The loop is not needed at all, filter can operate on the column of whole matrices at once. However, since you're working on rows, you'll have to transpose your matrix (and transpose the result):
mean_power_50 = filter(ones(1,50)/50,1, pote.').'; %work on the whole array
power_fading = pote - mean_power_50;
You can then split it into subarrays. However do not use numbered variables. Instead put these subarrays altogether into a cell array or other container:
%assuming that power_fading is 5000 columns:
x = mat2cell(power_fading, size(power_fading, 1), [1000 1001 1001 1001 997]); %any reason why it's not 1000 for each submatrix?
