To notch or not to notch

[bohrok2610]

When measuring low levels of THD or THD+N the recommendation is often to use a notch and LNA.

Here is a measurement at -2dBFS of a AK4493 dac with ES9822PRO. Without notch.

I get similar results with another ES9822PRO ADC so this is not device dependent. And just to show that this is not some heavy distortion cancellation here is the same setup but with at 2567Hz -3dBFS.

So very similar result.

Now let's use a -30dB notch + LNA (CosmosAPU). But before that here is a measurement without notch but at -32dBFS level.

Third harmonic is now clearly dominating over 2nd. Hard to tell if this is due to DAC or ADC.

And finally a measurement at -2dBFS using Cosmos APU +6dB with compensation.

The harmonic profile is very similar to previous measurement at -32dBFS.

Now the question: compared to the first measurement without notch why would this notch measurement be considered more accurate?

 

[altor]

Measurements with notch just eliminates (try to eliminate) ADC's distortion at high signal level (by reducing the main tone amplitude).
So the questions is what is dominate - ADC's distortion or DUT distortion.

 

[bohrok2610]

I understand that but IMO this only works if ADC distortions get lower in "linear" fashion when input level is lowered. As we have seen in the Return of the ESS hump thread this is not the case with ES9822PRO. So depending on the notch depth ADC distortions may actually be relatively higher than without notch.

Another issue is that with additional active elements in the measurement path (notch + LNA) the possibility of distortion cancellations (or additions) increases.

 

[altor]

if ADC distortions get lower in "linear" fashion when input level is lowered

Sure.

So depending on the notch depth ADC distortions may actually be relatively higher than without notch.

Yes, if 9822's hump is really exists (I didn't measured, but if will have a time, will do).