TSP (libtsp) - SPfRFFT

SPfRFFT

Routine

void SPfRFFT (float x[], int N, int Ifn)

Purpose

Fast Fourier transform of a real sequence

Description

If Ifn is greater than zero, this routine computes the discrete Fourier transform of N real data points. The first step in this process is to rearrange the data in-place. Let the input real sequence be x[0], x[1], ..., These values are rearranged to have the even-numbered coefficients first, followed by the odd-numbered coefficients. These rearranged values are considered to be the real and imaginary components of a complex sequence. This complex sequence is passed to an N/2 point fast Fourier transform. The final step is to post-process the output of the FFT to form the output values. Let the N/2+1 complex output values be

  (ar(0),ar(0)), ... , (ar(N/2), ai(N/2))

These output values are stored as follows.

    ar(0)     <->  x[0]
    ar(1)     <->  x[1]
     ...           ...
    ar(N/2-1) <->  x[N/2-1]
    ar(N/2)   <->  x[N/2]
    ai(1)     <->  x[N/2+1]
    ai(2)     <->  x[N/2+2]
     ...           ...
    ai(N/2-1) <->  x[N-1]

Note that ar(N/2) is stored in the place that would normally be reserved for ai(0). This is possible since ai(0) (and ai(N/2)) are known a priori to be zero.

If Ifn is less than zero, this routine computes a inverse discrete Fourier transform that results in N real data values. Let the N/2+1 complex input values be (ar(0),ai(0)), ... , (ar(N/2),ai(N/2)). If these values are to result in a real sequence after the inverse discrete Fourier transform, ai(0) and ai(N/2) must both be zero. The remaining N data values (N/2+1 real components and N/2-1 imaginary components) are stored in the array x as follows.

    ar(0)     <->  x[0]
    ar(1)     <->  x[1]
     ...           ...
    ar(N/2-1) <->  x[N/2-1]
    ar(N/2)   <->  x[N/2]
    ai(1)     <->  x[N/2+1]
    ai(2)     <->  x[N/2+2]
     ...           ...
    ai(N/2-1) <->  x[N-1]

Note that ar(N/2) is stored in the place that would normally be reserved for ai(0). The output of the inverse transform contains the N dreal data points stored sequentially in array x.

Parameters

<-> float x[]: Data array (N elements)
-> int N: Number of elements in x. N must be a power of 2.
-> int Ifn: Parameter specifying the direction of the transform. If Ifn > 0, x is the time series on input and the frequency response on output. If Ifn < 0, x is the frequency response on input and the time series on output.

Author / revision