how speed up or avoid loops

I think there must be ways to avoid for loops, but have not found the correct magic matlab syntax for doing so. Can someone suggest how I might make the following more efficient? -Jeff

% example.m
% how can I speed up inner loop(s) here?
% create random array of signals (Nchannel x Nframes)
% for each pair of signals compute difference wave
% for each wave, integrate over moving window Nwin wide
% will be flat epochs at beginning and end of integrated difference waves,
%   Hwin wide

nchan = 4;
nprs = nchan*(nchan-1)/2;
nfrm = 1000;
data = rand(nchan,nfrm)-0.5;
new = zeros(nprs,nfrm);

hwin = 10;
nwin = 2 * hwin + 1;
fnwin = single(nwin);

ipr = 0;
for ich = 1:nchan-1
    for jch = ich+1:nchan
        ipr = ipr + 1;
        dif = data(jch,:) - data(ich,:);
        rsum = 0;
        for ifr = hwin+1:nfrm-hwin
            sum = 0;
            for jfr = ifr-hwin:ifr+hwin
                sum = sum + dif(jfr);
            end
            new(ipr,ifr) = abs(sum/fnwin);
        end
    end
end

Answer 1

[Matt Fig]

This is much faster. Note that you are masking a very valuable MATLAB function by naming a variable 'sum' in your loop! Please stop doing this or you will end up spending an awful lot of time debugging why the SUM function is broken in your copy of MATLAB!

nchan = 4;
nprs = nchan*(nchan-1)/2;
nfrm = 1000;
data = rand(nchan,nfrm)-0.5;
new = zeros(nprs,nfrm);

hwin = 10;
nwin = 2 * hwin + 1;
fnwin = single(nwin);
ipr = 0;
idx = ones(nwin,nfrm-2*hwin);
idx(:,1) = 1:nwin;
idx = cumsum(idx,2);
for ich = 1:nchan-1
    for jch = ich+1:nchan
        ipr = ipr + 1;
        dif = data(jch,:) - data(ich,:);
        new(ipr,hwin+1:nfrm-hwin) = sum(dif(idx));
    end
end
new = abs(new)/fnwin;

[Jeff Eriksen]

a belated "Thak you", -Jeff