Hello all, This has stirred up more interest (also off-list) than I ever imagined it would ! OK, beers will go to (in no particular order) Niels: his comments on ADC/DAC reference voltages provide the key to understanding what happened. Veronica: for the most to-the-point one-liner. Florian: (off list) who got it right in the end. I guess that at least initially all of you put the blame of card X - either it's crap or it's defective. But the real culprit is card A, even if it did show quite well in the loopback test. And what happens is 100 Hz (rectified power grid 50 Hz) ripple on the DAC and ADC reference voltages, both derived from the same source. Most audio ADCs and DACs have two reference voltages (or one, and then the second is derived internally). In many cases these are the voltages that correspond to the min and max PCM values, and even if that is not the case the conversion factor of the ADC or DAC depends linearly on these voltages. If they are not well filtered, any fluctuations do generate AM, and the effects of the same eror on a ADC and DAC will cancel out. Cancellation also occurs if the sample clock (if shared by ADC and DAC) is modulated, but here the effect is delay modulation, which for a single tone translates into phase modulation proportional to the frequency of the signal. Since in this case the sideband levels do not depend on the measurement frequency, the sidebands are not the result of clock jitter. So Veronica is quite right: you can't trust a loopback test. And if, as in this case, the variations are the result of mains hum, you can't even trust a setup of two cards using the same mains power (although exact cancellation as in this case would be rare). It is also true that you have to be very careful when using a low-fi card to test a much better one. In this case the only purpose of the USB card was to provide a calibrated analog level, which is required for e.g. EIN measurement. The card is portable so I can easily take it to the lab and calibrate it against a high precision certified RMS meter which I can't take out without first filling out a hundred forms and then waiting two weeks. Meanwhile I did measure the signal quality of card X using a *decent* generator - it is extremely good. Ciao, -- FA O tu, che porte, correndo si ? E guerra e morte !