Input signal recovery from frequency-selective fading channels is a problem
of great theoretical and practical importance. In this paper, we present s
everal new blind algorithms that utilize second-order statistics for direct
multichannel equalization. The algorithms are based on the subspace extrac
tion of a preselected block column of the channel convolution matrix. For m
ultiuser system, user signal separation can be achieved based on partial in
formation of the composite channel response. These equalization algorithms
do not rely on the precise separation of signal and noise subspaces and the
refore tend to be less sensitive to channel order (or column rank) estimati
on errors. Equalization is directly achieved without channel identification
. Furthermore, equalizability conditions of these algorithms are discussed.