Jan-Paul @JP-Huijser, Matt @matt , and anybody else knowledgeable, I would appreciate your input into whether the approach described below makes any sense, or what I might be overlooking.
What I’m thinking of doing is using the harmonic visualizer to extract the harmonic amplitudes at discrete frequencies, and then subtract the reference channel amplitudes to arrive at the amplifier contribution. Is there something fundamentally (or subtly) wrong with this approach?
I looped the second channel of my small audio transformer into the QA403 right channel single ended loopback reference setup, and dialed in the gain of the amplifier in the left channel loop so that both channels see about a gain of 1 and read the same signal amplitude at 1kHz and 100 mV QA403 generator output (and, within errors, at all test frequencies from 100 Hz to 8 kHz.)
Here is an example of the output at 8 kHz; left channel (yellow) is QA403 single ended output feeding the amplifier, speaker output into 8 Ohm, and then through one channel of the small audio transformer single-ended into the QA403. The right channel is QA403 single ended - other transformer channel - single ended into QA403. There is no additional low pass filter in the loops.
So I’d take the amplifier contribution as the difference between the blue and red amplitudes, which in this case is easy, because the blue ones are so much larger (e.g. -67.4 dBc - 91.5dBc = -67.96dBc, so pretty much the same as the first number).
Up to 1 kHz there is hardly any difference at all, and at 5 kHz there is some. I defined, or tried to define, the measurement bandwidth to go up to 48 kHz, I ran the QA403 at 192kHz, so at 5kHz H9 should still be in the band, and at 8kHz H6, if the 48 kHz limit is really applied. I note that the visualizer bar graph quotes higher harmonics than H6 at 8 kHz, but maybe the bandwidth setting only affects the THD sum, or maybe only the noise integration. Which is why I tagged @matt .
Does any of this makes sense, or is the approach invalid, for example because too much switching energy is leaking or is aliased into these (seemingly sharp, though) harmonic lines? (I know I can’t measure the noise this way, evident from the yellow spectrum rising from 20 kHz to 100 kHz.)
I inadvertently ran the same measurement series (100Hz to 8kHz) with the 20 microHenry inductors in place, and interestingly the inductors seem to mostly increase the lower harmonics H2 and especially H3, and decrease the higher ones:
Puzzling, or maybe a lucky coincidence, is that the THD numbers (sums? How are they computed?) are almost the same with and without inductors, even though the individual harmonic amplitudes are quite different.
H2 +11, H3 +28(!), H4 ~same, H5 +9, H6 -8(!), H7 ~same, H8 -8, H9 -7, H10 (80kHz) -6.
.
