We are trying to get 1/4 scaling at each stage in DSP_fft16x16 function of C64x+ DSPLIB library. For example FFT size of 1024. with an additional 1/4 scaling of the input we would be able to get a total scaling of 1/1024 which is the typical FFT with 1/N scaling [1/N*sum(x[n] exp(-jnk/N)) where N =1024]
Basically the user manual of DSPLIB functions says there is 1/2 scaling inherently built in each stage of DSP_fft16x16 function and to get additional 1/2 scaling at each stage, we need to scale down the twiddle factors by 2, but when we tried this, it wasn't giving the expected 1/N scaling i.e. 1/1024 scaling if N =1024, instead it was giving a scaling of only 1/64 (including a scaling of 1/4 of the input) which means only the inherent 1/2 scaling has been applied and not the twiddle scaling by 2. Can you please help us solve this issue. Thanks in advance.
