E1DA Cosmos ADC

[IVX]

Guys, I don't want to repeat that I see nothing interesting in the entire idea of cascading ADCs, and especially cascading DACs(100.00% nonsense). The Stitcher is a purely DIY thing with no practical or commercial purpose. The only reason I made that is to explore how it will, in fact, work when implemented in the simplest way. And it works with inaudible artifacts to me.

 

[KSTR]

I certainly would agree that cascaded DACs don't offer significant benefits and are quite an effort to realize as the stitching is happening in the analog domain.

But stacked ADCs are relatively easy to implement and offer a very effective means to avoid headroom vs noise floor issues in the real world and thus are now implemented more and more in many devices (as most patents have expired).

 

[FrenchFan]

I agree, it's a nice toy to explore.

And, as you say, the artifacts are inaudible.

But if I ever have a microphone with DR >130dB , maybe I'll use it.

Just for the FUN .

 

[1231rq32r1qw32r]

Did anyone ever resolve that audio dropouts that occur in MacOS? I've tried various buffer settings and various apps - can't seem to resolve them

 

[jmag999]

I had the same issue when using REW. I downloaded the latest version of REW a few days ago and it's no longer happening. I don't know if it was a mac os version or rew version that fixed it for me.

 

[1231rq32r1qw32r]

Happens for me in every application

 

[jmag999]

I haven't tried it with any other applications, but when it occurred, I could see the microphone icon in the taskbar quickly flash off and then back on. Do you experience that?

 

[1231rq32r1qw32r]

No, I don’t. It’s dropping what seems like less than 1ms. Something weird going on. I’m using an official Apple USB C cable. I’ll try some other stuff

 

[1231rq32r1qw32r]

Fixed the dropout issue. I created and aggregate device (with the other device that I'm using for output) and I set the D1DA to be the clock source. Issue fixed

 

[IVX]

Interesting that Windows users didn't complain about that.

 

[1231rq32r1qw32r]

Two totally different ways of handling audio. I'm not even sure if windows can do separate devices for IO. Been a long time since I was forced to use Windows though...

 

[Cale]

Hi,

I am using an ADAU 1452 evaluation board from Analog Devices for my active speakers, which I really like because of the endless possibilites by the use of Sigma Studio.

The only drawback is that it is using an AD1938 codec on board for the analog outputs. With 108 dB DAC dynamic range and -94 dB THD+N performance could be better, though it is not problem in my current application. As input I use the SPDIF Input, therefore the ADC performance does not matter.
I am thinking of improving the DAC performance by using I2S to balanced analog out DAC PCB. To ensure that this combination is really better, I would like to measure the current performance and that of a potential replacement.

Therefore I bought the Cosmos ADCiso and have some question after my first trials. I am using balanced outputs with nominal 2 Vrms out for 0 dBFS. I measured it with an multimeter and the reading showed 1,78 Vrms, which should be correct, just 1 dB out of spec.
On the ADCiso I choose 2.7 V input setting and 1K ohm impedance. I connected the XLR cable to the left input only and the device is running in mono mode.

My first measurement of the spectrum looked like this.

I was surprised by a few things:

• Level should be higher than -10.73 dBFS
• THD is about 10 dB lower than spec of the pure codec
• SNR is 9 dB lower than spec of the pure codec (108 + 3 =111 because of balanced use)

After this I changed the input setting of the ADCiso to 4.5 volt and 1.66 K ohm impedance with the following result.

The level dropped more or less as expected. The SNR is 0.5 dB worse, but the THD improved by 9 dB (now in spec), which I did not expect. This means that already two things are different than expected, the too low level and the THD improvement with lower sensitivity of the ADCiso.

I started to measure the frequency response for 2.7 V input setting.

The low frequency rolloff was not there, when I measured it with an RME Interface and I decided to measure it for all sensitivities of the ADCiso.

As you can see the rolloff towards lower frequencies is dependent on the ADCiso input setting. I remembered that Amir has written, that the low input impedance of the ADCiso could be a problem for some devices.

Do you think that this could explain the low level and the strange THD behaviour as well? Would it help, if I would use a E1DA Cosmos Scaler in front?

Thank you!

 

[staticV3]

On the ADCiso I choose 2.7 V input setting and 1K ohm impedance. I connected the XLR cable to the left input only and the device is running in mono mode.

Mono mode sums the Left and Right input.

If the signal is only present on one input, then the signal amplitude will be 6dB lower than expected.

THD is about 10 dB lower than spec of the pure codec

The codec might not like the low 1k input impedance

As you can see the rolloff towards lower frequencies is dependent on the ADCiso input setting.

Yeah definitely caused by low damping factor.

This is by design BTW, so that Amps and DAC+Amps which don't need a buffer can be measured with maximum SNR.

A buffered input adds some noise, and can be added for DUTs that need it in the form of the Cosmos Scaler.

Would it help, if I would use a E1DA Cosmos Scaler in front?

Yes.

FYI, if you use just one of the Scaler's two inputs, then it'll automatically output the signal to both outputs so that the Cosmos ADC's Mono mode works seamlessly. No splitter cables/ Y cables required.

 

[mdsimon2]

AD1938 datasheet quotes unweighted, 20-20K, DR of 104 dB (typical) / 98 dB (min). 108 dB is A-weighted, average. Dynamic range is typically measured by sending a -60 dBFS signal, measuring the signal to noise ratio and then adding 60 dB to the result.

If you are only using one input, make sure the Cosmos is set to stereo mode. Mono mode should only be used if you are using a y-splitter to connect a single output into both inputs. That is likely part of your level issue. The other thing to consider is the low input impedance of the Cosmos will create a voltage divider. In practice this typically means you can use an input voltage setting one level lower than you think.

The low input impedance also affects THD. In general, you will get better THD numbers with higher impedance but noise will be worse. I posted some data on this from a MOTU Ultralite Mk5 here -> https://www.audiosciencereview.com/...xxmp-buffer-bypass-someone.32763/post-1147902.

Michael

 

[Cale]

Thank you very much for your help . After your advise I remeasured the spectrum, using an XLR Y - splitter to spread the signal to both inputs of theADCiso.
As input channel in REW I am still using the left output of ADCiso, sensitivity setting is back to 2.7 volts.

The Y-cable increased the level from -10,73 to -5,51 dBFS = + 5,22 dB. SNR stayed at about 102 dB, which makes me think (especially with your comments in mind) that this the real performance level of the DUT. THD+N increases from -85,3 to -70,8 dB fitting to the statement / measurements of mdsimon2 .
As you wrote, one option seems to be the scaler, another (limited) one to use different sensitivities / input impedance for SNR and THD.

 

[staticV3]

The Y-cable increased the level from -10,73 to -5,51 dBFS = + 5,22 dB.

+6dB from proper summing
-0.78dB from the even worse damping factor, alse evidenced by the drop in THD/THD+N

 

[Sokel]

Hi,

I am using an ADAU 1452 evaluation board from Analog Devices for my active speakers, which I really like because of the endless possibilites by the use of Sigma Studio.

The only drawback is that it is using an AD1938 codec on board for the analog outputs. With 108 dB DAC dynamic range and -94 dB THD+N performance could be better, though it is not problem in my current application. As input I use the SPDIF Input, therefore the ADC performance does not matter.
I am thinking of improving the DAC performance by using I2S to balanced analog out DAC PCB. To ensure that this combination is really better, I would like to measure the current performance and that of a potential replacement.

Therefore I bought the Cosmos ADCiso and have some question after my first trials. I am using balanced outputs with nominal 2 Vrms out for 0 dBFS. I measured it with an multimeter and the reading showed 1,78 Vrms, which should be correct, just 1 dB out of spec.
On the ADCiso I choose 2.7 V input setting and 1K ohm impedance. I connected the XLR cable to the left input only and the device is running in mono mode.

My first measurement of the spectrum looked like this.
View attachment 474321
I was surprised by a few things:

• Level should be higher than -10.73 dBFS
• THD is about 10 dB lower than spec of the pure codec
• SNR is 9 dB lower than spec of the pure codec (108 + 3 =111 because of balanced use)

After this I changed the input setting of the ADCiso to 4.5 volt and 1.66 K ohm impedance with the following result.
View attachment 474322
The level dropped more or less as expected. The SNR is 0.5 dB worse, but the THD improved by 9 dB (now in spec), which I did not expect. This means that already two things are different than expected, the too low level and the THD improvement with lower sensitivity of the ADCiso.

I started to measure the frequency response for 2.7 V input setting.
View attachment 474323
The low frequency rolloff was not there, when I measured it with an RME Interface and I decided to measure it for all sensitivities of the ADCiso.

View attachment 474324
As you can see the rolloff towards lower frequencies is dependent on the ADCiso input setting. I remembered that Amir has written, that the low input impedance of the ADCiso could be a problem for some devices.

Do you think that this could explain the low level and the strange THD behaviour as well? Would it help, if I would use a E1DA Cosmos Scaler in front?

Thank you!

Sometimes lows are affected by smaller FFT size, and in general it's better to use more than double the sample rate one (4x + is even better)
Give it a try and see. Also check THD+N for the very lows measuring this way.

 

[Cale]

Sometimes lows are affected by smaller FFT size, and in general it's better to use more than double the sample rate one (4x + is even better)
Give it a try and see. Also check THD+N for the very lows measuring this way.

After your post I tried this as well, but even extrem long FFTs did not make a difference. I testet this at 1 kHz and 100 Hz.
Interessting are the curves of THD vs frequency for different input sensitivities / input gains. This is an older measurement with just one channel of the ADCiso used but again this shows to my understanding that the eval board and the ADC iso are not a good match.
THD is shown in % this time to compensate the different input levels of each graph.

 

[nick_l44.1]

After your post I tried this as well, but even extrem long FFTs did not make a difference. I testet this at 1 kHz and 100 Hz.
Interessting are the curves of THD vs frequency for different input sensitivities / input gains. This is an older measurement with just one channel of the ADCiso used but again this shows to my understanding that the eval board and the ADC iso are not a good match.
THD is shown in % this time to compensate the different input levels of each graph.

View attachment 474469

This is the dependence of board distortion on load (ADC input resistance). You probably need a buffer between the board and the ADC. If you want to measure THD more accurately without a buffer, set the maximum input resistance on the ADC (10k).

 

[nanook]

The ADAU 1452 evaluation board has 50 Ohms series resistance and 10uF DC blocking in the outputs.
The low input resistance of the Cosmos ADCiso is a real drawback for many applications. I finally decided to get the scaler as a high Impedance buffer with the added benefit to have gain available and a convenient mono mode. Since then the gain of my ADCiso sits untouched in the 4.5V setting which is nice because I do not expect the DIP-switches to be very robust.