Offt

Using the Offt object requires a little basic knowledge of how the Fast Fourier Transform operates. Julius O. Smith has a good math tutorial about the FFT, and the FFTW page is probably the best source of FFT information currently on the web.

The specific FFT code used by the Offt object depends upon how RTcmix was compiled. The default compilation uses older code written by Laurent de Soras (available at musicdsp.org and other places). If RTcmix was compiled using the "--with-fftw" flag during configuration, then the FFTW v. 3 library is used. [note: the FFTW library needs to be compiled with the "--enable-float" flag to work in this object.]

Constructors

Offt(int fftsize[, unsigned int flags])


fftsize is the length of the buffer used by Offt to communicate data in and out of the object. Note that the number of FFT points will be half the size of the buffer. This is because the FFT operation is done 'in place' and the buffer is used to hold two values for each FFT data point. The fftsize should be a power of 2.
flags is an optional parameter for use with the FFTW library. It should be set to the token kRealToComplex or kComplexToReal depending upon whether a forwards or inverse FFT is desired. Both flags may be set simultaneously if both forward and inverse FFTs will be performed on the buffer. These flags are defined in RTcmix/genlib/Offt.h as:

   enum {
      kRealToComplex = 1,
      kComplexToReal = 2
   };

Access Methods

float* Offt::getbuf()


returns a pointer to the buffer used by Offt for the transform. This buffer is used both as a holder for sound samples going "in" to the FFT and coming "out" of the FFT. It is also used to hold spectral information once the FFT has been performed on incoming sound samples.


void Offt::r2c()


takes a buffer filled with fftsize (set in the constructor) real-valued samples (i.e. time-domain sound samples) and transforms it into fftsize/2 complex numbers, expressed as pairs of consecutive floats representing the real and imaginary parts of the frequencies that are integer multiples of the fundamental analysis frequency. These values in the range [0, Nyquist) -- that is, up to, but not including Nyquist. The real parts have even indices into the buffer; the imaginary parts have odd indices. The one wrinkle is that what would be the imaginary portion of 0 Hz -- in other words buf[1] -- is replaced by the real value of Nyquist. So the imaginary parts of DC and Nyquist are missing. (This is the same format used in earlier cmix FFT routines.) r2c() is also referred to as the "forward" FFT. The values in the buffer may be accessed using the getbuf() method described above.


void Offt::c2r()


takes a buffer filled with fftsize/2 complex numbers representing the spectrum of a signal and transforms it into a buffer filled with fftsize real-valued sound samples. This is the inverse of the r2c() operation described above, and is in fact often called the "inverse" FFT. The values in the buffer may be accessed using the getbuf() method described above.

Examples

#include <Ougens.h>

Offt *theFFT;
int fftsize;
float *fftbuf;

int MYINSTRUMENT::init(float p[], int n_args)
{

	...

	fftsize = 1024;
	theFFT = new Offt(fftsize);
	fftbuf = theFFT->getbuf();

	...

}

int MYINSTRUMENT::run()
{
	float out[2];
	int i;

	...

	// let's assume RTBUFSAMPS is equal to fftsize for this example
	// the best way to deal with fftbuf vs RTBUFSAMPS diffs is to
	// use the Obucket object
	for (i = 0; i < framesToRun(); i++) // read 'em in
	{
		fftbuf[i] = someSampleGeneratingProcess();
	}

	theFFT->r2c();
	// now do weird stuff to the buffer...
	...
	theFFT->c2r();

	for (i = 0; i < framesToRun(); i++)  // write 'em out
	{
		out[0] = fftbuf[i];
		rtaddout(out);
		increment);
	}

	...

}

See Also

Obucket