L∞ blind deconvolution for the generalized AR model

Applying the convex cost function L
∞
to the blind deconvolution of general non-minimum phase AR(u) models is studied. A simple and realizable constraint is proposed for the L∞ deconvolution. With this constraint, except for a gain, the model parameter is the unique solution of the L
∞
deconvolution. The strong consistency of the estimator of the model parameter defined by the sample version of L
∞
norm is presented. An algorithm is suggested for the iterative computation of the estimator. Simulation examples show the proposed approach works well for apprepriate blind equalization problems.