0

Recently I used two measurement systems to record a quantity. I made sure both use the same sampling frequency. One system was recording continuously while the other one was operated to only record certain events. Now I am trying to use cross-correlation to synchronize the measurements between the two measurment systems. Because of noise in the systems I don't expect to find 100% accurate fits. However until now I fail completely to fit the data by this method.

Here is an example:

  • 10001 datapoints from the continuously operated recording x
  • 640 datapoints from the triggered system y

I try to find the position of the y datapoints within x using

toZero=mean(x);
x=x-toZero;
y=y-toZero;
[c,lags] = xcorr(x,y);

subplot(3,1,1)
plot(lags,c);


[a,b]=max(c);


c1=diff(c);
[a1,b1]=max(c1);


subplot(3,1,2)
plot(lags(2:end),c1);


subplot(3,1,3)
plot(tx,x)
axis tight
grid on
hold on
plot(b-length(tx)+ty-ty(1)+1,y)
plot(b1-length(tx)+ty-ty(1)+2,y)
legend('Base','Max','MaxDiff')

in matlab. So I make sure the mean of the datastream is zero and then calculate the cross-correlation of the two functions. Then I search for the maximum of the cross-correlation function and shift x by the position of the maximum.

As that didn't work in addition I also tried to calculate the derivatice of the cross-corelation. Then I use the maximum of that result to shift x by the position of that maximum.

Here is the result: Cross-Correlation and it's derivative of x and y using measured data

In the first diagramm is the cross-correlation of x and y with the maximum and the 3rd highest peak marked. In the second diagramm is the derivative of the first diagramm. In the third diagramm is the time data:

  • blue is x
  • orangs is y shifted by the maximum position of the cross correlation
  • yellow is y shifted by the maximum position of the derivative of the cross-correlation
  • purple is y shifted by hand to the correct position (~3rd highest peak from 1st diagramm)

So now I wonder why I don't find the correct position using cross-correlation? For some reason the highest peak from the cross-correlation function is a much worse fit than the 3rd highest. So I wonder why the 3rd highest peak is not the highest?

Using random input data instead of my measured data however my code seems to work perfectly:

Cross-Correlation and it's derivative of x and y using random data

So, what is the problem with my recorded data then?

Here is the code for the 2nd example with the random data for x and y:

tx=1:1000;
x=randn(1,1000);
toZero=mean(x);
x=x-toZero;
y=x(100:150)-toZero;
ty=[1:length(y)];

[c,lags] = xcorr(x,y);


subplot(3,1,1)
title('Cross-Correlation')
plot(lags,c);


[a,b]=max(c);


c1=diff(c);
[a1,b1]=max(c1);


subplot(3,1,2)
title('Derivative of Cross-Correlation')
plot(lags(2:end),c1);


subplot(3,1,3)
title('Time Signals')
plot(tx,x)
axis tight
grid on
hold on
plot(b-length(tx)+ty-ty(1)+1,y,'linewidth',2)
plot(b1-length(tx)+ty-ty(1)+2,y)
legend('Base','Max','MaxDiff')
xlabel('Samples')
Matthias La
  • 380
  • 2
  • 9
  • looking at my cross-correlation function it looks like the amplitude gets lower and lower to higher sample numbers. Therefor the peak at the best fit is reduced and gets lower as the first peak which is at a lower sample number. But why is that? – Matthias La Nov 10 '21 at 10:52

1 Answers1

1

According to the answer from SergV to this post normalized cross-correlation is the way to go. That way the result is not influeced by the amplitudes of the given segments. I tried it out using his code and it worked great:

Using normalized cross-correlation to find correct position of time segment in datastream

Matthias La
  • 380
  • 2
  • 9