E1DA Cosmos ADC

[Paulvaso80]

The right channel has the resistor divider 10k/10k to make half of V-reference voltage, and that Vref is filtered by 47uF MLCC, if that cap will lose the connection, I guess we can see something like you show i.e. low freq. noise.

Ok! that sounds good, it could be a reason, but can I measure it from the XLRs? I haven't opened the drive yet.

 

[Paulvaso80]

Ok Ivan, here it is 7:44 AM and I really need to go to sleep (I stayed all night because of the time difference).

In a few hours I will connect again to see if you were able to re-evaluate the situation with this problem.
I really don't want to open the device.
I paid a lot to receive a product in perfect condition and that is really what I want.
If it's a simple problem to be fixed by software/firmware, then Ok.
But if in addition to paying, I have to work to repair the board or replace 1 mm smd components, then the situation does not please me.

I ask you, with all due respect, to please find a reasonable way to fix this.

Thank you very much, and excuse my bad English.

 

[Music1969]

If it is a hardware problem, it is almost impossible to return it on time to Amazon USA from my country before the deadline (May 25), and it would also be almost as expensive as the equipment itself.

You should lodge a return request with Amazon immediately.

It may be the case that as long as you show evidence that you post before the 30days, they MAY be understanding - but you won't know till you ask them.

Use the online chat with Amazon so that you have a written record, if they agree to anything.

 

[NDD]

Hi Paulvaso80,
Very sorry to hear about your noisy ADC, but it is also reassuring to know that I'm not the only one that has this problem. Your postings have motivated me to join and describe my issues in the hope we can get this fixed.


Hi Ivan.
I have the Cosmos ADC A variant. It worked fine initially, recently the right channel has become noisy.

The right channel led flickers continuously.
It is independent of the inputs being open circuit or short circuit.
It appears to be independent of ADC gain setting.
On power on the REW oscilloscope display show a very noisy signal. REW RTA showing a noise powe of -83dBFS
After an hour this quietens down to a noise spike every half second or so.
The REW RTA spectra in my case is flat, which would correspond to the very short duration of the noise spikes.
The spectra bounces around, increases when a noise spike comes along and decreases at a rate dependent on the averaging applied.


On power on

After 50 mins

I have found a temporary work around. I have made a simple 2W heater with a handful of power resistors which is placed under the ADC. After the ADC warms up the noise spikes disappear and I measure a stable noise power of -127dBFS A in mono mode. In this state the temperature of the top of the case is 28C and the temperature of the bottom of the case, measured with a radiometer after removal from the heater, is around 29C.

What is the issue here and how can this be permanently fixed?

 

[IVX]

NDD, I think you may try the same thing I recommended Paulvaso80 i.e.
Try to change ADC_IMT_SEL parameters in Tweak_Cosmos application to see if it helps. The right channel ADC_IMT_SEL2 and ADC_IMT_SEL4, the default values are 11;11, you can try 11;7, 7;11, 7;7, also you can try 15 instead of 7. Try to mod that online to see if that helps. ADC_IMT_SEL anyhow affects the noiseshaper loop stability, and the optimal value isn't always 11;11.

 

[NDD]

Thanks Ivan, I'll give that a go.

 

[NDD]

Hi Ivan,
Changing the ADC_INT_SEL ADC2 and ADC_INT_SEL ADC4 parameters makes a significant difference to the noise power, in particular ADC_INT_SEL ADC4, however the ADC was already set up with the values which gave the lowest noise. See summary table below.

Measured using the REW RTA.

What else could be causing this issue?

 

[IVX]

NDD, actually, nothing more, this issue is 100% ES9822 related. I have a few tens(about 20pcs) chips replaced from the new units, if 9822 can't pass my QC.
The first time I noticed 9822 changed its noise level during days when I sent the Cosmos ADC sample to @staticV3. He tested that and reported back about the noise mismatch vs my measurements. After that, he did modify ADC_INT_SEL ADC and the noise was a bit reduced. BTW, you can reflash the newest FW as well, because there is the noise a bit improved. https://drive.google.com/file/d/15n9tqRxPJmrOOEtsab6VNdi54Dsc20o6/view?usp=share_link

 

[LKA]

IVX, how to improve the THD 10k performance of the cosmos adc ? Both ASUS STXII and my modded XFi USB HD have better numbers.

 

[IVX]

LKA, you can try to mod the reference schematic like that:

ES9822 wins at max levels, like -.5dbfs.

 

[pma]

IVX, how to improve the THD 10k performance of the cosmos adc ? Both ASUS STXII and my modded XFi USB HD have better numbers.

How do you know what is the Victor generator distortion at 10kHz?? Any reference measurement, like AP, or notch measurement? Please take into account that in case of similar distortion of the generator and soundcard the distortion components may subtract (depending on their phase) and you get seemingly lower distortion with the worse ADC. This is definitely the case of your upgraded USB X-Fi HD. I found the same issue about 3 years ago with that sound card, which turned out to be a measurement error.
And how about the generator working into Cosmos low input impedance?? No rise of generator distortion?

You can also see much higher noise with the X-Fi HD. In your measurements, check the noise floor of D10s CCIF to Cosmos x X-Fi. Cosmos has much lower TD+N and same distortion. X-Fi may look seemingly “nicer” only because of higher noise floor. You also do not show THD+N vs. frequency, but THD vs. frequency. Again, Cosmos is better, because of much lower noise compared to the soundcard used.

BTW, I find the E1DA Cosmos ADC very useful tool. Sometimes, I need to use an input opamp buffer to get higher input impedance. An I have ceased using the mono mode, in stereo mode L channel I found the most consistent results, though with a slightly higher noise floor. This was my "best buy" of the affordable digital measuring tool. I agree it is aimed rather for a qualified user.

 

[onlyoneme]

Sometimes, I need to use an input opamp buffer to get higher input impedance.

Can you suggest something available on the market in the absence of Scaler? I have no skills to do it myself.

 

[LKA]

IVX, I have clipping at -6dBFS, so -9dBFS corresponds to 1.25Vrms or so.

PMA, my XFi has all opamps replaced (2xLME49720 + 2xOPA1656 + 1xforgot_the_name), so it is better than yours.
I'm talking about THD not THD+N. Cosmos ADC has superior THD+N at 1kHz but disappoints me below 100Hz and above 3kHz.


All measurements done 24b 96k

DAC D10s -> ADC Cosmos (input 1.3Vrms unbalanced)
CCIF IMD = -115dB
THD 10k = -119dB

DAC D10s -> ADC STXII
CCIF IMD = -115dB
THD 10k = -120dB

DAC D10s -> ADC XFi USB HD modded
CCIF IMD = -115dB
THD 10k = -120dB

D10s is the limiting component

===================================================

Viktor10k -> ADC Cosmos (input 1.3Vrms unbalanced)
THD 10k = -119dB

Viktor10k -> ADC STXII
THD 10k = -121dB

Viktor10k -> ADC XFi USB HD modded
THD 10k = -122dB

==================================================

 

[LKA]

I made a measurement amplifier (MAmp) capable of doing 0.000003% thd into 1kOhm at 10kHz (-150dB in simulation)

Three-pole compensated blameless clone

We plan on moving D9 on the Wolverine to the collector on the next pcb order. This was recently found. Jeremy

www.diyaudio.com

Cosmos ADC input switched to 10Vrms

Viktor10k -> MAmp -> ADC Cosmos (input 9Vrms unbalanced)
THD 10k = -119dB

Cosmos ADC: Whatever try I cannot go below -119dB with THD10k, which is worse than STXII (CS5381, release year 2002) and XFi (CS5361, release year 2002) can do

MAmp with passive notch, note: the graph is not compensated for the passive filter
Viktor10k -> MAmp_9Vrms -> Divider_306R+51R -> PNotch10k -> Amp_100x -> XFi_nocomp
H2 = -138.5dB +9dB comp = -129.5
H3 = -140.2dB +6dB comp = -134.2
H4 = -156.2dB +3dB comp = -153.2
THD10k = -128dB

DAC, ADC comparison http://stephan.win31.de/dac-adc-hist.htm

 

[phofman]

IME the REW coherent (i.e. vector) averaging reduces tiny harmonics more than the averaging based on amplitudes only. The reason is simple - due to noise the tiny harmonics vectors fluctuate both in amplitude and angle. Vector averaging (by principle) reduces components with unstable angle more. IME coherent averaging is great for identifying tiny harmonics (random noise unrelated to the fundamental gets suppressed a lot), but not so much for measuring their amplitude.

Also - please what is the cause for the raised noise floor around the fundamental in the Cosmos screenshot (no notch IIUC), compared to the flat line of the XFi notch spectrum (adjusted by REW for the notch params)? The second chart looks more like a measurement of a DAC-ADC chain with single common clock (i.e. exact absolutely stable frequency). Just wondering about the cause, not disputing anything.

 

[pma]

D10s/Cosmos clock stability of this combo is poor.

 

[phofman]

Still I do not understand the difference between the two spectrums. In both cases combination of a crystal-clocked ADC and of the same analog oscillator (i.e. very unstable compared to the crystal clock).

 

[pma]

Still I do not understand the difference between the two spectrums. In both cases combination of a crystal-clocked ADC and of the same analog oscillator (i.e. very unstable compared to the crystal clock).

I do not understand it either. But, none of us two has made the same test with the same devices, so who knows. The test description posted is not clear enough to me.
But, one test is without the notch (Cosmos), the other one with the notch (XFi).

Moreover, I have never seen such spectrum noise floor rise near the 10k testtone as shown by @LKA .

 

[pma]

Still I do not understand the difference between the two spectrums. In both cases combination of a crystal-clocked ADC and of the same analog oscillator (i.e. very unstable compared to the crystal clock).

Moreover, I have never seen such spectrum noise floor rise near the 10k testtone as shown by @LKA .

I get this from D10s + Cosmos at 10kHz. No noise floor elevation near 10kHz. THD+N is of 40dB better than @LKA shows here.

 

[LKA]

"I have never seen such spectrum noise floor rise near the 10k testtone as shown by LKA"
It is specific to Viktor's analog oscillator (1k,10k,...)

The second test "Viktor10k -> MAmp_9Vrms -> Divider_306R+51R -> PNotch10k -> Amp_100x -> XFi_nocomp" is just proof that the "Viktor10k -> MAmp" combo can go well below -120dB THD10k.

Mostly I measure power amps, the noise is much higher than at line level. My favorite tests are stepped THDvF graphs at levels 1W,10W,50W,100W,200W into 4R.