Benchmarks of Multi-Component Signal Analysis Methods

Juan M. Miramont (1) , Quentin Legros (2) , Dominique Fourer (3) , François Auger (1)

1 IREENA - Institut de Recherche en Energie Electrique de Nantes Atlantique UR 4642
2 DAO - Données, Apprentissage et Optimisation
3 IBISC - Informatique, BioInformatique, Systèmes Complexes

Non-stationary multicomponent signals are ubiquitous in real-world applications. They can be modeled as a superimposition of amplitude-and frequency-modulated components so-called the modes, which require dedicated techniques to be efficiently analyzed and disentangled. State-of-the-art methods use specific assumptions and paradigms which can produce very different results in specific use cases. Hence, this paper aims to present and discuss the advantages and the limitations of several promising recent approaches respectively applied to signal denoising, mode retrieval and instantaneous frequency estimation through a comparative evaluation benchmark. Our numerical experiments show the specific scenarios where each method is the more adapted in terms of quality of mode separation and reconstruction while also considering the computational efficiency.

Mots clés: non-stationary multicomponent signal; mode retrieval; comparative study; time-frequency; synchrosqueezing

Domaines: Traitement du signal et de l'image [eess.SP]

Juan M. Miramont, Quentin Legros, Dominique Fourer, François Auger. Benchmarks of Multi-Component Signal Analysis Methods. 31st European conference on Signal Processing ( EUSIPCO 2023 ), European Association for Signal Processing (EURASIP), Sep 2023, Helsinki, Finland. pp.1783--1787, ⟨10.23919/EUSIPCO58844.2023.10290093⟩. ⟨hal-04197887⟩