I'm running an FFT with mkl ComputeForward and sometimes, some of the bins are 0 + 0i.
Specifically, If I do a single, in place DFT of 20480 samples of a single 16K tone sampled at
1.024 MSps, all the bins have values except just one, which is zero. Curiously, the surrounding
bins seem ok.
...
-2.13270645e-13 -6.73710165e-13
-2.29536984e-13 -1.78759893e-12
0.00000000 0.00000000
-2.29536957e-13 1.78759893e-12
-2.13270550e-13 6.73710274e-13
-3.53691855e-14 2.86772993e-13
....
What can be causing this? I don't think it is a resolution problem because the failing bin should
fit in a float, I checked it doing the FFT with Octave.
