I have some signal data from my lab that I want to present but I want to be able to say whether the signals I have are significantly different or not. I thought I could use cross-correlation. I haven't seen cross-correlation, convolution, or DFFT, etc. since the university course I took on signal processing. I still have the Python and Matlab programs that will perform a cross correlation on .txt files. What I want to know is what should I be looking for if the signals are 'pretty close' or not significantly different under a Normal distribution - a peak at the centre? The thing is my data now doesn't depend on time - it is UV-vis absorption spectra of chemicals that I want to compare based on their wavelength. Is the y-axis the correlation coefficient between 0 and 1? And is it OK to replace the time-lag in seconds with say, wavelength?


Difficult to say unless you specify what your software calculates and how the ccf is normalised. Usually you would design it so a perfect match gave a ccf of unity at zero lag.

In general you expect to get a big peak in the centre with a FWHM that is roughly the sum of the spectral resolutions of your individual spectra.

Using wavelength on the x-axis is quite standard - e.g. Cross correlation as a means of measuring the redshift of galaxies, using a similar galaxy spectrum at rest as a template.

The x-axis of your CCF in such cases is a wavelength difference.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.