Fast converging robust adaptive arrays

A pseudo-median canceller is introduced as the canonical processor of a robust adaptive array method which significantly reduces the deleterious effects of non-Gaussian, real-world noise and interference (outliers) on typical array performance metrics such as (normalized) output noise power residue and signal-to-interference-plus-noise ratio (SINR). In addition, the proposed structure offers natural protection against signal cancellation, or equivalently, against an increase in the output noise power residue, when weight-training data contains desired signal components. The convergence rate is shown to be commensurate with sample matrix inversion (SMR) methods for Gaussian noise and interference, and convergence is essentially unaffected when outliers are added to the Gaussian weight-training data, while non-robust SMI methods slow significantly under the same circumstances.