UC Riverside

Identification of finite-impulse-response (FIR) and multiple-input multiple-output (MIMO) channelsdriven by unknown uncorrelated colored sources is a challenging problem. In this paper, a groupdecorrelation enhanced subspace (GDES) method is presented. The GDES method uses the idea ofsubspace decomposition and signal decorrelation more effectively than the joint diagonalizationenhanced subspace (JDES) method previously reported in the literature. The GDES method has a much better performance than the JDES method. The correctness of the GDES method is proved assuming that 1) the channel matrix is irreducible and column reduced and 2) the source spectral matrix has distinct diagonal functions. However, the GDES method has an inherent ability to trade off between the required condition on the channel matrix and that on the source spectral matrix. Simulations show that the GDES method yields good results even when the channel matrix is not irreducible, which is not possible at all for the JDES method.