How do I test for correlation between digital data series?

조회 수: 10 (최근 30일)
Chris
Chris 2013년 12월 4일
편집: Wayne King 2013년 12월 5일
This may be a dumb question. I have two digital data series:
a = [-1 1 -1 -1 -1 1 1 -1 -1 -1 1 1]; b = [1 -1 1 -1 1 1 -1 -1 -1 -1 1 -1];
What's the best way to determine whether they are correlated? Corr? Chi-square? If chi-square, how exactly (there are several functions)?
Thanks

채택된 답변

Wayne King
Wayne King 2013년 12월 5일
편집: Wayne King 2013년 12월 5일
xcorr() is in the Signal Processing Toolbox. You can calculate the cross correlation in the frequency domain and then invert.
a = [-1 1 -1 -1 -1 1 1 -1 -1 -1 1 1];
b = [1 -1 1 -1 1 1 -1 -1 -1 -1 1 -1];
npad = length(a)+length(b)-1;
crosspec = fft(a,npad).*conj(fft(b,npad));
xcr = fftshift(ifft(crosspec));
anorm = norm(a,2)^2;
bnorm = norm(b,2)^2;
xcr = xcr./sqrt(anorm*bnorm);
lags = -length(a)+1:length(a)-1;
stem(lags,xcr,'markerfacecolor',[0 0 1])
Or you can use corrcoef() -- again this is not the best for time series.
R = corrcoef(a,b);

추가 답변 (1개)

Wayne King
Wayne King 2013년 12월 4일
편집: Wayne King 2013년 12월 4일
The best way is to compute the cross correlation sequence.
The reason you want to do that is with time series data, you want to account for the fact that the two series may differ only by a shift.
a = [-1 1 -1 -1 -1 1 1 -1 -1 -1 1 1]; b = [1 -1 1 -1 1 1 -1 -1 -1 -1 1 -1];
[c,lags] = xcorr(a,b,'coeff');
stem(lags,c)
Look at some of the "Examples and How Tos" here
  댓글 수: 1
Chris
Chris 2013년 12월 4일
I get an error when I run that: "Undefined function 'xcorr' for input arguments of type 'char'." Still get an error even if I convert a & b to logicals.
Also, in my case there is no concern about a time shift, does that mean I can use some other method to test for correlation?

댓글을 달려면 로그인하십시오.

카테고리

Help CenterFile Exchange에서 Correlation and Convolution에 대해 자세히 알아보기

태그

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by