The Kolmogorov signal processor

Abstract

A new concept of signal processing architecture, with general guide-lines for design and learning is reported. Starting from the pioneer works reported by Kolmorogov [2] and J. Herault [4], the authors develop a signal processor that, instead of facing directly the problem of simultaneous multiple signal enhancement or multiparameter estimation, decomposes the signal processing task in single signal enhancement, or single parameter estimation, problems. The resulting framework is able to cope with complex signal processing designs with reduced complexity and in a more distributed manner. At the same time, the reader, through the work reported herein, may realise up to what degree attractive signal processing tools, as neural networks, the estimate/maximise algorithm and high order signal processing, show up in a natural way, when the reported processor is developed, designed and non-supervised learning is considered.