• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

E1DA Cosmos APU - Teardown and personal thoughts and measurements

Here is output of my SMSL SP400 headphone output (combination with SMSL D-6 DAC).

APU + ADC

I'm pretty happy with this out of headphone amp

I did the custom 1kHz notch compensation curve following Archimago method


View attachment 280643
A few (amateur) questions:
  • What ADC did you use? If it was the E1DA Cosmos ADC, I assume the APU was not really needed for an HPA (no issue with ADC impedance) at this level of noise and THD+N, correct?
  • The Archimago notch compensation curve gives the nice and "usual" FFT graph that you show, but it is not really needed either to get the correct N / THD+N results, correct?
  • Any reason why you choose to use the Dolph-Chebyshev windowing?
 
  • What ADC did you use? If it was the E1DA Cosmos ADC, I assume the APU was not really needed for an HPA (no issue with ADC impedance) at this level of noise and THD+N, correct?
The APU notch may be used, as it s here, to provide a more realistic load to the DAC/HPA + the notch allows a much deeper analysis of THD.
  • The Archimago notch compensation curve gives the nice and "usual" FFT graph that you show, but it is not really needed either to get the correct N / THD+N results, correct?
The Archimago compensation is an approximation of the (reverse) notch filter, so, as it compensates (although approximatively) for the notch filter, it is needed to get readable THD+N result.
  • Any reason why you choose to use the Dolph-Chebyshev windowing?
Audio Precision uses a proprietary "AP Equiripple" FFT windowing function, which they describe has "derived from Dolph Chebyshev".
So: why not to use the closest function from the AP's ?
It shouldn't make a difference, though.

By experience, what does make a difference is that REW should be set to use "AES17 notch" setting for THD+N (and especially for the "N") measurement with the APU

EDIT: FFT size 2M is complete overkill.
64k is far enough at 44.1 or 48kHz sampling frequency.
Also, the result is better if the ADC input level is close to 0dBFS. Here, with the compensation, it's difficult to say what the ADC level is, but -9.63dBFS, if that's the ADC level, is far from optimal.
 
Last edited:
Also, the result is better if the ADC input level is close to 0dBFS. Here, with the compensation, it's difficult to say what the ADC level is, but -9.63dBFS, if that's the ADC level, is far from optimal.
Archi was at -5 dBS

My ADC is at 4.5 Vrms input sensitivity

I guess I could lower it and then reduce headamp output and see if I can squeeze more SINAD but I'm already happy with the SINAD of DAC+amp.
 
Last edited:
Since if have APU i might as well try to squeeze out most SINAD i guess :p

Re-done properly with optimal gains - better than Archimago ! (closer to 0 dBFS) - gained 1dB in SINAD.

This is DAC + headamp (SMSL D-6 and SMSL SP400) - clean !

1682165168215.png
 
Is @Archimago on this forum anymore?

The calc in his picture is different to the calc in his caption.

Which one is correct?
View attachment 282506
REW gives N+D level A-weighted but as an absolute level not as a ratio, hence, we need to subtract the fundamental level there: 126.8-1.41 = 125.39db(A). The A-weighted result could be better for up to 3db, usually 2-2.5db, and you have that 125.4-123.3=2.1db.
 
Recently purchased two APU and and E1 ADC.

Two quick questions.

1. Can someone link to amazon the RCA to 3.5 inputs I'd need to connect the output of my MM turntable to the APU? I'm concerned I may pick something with the wrong terminations.
2. Can someone link to amazon the appropriately pinned out interconnect to run between APU and E1 ADC? I'm also concerned about selecting the wrong pinout.

Thanks!
 
So, for an unbalanced device going into 2 APU devices, I would just need 2 of these simple 3.5mm TS to RCA cables?

Screenshot 2023-07-31 192851.jpg
 
Can anyone (@IVX?) help explain the noise floor tilt in APU notch -> ADC measurement? I see a similar tilt in Archimago's measurements, but it looks particularly obvious here:

1710422037348.png



Frequency response is nearly perfect with calibration, so it's just the noise floor that seems to produce a tilt:
1710422391284.png
 
Can anyone (@IVX?) help explain the noise floor tilt in APU notch -> ADC measurement? I see a similar tilt in Archimago's measurements, but it looks particularly obvious here:

View attachment 356347


Frequency response is nearly perfect with calibration, so it's just the noise floor that seems to produce a tilt:
View attachment 356352
This is a very typical noise response from an active notch.
The reason is that the RC circuit is typically high resistance below the notch freq, while above notch the capacitors shunt the resistors, lowering impedance, shorting out the noise contribution from the resistors. The hump comes from the positive feedback which is used to make the notch very high Q. On top of that we have the typical 1/f rsisng noise slope of the opamps, including the ADC itself.

Noise plot for a typical notch filter (Hall notch type), with noise-less opamp:
1710452690650.png


Frequency response:
1710452855779.png
 
Last edited:
This is a very typical noise response from an active notch.
The reason is that the RC circuit is typically high resistance below the notch freq, while above notch the capacitors shunt the resistors, lowering impedance, shorting out the noise contribution from the resistors. The hump comes from the feedback which is used to make the notch very high Q. On top of that we have the typical 1/f noise slope of the opamps.

Noise plot for a typical notch filter (Hall notch type), with noise-less opamp:
View attachment 356431

Frequency response:
View attachment 356432

Thanks, Klaus, that makes sense. Unfortunately that has the potential to skew N, N+D, SNR and THD+N calculations if noise is dominant.
 
Back
Top Bottom