Deconvolution without system model or a new blind deconvolution

This paper is concerned with a new method of blind deconvolution. The deconvolution algorithms restore the input signal from the observed signal and "information" about the distorting system. This information may be expressed in terms of : transfer function frequency response state equation which are obtained from modelisation and/or identification. The blind deconvolution is a very powerfull concept when this phase of modelisation is wanted to be avoid. A first step have been done by authors who applied the homomorphic deconvolution [1]. This procedure does not required a system model but just a cut-off "quefrency" in order to separate the input signal cepstrum from the impulse response one. We describe here a new method which works with a input-output pair signals as information about the system, and a well known iterative procedure to achieve the deconvolution.