An Improved Subspace-Based Algorithm in the Small Sample Size Regime

A new method for subspace identification in array signal processing applications is proposed. The method is based on random matrix theory and provides consistent estimates even when the observation dimension increases without bound at the same rate as the number of observations. This guarantees a good behavior in finite sample size situations, where the number of sensors and the number of samples have the same order of magnitude. Consistency of the algorithm holds in situations where the signal and noise subspaces are asymptotically separable in the sense that, in the asymptotic sample eigenvalue distribution, signal and noise eigenvalues generate different spectral clusters