Thanks
MrAl for taking a shot, you was only one who dared to try
Looking at the number of needed operations FFT is always faster
FFT
Needs N/2*log
2(N/2) multiplications
N*log
2 N additions
DFT
N(N-1) multiplications
N*N additions
So in the best way DFT is as fast as FFT.
So the only drawback of FFT lies in the execusion of algoritm itself.
In DFT we may use paralell operations more efficiently, because result of next step does not depend on previous one. Superskalarity and floating point libraries can also be taken into account
In FFT it looks like that
**broken link removed**
-----------------------------------------------------------------
Today I had discussion as heated as I could have with the lecturer. And conclusions was that there have to be 2 conditions fulfilled to calculate DFT faster than FFT:
1) all summing operations have to be done on one clock circle (sic!)
2) CATCHE memory have to be multiport and make all operations during one clock cycle. The same memory cell cannot be written and read at the same time, so for one separate operation we need unused, new cell because all operations have to be done during one clock cycle. Theoretically we may create as big memory as we wish, so propability of overlapping is really small.