Optimal Filtering from the Point of View of Information Theory.

The filtration of disturbed signals can be regarded as a problem of information retrieval from these signals. It is investigated how optimal filters can be derived from information-theoretical considerations. The observer's limited ability to extract information from a signal has to be taken into account if useful results are to be obtained. Starting from an approach which has been proposed earlier, a new measure of extractable information in the case of sequential output of the estimated signal is presented. From this, a very general principle for optimal filtering can be derived.