Hi @bmsorin, it’s a bit hard to compare the two plots because one is absolute while the other is relative. Also, I think % distortion at these levels is really hard to deal with and that dB are better. But as a quick look the 3 kHz harmonic in REW is shown at 0.00022%, which is -113 dBc, the THDN is 0.00046%, which is -106.7 dB. THDN is also SNR when the bandwidth are the same. Normally, SNR refers to a bandwidth all the way up to Nyquist (see first link below). But if SNR bandwidth is constrained, the SNR and THD are just the opposites of each other (eg -100 dB THDN is 100 dB SNR). And practically, with modern converters they are the same because there’s not much going on between the 20 kHz of the measurement limit and the 24 kHz that is Nyquist–all the action is below 10 kHz).
So, in the REW plot, your THDN measurement is mostly set by the noise and it is -106.7 dB, but if the 3rd harmonic were gone, I think the THDN would be -108.
In the QA40x plot, the 3 kHz is about -117 dBc. The THDN is reported at -105.62, which is considerably better than the -117, so we know the THDN measurement is dominated by the noise. We can also see the hum at 50 Hz is -112 dB or so, which might in fact be degrading the THDN by a dB (if the noise were at -107, and your hum was at -117, the RMS sum of those values would be -106.58 dB).
So, I think there are two things to observe:
First, the THDN looks very close (-106.7 versus -105.62), and second, the hum on the QA40x is probably degrading the measurement by a dB. For the hum, I’d look at other grounding options, or potentially orientation and/or proximity to nearby AC-powered devices. Does the 50 Hz change if the put the QA40x on end? Upside down? If you move it to the floor with longer cables? The aim there is just to see what might change it. That will tell you if it’s radiated or conducted.
In loopback, the THDN of the QA403 should be very close to what is shown on the page 16 of the product spec (link 2). Note in the lower part of that graph you can see the note "Autoset 2 dB adder, Hann, 32k. So, to repro, use Hann with 32k FFT. But the autoset 2 dB sometimes isn’t clear. That means you want your full scale input to be the nearest input that is at least 2 dB higher than your incoming signal. So, if your incoming signal is 2 dBV, you’d want a full scale input of 4 dBV, but since that isn’t an option, you’d pick 6 dBV.
In link 3, I made some measurements on a Topping D10B (balanced) which you might find interesting too. It’s challenging measuring at these levels for sure. In the link 3, I used a balanced notch to (I hope) confidently determine the DAC limit around -117 dB THDN.