If the entire file fits in memory I recommend that you
- read the entire file as character
- extracts the "numerical chunks" (an appropriate job for regexp). See Lookahead Assertions in Regular Expressions
- convert to double
>> M = cssm()
M =
[4x3 double] [5x3 double]
where cssm.m contains
function M = cssm()
num_xpr = '[ 0-9\.\+\-e\r\n]++';
xpr = ['(?<=([-]{10,}) *[\r\n]+)',num_xpr,'(?=(Frequency)|($))'];
str = fileread( 'cssm.txt' );
cac = regexp( str, xpr, 'match' );
M = cell( 1, length( cac ) );
for jj = 1 : length( cac )
M{jj} = str2num( cac{jj} );
end
end
and cssm.txt contains the example of your question
 
or an alternative implementation:
The magic number 150 is the minimum number of characters that a block of numerical data must contain.
>> g = read_blocks_of_numerical_data( 'cssm.txt', 150 )
g =
[4x3 double] [5x3 double]
where read_blocks_of_numerical_data
function out = read_blocks_of_numerical_data(filespec,block_size,delim)
narginchk( 2, 3 )
buffer = fileread( filespec );
if nargin == 2
del_xpr = '[ ]+';
trl_xpr = '[ ]*';
else
del_xpr = ['([ ]*',delimiter,'[ ]*)'];
trl_xpr = ['([ ]*',delimiter,'?[ ]*)'];
end
num_xpr = '([+-]?(\d+(\.\d*)?)|(\.\d+))';
sen_xpr = '([EeDd](\+|-)\d{1,3})?'; % optional scientific
num_xpr = [ num_xpr, sen_xpr ];
nl_xpr = '((\r\n)|\n)';
row_xpr = cat( 2, '(^|', nl_xpr, ')[ ]*(' ...
, num_xpr, del_xpr, ')*' ...
, num_xpr, trl_xpr, '(?=' ...
, nl_xpr,'|$)' );
blk_xpr = ['(',row_xpr,')+'];
blocks = regexp( buffer, blk_xpr, 'match' );
is_long = cellfun( @(str) length(str)>=block_size, blocks );
blocks(not(is_long)) = [];
out = cell( 1, length( blocks ) );
for jj = 1 : length( blocks )
out{jj} = str2num( blocks{jj} );
end
end
