'Blind' does not mean visually challenged: extracting source signals from mixed data

This paper introduces the problem of blind source separation, a phrase that denotes a class of techniques aimed at estimating signals when the physical system through which they are sensed is not known. The solution to this problem thus entails both system identification and signal estimation. Actually, I show that any lack of information on the physical system must be replaced by information on signals, and that, although a reliable data model is lacking, many pieces of information are used to constrain it. I only introduce some basic principles, with just a few details on the techniques used in practice, but the bibliography can help the reader to deepen their understanding of the matter. Most of the material is introduced by examples.