Frequency-domain intrinsic component decomposition for multimodal signals with nonlinear group delays

Signals propagated in dispersive systems usually present multimodal and nonlinear group delay (GD) properties. The characterization of GDs and the signal decomposition are still challenging issues. Existing methods are mainly limited to identify monotonous GDs by estimating monotonous instantaneous frequencies. Bandpass filters are usually utilized to separate each mode, but the inherent defects of digital filters will decrease the reconstruction accuracy inevitably. In this paper, we propose the frequency-domain intrinsic component decomposition method (FICD). On one hand, it can characterize nonlinear and non-monotonic GDs in frequency domain by using different kernel functions. On the other hand, the method itself can act as a time-frequency filter to separate and reconstruct each mode simultaneously. One of the advantages of our method is that it can deal with close or overlapped signals with high precision. Finally, simulated examples are provided to verify the effectiveness and efficiency.