You have to keep in mind:
1.) xcorr is returning negative lags as well as positive, but for real-valued inputs the cross-correlation sequence is even
2.) you can return the lags argument
Create an example with a lag of 10 between the two vectors.
x = randn(100,1);
y = [zeros(10,1); x];
% xcorr returns 2*length of the longest vector-1 lags by default
% therefore lag 110 is the zero-th lag
[xc,lags] = xcorr(y,x);
xc = xc(110:end);
lags = lags(110:end);
[val,index] = max(abs(xc));
lags(index)
The above gives a lag of 10, which is the correct lag.
Also see the examples and how tos for estimating delay on this page:
< < https://www.mathworks.com/help/signal/convolution-and-correlation.html>
