Re: [Discuss-gnuradio] new peakfinding and averaging wxgui fft code (and dummy block has huge impact on usability fir filter)

[James Cooley]

Speaking of averaging and peak-detection, I'm interested in doing so 
outside of the gui.

Could you build an averager like this, with fir filters applied to the 
fft of the signal? Would it work? Is there a better way?

-take fft of signal (much as fft_sink does)

-Run fft output through an averager to smooth it out a bit:
FIR filter with taps [1/N,1/N,1/N,1/N,1/N,1/N...] up until N number of 
samples you want to average over. (A running average FIR filter)
averager = gr.fir_filter_fff(1, taps)

-run filter to find some peaks. Can this be done by constructing a FIR 
with taps [1,-1] and then noting where the 0 crossings take place? These 
taps should be equivalent to:
y[n] = -1*x[n] + 1*x[n+1]
which is like the slope.... derivative in continuous functions.... over 
a pair of samples.

-probably you'd want to also use a threshold on the original fft signal 
to weed out insignificant peaks?

Thanks,

-jamie




Martin Dvh wrote:

> Eric I am Recompiling now (again;-).
> I don't know if I will be doing the testing tonight (It is kind of 
> late here in the Netherlands)
>
> I did however also finish my modified fft_sink in wxgui and put in on 
> my website.
> I added the (default) option to average over several fft frames 
> (default 100)
> This gives a much quiter spectrum
>
> The most interesting feature I also added is a peakfinding algorithm.
> The spectrumwindow can now find peaks and shows the exact frequency 
> and signal level of these peaks.
>
> def make_fft_sink_f (fg, parent, label, fft_size, 
> input_rate,show_peak_info=False,show_peak_markers=False,navg_frames=100,peak_searchwidth=-1): 
>
>     #show_peak_info (default=False): if set to true the fft_window 
> will show the frequency and level of every big peak found in the spectrum
>     #show_peak_markers (default=False):  if set to true the fft_window 
> will show a marker at every big peak found in the spectrum
>     #navg_frames (default=100): number of fft frames to average before 
> showing it. If set to 100 or higher you get a much quiter spectrum 
> (less noisy and not changing very much every frame)
>     #peak_search_width(-1 gets you the default of 10 for a fft_size of 
> 1024): number of neigbours the peakfinding algorithm will look at for 
> determing if something is a peak.
>     #                                         A higher value means you 
> get less peaks detected
>     #Default the peakfinding algorithm determines a minimum 
> signalstrength threshold below which no peaks are detected
>     # you can override this with set_autopeakthreshold(False) and 
> set_peakthreshold(mythreshold) to use your own threshold
>     # the higher the threshold, the less peaks found
>
>
> I modified Makefile.am,fftsink.py and added peakfind.py
> You can all find them on:
> http://www.olifantasia.com/projects/gnuradio/mdvh/wxgui/
>
> I ported peakfind.py fromjava code from:
> #ported from  YALE - Yet Another Learning Environment
> #  Copyright (C) 2001-2004
> #      Simon Fischer, Ralf Klinkenberg, Ingo Mierswa,
> #          Katharina Morik, Oliver Ritthoff
> #      Artificial Intelligence Unit
> #      Computer Science Department
> #      University of Dortmund
> #      44221 Dortmund,  Germany
> #  email: address@hidden
> #  web:   http://yale.cs.uni-dortmund.de/java
> Which is GPL
>
> The files need to be put into:
> gr-wxgui/src/python/gnuradio/wxgui
>
> I also tried to make a diff against cvs.
> It is at 
> http://www.olifantasia.com/projects/gnuradio/mdvh/wxgui/fft_avg_and_peakfind.diff 
>
> But cvs diff does not seem to add any new local files. So you have to 
> still add peakfind.py manually.
>
> How do I tell cvs diff to include local new files ???
>
> greetings,
> Martin
>
>
>
> _______________________________________________
> Discuss-gnuradio mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/discuss-gnuradio