Hi Glenn, I had the exact same problem in a 16k design. I corrected it by changing the Max Coeff. Depth parameter to nstages + 1. In some discussions with everyone here about this, the conclusion seemed to be that there was a good reason for setting the default to 2^11. Best person to explain what that reason was is probably Aaron.
- Andrew On 10/20/08 1:27 AM, "G Jones" <[email protected]> wrote: > Hello, > I've made a spectrometer that uses a 2^11 point biplex (num inputs = 2^0) FFT > block, and it works great. I then changed it to 2^12 points and it looks good > except that when I put in a CW tone, I get the expected spike plus another > spike 2048 channels away. I've checked everything I can think of, so I started > digging into the blocks. The one thing that caught my mind is that each of the > fft_stage_n building blocks has the "Maximum Coefficient Depth" parameter set > to 2^11. I vaguely recall that this parameter allows conservation of resources > for large FFTs where the twiddle factors do not change by much, so you can > store fewer of them and reuse them. The parameter is set at 2^11 for all > stages, and it seems it would only be important at the 12th stage, which is > exactly the difference between my 2^11 point spectrometer and my 2^12 point > spectrometer. I assume any bugs in this with the green blocks would have gone > unnoticed because large FFTs previously crashed MATLAB. I will try rebuilding > the 2^11 point design with Maximum Coefficient Depth set to 2^10 to see if I > observe similar behavior. In the mean time, if you know how this parameter > should work, please explain. > Thanks, > Glenn >
