Hi all On 30 March 2011 16:14, Jack Hickish <[email protected]> wrote: > Hi Casperites, > > After a long day of debugging an FFT spectrum yesterday, I got down to > basics and compared the output of the 8-pt direct fft block with matlab's > fft function. When I used random numbers as the inputs it quickly became > apparant that they didn't match. > > After a little more investigation, it turns out that the twiddle factors > used in the FFT all fall in the imag(coeff) <= 0 range of the unit-circle, > apart from the twiddle_coeff_1 block, which "multiplies" by the coefficient > +j. Hence, it would appear, that an fft which uses a mixture of the special > case twiddle_coeff_1 block and twiddle_general blocks will give an incorrect > result. > > After switching the twiddle_coeff_1 block to multiply by -j the direct fft > seems to work correctly (at least for the 8pt and 16pt cases that I've > tested) > > I'm not sure when (if ever) the twiddle_coeff_1 block is used other than a > direct fft outside of one of the larger wideband real blocks, but here's a > heads up to the anyone using the the block in this way like I am. > > Cheers, > > Jack
Fixed and in the kat git fork on casper. Regards Andrew
