@cindex FFT @cindex Fast Fourier Transforms, see FFT @cindex Fourier Transforms, see FFT @cindex Discrete Fourier Transforms, see FFT @cindex DFTs, see FFT This chapter describes functions for performing Fast Fourier Transforms (FFTs). The library includes radix-2 routines (for lengths which are a power of two) and mixed-radix routines (which work for any length). For efficiency there are separate versions of the routines for real data and for complex data. The mixed-radix routines are a reimplementation of the @sc{fftpack} library of Paul Swarztrauber. Fortran code for @sc{fftpack} is available on Netlib (@sc{fftpack} also includes some routines for sine and cosine transforms but these are currently not available in GSL). For details and derivations of the underlying algorithms consult the document @cite{GSL FFT Algorithms} (@pxref{FFT References and Further Reading}) @menu * Mathematical Definitions:: * Overview of complex data FFTs:: * Radix-2 FFT routines for complex data:: * Mixed-radix FFT routines for complex data:: * Overview of real data FFTs:: * Radix-2 FFT routines for real data:: * Mixed-radix FFT routines for real data:: * FFT References and Further Reading:: @end menu @node Mathematical Definitions @section Mathematical Definitions @cindex FFT mathematical definition Fast Fourier Transforms are efficient algorithms for calculating the discrete fourier transform (DFT), @tex \beforedisplay $$ x_j = \sum_{k=0}^{N-1} z_k \exp(-2\pi i j k / N) $$ \afterdisplay @end tex @ifinfo @example x_j = \sum_@{k=0@}^@{N-1@} z_k \exp(-2\pi i j k / N) @end example @end ifinfo The DFT usually arises as an approximation to the continuous fourier transform when functions are sampled at discrete intervals in space or time. The naive evaluation of the discrete fourier transform is a matrix-vector multiplication @c{$W\vec{z}$} @math{W\vec@{z@}}. A general matrix-vector multiplication takes @math{O(N^2)} operations for @math{N} data-points. Fast fourier transform algorithms use a divide-and-conquer strategy to factorize the matrix @math{W} into smaller sub-matrices, corresponding to the integer factors of the length @math{N}. If @math{N} can be factorized into a product of integers @c{$f_1 f_2 \ldots f_n$} @math{f_1 f_2 ... f_n} then the DFT can be computed in @math{O(N \sum f_i)} operations. For a radix-2 FFT this gives an operation count of @math{O(N \log_2 N)}. All the FFT functions offer three types of transform: forwards, inverse and backwards, based on the same mathematical definitions. The definition of the @dfn{forward fourier transform}, @c{$x = \hbox{FFT}(z)$} @math{x = FFT(z)}, is, @tex \beforedisplay $$ x_j = \sum_{k=0}^{N-1} z_k \exp(-2\pi i j k / N) $$ \afterdisplay @end tex @ifinfo @example x_j = \sum_@{k=0@}^@{N-1@} z_k \exp(-2\pi i j k / N) @end example @end ifinfo @noindent and the definition of the @dfn{inverse fourier transform}, @c{$x = \hbox{IFFT}(z)$} @math{x = IFFT(z)}, is, @tex \beforedisplay $$ z_j = {1 \over N} \sum_{k=0}^{N-1} x_k \exp(2\pi i j k / N). $$ \afterdisplay @end tex @ifinfo @example z_j = @{1 \over N@} \sum_@{k=0@}^@{N-1@} x_k \exp(2\pi i j k / N). @end example @end ifinfo @noindent The factor of @math{1/N} makes this a true inverse. For example, a call to @code{gsl_fft_complex_forward} followed by a call to @code{gsl_fft_complex_inverse} should return the original data (within numerical errors). In general there are two possible choices for the sign of the exponential in the transform/ inverse-transform pair. GSL follows the same convention as @sc{fftpack}, using a negative exponential for the forward transform. The advantage of this convention is that the inverse transform recreates the original function with simple fourier synthesis. Numerical Recipes uses the opposite convention, a positive exponential in the forward transform. The @dfn{backwards FFT} is simply our terminology for an unscaled version of the inverse FFT, @tex \beforedisplay $$ z^{backwards}_j = \sum_{k=0}^{N-1} x_k \exp(2\pi i j k / N). $$ \afterdisplay @end tex @ifinfo @example z^@{backwards@}_j = \sum_@{k=0@}^@{N-1@} x_k \exp(2\pi i j k / N). @end example @end ifinfo @noindent When the overall scale of the result is unimportant it is often convenient to use the backwards FFT instead of the inverse to save unnecessary divisions. @node Overview of complex data FFTs @section Overview of complex data FFTs @cindex FFT, complex data The inputs and outputs for the complex FFT routines are @dfn{packed arrays} of floating point numbers. In a packed array the real and imaginary parts of each complex number are placed in alternate neighboring elements. For example, the following definition of a packed array of length 6, @example double x[3*2]; gsl_complex_packed_array data = x; @end example @noindent can be used to hold an array of three complex numbers, @code{z[3]}, in the following way, @example data[0] = Re(z[0]) data[1] = Im(z[0]) data[2] = Re(z[1]) data[3] = Im(z[1]) data[4] = Re(z[2]) data[5] = Im(z[2]) @end example @noindent The array indices for the data have the same ordering as those in the definition of the DFT---i.e. there are no index transformations or permutations of the data. A @dfn{stride} parameter allows the user to perform transforms on the elements @code{z[stride*i]} instead of @code{z[i]}. A stride greater than 1 can be used to take an in-place FFT of the column of a matrix. A stride of 1 accesses the array without any additional spacing between elements. To perform an FFT on a vector argument, such as @code{gsl_vector_complex * v}, use the following definitions (or their equivalents) when calling the functions described in this chapter: @example gsl_complex_packed_array data = v->data; size_t stride = v->stride; size_t n = v->size; @end example For physical applications it is important to remember that the index appearing in the DFT does not correspond directly to a physical frequency. If the time-step of the DFT is @math{\Delta} then the frequency-domain includes both positive and negative frequencies, ranging from @math{-1/(2\Delta)} through 0 to @math{+1/(2\Delta)}. The positive frequencies are stored from the beginning of the array up to the middle, and the negative frequencies are stored backwards from the end of the array. Here is a table which shows the layout of the array @var{data}, and the correspondence between the time-domain data @math{z}, and the frequency-domain data @math{x}. @example index z x = FFT(z) 0 z(t = 0) x(f = 0) 1 z(t = 1) x(f = 1/(N Delta)) 2 z(t = 2) x(f = 2/(N Delta)) . ........ .................. N/2 z(t = N/2) x(f = +1/(2 Delta), -1/(2 Delta)) . ........ .................. N-3 z(t = N-3) x(f = -3/(N Delta)) N-2 z(t = N-2) x(f = -2/(N Delta)) N-1 z(t = N-1) x(f = -1/(N Delta)) @end example @noindent When @math{N} is even the location @math{N/2} contains the most positive and negative frequencies (@math{+1/(2 \Delta)}, @math{-1/(2 \Delta)}) which are equivalent. If @math{N} is odd then general structure of the table above still applies, but @math{N/2} does not appear. @node Radix-2 FFT routines for complex data @section Radix-2 FFT routines for complex data @cindex FFT of complex data, radix-2 algorithm @cindex Radix-2 FFT, complex data The radix-2 algorithms described in this section are simple and compact, although not necessarily the most efficient. They use the Cooley-Tukey algorithm to compute in-place complex FFTs for lengths which are a power of 2---no additional storage is required. The corresponding self-sorting mixed-radix routines offer better performance at the expense of requiring additional working space. All the functions described in this section are declared in the header file @file{gsl_fft_complex.h}. @deftypefun int gsl_fft_complex_radix2_forward (gsl_complex_packed_array @var{data}, size_t @var{stride}, size_t @var{n}) @deftypefunx int gsl_fft_complex_radix2_transform (gsl_complex_packed_array @var{data}, size_t @var{stride}, size_t @var{n}, gsl_fft_direction @var{sign}) @deftypefunx int gsl_fft_complex_radix2_backward (gsl_complex_packed_array @var{data}, size_t @var{stride}, size_t @var{n}) @deftypefunx int gsl_fft_complex_radix2_inverse (gsl_complex_packed_array @var{data}, size_t @var{stride}, size_t @var{n}) These functions compute forward, backward and inverse FFTs of length @var{n} with stride @var{stride}, on the packed complex array @var{data} using an in-place radix-2 decimation-in-time algorithm. The length of the transform @var{n} is restricted to powers of two. For the @code{transform} version of the function the @var{sign} argument can be either @code{forward} (@math{-1}) or @code{backward} (@math{+1}). The functions return a value of @code{GSL_SUCCESS} if no errors were detected, or @code{GSL_EDOM} if the length of the data @var{n} is not a power of two. @end deftypefun @deftypefun int gsl_fft_complex_radix2_dif_forward (gsl_complex_packed_array @var{data}, size_t @var{stride}, size_t @var{n}) @deftypefunx int gsl_fft_complex_radix2_dif_transform (gsl_complex_packed_array @var{data}, size_t @var{stride}, size_t @var{n}, gsl_fft_direction @var{sign}) @deftypefunx int gsl_fft_complex_radix2_dif_backward (gsl_complex_packed_array @var{data}, size_t @var{stride}, size_t @var{n}) @deftypefunx int gsl_fft_complex_radix2_dif_inverse (gsl_complex_packed_array @var{data}, size_t @var{stride}, size_t @var{n}) These are decimation-in-frequency versions of the radix-2 FFT functions. @end deftypefun @comment @node Example of using radix-2 FFT routines for complex data @comment @subsection Example of using radix-2 FFT routines for complex data Here is an example program which computes the FFT of a short pulse in a sample of length 128. To make the resulting fourier transform real the pulse is defined for equal positive and negative times (@math{-10} @dots{} @math{10}), where the negative times wrap around the end of the array. @example @verbatiminclude examples/fft.c @end example @noindent Note that we have assumed that the program is using the default error handler (which calls @code{abort} for any errors). If you are not using a safe error handler you would need to check the return status of @code{gsl_fft_complex_radix2_forward}. The transformed data is rescaled by @math{1/\sqrt N} so that it fits on the same plot as the input. Only the real part is shown, by the choice of the input data the imaginary part is zero. Allowing for the wrap-around of negative times at @math{t=128}, and working in units of @math{k/N}, the DFT approximates the continuum fourier transform, giving a modulated sine function. @iftex @tex \beforedisplay $$ \int_{-a}^{+a} e^{-2 \pi i k x} dx = {\sin(2\pi k a) \over\pi k} $$ \afterdisplay @end tex @sp 1 @center @image{fft-complex-radix2-t,2.8in} @center @image{fft-complex-radix2-f,2.8in} @quotation A pulse and its discrete fourier transform, output from the example program. @end quotation @end iftex @node Mixed-radix FFT routines for complex data @section Mixed-radix FFT routines for complex data @cindex FFT of complex data, mixed-radix algorithm @cindex Mixed-radix FFT, complex data This section describes mixed-radix FFT algorithms for complex data. The mixed-radix functions work for FFTs of any length. They are a reimplementation of Paul Swarztrauber's Fortran @sc{fftpack} library. The theory is explained in the review article @cite{Self-sorting Mixed-radix FFTs} by Clive Temperton. The routines here use the same indexing scheme and basic algorithms as @sc{fftpack}. The mixed-radix algorithm is based on sub-transform modules---highly optimized small length FFTs which are combined to create larger FFTs. There are efficient modules for factors of 2, 3, 4, 5, 6 and 7. The modules for the composite factors of 4 and 6 are faster than combining the modules for @math{2*2} and @math{2*3}. For factors which are not implemented as modules there is a fall-back to a general length-@math{n} module which uses Singleton's method for efficiently computing a DFT. This module is @math{O(n^2)}, and slower than a dedicated module would be but works for any length @math{n}. Of course, lengths which use the general length-@math{n} module will still be factorized as much as possible. For example, a length of 143 will be factorized into @math{11*13}. Large prime factors are the worst case scenario, e.g. as found in @math{n=2*3*99991}, and should be avoided because their @math{O(n^2)} scaling will dominate the run-time (consult the document @cite{GSL FFT Algorithms} included in the GSL distribution if you encounter this problem). The mixed-radix initialization function @code{gsl_fft_complex_wavetable_alloc} returns the list of factors chosen by the library for a given length @math{N}. It can be used to check how well the length has been factorized, and estimate the run-time. To a first approximation the run-time scales as @math{N \sum f_i}, where the @math{f_i} are the factors of @math{N}. For programs under user control you may wish to issue a warning that the transform will be slow when the length is poorly factorized. If you frequently encounter data lengths which cannot be factorized using the existing small-prime modules consult @cite{GSL FFT Algorithms} for details on adding support for other factors. @comment First, the space for the trigonometric lookup tables and scratch area is @comment allocated by a call to one of the @code{alloc} functions. We @comment call the combination of factorization, scratch space and trigonometric @comment lookup arrays a @dfn{wavetable}. It contains the sine and cosine @comment waveforms for the all the frequencies that will be used in the FFT. @comment The wavetable is initialized by a call to the corresponding @code{init} @comment function. It factorizes the data length, using the implemented @comment subtransforms as preferred factors wherever possible. The trigonometric @comment lookup table for the chosen factorization is also computed. @comment An FFT is computed by a call to one of the @code{forward}, @comment @code{backward} or @code{inverse} functions, with the data, length and @comment wavetable as arguments. All the functions described in this section are declared in the header file @file{gsl_fft_complex.h}. @deftypefun {gsl_fft_complex_wavetable *} gsl_fft_complex_wavetable_alloc (size_t @var{n}) This function prepares a trigonometric lookup table for a complex FFT of length @var{n}. The function returns a pointer to the newly allocated @code{gsl_fft_complex_wavetable} if no errors were detected, and a null pointer in the case of error. The length @var{n} is factorized into a product of subtransforms, and the factors and their trigonometric coefficients are stored in the wavetable. The trigonometric coefficients are computed using direct calls to @code{sin} and @code{cos}, for accuracy. Recursion relations could be used to compute the lookup table faster, but if an application performs many FFTs of the same length then this computation is a one-off overhead which does not affect the final throughput. The wavetable structure can be used repeatedly for any transform of the same length. The table is not modified by calls to any of the other FFT functions. The same wavetable can be used for both forward and backward (or inverse) transforms of a given length. @end deftypefun @deftypefun void gsl_fft_complex_wavetable_free (gsl_fft_complex_wavetable * @var{wavetable}) This function frees the memory associated with the wavetable @var{wavetable}. The wavetable can be freed if no further FFTs of the same length will be needed. @end deftypefun @noindent These functions operate on a @code{gsl_fft_complex_wavetable} structure which contains internal parameters for the FFT. It is not necessary to set any of the components directly but it can sometimes be useful to examine them. For example, the chosen factorization of the FFT length is given and can be used to provide an estimate of the run-time or numerical error. The wavetable structure is declared in the header file @file{gsl_fft_complex.h}. @deftp {Data Type} gsl_fft_complex_wavetable This is a structure that holds the factorization and trigonometric lookup tables for the mixed radix fft algorithm. It has the following components: @table @code @item size_t n This is the number of complex data points @item size_t nf This is the number of factors that the length @code{n} was decomposed into. @item size_t factor[64] This is the array of factors. Only the first @code{nf} elements are used. @comment (FIXME: This is a fixed length array and therefore probably in @comment violation of the GNU Coding Standards). @item gsl_complex * trig This is a pointer to a preallocated trigonometric lookup table of @code{n} complex elements. @item gsl_complex * twiddle[64] This is an array of pointers into @code{trig}, giving the twiddle factors for each pass. @end table @end deftp @noindent The mixed radix algorithms require additional working space to hold the intermediate steps of the transform. @deftypefun {gsl_fft_complex_workspace *} gsl_fft_complex_workspace_alloc (size_t @var{n}) This function allocates a workspace for a complex transform of length @var{n}. @end deftypefun @deftypefun void gsl_fft_complex_workspace_free (gsl_fft_complex_workspace * @var{workspace}) This function frees the memory associated with the workspace @var{workspace}. The workspace can be freed if no further FFTs of the same length will be needed. @end deftypefun @comment @deftp {Data Type} gsl_fft_complex_workspace @comment This is a structure that holds the workspace for the mixed radix fft @comment algorithm. It has the following components: @comment @comment @table @code @comment @item gsl_complex * scratch @comment This is a pointer to a workspace of @code{n} complex elements, @comment capable of holding intermediate copies of the original data set. @comment @end table @comment @end deftp @noindent The following functions compute the transform, @deftypefun int gsl_fft_complex_forward (gsl_complex_packed_array @var{data}, size_t @var{stride}, size_t @var{n}, const gsl_fft_complex_wavetable * @var{wavetable}, gsl_fft_complex_workspace * @var{work}) @deftypefunx int gsl_fft_complex_transform (gsl_complex_packed_array @var{data}, size_t @var{stride}, size_t @var{n}, const gsl_fft_complex_wavetable * @var{wavetable}, gsl_fft_complex_workspace * @var{work}, gsl_fft_direction @var{sign}) @deftypefunx int gsl_fft_complex_backward (gsl_complex_packed_array @var{data}, size_t @var{stride}, size_t @var{n}, const gsl_fft_complex_wavetable * @var{wavetable}, gsl_fft_complex_workspace * @var{work}) @deftypefunx int gsl_fft_complex_inverse (gsl_complex_packed_array @var{data}, size_t @var{stride}, size_t @var{n}, const gsl_fft_complex_wavetable * @var{wavetable}, gsl_fft_complex_workspace * @var{work}) These functions compute forward, backward and inverse FFTs of length @var{n} with stride @var{stride}, on the packed complex array @var{data}, using a mixed radix decimation-in-frequency algorithm. There is no restriction on the length @var{n}. Efficient modules are provided for subtransforms of length 2, 3, 4, 5, 6 and 7. Any remaining factors are computed with a slow, @math{O(n^2)}, general-@math{n} module. The caller must supply a @var{wavetable} containing the trigonometric lookup tables and a workspace @var{work}. For the @code{transform} version of the function the @var{sign} argument can be either @code{forward} (@math{-1}) or @code{backward} (@math{+1}). The functions return a value of @code{0} if no errors were detected. The following @code{gsl_errno} conditions are defined for these functions: @table @code @item GSL_EDOM The length of the data @var{n} is not a positive integer (i.e. @var{n} is zero). @item GSL_EINVAL The length of the data @var{n} and the length used to compute the given @var{wavetable} do not match. @end table @end deftypefun @comment @node Example of using mixed-radix FFT routines for complex data @comment @subsection Example of using mixed-radix FFT routines for complex data Here is an example program which computes the FFT of a short pulse in a sample of length 630 (@math{=2*3*3*5*7}) using the mixed-radix algorithm. @example @verbatiminclude examples/fftmr.c @end example @noindent Note that we have assumed that the program is using the default @code{gsl} error handler (which calls @code{abort} for any errors). If you are not using a safe error handler you would need to check the return status of all the @code{gsl} routines. @node Overview of real data FFTs @section Overview of real data FFTs @cindex FFT of real data The functions for real data are similar to those for complex data. However, there is an important difference between forward and inverse transforms. The fourier transform of a real sequence is not real. It is a complex sequence with a special symmetry: @tex \beforedisplay $$ z_k = z_{N-k}^* $$ \afterdisplay @end tex @ifinfo @example z_k = z_@{N-k@}^* @end example @end ifinfo @noindent A sequence with this symmetry is called @dfn{conjugate-complex} or @dfn{half-complex}. This different structure requires different storage layouts for the forward transform (from real to half-complex) and inverse transform (from half-complex back to real). As a consequence the routines are divided into two sets: functions in @code{gsl_fft_real} which operate on real sequences and functions in @code{gsl_fft_halfcomplex} which operate on half-complex sequences. Functions in @code{gsl_fft_real} compute the frequency coefficients of a real sequence. The half-complex coefficients @math{c} of a real sequence @math{x} are given by fourier analysis, @tex \beforedisplay $$ c_k = \sum_{j=0}^{N-1} x_j \exp(-2 \pi i j k /N) $$ \afterdisplay @end tex @ifinfo @example c_k = \sum_@{j=0@}^@{N-1@} x_j \exp(-2 \pi i j k /N) @end example @end ifinfo @noindent Functions in @code{gsl_fft_halfcomplex} compute inverse or backwards transforms. They reconstruct real sequences by fourier synthesis from their half-complex frequency coefficients, @math{c}, @tex \beforedisplay $$ x_j = {1 \over N} \sum_{k=0}^{N-1} c_k \exp(2 \pi i j k /N) $$ \afterdisplay @end tex @ifinfo @example x_j = @{1 \over N@} \sum_@{k=0@}^@{N-1@} c_k \exp(2 \pi i j k /N) @end example @end ifinfo @noindent The symmetry of the half-complex sequence implies that only half of the complex numbers in the output need to be stored. The remaining half can be reconstructed using the half-complex symmetry condition. This works for all lengths, even and odd---when the length is even the middle value where @math{k=N/2} is also real. Thus only @var{N} real numbers are required to store the half-complex sequence, and the transform of a real sequence can be stored in the same size array as the original data. The precise storage arrangements depend on the algorithm, and are different for radix-2 and mixed-radix routines. The radix-2 function operates in-place, which constrains the locations where each element can be stored. The restriction forces real and imaginary parts to be stored far apart. The mixed-radix algorithm does not have this restriction, and it stores the real and imaginary parts of a given term in neighboring locations (which is desirable for better locality of memory accesses). @node Radix-2 FFT routines for real data @section Radix-2 FFT routines for real data @cindex FFT of real data, radix-2 algorithm @cindex Radix-2 FFT for real data This section describes radix-2 FFT algorithms for real data. They use the Cooley-Tukey algorithm to compute in-place FFTs for lengths which are a power of 2. The radix-2 FFT functions for real data are declared in the header files @file{gsl_fft_real.h} @deftypefun int gsl_fft_real_radix2_transform (double @var{data}[], size_t @var{stride}, size_t @var{n}) This function computes an in-place radix-2 FFT of length @var{n} and stride @var{stride} on the real array @var{data}. The output is a half-complex sequence, which is stored in-place. The arrangement of the half-complex terms uses the following scheme: for @math{k < N/2} the real part of the @math{k}-th term is stored in location @math{k}, and the corresponding imaginary part is stored in location @math{N-k}. Terms with @math{k > N/2} can be reconstructed using the symmetry @c{$z_k = z^*_{N-k}$} @math{z_k = z^*_@{N-k@}}. The terms for @math{k=0} and @math{k=N/2} are both purely real, and count as a special case. Their real parts are stored in locations @math{0} and @math{N/2} respectively, while their imaginary parts which are zero are not stored. The following table shows the correspondence between the output @var{data} and the equivalent results obtained by considering the input data as a complex sequence with zero imaginary part, @example complex[0].real = data[0] complex[0].imag = 0 complex[1].real = data[1] complex[1].imag = data[N-1] ............... ................ complex[k].real = data[k] complex[k].imag = data[N-k] ............... ................ complex[N/2].real = data[N/2] complex[N/2].imag = 0 ............... ................ complex[k'].real = data[k] k' = N - k complex[k'].imag = -data[N-k] ............... ................ complex[N-1].real = data[1] complex[N-1].imag = -data[N-1] @end example @noindent Note that the output data can be converted into the full complex sequence using the function @code{gsl_fft_halfcomplex_unpack} described in the next section. @end deftypefun The radix-2 FFT functions for halfcomplex data are declared in the header file @file{gsl_fft_halfcomplex.h}. @deftypefun int gsl_fft_halfcomplex_radix2_inverse (double @var{data}[], size_t @var{stride}, size_t @var{n}) @deftypefunx int gsl_fft_halfcomplex_radix2_backward (double @var{data}[], size_t @var{stride}, size_t @var{n}) These functions compute the inverse or backwards in-place radix-2 FFT of length @var{n} and stride @var{stride} on the half-complex sequence @var{data} stored according the output scheme used by @code{gsl_fft_real_radix2}. The result is a real array stored in natural order. @end deftypefun @node Mixed-radix FFT routines for real data @section Mixed-radix FFT routines for real data @cindex FFT of real data, mixed-radix algorithm @cindex Mixed-radix FFT, real data This section describes mixed-radix FFT algorithms for real data. The mixed-radix functions work for FFTs of any length. They are a reimplementation of the real-FFT routines in the Fortran @sc{fftpack} library by Paul Swarztrauber. The theory behind the algorithm is explained in the article @cite{Fast Mixed-Radix Real Fourier Transforms} by Clive Temperton. The routines here use the same indexing scheme and basic algorithms as @sc{fftpack}. The functions use the @sc{fftpack} storage convention for half-complex sequences. In this convention the half-complex transform of a real sequence is stored with frequencies in increasing order, starting at zero, with the real and imaginary parts of each frequency in neighboring locations. When a value is known to be real the imaginary part is not stored. The imaginary part of the zero-frequency component is never stored. It is known to be zero (since the zero frequency component is simply the sum of the input data (all real)). For a sequence of even length the imaginary part of the frequency @math{n/2} is not stored either, since the symmetry @c{$z_k = z_{N-k}^*$} @math{z_k = z_@{N-k@}^*} implies that this is purely real too. The storage scheme is best shown by some examples. The table below shows the output for an odd-length sequence, @math{n=5}. The two columns give the correspondence between the 5 values in the half-complex sequence returned by @code{gsl_fft_real_transform}, @var{halfcomplex}[] and the values @var{complex}[] that would be returned if the same real input sequence were passed to @code{gsl_fft_complex_backward} as a complex sequence (with imaginary parts set to @code{0}), @example complex[0].real = halfcomplex[0] complex[0].imag = 0 complex[1].real = halfcomplex[1] complex[1].imag = halfcomplex[2] complex[2].real = halfcomplex[3] complex[2].imag = halfcomplex[4] complex[3].real = halfcomplex[3] complex[3].imag = -halfcomplex[4] complex[4].real = halfcomplex[1] complex[4].imag = -halfcomplex[2] @end example @noindent The upper elements of the @var{complex} array, @code{complex[3]} and @code{complex[4]} are filled in using the symmetry condition. The imaginary part of the zero-frequency term @code{complex[0].imag} is known to be zero by the symmetry. The next table shows the output for an even-length sequence, @math{n=6} In the even case there are two values which are purely real, @example complex[0].real = halfcomplex[0] complex[0].imag = 0 complex[1].real = halfcomplex[1] complex[1].imag = halfcomplex[2] complex[2].real = halfcomplex[3] complex[2].imag = halfcomplex[4] complex[3].real = halfcomplex[5] complex[3].imag = 0 complex[4].real = halfcomplex[3] complex[4].imag = -halfcomplex[4] complex[5].real = halfcomplex[1] complex[5].imag = -halfcomplex[2] @end example @noindent The upper elements of the @var{complex} array, @code{complex[4]} and @code{complex[5]} are filled in using the symmetry condition. Both @code{complex[0].imag} and @code{complex[3].imag} are known to be zero. All these functions are declared in the header files @file{gsl_fft_real.h} and @file{gsl_fft_halfcomplex.h}. @deftypefun {gsl_fft_real_wavetable *} gsl_fft_real_wavetable_alloc (size_t @var{n}) @deftypefunx {gsl_fft_halfcomplex_wavetable *} gsl_fft_halfcomplex_wavetable_alloc (size_t @var{n}) These functions prepare trigonometric lookup tables for an FFT of size @math{n} real elements. The functions return a pointer to the newly allocated struct if no errors were detected, and a null pointer in the case of error. The length @var{n} is factorized into a product of subtransforms, and the factors and their trigonometric coefficients are stored in the wavetable. The trigonometric coefficients are computed using direct calls to @code{sin} and @code{cos}, for accuracy. Recursion relations could be used to compute the lookup table faster, but if an application performs many FFTs of the same length then computing the wavetable is a one-off overhead which does not affect the final throughput. The wavetable structure can be used repeatedly for any transform of the same length. The table is not modified by calls to any of the other FFT functions. The appropriate type of wavetable must be used for forward real or inverse half-complex transforms. @end deftypefun @deftypefun void gsl_fft_real_wavetable_free (gsl_fft_real_wavetable * @var{wavetable}) @deftypefunx void gsl_fft_halfcomplex_wavetable_free (gsl_fft_halfcomplex_wavetable * @var{wavetable}) These functions free the memory associated with the wavetable @var{wavetable}. The wavetable can be freed if no further FFTs of the same length will be needed. @end deftypefun @noindent The mixed radix algorithms require additional working space to hold the intermediate steps of the transform, @deftypefun {gsl_fft_real_workspace *} gsl_fft_real_workspace_alloc (size_t @var{n}) This function allocates a workspace for a real transform of length @var{n}. The same workspace can be used for both forward real and inverse halfcomplex transforms. @end deftypefun @deftypefun void gsl_fft_real_workspace_free (gsl_fft_real_workspace * @var{workspace}) This function frees the memory associated with the workspace @var{workspace}. The workspace can be freed if no further FFTs of the same length will be needed. @end deftypefun @noindent The following functions compute the transforms of real and half-complex data, @deftypefun int gsl_fft_real_transform (double @var{data}[], size_t @var{stride}, size_t @var{n}, const gsl_fft_real_wavetable * @var{wavetable}, gsl_fft_real_workspace * @var{work}) @deftypefunx int gsl_fft_halfcomplex_transform (double @var{data}[], size_t @var{stride}, size_t @var{n}, const gsl_fft_halfcomplex_wavetable * @var{wavetable}, gsl_fft_real_workspace * @var{work}) These functions compute the FFT of @var{data}, a real or half-complex array of length @var{n}, using a mixed radix decimation-in-frequency algorithm. For @code{gsl_fft_real_transform} @var{data} is an array of time-ordered real data. For @code{gsl_fft_halfcomplex_transform} @var{data} contains fourier coefficients in the half-complex ordering described above. There is no restriction on the length @var{n}. Efficient modules are provided for subtransforms of length 2, 3, 4 and 5. Any remaining factors are computed with a slow, @math{O(n^2)}, general-n module. The caller must supply a @var{wavetable} containing trigonometric lookup tables and a workspace @var{work}. @end deftypefun @deftypefun int gsl_fft_real_unpack (const double @var{real_coefficient}[], gsl_complex_packed_array @var{complex_coefficient}, size_t @var{stride}, size_t @var{n}) This function converts a single real array, @var{real_coefficient} into an equivalent complex array, @var{complex_coefficient}, (with imaginary part set to zero), suitable for @code{gsl_fft_complex} routines. The algorithm for the conversion is simply, @example for (i = 0; i < n; i++) @{ complex_coefficient[i].real = real_coefficient[i]; complex_coefficient[i].imag = 0.0; @} @end example @end deftypefun @deftypefun int gsl_fft_halfcomplex_unpack (const double @var{halfcomplex_coefficient}[], gsl_complex_packed_array @var{complex_coefficient}, size_t @var{stride}, size_t @var{n}) This function converts @var{halfcomplex_coefficient}, an array of half-complex coefficients as returned by @code{gsl_fft_real_transform}, into an ordinary complex array, @var{complex_coefficient}. It fills in the complex array using the symmetry @c{$z_k = z_{N-k}^*$} @math{z_k = z_@{N-k@}^*} to reconstruct the redundant elements. The algorithm for the conversion is, @example complex_coefficient[0].real = halfcomplex_coefficient[0]; complex_coefficient[0].imag = 0.0; for (i = 1; i < n - i; i++) @{ double hc_real = halfcomplex_coefficient[2 * i - 1]; double hc_imag = halfcomplex_coefficient[2 * i]; complex_coefficient[i].real = hc_real; complex_coefficient[i].imag = hc_imag; complex_coefficient[n - i].real = hc_real; complex_coefficient[n - i].imag = -hc_imag; @} if (i == n - i) @{ complex_coefficient[i].real = halfcomplex_coefficient[n - 1]; complex_coefficient[i].imag = 0.0; @} @end example @end deftypefun @comment @node Example of using mixed-radix FFT routines for real data @comment @subsection Example of using mixed-radix FFT routines for real data Here is an example program using @code{gsl_fft_real_transform} and @code{gsl_fft_halfcomplex_inverse}. It generates a real signal in the shape of a square pulse. The pulse is fourier transformed to frequency space, and all but the lowest ten frequency components are removed from the array of fourier coefficients returned by @code{gsl_fft_real_transform}. The remaining fourier coefficients are transformed back to the time-domain, to give a filtered version of the square pulse. Since fourier coefficients are stored using the half-complex symmetry both positive and negative frequencies are removed and the final filtered signal is also real. @example @verbatiminclude examples/fftreal.c @end example @iftex @sp 1 @center @image{fft-real-mixedradix,3.4in} @center Low-pass filtered version of a real pulse, @center output from the example program. @end iftex @node FFT References and Further Reading @section References and Further Reading A good starting point for learning more about the FFT is the review article @cite{Fast Fourier Transforms: A Tutorial Review and A State of the Art} by Duhamel and Vetterli, @itemize @asis @item P. Duhamel and M. Vetterli. Fast fourier transforms: A tutorial review and a state of the art. @cite{Signal Processing}, 19:259--299, 1990. @end itemize @noindent To find out about the algorithms used in the GSL routines you may want to consult the document @cite{GSL FFT Algorithms} (it is included in GSL, as @file{doc/fftalgorithms.tex}). This has general information on FFTs and explicit derivations of the implementation for each routine. There are also references to the relevant literature. For convenience some of the more important references are reproduced below. @noindent There are several introductory books on the FFT with example programs, such as @cite{The Fast Fourier Transform} by Brigham and @cite{DFT/FFT and Convolution Algorithms} by Burrus and Parks, @itemize @asis @item E. Oran Brigham. @cite{The Fast Fourier Transform}. Prentice Hall, 1974. @item C. S. Burrus and T. W. Parks. @cite{DFT/FFT and Convolution Algorithms}. Wiley, 1984. @end itemize @noindent Both these introductory books cover the radix-2 FFT in some detail. The mixed-radix algorithm at the heart of the @sc{fftpack} routines is reviewed in Clive Temperton's paper, @itemize @asis @item Clive Temperton. Self-sorting mixed-radix fast fourier transforms. @cite{Journal of Computational Physics}, 52(1):1--23, 1983. @end itemize @noindent The derivation of FFTs for real-valued data is explained in the following two articles, @itemize @asis @item Henrik V. Sorenson, Douglas L. Jones, Michael T. Heideman, and C. Sidney Burrus. Real-valued fast fourier transform algorithms. @cite{IEEE Transactions on Acoustics, Speech, and Signal Processing}, ASSP-35(6):849--863, 1987. @item Clive Temperton. Fast mixed-radix real fourier transforms. @cite{Journal of Computational Physics}, 52:340--350, 1983. @end itemize @noindent In 1979 the IEEE published a compendium of carefully-reviewed Fortran FFT programs in @cite{Programs for Digital Signal Processing}. It is a useful reference for implementations of many different FFT algorithms, @itemize @asis @item Digital Signal Processing Committee and IEEE Acoustics, Speech, and Signal Processing Committee, editors. @cite{Programs for Digital Signal Processing}. IEEE Press, 1979. @end itemize @comment @noindent @comment There is also an annotated bibliography of papers on the FFT and related @comment topics by Burrus, @comment @itemize @asis @comment @item C. S. Burrus. Notes on the FFT. @comment @end itemize @comment @noindent @comment The notes are available from @url{http://www-dsp.rice.edu/res/fft/fftnote.asc}. @noindent For large-scale FFT work we recommend the use of the dedicated FFTW library by Frigo and Johnson. The FFTW library is self-optimizing---it automatically tunes itself for each hardware platform in order to achieve maximum performance. It is available under the GNU GPL. @itemize @asis @item FFTW Website, @uref{http://www.fftw.org/} @end itemize @noindent The source code for @sc{fftpack} is available from Netlib, @itemize @asis @item FFTPACK, @uref{http://www.netlib.org/fftpack/} @end itemize