Spatial smoothing and blind source separation

Abstract

The paper focuses on the role of spatial smoothing in blind source separation (BSS). Source separation has applications that range from radar signal processing to high data rate communications, and to audio engineering. There have been numerous algorithms proposed for source separation in the literature. BSS and principal component analysis (PCA) have also been of particular interest to radar researchers and to audio experts. BSS and PCA are array processing techniques that attempt to decompose an incoming mixture of signals into basic components. A blind approach to signal separation implies little knowledge about the array manifold and other parameters of noise level, angles of arrival, and spectral shape. There are scenarios where it is desired to separate correlated sources using array processing. In situations where little is known about the array manifold, a blind separation approach is the only option. The paper exploits spatial smoothing, a well known technique in parametric array processing applied to BSS of correlated sources, particularly in a convolutive channel.