Blind separation of weak signals in non-isotropic spatio-temporal colored noise with possibly critically loaded uncalibrated arrays

A blind source separation (BSS) algorithm optimal in output signal-to-interference (SIR) for critically loaded arrays operating at possibly negative SNIR is proposed. The proposed technique uses generalized eigen-decomposition (i.e. spectral estimation) of a first-order matrix pencil of spatial fourth-order cumulant matrices of novel definition. The algorithm has demonstrated insensitivity to non-isotropic spatially and/or temporally correlated noise without the need for whitening processes. To illustrate the potential effectiveness of the proposed approach examples are provided involving narrowband statistically independent co-channel signal mixtures of analog and digital modulations with received SNIR < 0 dB, with noise environment and sensing array manifold completely unknown.