[hpsdr] DSP theory, interpreting IQ streams as audio, and way too much math - is this a good approach?

[Alberto I2PHD]

On 6/23/2012 7:01 AM, Chris Albertson wrote:
> So you have decided NOT to use an FFT?    You can save a lot of CPU
> cycles if you do dump the data stream into FFTW.
>
> The reason is because there is this neat trick, if you take the FFT,
> multiply that by an implies function and then take the inverse FFT you
> have just applied any number of convolutions.  This pays off in a big
> way if you want to do more with the data then just detect one USB
> signal.
That is correct, but do not forget that using FFTs is like applying a "circular"convolution to the signal,
while what you really want is a "linear" convolution.  So, before taking the inverse FFT, you must apply
either the "overlap and save" or the "overlap and discard" algorithm.

They are quite well described in the Openheimer & Shafer well known textbook on Digital Signal Processing.

-- 
/*73 Alberto I2PHD*/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openhpsdr.org/pipermail/hpsdr-openhpsdr.org/attachments/20120624/1d7fffce/attachment-0004.htm>