Hi
We are using the 16x32 fft from the DSP lib and are scaling the input to Q23 in order to prevent overflow of the 256 pnt FFT output. I noticed that there is a cfft_NOSCALE() function for the C55x and I'm wondering if there is something similar for the C64x. Our application requires more dynamic range than Q23 and we don't mind if the FFT spectra are clipped. Basically, we'd like the DSP_fft16x32() to saturate on additions when it correlates with the twiddle factors.
I understand that this would be much slower, but we can afford the MIPs.
Cheers2u