Blind deconvolution by spectral weighted least-squares technique

A new algorithm is presented for blind deconvolution of nonminimum phase sy stems. It is based on the gradient minimisation of a criterion designed as the weighted square of the difference between two statistical spectra. Its main advantages are that it does not require signal preprocessing (such as spectral prewhitening) nor source information, except for statistics up to the fourth-order, and has a high convergence speed and fairly limited compu tational requirements.