E1DA Cosmos ADC

[restorer-john]

Yes, 9039Q2M has separated R and L THD compensation, but you know what, all these coefficients are nearly zero.. In my mind, ESS has reached something revolutionary with that IC! A huge progress.

Unbelievable. When does Cosmos ADC get it?

 

[IVX]

?? On the screenshot with nearly -150db of THD is the output of 9039S, DAC+HPA dongle.

 

[pma]

We all know that E1DA Cosmos ADC is a great converter with superb parameters, especially when considering the price. We also know that there are some caveats, like very low input impedance, that may limit its use. I am now describing another issue. Please forgive me if it was already discussed.

Cosmos ADC is very good for measurements of power amplifiers, where its low input impedance does not matter. Most of power amplifiers have single ended output, HOT output terminal yields a signal against common signal ground. Thus, the input of Cosmos ADC is driven from single-ended and not from the balanced signal source. And there comes quite a big issue with low frequency distortion of input coupling electrolytic capacitors in the Cosmos ADC.

When measuring low noise amplifier like Purifi, I need to measure per-partes (with dividers) and also to utilize Cosmos without the external divider. Below is the explanation, measurement of THD vs. frequency at 10W/4ohm, which equals to 6.325V input signal. 10V setting of Cosmos is used and I repeat that Cosmos is driven from the single-ended source. The result is below as a yellow trace.

One can see enormous rise of distortion below 70Hz. This is not an amplifier distortion, this is a Cosmos ADC input capacitor distortion. To give a proof, there is a second trace, the green one. Same amplifier, same signal, same level, but there are 7.5k resistors in series with Cosmos ADC input. 7.5k makes with 1.74k Cosmos input impedance the voltage divider and the resulting input range is now 53.1V (instead 10V). The low frequency distortion has disappeared, because the input capacitor of Cosmos ADC now sees 7.5k + 1.74k load resistor so the LF distortion corner is shifted to about 5x lower frequency. But, due to lower input sensitivity, the inherent noise now limits the THD measurement threshold. So, due to capacitor distortion we are unable to make a reliable measurement of low noise, low distortion amplifiers with single ended output.

The situation is better when we measure amplifiers with balanced output. The capacitor even harmonics distortion, which is mostly H2, is then almost eliminated. I have enough measurements to make my statements.

 

[SIY]

I have enough measurements to make my statements.

How do you know it's capacitor distortion specifically as opposed to some other input issue? Just the corner frequency shift? That's hard to see in these graphs.

 

[pma]

How do you know it's capacitor distortion specifically as opposed to some other input issue? Just the corner frequency shift? That's hard to see in these graphs.

Same preamp as a source, used as single-ended drive and balanced drive of Cosmos ADC. With balanced drive, H2 drops of about 12dB.

 

[SIY]

Same preamp as a source, used as single-ended drive and balanced drive of Cosmos ADC. With balanced drive, H2 drops of about 12dB.

View attachment 385469

Well, certainly the distortion source is symmetrical. But how did you isolate it to the capacitor?

 

[KSTR]

I will check the issue next time I open my Cosmos (to install an SPDIF output), where I'm also going to replace the caps with oversized Nichicon Muse bipolar ones (330uF/50V) plus a shorting option (to be able to measure DC).
We do already know the caps are weak point of the product, if only in terms of withstanding DC voltage...

 

[SIY]

if only in terms of withstanding DC voltage.

What is their rating? And dielectric?

 

[pma]

Well, certainly the distortion source is symmetrical. But how did you isolate it to the capacitor?

The LF even order H2 distortion disappears when you bypass the input capacitor and use external one. Contrary to popular belief, electrolytic capacitors have LF distortion that starts well above the high pass corner. I have shown some measurements in the past.

 

[SIY]

The LF even order H2 distortion disappears when you bypass the input capacitor and use external one.

OK, that's what I was looking for, thanks! Are these bipolar or single ended and biased?

 

[dominikz]

We all know that E1DA Cosmos ADC is a great converter with superb parameters, especially when considering the price. We also know that there are some caveats, like very low input impedance, that may limit its use. I am now describing another issue. Please forgive me if it was already discussed.

Cosmos ADC is very good for measurements of power amplifiers, where its low input impedance does not matter. Most of power amplifiers have single ended output, HOT output terminal yields a signal against common signal ground. Thus, the input of Cosmos ADC is driven from single-ended and not from the balanced signal source. And there comes quite a big issue with low frequency distortion of input coupling electrolytic capacitors in the Cosmos ADC.

When measuring low noise amplifier like Purifi, I need to measure per-partes (with dividers) and also to utilize Cosmos without the external divider. Below is the explanation, measurement of THD vs. frequency at 10W/4ohm, which equals to 6.325V input signal. 10V setting of Cosmos is used and I repeat that Cosmos is driven from the single-ended source. The result is below as a yellow trace.

View attachment 385451

One can see enormous rise of distortion below 70Hz. This is not an amplifier distortion, this is a Cosmos ADC input capacitor distortion. To give a proof, there is a second trace, the green one. Same amplifier, same signal, same level, but there are 7.5k resistors in series with Cosmos ADC input. 7.5k makes with 1.74k Cosmos input impedance the voltage divider and the resulting input range is now 53.1V (instead 10V). The low frequency distortion has disappeared, because the input capacitor of Cosmos ADC now sees 7.5k + 1.74k load resistor so the LF distortion corner is shifted to about 5x lower frequency. But, due to lower input sensitivity, the inherent noise now limits the THD measurement threshold. So, due to capacitor distortion we are unable to make a reliable measurement of low noise, low distortion amplifiers with single ended output.

The situation is better when we measure amplifiers with balanced output. The capacitor even harmonics distortion, which is mostly H2, is then almost eliminated. I have enough measurements to make my statements.

Thanks a lot!
Do you get the same LF rise if you use the 2,5mm input on the ADC?
That one allows up to 43Vrms input level (with bottom switches set for 10V sensitivity) and has 13.6kΩ input impedance. I suspect the 2,5mm input might have the input series resistor built-in.

 

[restorer-john]

plus a shorting option (to be able to measure DC).

You'll have DC offset issues/compensation to deal with, especially linearity around bi-polar-zero.

 

[KSTR]

You'll have DC offset issues/compensation to deal with, especially linearity around bi-polar-zero.

Offset is taken care of by software nulling prior to measurement, or in any other way determined by the task.

Why should there be (low-level) linearity problems? It's a D/S ADC, that is, basically no linearity errors.
I'm doing this already with my ADI-2 Pro (AK5574) where I shorted the input caps. This is working really well.

But, @IVX, to properly implement that the ADC default DC filter should be switched off -- which might already be the case (?) as I see DC offset in the output at the moment, quite a bit on the right channel.

 

[KSTR]

OK, that's what I was looking for, thanks! Are these bipolar or single ended and biased?

Physically very small low voltage bipolar e-caps, I didn't take notes though. With ground referenced input they do have DC bias on them, as their output side (FDA summing nodes) is at Vcom bias potential of the ADC chip, +2.25V or thereabouts.

 

[SIY]

Physically very small low voltage bipolar e-caps, I didn't take notes though. With ground referenced input they do have DC bias on them, as their output side (FDA summing nodes) is at Vcom bias potential of the ADC chip, +2.25V or thereabouts.

I wonder if the low voltage rating is the issue here?

 

[LKA]

I recommend replacing the 220u/10V capacitors with 470uF/16V (C1 C2 on my schematic). I personally use 1000u/16v, I get much better LF performance when using the SE input.

 

[LKA]

PMA, have you tried a balanced divider for the SE output ?

 

[pma]

PMA, have you tried a balanced divider for the SE output ?

I use nothing else than the balanced divider (even for SE), since the beginning. See the green trace in the post no. 1543 above.

 

[pma]

I recommend replacing the 220u/10V capacitors with 470uF/16V (C1 C2 on my schematic). I personally use 1000u/16v, I get much better LF performance when using the SE input.

You know, the capacitors are a basic issue (and remember that even in simulation there must be a return path, thus R6, R7 for simulation purposes with balanced input. Microcap knows and would refuse to start analysis without the ground return path).

Big C1 and C2 may keep charge and their behaviour is unpredictable (long sweeps, starting at 2 Hz, previous initial conditions).

The design should be direct coupled (with a necessary DC shift) to get good LF parameters, or use much higher input impedance and polypropylene capacitors.

 

[pma]

Just an idea - this should work? To eliminate coupling capacitors. Component values as an example for 10V range.