On the Blind Separation of Signals from Moving Sources Using Switching and Tuning

In this paper, we present a conceptually simple solution to the problem of blind separation of signals whose sources are moving. Essentially, a set of beams are cast, one for each signal. These are based on an adaptive version of the minimum-variance distortion less response beam former. Each beam pattern is such that the gain is a maximum along the direction-of-arrival of a signal, and nulls along the directions of all other signals. The weights are switched whenever there is spatial coherence between the signals. The arrival-directions of all signals are determined using the ESPRIT algorithm.