Multi-frequency signal detector and classifier

Abstract

A method and apparatus for detecting and classifying signals that are the additive combination of a few constant-amplitude sinusoidal components, herein called N-tones. The attributes of this method and apparatus include provision of superior classification performance with an algorithm of low computational complexity. The method includes filtering to remove extraneous signal components, separation of the incoming signal into one or more output streams and segmenting these output streams into blocks of successive data samples. The magnitude of the data within each block is estimated along with the frequencies of the dominant spectral components. The frequency estimates are then used as part of the configuration process of a set of notch filters for removal of the identified dominant spectral components. The newly-configured notch filters are then applied to the same data blocks and the magnitudes of their outputs are estimated. The estimated magnitudes from before and after the notch filtering are then passed on to a block classifier along with the aforementioned frequency estimates. This block classifier tests its inputs to ascertain if the input signal conforms to prespecified conditions of tonal purity, absolute component magnitude, relative component magnitude and component frequency. Finally, a timing classifier is applied to monitor the time course of block classifications and test the candidate N-tones for conformance with application-specific requirements regarding signal persistence and longevity.