Blind equalization by alternating minimization for applications to mobile communications

Fast fading channels in a modern digital mobile communication environment call for fast channel identification and equalization. Alternating minimization is proposed as a general framework to accomplish joint data estimation and channel identification. Under this framework, a cost function is minimized through alternating two minimization steps which turn out to be data estimation and channel identification, and algorithms derived from this scheme is guaranteed to be convergent. These two minimization steps can be implemented using many previously proposed sequence estimation and channel identification methods (such as the Viterbi and LMS algorithms) as well as many optimization methods. A simple blind equalization algorithm is derived based on the steepest descent method. This algorithm degenerates into a simple sequence estimator if the channel response is known. The computational complexity is at most linear with channel memory, yet simulation shows that the blind algorithm suffers only 3 to 5 dB SNR loss when compared with the Viterbi algorithm with known channel response