Neural network based novelty filtering for signal detection enhancement

Discusses the analytical basis for designing an adaptive novelty filter (ANF) based on a multilayer feedforward neural network in order to enhance the detectability of weak, transient signals in the presence of comparatively high-level background noise or interference which has unknown, uncharacterized, or time-varying statistical properties. The ANF serves as a front end preprocessor to any device which performs signal detection, estimation or classification. The ideal ANF would selectively filter out the noise while passing the signal without attenuation or distortion. The conditions under which the novelty filtering effect is most pronounced are presented.<
>