Degrees-of-freedom estimation in the Student-t noise model

Abstract

The Student-t noise model was introduced as a robust alternative to the commonly used stationary, Gaussian noise model that constitutes the basis for many signal processing procedures. This model may also be used to generalize the matched filter in order to make it less outlier-sensitive. The generalized Student-t model includes an additional set of degrees-of-freedom parameters nu_j along with the power spectral density S(f_j). These d.o.f. parameters effectively specify the degree of heavy-tailedness of the noise that is accounted for or expected through the model. While there are many off-the-shelf procedures available for estimating the PSD S(f_j), it is not so obvious how one would go about estimating the d.o.f. parameters. We decribe a first attempt at eliciting these parameters from empirical measurements using some LIGO data.