Re: How do YOU handle this disparity in fft?

[Brian Kaczynski]

> Regarding the DC term in the FFT --

I've always thought that any linear trend in the data was supposed to
be removed (least squares, least median, ...) before the FFT was done.
In which case any DC component is going to be small.

One comment about removing DC or linear trend -- If you use a window, you want to make sure the trend is removed AFTER the window is applied, not before.  This is not always trivial though... Supposing w[n] is your window and x[n] is your signal, to ensure DC is removed AFTER windowing you would do this:

x_DC = sum(w.*x)/sum(w);

x_windowed = w.*(x - x_DC);

You can prove this pretty easily by writing the sums out.

I haven't bothered to derive the procedure for removing linear trend with windowing -- perhaps someone else can do that and enlighten us!

-Brian

 

If you are doing FFT in real time, it takes a bit of planning to remove
a linear trend in real time.

Gord

_______________________________________________
Help-octave mailing list
address@hidden
https://mailman.cae.wisc.edu/listinfo/help-octave