Yet More FFT Stuff

Using a variable bandwidth filter (constant % of frequency) and scanning it from low to high frequency should be able to pull any repetitive signals out of noise. I used this in getting a reconstructed HADCRUT3 waveform out of the sum of cosines.

http://climateandstuff.blogspot.co.uk/2011/05/is-it-trend.html


But just how well does it compare to a FFT of the same data.

FFT plot from spice programme (Blackmann-Harris window)

The bandpass plot is based on 4000 points and the FFT on 4096.
All 5 frequencies were found in both Band pass and FFT forms.The FFTs show correct amplitudes but the filtering shows a drop off with frequency

Next a comparison on hadcrut4 NH data 1850 to 2011 data length 1930 months

The interpretation of the band pass filtered for is easier but is it more reliable? The reported frequency needs an offset adjustment

This test at least proves that the bandpass plugin works!

Bandpass plugin is available from Web:reg:
http://www.web-reg.de/bp_addin.html

Spice programme:
http://www.linear.com/designtools/software/#LTspice


This site has other useful goodies!