FIR channel estimation through generalized cumulant slice weighting

We study the problem of linear channel estimation with unknown channel input signals. Our work stems from the weighted slice algorithm, in which cumulant slices of different orders are linearly combined to produce an estimate of a single-input single-output channel. Aiming to improve the reliability of estimation, the new approach incorporates a certain matrix structure constraint into its weight computation criterion. It provides more reliable results for channels with weak leading coefficients. This generalized algorithm maintains the advantage of Fonollosa and Vidal's algorithm in handling the situation with channel order overestimation. Its natural extension to multiple-input multiple-output channel estimation is presented along with the channel identifiability conditions required.