System identification toolbox - boundaries in Estimation of Process models

I would like to have some information about the algorithm behind the estimation of process models within the system identification toolbox. Do anyone know where to find such documentation? The help section is not enough.

I have created a regression model and I want to estimate some parameters - a classical grey box problem set-up. The system identification returns a few negative parameter values (and a few positive). However, all parameters must be positive. In order to achieve this, I have tried to set the lower interval boundaries for all parameters to 0. A new system identification with these lower boundaries returns the same parameters as the first one, only that all originally negative parameters are 0. I find this a bit strange. I thought that changing the originally negative parameter values would result in a completely new optimal parameter set (hence affecting also the positive parameters). In order to understand the optimal parameter set I would need to know how the algorithm works I think.