find and indexing fun (was Re: Efficient Octave Code)

[Joan Picanyol i Puig]

* Bill Denney <address@hidden> [20061021 16:31]:
> rockster8 wrote:
> > for i=1:50
> >    for j=1:50
> >       for k=1:10
> >          Fk(i,j) = F(i,j) .* (w(i,j) == k);
> >       end
> >    end
> > end
> >
> > where the following are initialized as:
> > Fk(1:50,1:50) = 0;
> > F(1:50,1:50) = 0;
> > w(1:50,1:50) = randomly generated numbers..
> >   
> Well, first off, since you're multiplying 0 (F) times a boolean, Fk will 
> always be zero, so
> 
> Fk = zeros (size (F));
> 
> would work.  Assuming that F weren't 0, you could do something like this:
> 
> for k = 1:10
>   mask = (w == k);
>   Fk(mask) = F(mask);
> endfor

How do you do that for more than one dimension? I have the following
piece of code, from which I don't know how to get the for loops out:

    % flatten out pcsi's 3rd dimension (input channels) by vertical
    % concatenation
    lcsi = zeros (ln_taps * ln_input_channels, ln_OFDM_symb);
    for li = 1:ln_input_channels
        lidx = 1:ln_taps;
        lidx = lidx + repmat (ln_taps * (li - 1), 1, ln_taps),
        lcsi (lidx,:) = pcsi(:,:,li);
    end
    lcsi,

    % flatten out ptaps' 2nd dimension (input channels) by vertical
    % concatenation
    ltaps = reshape(ptaps.', ln_echoes * ln_input_channels, 1);
    lchannels = repmat(ltaps, 1, ln_OFDM_symb),
    % apply CSI to channels
    % funny indexing caused by weird find behaviour
    lchannels(lchannels ~= 0) = lchannels(lchannels ~= 0) .* lcsi(find (lcsi)),

    lchannel = zeros (ln_echoes, ln_OFDM_symb);
    for li = 1:ln_input_channels
        lidx = 1:ln_echoes;
        lidx = lidx + repmat (ln_echoes * (li - 1), 1, ln_echoes);
        lchannel(:,:) = lchannel + lchannels(lidx,:);
    end
    lchannel = lchannel / ln_input_channels,

For those in communications, this should equalize an OFDM signal through
a MISO channel (but it doesn't).

tks
--
pica