[hpsdr] Morse transcription project

[Frank Brickle]

On Sun, Dec 5, 2010 at 11:35 AM, Rob Frohne <frohro at wallawalla.edu> wrote:

> ...Some things are easier to see or compute in each domain, and if you don't look in both coordinate systems, you will surely be missing an insight, or doing something inefficiently.

To some extent it's Practical DSP 101, basically. The overhead of
((FFT or IFFT) + multiplication) is often less than a single
convolution, so you usually win by taking the signal to whichever side
(time or freq) lets you multiply rather than convolve. The only way
you can win bigger is by eliminating the operation that would require
convolution altogether. (That's why, at least conceptually, using IIR
rather than FIR filters in multirate systems seems like such a
promising idea.)

A good example would be enhanced spectral resolution -- a filtering
effect that would require convolution on the frequency side -- by
windowing (multiplying) data in the time domain. This shows up in all
sorts of places such as FFT filter banks, spectral line estimation,
etc.

73
Frank
AB2KT

-- 
Pourtant, j'étais fort mauvais poète.
Je ne savais pas aller jusqu'au bout.
-- Blaise Cendrars, "Prose du Transsibérien et de la petite Jehanne de France"

 1291572313.0