| The blind deconvolution problem addressed in this dissertation is concerned with the task of recovering a time series or an image from its filtered version without knowledge of the dynamics of the underlying filter. In this dissertation, a solution to the blind deconvolution problem is presented which is based on the constrained maximization (or minimization) of the deconvolved signal's fourth or higher order moment. More importantly, effective algorithms for achieving the constrained maximization are developed that are applicable to sequences or arrays whose elements are independent realizations of an underlying random variable with non-Gaussian distribution. These algorithms do not require any a priori information about the statistics of the original signal nor place any restriction to the underlying filter as do other algorithms. Moreover, these algorithms converge to improved solutions in fewer iterations than do other algorithms. |
