FFT convolution with overlap add

Uli wrote:
> Hi all,
>
> thanks first for all your comments.
> As I honestly have said I'm a newbie to DSP. The book of
> www.dspguide.com is now what I understood so far. And I have directly
> used the code samples there. They are given in BASIC and I have
> transferred them to DELPHI, my favourite programming language.
> So I have declared arrays of the described sizes with double
precision.

I haven't seen the code snippets you refer to. Usually, such
snippets are shaved to the bone what memory management,
optimizations and numerical consistecy is concerned. If you have
used a textbook code example to handle bulky problems like what
you indicate here, I am not at all suprised you end up with
problems.

Having said that, this is a very useful experience. I whish more
people met practical problems like these when they programmed.
These are the kind of real-world problems that make or brake a
project. Anticipating them early is a huge advantage in practical
project work.

I think DELPHI is a descendant of what used to be Borland Turbo
Pascal? If so, there is no obvious reason why you would want to
change to C/C++ or FORTRAN. The old pascal compiler (TP 6.0,
1990 vintage) was comparable with the C compiler of the time.
I didn't swap from pascal to C until I attempted to implement
production code.

I haven't studied your code, I'm pretty sure most of the problem
has to do with memory management. I suggest you try to handle this
with some decent data I/O buffers. It ought to be possible to
achieve a run-time less than one hour, given the computer, compiler
and your experience. 

Rune