A structured least-squares approach to blind channel identification and equalization

This paper represents the blind channel identification problem as a structured least-squares estimation problem. The noisy observed sequence is approximated by another sequence that has "noise-free" structure. A solution to this problem for scalar valued signals and observations (single-input single-output systems) has been given by De Moor. We generalize the solution to the case of matrix valued sequences (multiple-input multiple output systems). The channel estimation algorithm also produces "noise-free" observations which can be used in conjunction with the channel estimate for equalization. Simulation results show that both channel and source estimates of the new method compare favorably with multichannel linear prediction based estimates.