Most Efficient Way to Smooth Wavelet Spectral Matrix over Time and Frequency

I want to smooth a 4 dimensional matrix (channel i x channel j x frequency x time) containing wavelet coefficients across time and frequency for each channel ij pair.

Dimenions of S:

size(S)

ans =

           3           3          67        4501

My solution:

% Smoothing over time
k = 50;
for chani = 1:size(S,1)
    for chanj = 1:size(S,2)
        for freqi = 1:size(S,3)
            temporalsmoothedS(chani,chanj,freqi,:) = movmean(squeeze(S(chani,chanj,freqi,:)),k);
        end
    end
end
clear k chani chanj freqi

% Smoothing over frequency
k = 3;
for chani = 1:size(S,1)
    for chanj = 1:size(S,2)
        for timei = 1:size(S,4)
            doublesmoothedS(chani,chanj,:,timei) = movmean(squeeze(S(chani,chanj,:,timei)),k);
        end
    end
end
clear k chani chanj timei temporalsmoothedS

Is there a way to do this better? Perhaps do time and frequency at the same time? would conv2 work? Do I absolutely need all the loops? Thanks.

Answer 1

[William Rose]

@Anas Khan, I think your approach is good. It has the virtues of being readable and easy to follow. You do not need

clear k chani chanj freqi

since k, chani, chanj will all be assigned new values in a line or two, and it is not necessary to remove freqi from memory.

You also don't need

clear k chani chanj timei temporalsmoothedS

for similar reasons. And in the case of temporalsmoothedS, you have computed it and then erased it without ever using it for anything - like plotting, saving to a file, etc.