Properties of matrix polynomials and MIMO channel identifiability from second order statistics

Blind channel identification and equalization have attracted much attention recently. Their potential application in wireless communication systems has been explored. For multiple user systems, several algorithms known to date require the assumption that the linear system has a full rank convolution matrix. In fact, this assumption requires that the system be irreducible and column reduced. In this paper, we show that some of these algorithms can remain effective even when the rank condition is related. Several algorithms, including the outer-product decomposition, can still be applied with care for irreducible (but not column reduced) MIMO linear systems