If I have a signal that I'd like to filter very precisely (say, a few Hz bandwidth), I feel like the "best" one can possibly do is to take a time sample long enough to get a discrete transform frequency resolution of the same width, erase all but the frequency component of interest, and then transform back. But am I supposing the Fourier transform is more fundamental than it really is? Could I use, say, a recursive filter or other set of basis functions, to have a few Hz passband over timescales much shorter than the transform-limit? Or is the (Fourier) transform limit really a fundamental limit? Thanks!