[edit: Add comment about circular correlation; correct typos.]
You need to use a circular correlation, because that treats each signal as if it wraps around on itself. Unfortunately, xcorr() does not have a 'circular' option. Therefore try the following to get the circular correlation function:
Use the inverse FFT of the point-by-point product of the FFT and the conjugate FFT of the signals.
circcorr_ab = ifft(fft(a).*conj(fft(b)));
Here is your code using the function above.
sig1 = zeros(1,100);
sig1(71:90) = 1;
sig2 = zeros(1,100);
sig2([21:40]) = 1;
[xc1,l1] = xcorr(sig1,sig2);
[mx1,ix1] = max(xc1);
subplot(311), plot(1:100,sig1,'.r-');hold on;grid on;
plot((1:100)+l1(ix1),sig2,'xb-');hold on;grid on;
% Make a new sig2
sig2 = zeros(1,100);
sig2([1:10 91:100]) = 1;
% Compute circular cross correlation with inverse FFT
xc2 = ifft(fft(sig1).*conj(fft(sig2)));
l2=-length(sig1):length(sig1);
% Find max of circulation cross correlation
[mx2,ix2] = max(xc2);
% Plot shifted signal
subplot(312)
plot(1:100,sig1,'.r-');hold on;grid on;
plot((1:100)+l2(ix2),sig2,'xb-');hold on;grid on;
% Middle plot looks incomplete
% Extend sig2, then plot sig1 and the extended sig2
sig2ext=[sig2,sig2(1:abs(l2(ix2)))];
% Plot sig1 and extended sig2
subplot(313)
plot(1:100,sig1,'.r-');hold on;grid on;
plot((1:(100+abs(l2(ix2))))+l2(ix2),sig2ext,'xb-');hold on;grid on;
The middle plot looks incomplete, so I extended sig2, then plotted sig1 and extended sig2.
