Optimally regularized channel tracking techniques for sequence estimation based on cross-validated subspace signal processing

New methods for maximum likelihood sequence estimation based on the Viterbi algorithm (VA) are presented. In the proposed scheme, the channel estimato r and the Viterbi processor operate concurrently. At any gi given time-step , the sequence provided to the channel estimator comes from the survivor wi th the best metric value, These already known modifications of the traditio nal decision-directed VA cause large variance in the estimated channel coef ficients. In fact, sequences with a high error rate may be used to perform estimation, and also the adjustment term of the channel tracking algorithm may exhibit abrupt changes caused by a "survivor swap," (that is by the eve nt in which a different survivor has the best metric at step n with respect to the step n - 1), The proposed regularization procedure forces the chann el vector to lie in the appropriate a priori known subspace: while the vari ance decreases, a certain amount of bias is introduced. The variance-bias t radeoff is then automatically adjusted by means of a cross-validation "shri nkage" estimator, which is at the same time optimal in a "small sample" pre dictive sum of squares sense and asymptically model mean-squared-error opti mal, The method is shown by means of hardware experiments on a wide-band ba se station extremely more effective than per survivor processing, minimum s urvivor processing, and traditional decision-directed approaches.