This thesis addresses the blind deconvolution problem in which the input sig-nals to a multi-input multi-output system and the associated system transfer functions are to be identified from the output signals only, without any a priori knowledge of both. A class of algorithm based on optimization of a cumulant-based objective function is proposed and analysed. It is shown that when the input signals are non-Gaussian i.i.d. and mutually uncorrelated, the objective function exhibits only desirable extrema and saddle points. Hence, simple gradi-ent ascent algorithm can be used for global optimization. Further, a descending algorithm which minimizes maximum distortion and inter-symbol interference simultaneously is proposed. It is shown that under normal conditions, the con-vergence rate of...[ Read more ]

This thesis addresses the blind deconvolution problem in which the input sig-nals to a multi-input multi-output system and the associated system transfer functions are to be identified from the output signals only, without any a priori knowledge of both. A class of algorithm based on optimization of a cumulant-based objective function is proposed and analysed. It is shown that when the input signals are non-Gaussian i.i.d. and mutually uncorrelated, the objective function exhibits only desirable extrema and saddle points. Hence, simple gradi-ent ascent algorithm can be used for global optimization. Further, a descending algorithm which minimizes maximum distortion and inter-symbol interference simultaneously is proposed. It is shown that under normal conditions, the con-vergence rate of the proposed algorithm is super-exponential. Generalization of the algorithm when the input signals are not i.i.d. is also considered. Computer simulation results are presented to demonstrate the validity of the proposed algorithm.