[LAD] Jack buffer management for block convolution

[hermann]

Am Donnerstag, den 10.06.2010, 23:18 +1000 schrieb Damien Zammit:
> Hi,
> 
> I have written a FFTW based block convolution jack client, but I am
> having difficulty processing the last jackframes%L samples of the jack
> buffer.  I have read that it might be an idea to collect the samples
> into a second buffer and process them independently of the jack
> process callback, but I have no clue how to implement that.  Any
> suggestions would be appreciated.
> 
> I want to use the Parks-McClellin algorithm to generate some FIR
> filter coefficents with some user configurable parameters and have it
> process the filters using block convolution.  Possibly as an addon for
> calf?
> 
> So far my jack client works, and the block convolution overlap-add
> algorithm works as per Oppenheim and Schafer, but the last few samples
> of each jack buffer are zeroes.
> 
> It would be nice to be able to have FIR filters longer than the size
> of the jack buffer too.  Any ideas?
> 
> ---
> Damien Zammit

Hi

here is the implementation I used for a time domaine convolution with a
fixed filter kernel length 

inline void convolver_filter(float* input, float* output, int sf,
unsigned int iconvolvefilter)
{
        // select kernel filter
	if (iconvolvefilter >= sizeof(filters) / sizeof(filters[0])) {
		iconvolvefilter = 0;
	}
        // push last samples to the start
	for (int i=0; i < 45; i++) {
		result[i] = result[sf+i];
	}
        // set the rest to zero
	for (int i=45; i < sf+46; i++) {
		result[i] = 0;
	}

// Do convolution:
	for (int i = 0; i < sf; i++) {
		for (int j = 0; j < 45; j++) {
			result[i+j] += input[i] * filters[iconvolvefilter][j];
		}
	}
	for (int i = 0; i < sf; i++) {
		*output++ = result[i];
	}
}

result[] is a internal buffer with length = jack_buffer_size + kernel
filter length.
sf is the jack_buffer_size.

greats  hermann