Fast Fourier Transformation


Algorithm description

The following implementation uses a nonrecursive Fast Fourier Transformation algorithm to multiply polynomials. If we view the elements of a vector (a0,...,an-1) to be transformed as coefficients of a polynomial a(x), then the Fourier transform is the same as computing a(wj) for 0 <= j < N. This is also equivalent to computing the remainder by making use of a principle called balancing. In this case we compute the remainders with the help of a process which is structured like a binary tree. Since the complex description of this algorithm is beyond the scope of this document, refer to the bibliography for further information.

Algorithm implementation

 1  procedure fft(arr: in out coeff_array; m: Integer) 
 2  is
 3      i,j,k,l,n,ndiv2,pow2,pow2m1,index : Integer;
 4      r,s,t : complex;
 5  begin
 6      n := 2**m;
 7      ndiv2 := n/2;
 8      j := 1;
 9      for i in 1..n-1 loop
11          if i<j then 
12              t := arr(j);
13              arr(j) := arr(i);
14              arr(i) := t;
15          end if;
16          k := ndiv2;
17          discrete a:=k in reverse j..k
18              new a:=a/2
19          loop
20              j := j-k;
21              a := a/2;
22              k := a;
23          end loop;
24          j := j+k;
25      end loop;
26      for l in 1..m loop
27          pow2 := 2**l;
28          pow2m1 := pow2/2;
29          r := (1.0,0.0);  
30          s := (cos(pi/double(pow2m1)),sin(pi/double(pow2m1)));
31          for j in 1..pow2m1 loop
32              discrete i:=j in j..n
33                  new i:=i+pow2
34              loop
35                  index := i+pow2m1;
36                  t := arr(index)*r;
37                  arr(index) := arr(i)-t;
38                  arr(i) := arr(i)+t;
39                  i := i+pow2;                
40              end loop;
41              r := r*s;
42          end loop;        
43      end loop;        
44  end fft;

[ Sourcefile ]

Notes on design

In our example above we assume complex arithmetic and w=e2*PI*i/j is expressed in terms of sines and cosines. 'arr' contains the input coefficients and the elements of the transform are computed in-place.

Worst Case Performance

The complexity of the non-recursive Fast Fourier Transform is

N log N.

Selected Bibliography

[Sav76, p. 342]


[ Samples Overview | Bibliography | WOOP Project ]