Local Adaptive Time-Frequency Bidirectional Synchrosqueezing Transform

Abstract

Extracting time-frequency (TF) features is an essential way to describe the characteristics of non-stationary signals. The existing synchrosqueezing methods post-squeeze the TF representation in a single direction to preserve the reconstruction ability, thus they are only effective for single-direction weakly modulated components. To go beyond this limitation, we introduce a new criterion based on the concentration to locally category the components into harmonic-like and impulsive-like parts. Thereby simultaneously squeezing the TF plane in two directions and ensuring the recovery of original signals. We validate and compare the proposed approach with existing methods in both artificial and real-world signals.