Re: Least squares curve fitting

but rather than using the normal equations you should use the linear least
squares solver using the Householder QR decomposition which is much more stable and can also be implemented in a very compact program on any minicomputer.
you can download lots a variants of linar least squares solvers from http://plato.la.asu.edu/topics/problems/nlolsq.html the codes from lawson - hanson should perfectly do the job, there is no need here to involve the much more complex codes using the svd.

I'd also suggest considering Gentleman's algorithm, which is also a QR but uses (square root free) Givens rotations. Its big advantage for your application is that it works row-wise, which means you can do updates in real time. Google "statlib Gentleman" to find it.
.