Instantaneous frequency modes separation via a spectrogram-radon based approach

The instantaneous frequency estimation is required in various applications involving radar, audio and seismic signals. It becomes a hard task in case of multicomponent signals having interfering single modes. The existing non-linear methods, such as reassignment and synchrosqueezing, enhance the readability of time-frequency representations, but they are effective only under the assumption of separability of the modes, i.e. they fail to discriminate modes in case of strong interference. In this work, a non-parametric approach for separating the single modes of a multicomponent frequency modulated signal is presented. It is mainly based on the Radon Transform of signal spectrogram which has the ability to discriminate between the single modes. Hence, the inverse Radon transform of a small selection of Radon distribution allows us to separate each component by isolating the ridge curves even in the non-separable case.