Analysis and comparative evaluation of techniques for multichannel blind deconvolution

Deterministic multichannel blind deconvolution is an important problem arising in numerous areas of engineering. Two different approaches to solving this problem, maximum likelihood techniques (like IQML) and subspace techniques (like EVAM), have been proposed previously. These methods are theoretically elegant and computationally efficient, and questions arise as to what the properties of these estimators are. We attempt to answer some of these questions. We show that the subspace based EVAM estimator is a coarse approximation of the lQML estimator. We present a new iterative scheme to compute the ML estimator; and Cramer-Rao bounds for the channel and input estimates. In addition, we present a Monte-Carlo comparison study of the two estimators and establish the superiority of ML based techniques.