Blind channel identification using higher order statistics

The focus of this dissertation is centered on the blind channel identification for digital communication systems by exploiting higher order statistics. Three novel approaches are developed, providing closed-form solutions to finite impulse response channel estimation. They are investigated based on single-input single-output (SISO) systems and then extended to more complicated multiple-input multiple-output (MIMO) systems.; We first introduce a cumulant subspace algorithm which relies on nullspace decomposition of several well-defined cumulant matrices. A new MIMO channel identifiability condition with relaxed constraint is established. However, this subspace method needs to know the channel order accurately to guarantee the uniqueness of channel estimates. Next, a new algorithm is developed as a systematic generalization of the Weighted Slice (WS) algorithm which is used for SISO channel estimation. The generalized algorithm improves the reliability of channel estimates and can work properly with an over-estimated channel order. Its extension to MIMO systems is also discussed. Finally, we propose a simple matrix pencil approach motivated by some recent works on blind source separation. Channel estimates can be obtained from nontrivial generalized eigenvectors of a cumulant matrix pencil. Compared with the above two algorithms, this method allows the weak MIMO identifiability condition established in the subspace method as long as we know an approximate upper bound of the channel order.