Statistically robust signal subspace identification

The problem of signal subspace identification in the presence of transient, high-power noise or non-Gaussian noise is considered. To overcome such problems, an algorithm that results in a statistically robust singular value decomposition is proposed. This algorithm is derived from the connection between least-squares regression and the singular value decomposition. The robust singular value decomposition is then applied to the problem of estimation of the eigenstructure of a covariance matrix from raw data. The result of a Monte Carlo simulation study are presented to illustrate the effectiveness of the approach