Detection and Classification of Noisy Non-Stationary Acoustic Signals Using Global Dynamical Models

RIA scientists have developed a numerical scheme for the detection and classification of complicated signals in highly noisy environments (negative SNR). This method performs DIC using nonlinear signal information based on techniques developed from the theory of Chaotic dynamics. Preliminary tests on real-world data sets, recently supplied by Navy personnel, indicate that this method will likely outperform most current linear classification methods, provided the signal of interest contains some nonlinear correlations. In addition, this method has been modified to perform DIC on transient signals (e.g. active pulses), with correspondingly good performance. These algorithms are quite computationally efficient and can be implemented on any standard workstation environment. This research program will further refine these methods, measure operating characteristics, characterize performance on a variety of real-world data sets, and especially to modify the basic scheme for a variety of other problems of interest to the Army. Anticipated benefits to the Army will be electronic signal recognition and classification technologies with significantly improved performance characteristics especially suited to high noise environments. Commercial applications can be directly made to signal and image processing, forecasting of economic and geophysical systems (e.g. earthquakes), voice recognition, and biomedical technology (e.g. recognizing and characterizing physiological states.) Immediate application can likely be made to a variety of DoD and commercial application including secure voice communications, detection of precursors to heart attacks, detection of precursors to earthquakes, classification of economic trends, and classification of military/geopolitical modeling systems (e.g. self-organization in force deployments).