Spectrogram Filtering and Ridge Graph Fitting Based Time Frequency Analysis

Since a frequency modulation signal can be approximated by polynomial chirplet in a local time window, polynomial chirplet transform has been applied to acoustic signal processing, radar Doppler analysis and gravity wave analysis. However, the direct implementation of a polynomial chirplet transform has extremely high computational cost due to its high dimensional polynomial chirplet parameter space. In this paper, we propose a spectrogram time-frequency filtering and ridge graph polynomial fitting approach to estimate polynomial chirplet parameters for the time-frequency analysis. In the proposed method, a low dimensional spectrogram ridge graph fitting is developed to extract high dimensional polynomial chirplet parameters for the computational cost reduction. Furthermore, the spectrogram filtering in the time-frequency space is proposed to improve the reliability of spectrogram ridge extraction, and a ridge interpolation technique is recommended to improve the accuracy of ridge extraction. Test results show that the proposed method has a low computational cost, high reliability and accuracy for extracting polynomial chirplet parameters.