CODEWITHSUNDEEP
cpolynomial-mathcurve-fitting
goldenmean
goldenmean

Reputation: 18956

Any alternate approximation to this method in matlab?

I am trying to convert below matlab/octave function to C(Conventional way - understand the matlab function and code it in C from scratch). It is fitting a data to a gaussian curve using polynomial fitting.

   function y=func(data)
   N=128;
   y1=gausswin(N,4);
   x1=[0:1/N:1-1/N]';
   P=polyfit(x1,y1,12);      
   y=polyval(P,data);

But when I checked the functions polyfit, that seemed a lot of work as it involves lot of calls to further octave library functions. It computes a Vandermonde matrix first, then performs some QR decomposition of it, and computes norm of the vector etc...

  1. What other options/processing I can utilize to have similar functionality(approximation of the actual operation happeneing above) but with some simpler curve fitting or interpolation methods.

Any pointers would be useful.

Upvotes: 0

Views: 1295

Answers (1)

Egon
Egon

Reputation: 4787

Besides the point of the practical value of fitting such a polynomial to a gaussian, you can just analyze the behavior of your code:

N=128;
y1=gausswin(N,4);
x1=[0:1/N:1-1/N]';
P=polyfit(x1,y1,12);

The output of this section will always be the same, so you can execute this in MATLAB or Octave and just extract the polynomial P for usage in your C code where you include it as a constant. It's less flexible than rewriting everything in C, but it's also faster.

Otherwise, you might want to take a look at BLAS: BLAS defines an API for libraries used for linear algebra such as LAPACK (which is used by MATLAB). I suspect a lot of these libraries will implement the basic operations you need.

Addition: If you have no little experience with numerical computing or just want a lot of work taken out of your hands, you might want to consider Matlab Coder.

Upvotes: 1

Related Questions