Semi-blind equalization by least squares smoothing

The problem of channel equalization in communication systems with co-channel and intersymbol interference is considered in this paper. Training based approaches to this problem may require synchronization among users and often need a large number of training symbols. On the other hand, most blind channel estimation methods are restricted to the case that the multi-user channel is invertible, i.e., each user's signal and can be recovered in the absence of noise. A new semi-blind equalization algorithm is proposed based on least squares smoothing techniques that simultaneously reduce the number of interfering signals and the amount of intersymbol interference for the signal-of-interest. As a result, the proposed approach requires considerably less number of known symbols for equalization. Further, it also removes-to a large degree-the requirement of the existence of channel inverse for interfering users.