Here is a close-up of the spectrum of the 1.952819824 MHz signal at bin number "3199.5":

This is the familiar "Christmas Tree" shape that occurs in DFT spectra around sines which fall somewhere between two bins. In this case, the skirts are symmetrical since it falls exactly in the middle of two bins, but the otherwise the skirt will be taller on the side of the bin closest to the actual frequency.

The Moral of The Story: Data Windows have their place, but they have a significant downside: it's hard to see the trees for the forest (not to mention the grass!) If you possibly can, test the A/D in your system using the Magic Frequency of your choice rather than with an arbitrary frequency and a Data Window. Having the right relative frequency of the input signal relative to the clock frequency is the key, NOT having the exact absolute frequency. Therefore, you must phase-lock the frequency reference of your clock generator with the frequency reference of your signal generator. Otherwise you are unlikely to hit the exact center of a DFT bin.

This isn't "cheating"--it's simply a matter of not planting extra trees that obscure your forest!