Why do "corr(x,y)" and "xcorr(x,y)" give different answers when x and y are simulated discrete samples of radar signals of form a*e^ib, with partially correlated randomized b values?

To calculate the correlation between discrete samples from two simulated radar signals (column vectors of equal length), there appear to be two approaches in Matlab. One is the function "corr(x,y)" in the statistics toolbox and the other is "xcorr(x,y)" in the signal processing toolbox. For a given set of signal samples x,y these two functions give different results and the reason for the difference is not apparent.

Answer 1

[Honglei Chen]

It would be helpful if you could provide a script showcasing the difference and explain why you think the two should be the same.

From the documentation, corr and xcorr are quite different. corr computes the correlation coefficients between the two sequences, while xcorr computes the correlation of one sequence with another sequence at different delays. So to me it's expected that the two results are different.

HTH

[Paul Smith]

My description of the issue was incomplete. I used xcorr(x,y,0,'coeff') which should give the correlation coefficient between the two singals at zero lag.

[Honglei Chen]

Thanks for the clarification. The difference is that corr removes the mean of the signal while xcorr does not. If you compare

corr(x,y)

and

xcorr(x-mean(x),y-mean(y),0,'coeff')

the two should be the same.

HTH

[Paul Smith]

Many thanks for the clarification!

PLS