Code:
#include iostream
#include math.h
using namespace std;
#define PI 3.14159265
class DFT_Coefficient
{
public:
double real, img;
DFT_Coefficient()
{
real = 0.0;
img = 0.0;
}
};
int main(int argc, char **argv)
{
int N = 10;
cout << "Discrete Fourier Transform using naive method\n";
cout << "Enter the coefficient of simple linear function:\n";
cout << "ax + by = c\n";
double a, b, c;
cin >> a >> b >> c;
double function[N];
for (int i = 0; i < N; i++)
{
function[i] = (((a * (double) i) + (b * (double) i)) - c);
//System.out.print( " "+function[i] + " ");
}
cout << "Enter the max K value: ";
int k;
cin >> k;
double cosine[N];
double sine[N];
for (int i = 0; i < N; i++)
{
cosine[i] = cos((2 * i * k * PI) / N);
sine[i] = sin((2 * i * k * PI) / N);
}
DFT_Coefficient dft_val[k];
cout << "The coefficients are: ";
for (int j = 0; j < k; j++)
{
for (int i = 0; i < N; i++)
{
dft_val[j].real += function[i] * cosine[i];
dft_val[j].img += function[i] * sine[i];
}
cout << "(" << dft_val[j].real << ") - " << "(" << dft_val[j].img << " i)\n";
}
}
Output:
Discrete Fourier Transform using naive method
Enter the coefficient of simple linear function:
ax + by = c
1 2 3
Enter the max K value: 20
The coefficients are:
(105) - (-1.03386e-005 i)
(105) - (-1.03386e-005 i)
(105) - (-1.03386e-005 i)
(105) - (-1.03386e-005 i)
(105) - (-1.03386e-005 i)
(105) - (-1.03386e-005 i)
(105) - (-1.03386e-005 i)
(105) - (-1.03386e-005 i)
(105) - (-1.03386e-005 i)
(105) - (-1.03386e-005 i)
(105) - (-1.03386e-005 i)
(105) - (-1.03386e-005 i)
(105) - (-1.03386e-005 i)
(105) - (-1.03386e-005 i)
(105) - (-1.03386e-005 i)
(105) - (-1.03386e-005 i)
(105) - (-1.03386e-005 i)
(105) - (-1.03386e-005 i)
(105) - (-1.03386e-005 i)
(105) - (-1.03386e-005 i)
More C++ Programs: