Closed tknopp closed 2 years ago
Usually, it gives contains the same values in each dimension, except for the case that the oversampling factor sigma
(and thus the kernel shape parameter b
) is different for each dimension, see the definition of PHI
https://github.com/NFFT/nfft/blob/develop/include/infft.h#L209
This seems to be a very rare case, though.
Ah, yes, good point. Missed that use case.
Hi,
probably, I am missing something but IMHO the look up table calculated here
https://github.com/NFFT/nfft/blob/develop/kernel/nfft/nfft.c#L5776
is redundant, i.e. the same for all dimensions. This is because the argument is multiplied with
n[t]
which cancels out inPHI
.I just tried this out over at NFFT.jl and it just worked. Here is the important piece: https://github.com/JuliaMath/NFFT.jl/blob/master/src/precomputation.jl#L199 Compared to the previous implementation I removed the
n[t]
fromstep
and callPHI
withn=1
.Cheers,
Tobi