Hey Lawrence,

non-linear curve fitting is not so bad as you think. The finest program
I've ever seen (and a had a lot of...) is the NLREGR v 3.4 available from
Simtel (I mean downloadable). Knowing the matter, it is very stable and
reliable. You could also some things from the Linux/GNU scene: e. g.
FUDGIT or GNUPLOT (it can also fit even if it could be fooled). The point
is at the curve fitting: one must have a good idea how the curve looks (I
mean the term of the curve) and one should guess the initial parametes (e.
g. using mean calculation or linear regression). I dealt a lot of
thermistors and other beasts (e. g. rotating valve: describe the amount of
fluid depending on the opening angle - not a nice curve I promise).

The bad news is that in most cases floating point arithmetic is needed.

I hope this helps.

Imre

PS: If you want only interpolate based on given points and you are not
interested in the formula of the WHOLE curve that is a completely
different matter. If u want things simple (but relatively good) use
Newton's divided difference method. If u want to go as precise as possible
there is only one winner: the SPQS method to be found in Netlib. However
it is a several 10k of FORTRAN code (I put it onto awk and Euphoria, too).
This tool is a bless!


On Fri, 15 Nov 2002 llile@SALTONUSA.COM wrote:

> That's one way to deal with it.
>
> A typical problem for me is interpolating thermistor data.  The thermistor
> curve comes in chart form from the manufacturer, in very un-handy
> increments.  It's quite a complex curve, and I think curve fitting it
> would be a long involved task (unless Excel does this automatically
> somehow? )  Linear curve fitting would be a cinch.   The only way I know
> how to curve fit a nonlinear is, guess at the type of equation to use,
> plot it next to your data, sum the squares of the differences,  then run a
> linear regression on the results and use a Tools:GoalSeek on the results
> to approach the minimum, then guess again at the type of equation to use
> until you like the fit..  It is a time-consuming process, usually.
>
> Is there a more efficient way?
>
>
> -- Lawrence Lile
>
>
>
>
>
>
>
> "Gordon Williams" <g_will@cyberus.ca>
> 11/15/02 01:21 PM
>
>
>         To:     <llile@SALTONUSA.COM>
>         cc:
>         Subject:        Re:      Re: [EE]:Pressure sensor confusion
>
>
> The way that I usually handle it is to plot the experimental data, put my
> choice of curve fit on the graph and have it show the equation for the
> curve.
>
> The equation then can be used for interpolation or extrapolation.
>
> Gordon Williams
>
> --
> http://www.piclist.com hint: The PICList is archived three different
> ways.  See http://www.piclist.com/#archives for details.
>
>
>

--
http://www.piclist.com#nomail Going offline? Don't AutoReply us!
email listserv@mitvma.mit.edu with SET PICList DIGEST in the body