Non Stationary Signal Analysis, Energy Demodulation and the Multicomponent AM–FM Signal Model

1This work was submitted towards the ONR Young Investigator Program for the fiscal year 2002. This report addresses the problem of non-stationary signal analysis and multicomponent AM–FM signal sep-aration and demodulation, which are important and significant problems in the area of non stationary signal modeling. This problem appears in various guises in the areas of communications, where it appears in the prob-lem of the cochannel or adjacent channel problem, in biomedical signal processing, where it appears in the area of heart-beat sound modeling, radar signal processing, where it is used to model high frequency radar clutter, in speech processing where it appears in the form of AM–FM speech analysis and synthesis, and image process-ing, where it finds application in image demodulation and texture analysis. Although numerous multicomponent AM–FM signal analysis algorithms exist, these algorithms are unable to effect signal separation and demodu-lation in the general case and work only in narrow ranges of component spectral separation or relative power ratios. These algorithms specifically experience singularity problems when the components overlap spectrally, when one of the components is much stronger than the others or when the component instantaneous frequency tracks overlap. Numerous component enumeration methods have been studied for stationary multicomponent sinusoidal signals exist but such a framework does not exist for non-stationary signal component enumeration