Thread Subject: error propagation and weigted fit

I am trying to do a weighted, non-linear fit and I am confused about how to translate my estimated errors into weight. Any help would be greatly appreciated.

I have a number of NMR spectra and am fitting the change in peak volumes over a number of parameters. The problem is that my function fits the ratio of two peaks (V/V0).

So far I have determined an estimate of the noise in each spectrum by sampling "background" points and taking the RMSD from the mean of these points. I then multiplied this number by the square-root of the peak footprint, since my noise can be either positive or negative.

This is where I get stuck. I've been working under the assumption that I should weight by my signal/"noise" and I'm not sure if that is a valid assumption. Secondly, I'm not sure how the signal to noise propagates. I assume I multiply the noise for V and V0, but do I use V/V0 as my signal or V*V0?

Additional detail: I'm fitting using lsqnonlin. Not sure if that makes a difference.

Thanks in advance for any help you can offer.