> FYI -- Tom
>
> ------------------------------------------------------------------------
>
> Subject: Re: Signal analysis algorithms book.
> Date: Mon, 30 Mar 1998 12:15:46 -0800 (PST)
> From: "Dr. C. Allen Butler" <allen@va.wagner.com>
> To: Tom Powers <powers@cebaf.gov>
>
> The book is called "Numerical Recipes - The Art of Scientific
> Programming" by Press et. al.
>
> In my old version Chapter 12 is entitled Fourier Transform Spectral
> Methods and has lots of stuff related to FFT algorithms. I'm sure
> Amazon has a newer edition and I recall it having "in C" in the title
> somewhere (as opposed to Numerical Recipes in FORTRAN).
>
> Dr. C. Allen Butler (757)-727-7700 Voice
> Daniel H. Wagner Assoc. (757)-722-0249 FAX
> 2 Eaton St., Ste. 500 allen@va.wagner.com
> Hampton, VA 23669
>
> On Mon, 30 Mar 1998, Tom Powers wrote:
>
> > Allen -
> >
> > A week or so ago you mentioned a book of pre-written algoithms for
> > numerical analysis, etc. Of course I didn't write it down and now I am
> > looking for an efficient FFT routine.
> >
> > Could you please email more some more info this book (and/or a few
> > others) to the list so that there is a record.
> >
> > Thanks
> >
> > Tom
> >
> >
Hi,
I have used this book many times (in FORTRAN) and the C version has been out
for a long time also.
I believe a FREE copy of it, or chapters of it, is available on the web for
downloading, FYI.
I have also read MANY criticisms of the routines and approaches used in these
books. According to their critics, some of the routines do not work out of
the box, or are unstable. Others may work but are very inferior to the state
of the art. The critics recommend using routines from locations like netlib
and its cousins. You can search on netlib to find out about it.
My experience has been that Numerical Receipes gives you a) a very
easy-to-read description of the problem and some solutions, and b) a
quick-and-dirty solution. I have many times just taken their routines and
used them, and they have worked every time but one for me (the singular value
decomposition routine seemed to fail when I used it). Downloading something
from netlib (and getting it implemented) seems to be more work, however, it
may be a higher quality product and in some cases descriptions of the
routines are available in published journals. In any case, if it's a critical
application you should plan on a thorough testing phase.... It sounds like
the obvious thing to do, but somehow it doesn't always happen....
If people have experiences, good or bad, with any of these sources, I'd be
interested to hear about them.
- Will Brooks