As a DSP newbie, I've stumbled into doing band pass filtering (of pre-recorded IQ data) in the frequency domain - which allows brick wall filtering (I'm just setting the IQ values of each consecutive FFT 'batch' outside my frequency band of interest to zero and it seems to work a treat for what I've done so far). Doing filtering, e.g. a FIR, in the time domain seems to need many taps, a convolution function, and still not be as 'clean' as a brickwall filter. So, why/when is time domain frequency filtering the preferred way of doing it? Real time benefits? Works better with a low bandwidth SDR if subsequently transmitting? I should say that all my DSPing is being done in MatLab with no signal processing or other toolboxes - so I don't have access to FIR filter or convolution functions. If I want them, I have to write them or find them on-line somewhere - which I don't mind doing, if there's a good reason.
Asked
Active
Viewed 316 times
0
-
That's not brick-wall filtering. It's actually a sinc-shaped filterbank and has terrible sidelobes. You just sample the spectrum just so that you hit the zeros of the side lobes! – Marcus Müller Apr 07 '22 at 14:05
-
3Does this answer your question? Why is it a bad idea to filter by zeroing out FFT bins? – Marcus Müller Apr 07 '22 at 14:07
-
And as detailed further here as well: https://dsp.stackexchange.com/questions/31905/difference-between-frequency-sampling-and-windowing-method/31909#31909 – Dan Boschen Apr 08 '22 at 01:06