Matrix indexing: fast way to compute covariance of N different MxL matrices that are saved in NxMxL matrix

I have a NxMxL matrix called bigmatrix That I can split into N temporary submatrices. On each temporary submatrix I need to calculate the covariance. I do this in the following way:

for k=1:N

    submatrix=squeeze(bigmatrix(k,:,:)).';
       R=submatrix*submatrix';
  end

The covariance is calculated really fast. My problem is loading the submatrix. This takes 6 seconds. Is there a faster way to do this? The matrix indexing is slowing me down a lot.

Thanks!

Answer 1

[Matt J]

Using MTIMESX ( Download ), we can eliminate the loop altogether,

   data=permute(bigmatrix,[2,3,1]);

   R=mtimesx(data,'t',data);

If you build bigmatrix as MxLxN in the first place, you can avoid the overhead of permute(). Even without mtimesx, I expect an MxLxN data organization will accelerate your loop

for k=1:N
     submatrix=data(:,:,k);
     R=submatrix.'*submatrix;
end