samer@0: % additive_split - Arrow to process data with adaptive zero mean
samer@0: %
samer@0: % additive_split ::
samer@0: %    arrow( {[N,M]}, {[N,M]}, S1) ~'arrow to process N channel zero mean data',
samer@0: %    arrow( {[N]}, {[N]}, S2)     ~'arrow to process estimated mean',
samer@0: %    options {}                   ~'options passed to azeromean'
samer@0: % -> arrow( {[N,M]}, {[N,M]}, pair( pair([N],pair(S1,S2)), empty)).
samer@0: %
samer@0: % This arrow uses the azeromean arrow to adaptively estimate the mean of
samer@0: % the N channels of input data, using a gaussian data model. The mean is
samer@0: % subtracted out. The two components are then processed independently by
samer@0: % the two supplied arrows and added together to create the output.
samer@0: 
samer@0: function o=additive_split(a1,a2,varargin)
samer@37: 	opts=options(varargin{:});
samer@0: 
samer@0: 	o = states(azeromean(gaussian,opts))*(a1+a2)*aplus;
samer@0: end
samer@0: 
samer@0: function o=aplus, o=arrf(@plusfn,2,1); end
samer@0: function f=plusfn(sz)
samer@0: 	if all(sz{1}==sz{2}), f=@plus;
samer@0: 	elseif sz{1}(1)==sz{2}(1) && sz{1}(2)>sz{2}(2)
samer@0: 		f=@(a,b)a+repmat(b,1,size(a,2));
samer@0: 	elseif sz{1}(1)==sz{2}(1) && sz{1}(2)<sz{2}(2)
samer@0: 		f=@(a,b)b+repmat(a,1,size(b,2));
samer@0: 	else
samer@0: 		error('additive_split:incompatible sizes');
samer@0: 	end
samer@0: end