hi, i am implementing a simple program on the dsk 5402, which involves a fourier transform. Reading the application notes i have found out that there are some functions in the dsplib that do the hard work ('rfft256' and 'cbrev' for example). When i have tested them and compared the results with the matlab outputs, i have found out that the results are WRONG: the real part is quite the same (differences of 2-3 can be explained by the rounding error) but in the imaginary part, there is a block of indexes in the middle of the output buffer that are completely wrong (while the first indexes, and the last ones obviously, are o.k.). I found that in any length of the fft there is a block in the middle of the imaginary part which is wrong, and its size increases as the fft length goes larger. i must emphasize that i have been careful in the alignment needed, and i used cbrev as specified in the notes. Did any one met the same problems ? Where can i find any updated versions of the dsplib ? Thanks, Raffaele DaTo Italy. |
problems with fft routines in the dsplib of the dsk 5402
Started by ●June 3, 2002