Consequences of insufficient reconstruction filter in a DAC

[KSTR]

People are used to measure out of band with white noise, see tons of such pics floating around and also Amir's stopband ones. KSTR always amazes me, I have never seen so clear and clean measurements of that before (obviously not using noise).

Actually, I used REW's periodic white noise with band-limiting for these plots which gives a flat response in the RTA without any windowing when the FFT size matches the period. For the stopband plot of the ADC the noise starts at fs/2, for the passband it stopped there. For the DAC a full range source noise was used.

The hard part was to get that all work with two devices running at different sample rates (though one at multiples of the other) and with the devices not even sync'ed (as that can't be done at different rates, the ADI-2 don't have a 10MHz word-clock input **). The trick I used was heating/cooling the master clock oscillator of one device to bring it very close to the other, giving a low beating frequency and enough time to capture the FFT (with some averaging applied) at a point with no leakage artifacts.
For the ADC plots, I added some amount of smoothing to touch it up a little and I had to apply singificant gain to the stop-band curves to stitch them to their passbands, which is why the noise floor appears to be low on those.

**) I do have ideas how to fix this, the key is to feed both devices from the same 25MHz master clock oscillator... which would be a significant hardware hack not without risk to ruin the interfaces...

It's a pity that REW can'be scripted but then again, for a freeware it's absolutely phenomenal and in many regards on par with high $$$ AP, ListenInc, etc

 

[MC_RME]

There should be a much easier way. You can sync the second ADI from the first by using ADAT 1 out to ADAT 2 in, ADAT 2 in set as clock source in unit 2. ADAT is SMUX, so you can easily use the second unit in Double and Quad Speed based on the 'Single Speed' ADAT carrier coming in.

Of course in this setup only the second unit runs with REW, the first needs to be fed by a different computer or iPhone/iPad ( I use Neutron. I know I mentioned this several times already, but it is such a useful combination...). I just tried it - works.

 

[MC_RME]

Using the correct noise signal makes the biggest difference. I followed your advice and used the full range periodic white noise of REW with 8k length. Analysing this in HpW @ 32k sample size (4 times the sample rate) gives a noise free, straight line. No line smoothing or so needed. To reduce the (analog) noise at lowest levels 10x averaging was used. The frequency errors near sf/2 are nearly completely gone with one unit DA/DA at 44.1 kHz - didn't expect that! And using two units synced (see above) gives very nice stopband graphics.

 

[melowman]

Awesome curves! I'm not a measurement technical expert +++ but I had the intuition that some noise was used by KSTR..
Got to wait 2021 to make these curves possible and I am glad Matthias leaned something new. You help a lot ASR and I'm glad ASR also helped you back.

 

[KSTR]

There should be a much easier way. You can sync the second ADI from the first by using ADAT 1 out to ADAT 2 in, ADAT 2 in set as clock source in unit 2. ADAT is SMUX, so you can easily use the second unit in Double and Quad Speed based on the 'Single Speed' ADAT carrier coming in.

Of course in this setup only the second unit runs with REW, the first needs to be fed by a different computer or iPhone/iPad ( I use Neutron. I know I mentioned this several times already, but it is such a useful combination...). I just tried it - works.

Ah, thanks for that syncing tip, Matthias! Will try ASAP.
Your plots also show the higher stop-band level and ripple of SD-LD filter, this was a thing I was not 100% sure about whether it is real or not. In general we have good agreement in the curves, always a good thing to establish some confidence.

REW can run multiple independent instances (select option at install-time) so there is no need for a second computer for the second interface.

 

[MC_RME]

REW can, but ASIO can't....

The slightly higher ripple of SD LD is documented in the AK4493 data sheet.

 

[KSTR]

REW can, but ASIO can't....

Correct... had remembered this wrong. The low level driver seems to handle it, one can set up both intefaces with different settings but once ASIO connects to that driver layer it forces a common sample rate.
Though it might not impossible to hack that...

 

[MC_RME]

And finally: @amirm could use this simple trick as well to make the AP stopband measurements look like drawn from MathLab...

Edit: the attached zip includes four noise files made with the REW generator. 44.1 and 48 kHz each as no intersample peaks and with intersample peaks. The 'no ISP' is simply reduced in level to not cause these. The 'with ISP' produces about +4.3 dBFS and can be used to test the filters when overdriven.