I have a clue: make the gain of the ADC inversely proportional to the amplitude of the signal. That is, a more confident measurement is taken when the interference signal is lowest, and the least confident measurement when the interfering signal is high. Also, add artificial noise when the interfering signal is high, to swamp it and make it disappear from the analysis. Finally, use instead of the Fourier transform, the "most orthogonal" function to the interfering signal (which is smooth, it's not noise-like).
Does that go by any name? How do I find the orthogonal function?
Another strategy: Sample the signal only at the confident measurement area, then suppose that this yields half the frequency of the signal.
Multiplying the non-confident sampling area by (-1)^n (analogically) makes it possible to turn the non-confident sampling interval into a confident sampling interval.
I'm also interested in knowing what would the best ADC/op amp for this. I'm new to the switched capacitor, sigma-delta, or any other architecture.