MATLAB: Using fminsearch to tiffing distribution

Question

Hi all, my problem is this: i have a function F composed by Noise(gaussian distribution) + Value(lognormal distribution). The function F is the convolution of the two distributions. In my first script i have the function convolution.

    function f = convol(mus,sis)
        x=(1:16384);
        mur=1643.7;     %Mean
        sir=93.94;      %Sigma
        %****Noise Normal****    
        frum = exp( (-(x-mur).^2)./(2*sir*sir) );
        frum = frum / sum(frum); %Normalized



        %****SignalLognormal****
        fsig = exp( (-(log(x)-mus).^2)./(2*sis*sis) );
        fsig = fsig / sum(fsig); %Normalized
        f = conv(fsig,frum,'full');
        f = f(1:16384);
        f = f / sum(f); %Normalized

In this second script i calculate the ols error, fitting this distribution F with one that i have (called mySignal)

function diff = errorconv(PAR)
    f2 = mySignal./ sum(mySignal); %Normalized
    mus = PAR(1);
    sis = PAR(2);
    f = convol( mus, sis ); %Call the first script
    diff = sum( (f- f2 ).^2 ); %Calculate the ols

My question is: is there a method to use fminsearch to make this calculation automatically? Something like fminsearch(diff, PAR); beacause my target is find the mu and sigma of function fsig such that convolution f is the best fit of f2 (my signal) starting from PAR = [value,value]; Thanks for help Luca

Answer 1

function main
    par0(1)=value1;
    par0(2)=value2;
    par=fminsearch(@(par)errorconv(par,mySignal),par0);
function diff = errorconv(PAR,mySignal)
    f2 = mySignal./ sum(mySignal); %Normalized



    mus = PAR(1);
    sis = PAR(2);
    f = convol( mus, sis ); %Call the first script
    diff = sum( (f- f2 ).^2 ); %Calculate the ols
function f = convol(mus,sis)
        x=(1:16384);
        mur=1643.7;     %Mean
        sir=93.94;      %Sigma
        %****Noise Normal****    
        frum = exp( (-(x-mur).^2)./(2*sir*sir) );
        frum = frum / sum(frum); %Normalized
        %****SignalLognormal****
        fsig = exp( (-(log(x)-mus).^2)./(2*sis*sis) );
        fsig = fsig / sum(fsig); %Normalized
        f = conv(fsig,frum,'full');
        f = f(1:16384);
        f = f / sum(f); %Normalized

Best wishes

Torsten.