According to Lennart Augustsson:
>
>So I had a look at how to do sample rate conversion.  Basically all you do
>is to form the convolution of the input signal and the sinc function for
>each sample you want.

Ugh, that sounds very heavy on the math.  Depending on the direction
you are doing the conversion (i.e. oversampling or undersampling) then
there are some quick tricks you can pull.  To increase the sample rate
by n you first filter the input to cut off anything above
(sample_rate/n) and then replicate each sample n times - if you want
to get snazzy you can use a linear interpolation to generate samples
if you are increasing the sample rate significantly.  To decrease the
sample rate (decimate) you first filter to remove components greater
than the target sample frequency and then simply throw away 1 in n
samples.

>  Then you have to make tradeoffs for this
>to be computationally feasible, of course.  Even so, you need a fair number
>of multiply-adds per sample to produce, I'm afraid.  At least if you
>want good quality.
>

You will need some multiply-adds but if you use a decent filter say,
a 2-pole elliptical, you can get a decent roll off with only a small
computational overhead.

Brett"Yeah, I used to work on sonar processing software, how can you tell?"Lymn

-- 
===============================================================================
Brett Lymn, Computer Systems Administrator, British Aerospace Australia
===============================================================================