MATLAB code to C++ conversion

hi,

I have tried a lot to convert this m file to C++ (.cpp) file. But I am having some issues. I have tried a lot to make dynamic arrays as inputs but I have always got error. I wish someone could just convert this code to .cpp file for me using MATLAB in built C++ converter. Please note that the inputs y and x have to be dynamic and complex arrays, tol and L to be real double scalars. Please help me with it.

Below is the code!

%#eml

function [b, a, f] = mpencil(y, x, tol, L)

%eml.varsize('x', [1 10]);
%leny = inf;

%assert (isa(y,'double') && ~isreal(y) && all(size(y) == [1 5]) && isa(x,'double') && ~isreal(x) && isvector(x) && isa(tol,'double') && isa(L,'double') );
assert( isa(y,'double') && ~isreal(y) && all(size(y) == [1 inf]) && isa(x,'double') &&   ~isreal(x)   && all(size(y) == [1 size(y,2)]) && isa(tol,'double') && isa(L,'double')  );

% function [B, A, F] = mpencil(Y, X, [TOL, L])
%MPENCIL Fits a sum of exponential functions to data by a Matrix Pencil method
%
% B = MPENCIL(Y,X) approximates Y(i) by a weighted sum of
% exponentials sum_j=1^p A(j)*exp(B(j)*X(i)). The X(i) are
% assumed equally spaced.
%
% The number of exponential is defined with the third optional
% parameter TOL. If TOL < 1 (and > 0) the number of terms is the same
% as the number of such singular values (of the data) that are larger
% than TOL*biggest singular value. If TOL >= 1 and an integer, the
% number of exponential terms is TOL. The default TOL value is 1e-3.
%
% The last optional parameter L is a parameter for the Matrix Pencil
% method. Values between N/3 and N/2 are preferred where N is the
% number of data (Y).
% The default L is the ceil(1/2*(ceil(N/3)+floor(N/2))).
%
% The A(i) and the fitted values F(i) are optionally returned by
% [B,A,F] = MPENCIL(Y,X).

% References:
% Sarkar, T.K. and Pereira, O.
% Using the Matrix Pencil Method to Estimate the Parameters of a Sum
% of Complex Exponentials.
% IEEE Antennas and Propagation Magazine, Vol. 37, No 1, Feb. 1995

% Matti Taskinen, Rolf Nevanlinna Institute, University of Helsinki
% 06.03.1996. Latest revision 01.04.1996.

dd = x;
x = zeros(size(y));
x = dd;
% Check the vector sizes: 
N = size(y, 1);
if size(x, 1) ~= N
error('length(Y) should be length(X)');
end

if size(y, 2) ~= 1
error('Y should be column vector');
end

if size(x, 2) ~= 1
error('X should be column vector');
end

% Fill in the missing parameters:
if nargin < 4
    L1=ceil(1/3*N);
    L2=floor(1/2*N);
L = ceil((L1 + L2) / 2);
if nargin < 3
    tol = 1e-3;
end
end

% Check the parameter values:
M_given = round(tol) == tol;
if (tol < 0) || ((~M_given) && (tol > 1))
error('TOL should be either >= 1 and an integer or < 1 and > 0');
end

if L > N/2
error('L shoud be < N/2');
end

if M_given && (L < tol)
error('TOL should be <= L');
end

% X are assumed to be equally spaced:
T = diff(x(1:2));

Y = zeros(N-L, L+1);
Y = complex(Y,Y);
ind = 0:N-L-1;
for j = 1:L+1
Y(:,j) = y(ind+j);
end

[U,S,V] = svd(Y,0); % Economy size!

if M_given
% The number of exponential terms may be given in TOL:
M = tol;
else
% Otherwise figure it out from the singular values:
D = diag(S);
for M = 1:length(D)-1
    if (abs(D(M+1)/D(1)) <= tol)
        break;
    end
end
end

SM = S(:,1:M);

VM = V(:,1:M);
V1 = VM(1:L,:);
V2 = VM(2:L+1,:);

Y1 = U*SM*V1';
Y2 = U*SM*V2';

A = pinv(Y1)*Y2;
z = eig(A);
z = z(1:M);
b = 1/T * log(z);

if nargout > 1
    Z = exp(x*b.');
    a = Z\y;
    if nargout > 2
        f = Z*a;
    end
end

Haris

[haris]

Kindly also tell me how to write assert commands to be able to declare both x and y as complex dynamic arrays

Answer 1

[Rick Rosson]

Can you please answer a few questions:

1. What version of MATLAB (i.e. Rnnnnx) are you using?
2. Do you have access to Real-Time Workshop?
3. Do you want dynamic memory allocation for the variable-sized arrays, or can you specify a fixed upper bound on the size of each variable-size array?

Thanks!

Rick

[haris]

Hi,

Thanks for replying.

I am using MATLAB R2010b on 64 bit Windows 7.

Yes I have access to real time workshop.

Well, in the given function, I want x and y to be column vectors that can have any number of rows. It would be much better if there is no upper bound.

Answer 2

[Rick Rosson]

I would recommend trying a three-phase approach:

1. Generate code assuming fixed size arrays for all variables. Choose a size for x and y that are typical of your ultimate application, and easily manageable for testing and debugging purposes. Create build script to make the code generation process repeatable, and create a test script to do at least an adequate level of testing to convince yourself that the code is working correctly both interpreted and compiled.
2. Modify the MATLAB code and/or your build script (create one if you don't have one already) so that you can generate code for variable size arrays with fixed upper bound. Generate code and compile to a MEX function so that you can test the code for runtime errors within the MATLAB environment.
3. Once you have successfully generated code for fixed upper bound, then consider whether you really need dynamic memory allocation. If so, relax the upper bound conditions and re-generate the code with dynamic memory turned "on".

Generating C code from MATLAB can be challenging. It's much easier to take it one step at-a-time, rather than trying to get everything to work on the first attempt. Crawl before you walk, walk before you jog, jog before you sprint...