Matlab matched filtering help

Hello

Here is a code that I wrote to analyze the prp(pulse repetition pattern) of a signal I am getting from an oscilloscope.

      [signal_1,path_1] = uigetfile('*.csv','Select the .csv file to analyze');
  if isequal(signal_1,0)
     disp('User selected Cancel');
  end
  sig = csvread(fullfile(path_1, signal_1),0,3);

amp_1 = sig(1:2000000,2);
ts_1 = sig(2,1) - sig(1,1);
time_1 = ([0:2000000 - 1]*ts_1)';
ax1 = plot(time_1,amp_1);grid;
axis([0 max(time_1) -0.1 1.8]);
ax_handle = gca;
ii = 1;
fig_handle = gcf;

xlimit_orig = get(ax_handle, 'Xlim');
  win_xlimit = [0 0.001];
  offset = 0.001;
 starter = uicontrol('parent',fig_handle,'Style','ToggleButton','String','start pan','Value',0,'Position',[150, 5, 60, 20]);
 handle = uicontrol('parent',fig_handle,'Style','ToggleButton','String','pause/run','Value',1,'Position',[100, 5, 60, 20]);
 getvalue = uicontrol('parent',fig_handle,'Style','ToggleButton','String','get value','Value',1,'Position',[20, 5, 60, 20]);
 %Iterativley change the xlimit for the axes to pan across the figure

       while win_xlimit(1) <= xlimit_orig(2) && ii <= 9
           pause(5);
           if get(starter,'Value') == 1
           if get(handle,'Value') == 1
               set(ax_handle, 'Xlim', win_xlimit);
               win_xlimit = win_xlimit + offset;
           end   

           end
           if get(getvalue,'Value') == 0
                   [x(ii),y(ii)] = ginput(1);

                   ii = ii + 1;
                   set(getvalue,'Value',1);
           end

       end

   prp = (x(2:ii-1)-x(1:ii-2))*1e+06;
   disp('approximate prp is');
   disp(prp);

The signal is huge (as you see I am reading in 2 million samples of my raw data). I am panning the signal in x-axis to find the pulses of interest and then conditionally use the ginput command to get values when needed. The problem is that ginput requires manual interference. I am trying to see if I can avoid this by implementing a matched filter that will retain the pulses and remove the noise. Can anybody please help on how to start designing a matched filter for a signal like this?

Thanks

Answer 1

[Daniel Shub]

Depending on how noisy your data is you might just be able to use a simple threshold ...

time_1(amp_1 > threshold)

where you set an appropriate threshold.

If your data is noisy, then you probably want to do a crosscorrelation:

[y, lags] = xcorr(amp_1, template);
lags(y > threshold)

where you pick the appropriate threshold and template.

Answer 2

[Honglei Chen]

Hi vsee

Here is an earlier post regarding matched filter

http://www.mathworks.com/matlabcentral/answers/4502-matched-filter

Since you have a huge data set, it may be better for you to process them in a streaming way. If you have Phased Array System Toolbox, you can take a look at

>> doc phased.MatchedFilter

HTH

[vsee]

Thanks for the reply.

sig_1 is my original signal, sig_2 is the flipped signal. I used the flipud since sig_1 is a column vector and I plotted them.

http://www.yorkphoto.com/york/slideshow/AlbumID=4173891002/PictureID=249921602002/a=100160453_100160453/

If you can see there are pulses and noise. How do I use the filter command for extracting my signal?

[Honglei Chen]

If you want to perform the filtering, you can do y=filter(sig_2,1,sig_1). However, I'm not quite sure what exactly you want to achieve. As it is, the signal is already quite distinguishable from the noise. If you just want to remove the high frequency noise, then matched filter may not be the tool you want to use.

[vsee]

In other words, if

y = filter(sig_2, a, sig_1);

How do I select a?

Thanks

[vsee]

True, the signal is distinguishable from the noise. However, I am only plotting 10000 points of the full signal. The full signal has multiple pulses staggered all over. What I am trying to achieve is an algorithm that can retain the pulses and get rid of the noise completely so I can calculate the prp.

Thanks again

[Honglei Chen]

In this particular case, because you are reverting signal, it is an FIR filter so a is always 1.

I don't think there is an algorithm exist to retain the signal while completely get rid of the noise. There will always be some tradeoff between losing some signal vs. removing noise. If your goal is to make the pulse look clean, what you can do is to pass the pulse through a lowpass filter to get rid of some noise. However, your pulses then will not look like perfect rectangles.