A second-order statistics-based optimization approach for blind MIMO system identification

We consider the problem of identifying a Multiple-Input Multiple-Output (MIMO) finite impulse response system excited by colored inputs with known statistics. Among other applications this problem appears in the context of CDMA communications systems with spatial and temporal diversity. We propose a novel approach that optimizes a criterion involving spectra and cross-spectra of the system output. Simulation results indicate that the proposed scheme works well, even for large order systems, and is robust to noise and channel length mismatch.