[hpsdr] question on using FFTs in SDR

[Alberto I2PHD]

Greg Overkamp wrote:
> 
> I have a couple questions on how the FFT is used in
> software defined radios to do filtering.
> 
> 1) Is the time-domain data input to the FFT, and the
> time domain output of IFFT, overlapped in time? If so,
> what is the extent of the overlap?
> 
> 2) Is a window function applied to the time domain
> data before the FFT?
> 

  Greg,

   in addition to the excellent reference given by Bob, I can tell you as an example what I do in my program Winrad.
Forgetting the spectrum and waterfall displays, which run on different threads and different flow of data, for the audio 
processing I have input buffers of 512 samples, which I queue into a buffer of 2048 samples, which is the input to an 
FFT. The resulting frequency-domain buffer is multiplied by another 2048 sample buffer, obtained by doing an FFT of a 
1537 point kernel (padded to 2048 with zeros) which represent the bandpass filter that I want to apply. Before doing the 
FFT of the kernel, it is windowed with a modified Blackmann window. No window is applied to the audio data, as there is 
no need for it.

After the multiplication, I do an inverse FFT and take just the rightmost 512 samples of the resulting time series. This 
is what in the literature is called "overlap-and-save".

73  Alberto  I2PHD


 1173894675.0