function m_full = getFullMask(m,frame_d,delay,kernel)
%GETFULLMASK Convert frame rate mask to a sample-by-sample mask
% 
%   M_FULL = IOSR.BSS.GETFULLMASK(M,FRAME_D) expands the time-frequency
%   mask M, which has one unit for each frequency channel and frame of
%   length FRAME_D (in samples), to a sample-by-sample mask. The mask M is
%   a time-frequency mask, with one column for each frequency channel, and
%   one row for each time frame. The resulting mask will have dimensions
%   [FRAME_D*size(M,1) size(M,2)].
% 
%   M_FULL = IOSR.BSS.GETFULLMASK(M,FRAME_D,DELAY) removes a delay from
%   each frequency channel in the mask. DELAY is a vector, with the same
%   number of elements as M has columns, containing a delay (in samples)
%   that is removed from the corresponding frequency channel. The mask is
%   subsequently zero-padded.
% 
%   M_FULL = IOSR.BSS.GETFULLMASK(M,FRAME_D,DELAY,KERNEL) allows smoothing
%   to be applied to the full mask. By default, the full mask contains
%   rectangular transitions at unit boundaries. Specifying KERNEL allows
%   the transitions to be smoother, by convolving the full mask with a
%   two-dimensional kernel (dimensions [frequency time]). The central part
%   of the convolution is returned, so the centre of the KERNEL should be
%   in the centre of the matrix. The kernel is normalised in order to
%   ensure zero gain at DC.
% 
%   See also IOSR.BSS.RESYNTHESISE.

%   Copyright 2016 University of Surrey.


    if nargin < 2
        error('iosr:getFullMask:nargin','Not enough input arguments')
    end

    numchans = size(m,2);
    frameCount = size(m,1);

    if nargin < 3
        delay = zeros(1,numchans);
    end
    if nargin < 4
        kernel = 1;
    end

    % Create the sample-by-sample mask
    m_full = zeros(frameCount*frame_d,numchans);
    for i = 1:numchans
        for j = 1:frameCount
            m_full(((j-1)*frame_d+1):((j-1)*frame_d+1)+frame_d-1,i) = m(j,i); 
        end
        m_full(:,i) = [m_full(delay(i)+1:end,i); zeros(delay(i),1)];
    end

    % convolve with kernel
    kernel = kernel./sum(abs(kernel(:)));
    m_full = conv2(m_full,kernel,'same');
    
end