Getting started

[Dirkjan Krijnders]

Hi, For a course on university I need to write some signal processing software in C++. Now, I have some experience with c++, e.g. somewhat more than the basics.
For a start I would like to take the Fourier transform of a sinus. This should be pretty straight forward:
int main(){
    float *vector1;
    float *vector2;
    _sSCplx *vectorfft;

    ofstream file ("output.m");

    long int max_1=50000;
    long int FFTWinSize=20;
    bool RealFFT=true;
    clDSPOp dsp;

    vector1=new float[max_1];
    vector2=new float[max_2];
    vectorfft=new _sSCplx[max_1];

    generateSine(vector1,max_1,(max_1/t_max),700);
    dsp.FFTInitialize(FFTWinSize,RealFFT);
    dsp.FFTo(vectorfft,vector1);

    file << "data = [";
    for (int i=0;i<max_1;i++){
                file << vectorfft[i].R <<"+"<<vectorfft[i].I << "i ";
    }
    file << "];" << endl;

    file << "sine = [";
    for (int i=0;i<max_1;i++){
                file << vector1[i] << " ";
    }
    file << "];" << endl;

    file.close();
    exit(0);
}

void generateSine(float *y,int samples,float Fs,int freq){
    float t;
    cout << "Sine parameters:" << samples << " Sample Frequency: " << Fs << " Frequency: " <<freq << endl;
    for (int i=0;i<samples;i++) {
        t=((float)i/(float)Fs);
        y[i]=sin(2*freq*3.1415*t);
//        cout << t<< ":" << y[i] << endl;
    }
}

This should output the sine and its FFT to a matlab file. But, instead of a nice peak I get non-frequency dependant noise, though it seems to dependant on the windowsize. Any clues or suggestions?
Thanks in advance!

Dirkjan Krijnders
student Applied Physics
Unversity of Twente

[Jussi Laako]

I'll write example program at weekend, but few things caught my eyes immediately.

> long int max_1=50000;
> long int FFTWinSize=20;

You have sample set of 50000, but window size of 20. First of all window size must be N=2^x where x is integer. Second, you use very short window of 20 samples for large data set.

You could use FFTWinSize=max_1=65536 and thus you would get max_1/2+1 complex values as result, as you are doing real-to-complex transform.

Best regards,

- Jussi Laako

[Jussi Laako]

I have committed working example to CVS. It's work/example1.cc, you'll also need updated DSPVector.hh to compile it.

See http://cvs.sourceforge.net/cgi-bin/viewcvs.cgi/libdsp/libDSP/