This looks like the sort of correlation problem that is best done in the
Fourier domain - particularly so if the reference spectra are contant.
Then you get the entire correlation at every shift with a cost 2NlogN
transform and some book keeping where N is the length of your array.

Regards,
Martin Brown
.


Martin, can you explain further what you mean when you say it should be in
the Fourier domain? Can you give me some references?

Thanks,
Matt