Using Cross Corelation to lower influence of ADC for DAC measurements

[FrenchFan]

May I ask you to measure without weighting ?
It's a detail for sure, but this kind of measure is always unweighted here.

OK I do ASAP

 

[FrenchFan]

May I ask you to measure without weighting ?
It's a detail for sure, but this kind of measure is always unweighted here.

It enought only correlation

 

[Rja4000]

I didn't have the recording from yesterday, but this should be similar: https://we.tl/t-KzWymSK46j

Which DAC is it ?
D50 III in mono mode (both outputs summed) ?
'cause I have similar results in that case with notch.
Just to compare....

 

[FrenchFan]

Seems to work for me. Effect is most easily seen with open inputs, eliminating most common mode content. Measurement titles below show total input rms level for the various tests, starting with an rms average as a reference.

View attachment 385128

for me with E1DA cosmos ADC give exactly -15db with 1000 correlation

 

[pkane]

Which DAC is it ?
D50 III in mono mode (both outputs summed) ?
'cause I have similar results in that case with notch.
Just to compare....

It's SMSL DO300. Just the right channel output, no summing.

 

[DonH56]

what would you get with longer fft's comparable with the number of samples processed in total?

Longer FFs provide narrower bin widths, so the noise floor will drop, but the RMS value we are comparing will not change.

Edit: John @JohnPM already answered, should've read ahead.

 

[Rja4000]

It's SMSL DO300. Just the right channel output, no summing.

I need to try one ;-)

 

[pkane]

I need to try one ;-)

I like it. Bought a used one for about half the price.

 

[Rja4000]

I like it. Bought a used one for about half the price.

I need the same ;-)

 

[nanook]

In simpler terms, what happens to the CC averaging output if the ADC channel input signals are identical except for polarity? Does it (partially) cancel?

My expectation is that the correlation gives negative values for these signal components and the FFT will consider this as an amplitude and it will thus have a non zero rms value.

But as mentioned in post #112 I do see a real benefit for those systematic ADC imperfections that are polarity dependent.
Edit: I would therefore promote the proposal @KSTR made to have a button to invert the captured data of one of the channels (personally I'd prefer to have the option to invert the reference-channel)
Edit: removed systematic because I think this applies for every imperfection that is polarity dependent

 

[JohnPM]

What equation did you take

Averaging re1*re2 + im1*im2 for each FFT bin, the average being a proxy for the squared magnitude at that bin frequency with the proviso that it could be negative.

 

[FrenchFan]

Averaging re1*re2 + im1*im2 for each FFT bin, the average being a proxy for the squared magnitude at that bin frequency with the proviso that it could be negative.

(a+Jb)*conj(a+Jb) = (a+jb)*(a-Jb) =a^2+b^2
Good for me
after, i don't understand
sqrt(SUM(a^2+b^2)/nb-correl) right for you .
If this , we continue
I try to do the simulation with same level , -134dBFS / 64k

 

[JohnPM]

In simpler terms, what happens to the CC averaging output if the ADC channel input signals are identical except for polarity? Does it (partially) cancel?

OK, finally simple enough for me to understand.

Yes, that works. Here it is on that Apollo Solo again for 1000 cross correlations with a synthetic 1 kHz tone injected with the appropriate polarity on each input, with some H2 and H3 to confirm they survive the processing. There isn't a lot of common mode pickup on those open inputs but enough to show it works.

 

[JohnPM]

after, i don't understand
sqrt(SUM(a^2+b^2)/nb-correl) right for you .

Once the sets of re1*re2 + im1*im2 have been summed and divided by the number of cross correlations plot 10*log10 of their absolute values, since it is possible for them to be negative at frequencies where there is no input.

 

[JohnPM]

Thinking that to the end, a gain fine-trim (for one channel would be handy to get deepest null there.

The virtual balanced input uses the FS voltage calibration values for each input to adjust the level before the sum/difference operation. The same will apply for the inversion option.

 

[nanook]

Yes, that works.

Ok, now I'm bit puzzled ...
Is the improvement observed really due to cross-correlation, or is it in the end rather a subtraction of the common-mode signal present on the two inputs?
(which would still be a nice result)

 

[JohnPM]

Is the improvement observed really due to cross-correlation, or is it in the end rather a subtraction of the common-mode signal present on the two inputs?

Both.

 

[KSTR]

At any rate, a consistent 10..15dB reduction would be a rather terrific improvement!

 

[FrenchFan]

Once the sets of re1*re2 + im1*im2 have been summed and divided by the number of cross correlations plot 10*log10 of their absolute values, since it is possible for them to be negative at frequencies where there is no input.

for me is not good , look simulation , i have exactly attenuation = 5*log(N) for these levels with 2noise uncorrelated ~ 134dBFS
I have same reaction using E1DA ADC

 

[FrenchFan]

I like it. Bought a used one for about half the price.

For your DAC output measure use E1DA ADC ???