Residual-optimized general linear chirplet transform: A method for time–frequency feature extraction

Abstract

Time–frequency analysis (TFA) is an essential sub-area in signal processing, and many of its methods are widely used in various fields. Current methods cannot address the multi-component signals well when facing strong noise, which makes it difficult to describe the accurate time–frequency trajectory. This paper proposes a TFA method named residual optimized general linear chirplet transform (ROGLCT) to achieve better time–frequency energy concentration and robustness. Firstly, the beetle antennae search (BAS) algorithm with Rényi entropy was utilized as the fitness function to optimize GLCT parameters. Then, the maximum energy modes of the optimized GLCT time–frequency distribution are iteratively separated. Finally, the separated modes are squeezed to rebuild the spectrum. Numerical simulation and actual signals (echolocation and vibration signals) verify the proposed method’s effectiveness. Compared with other advanced methods, ROGLCT can overcome noise interference and depict clear, energy-concentrated time–frequency representation (TFR) in complex environments.