14 Jun
2010
14 Jun
'10
12:21 p.m.
On Mon, Jun 14, 2010 at 02:17:30PM +1000, Damien Zammit wrote:
The algorithm I am using is the standard overlap-add
method of block
convolution as defined in Oppenheim and Schafer 1989. Basically, I take a
FIR filter of length M taps, zero pad to N=L+M-1 where L is the length of my
sub-block of convolution. Also zero pad L inputs up to to N, then convolve
these N inputs with the zero padded filter.
The first M-1 outputs are added to the last M-1 outputs from the previous
block, and the output of the current block is truncated to L samples (from
N).
If the lenght of your filter is M, and the jack period is P,
then in each process callback you have to compute the convolution
of M filter taps and P samples. The result has M + P - 1 samples,
no matter how it is done.
The position of this block of M + P - 1 samples in the output
stream advances by steps of P samples. You need an extra buffer
to store and accumulate the part of each convolution that can't
be output in the current cycle.
Ciao,
--
FA
O tu, che porte, correndo si ?
E guerra e morte !