c@409: KISS FFT - A mixed-radix Fast Fourier Transform based up on the principle, c@409: "Keep It Simple, Stupid." c@409: c@409: There are many great fft libraries already around. Kiss FFT is not trying c@409: to be better than any of them. It only attempts to be a reasonably efficient, c@409: moderately useful FFT that can use fixed or floating data types and can be c@409: incorporated into someone's C program in a few minutes with trivial licensing. c@409: c@409: USAGE: c@409: c@409: The basic usage for 1-d complex FFT is: c@409: c@409: #include "kiss_fft.h" c@409: c@409: kiss_fft_cfg cfg = kiss_fft_alloc( nfft ,is_inverse_fft ,0,0 ); c@409: c@409: while ... c@409: c@409: ... // put kth sample in cx_in[k].r and cx_in[k].i c@409: c@409: kiss_fft( cfg , cx_in , cx_out ); c@409: c@409: ... // transformed. DC is in cx_out[0].r and cx_out[0].i c@409: c@409: free(cfg); c@409: c@409: Note: frequency-domain data is stored from dc up to 2pi. c@409: so cx_out[0] is the dc bin of the FFT c@409: and cx_out[nfft/2] is the Nyquist bin (if exists) c@409: c@409: Declarations are in "kiss_fft.h", along with a brief description of the c@409: functions you'll need to use. c@409: c@409: Code definitions for 1d complex FFTs are in kiss_fft.c. c@409: c@409: You can do other cool stuff with the extras you'll find in tools/ c@409: c@409: * multi-dimensional FFTs c@409: * real-optimized FFTs (returns the positive half-spectrum: (nfft/2+1) complex frequency bins) c@409: * fast convolution FIR filtering (not available for fixed point) c@409: * spectrum image creation c@409: c@409: The core fft and most tools/ code can be compiled to use float, double, c@409: Q15 short or Q31 samples. The default is float. c@409: c@409: c@409: BACKGROUND: c@409: c@409: I started coding this because I couldn't find a fixed point FFT that didn't c@409: use assembly code. I started with floating point numbers so I could get the c@409: theory straight before working on fixed point issues. In the end, I had a c@409: little bit of code that could be recompiled easily to do ffts with short, float c@409: or double (other types should be easy too). c@409: c@409: Once I got my FFT working, I was curious about the speed compared to c@409: a well respected and highly optimized fft library. I don't want to criticize c@409: this great library, so let's call it FFT_BRANDX. c@409: During this process, I learned: c@409: c@409: 1. FFT_BRANDX has more than 100K lines of code. The core of kiss_fft is about 500 lines (cpx 1-d). c@409: 2. It took me an embarrassingly long time to get FFT_BRANDX working. c@409: 3. A simple program using FFT_BRANDX is 522KB. A similar program using kiss_fft is 18KB (without optimizing for size). c@409: 4. FFT_BRANDX is roughly twice as fast as KISS FFT in default mode. c@409: c@409: It is wonderful that free, highly optimized libraries like FFT_BRANDX exist. c@409: But such libraries carry a huge burden of complexity necessary to extract every c@409: last bit of performance. c@409: c@409: Sometimes simpler is better, even if it's not better. c@409: c@409: FREQUENTLY ASKED QUESTIONS: c@409: Q: Can I use kissfft in a project with a ___ license? c@409: A: Yes. See LICENSE below. c@409: c@409: Q: Why don't I get the output I expect? c@409: A: The two most common causes of this are c@409: 1) scaling : is there a constant multiplier between what you got and what you want? c@409: 2) mixed build environment -- all code must be compiled with same preprocessor c@409: definitions for FIXED_POINT and kiss_fft_scalar c@409: c@409: Q: Will you write/debug my code for me? c@409: A: Probably not unless you pay me. I am happy to answer pointed and topical questions, but c@409: I may refer you to a book, a forum, or some other resource. c@409: c@409: c@409: PERFORMANCE: c@409: (on Athlon XP 2100+, with gcc 2.96, float data type) c@409: c@409: Kiss performed 10000 1024-pt cpx ffts in .63 s of cpu time. c@409: For comparison, it took md5sum twice as long to process the same amount of data. c@409: c@409: Transforming 5 minutes of CD quality audio takes less than a second (nfft=1024). c@409: c@409: DO NOT: c@409: ... use Kiss if you need the Fastest Fourier Transform in the World c@409: ... ask me to add features that will bloat the code c@409: c@409: UNDER THE HOOD: c@409: c@409: Kiss FFT uses a time decimation, mixed-radix, out-of-place FFT. If you give it an input buffer c@409: and output buffer that are the same, a temporary buffer will be created to hold the data. c@409: c@409: No static data is used. The core routines of kiss_fft are thread-safe (but not all of the tools directory). c@409: c@409: No scaling is done for the floating point version (for speed). c@409: Scaling is done both ways for the fixed-point version (for overflow prevention). c@409: c@409: Optimized butterflies are used for factors 2,3,4, and 5. c@409: c@409: The real (i.e. not complex) optimization code only works for even length ffts. It does two half-length c@409: FFTs in parallel (packed into real&imag), and then combines them via twiddling. The result is c@409: nfft/2+1 complex frequency bins from DC to Nyquist. If you don't know what this means, search the web. c@409: c@409: The fast convolution filtering uses the overlap-scrap method, slightly c@409: modified to put the scrap at the tail. c@409: c@409: LICENSE: c@409: Revised BSD License, see COPYING for verbiage. c@409: Basically, "free to use&change, give credit where due, no guarantees" c@409: Note this license is compatible with GPL at one end of the spectrum and closed, commercial software at c@409: the other end. See http://www.fsf.org/licensing/licenses c@409: c@409: A commercial license is available which removes the requirement for attribution. Contact me for details. c@409: c@409: c@409: TODO: c@409: *) Add real optimization for odd length FFTs c@409: *) Document/revisit the input/output fft scaling c@409: *) Make doc describing the overlap (tail) scrap fast convolution filtering in kiss_fastfir.c c@409: *) Test all the ./tools/ code with fixed point (kiss_fastfir.c doesn't work, maybe others) c@409: c@409: AUTHOR: c@409: Mark Borgerding c@409: Mark@Borgerding.net