly, we shouldn't present, limit , ourselves to h2...
A measurement at 192kHz/24kHz, BD 96kHz... down to at least h5/6?
;-)

Looks like 50 harmonics(rather 48 beacuse of Fs=96k) at 1000Hz.. The BW was 20-50000Hz. Does Stepped sweep have a special setting for the harmonics?

 

[morillon]

Excuse my language barrier...
I just wanted to be sure that the differences were so small, It wasn't due to a very low number of harmonics being considered in the THD measurement (I believe it's generally considered up to h9).

(But also, only displaying h2 isn't very useful... you need at least up to h5 or h7... even and odd harmonics.)
;-)
(Maybe observing the chemical capacitors bypassed by an small MKP or better yet a styroflex or Teflon would be funny.)

 

[David_M]

Can you please help explain this linearity curve from Wolf? Is it inline with the ESS adc specs?

 

[IVX]

Can you please help explain this linearity curve from Wolf? Is it inline with the ESS adc specs?



View attachment 538762

I did explain that 5 years ago. The max X-axis is about 15dbu, at -100dbu i.e. at -115dbfs ADC's noise starts to dominate, and we can't see the linearity. Cosmos ADC grade A has -128..-128.9db(A) or 125.5..126.4db unweighted, exactly at the level that we may expect to affect the linearity reading, with error reached +.5db. I performed the linearity test with no noise involved(BPF filtered) and got a lot better result.

 

[megapro17]

When it will be available for purchase? Any pre-order?

 

[IVX]

The soldering service queue is 2 weeks.

 

[David_M]

I did explain that 5 years ago. The max X-axis is about 15dbu, at -100dbu i.e. at -115dbfs ADC's noise starts to dominate, and we can't see the linearity. Cosmos ADC grade A has -128..-128.9db(A) or 125.5..126.4db unweighted, exactly at the level that we may expect to affect the linearity reading, with error reached +.5db. I performed the linearity test with no noise involved(BPF filtered) and got a lot better result.View attachment 538781

Thanks for the answer... so if Wolf BP filtered the ADC output signal, he would have gotten the above result, correct?

 

[KSTR]

Thanks for the answer... so if Wolf BP filtered the ADC output signal, he would have gotten the above result, correct?

The lower you go in level, the more the result is affected by any noise/hum or whatever is present in the signal. Thus, the more narrowband a filter you use and the more the
test frequency is away from any mains or other spurious content as seen in a full-resolution FFT, the better.

But, linearity measurement of delta-sigma devices, DAC or ADC, is basically useless because D/S is linear by design, it cannot be nonlinear in the way other conversion principles can be, unless there is design error on the digital input side. See my thread https://www.audiosciencereview.com/...ve-sample-values-are-always-offset-by-1.6865/ for an actual case.

And by using dithering it can be linear well below the lowest bit level.