Robust regression applied to estimation of object parameters from astronomical speckle interferometry

We use robust nonlinear regression techniques to estimate the separation, brightness ratio, and orientation of binary stars. We perform the regression by fitting a binary-star model to a measurement of the object complex visibility obtained from infrared speckle data. Examination of x" for subplanes of the parameter range provides predictions of the accuracy required in the initial guesses at the parameters. We use the predictions to design an effective global search. The results of the global least-squares fit provide the necessary starting point for the robust regression. Using observational data, we show the robust regression solution to be superior to the conventional least-squares solution.