> From: Scott Dattalo <scott@DATTALO.COM>
> On Thu, 6 Apr 2000, Gregg Lobdell 234-0884 wrote:
> > I am looking for a PIC (16F84 or similar) assembler routine to compute
> > arcsin with 32 bit floating point numbers.  <snip>
>
> Since you've ruled out look up tables, then I won't bother refering you to my
> web pages where there are solutions for sine, arcsine, arctan and sqrt. Instead
> I'll ask why do you need 32 bit floating point results for something controlled
> by a pic? What's your application?

Application is a platform for astrophotography.  Geometry of the
platform induces a "tangent error" when driven at a constant rate.  I
want to use the PIC to vary the step interval to eliminate this source
of error in the system.

Actually, Scott, I did find your pages, and they were helpful, but I
do need more precision.  I've simulated it at multiple levels of
precision (using higher level language) and I do need it.

> From: Spehro Pefhany <speff@INTERLOG.COM>
>
> CORDIC algorithms (do a web search) would probably be the best choice
> if you really need ~32 significant bits of result.
>
> If you have any application where the numbers have *physical* meaning,
> it's unlikely you'd need more than a 16 segment/16 bit table that
> might yield 12 or 13 bits of accuracy.(~0.02%). IMHO, of course.

I did a web search, then a literature search.  In the lit search I
came across references to CORDIC, but forgot to back to the web and
try it there.  Thanks for the reminder, I found several good hits.

I want to drive my platform a long time (1-2 hours) and the small
steps (about 4/sec), the cumulative error is a killer without the
higher precision.  Remember 32 bit floats have 24 bits precision and 8
bits exponent.

More comments on this topic welcome.

Thanks,

Gregg Lobdell                             Voice: 425-234-0884
Boeing Commercial Airplane Group            Fax: 425-234-5775
Seattle, WA, USA                       InterNet: gml@ata.ca.boeing.com