Nonstationary signal decomposition using quadratic time–frequency distributions
Loading...
Files
Published Version
Supplementary Material
Date
2025-07-12
Authors
O'Toole, J. M.
Stevenson, N. J.
Journal Title
Journal ISSN
Volume Title
Publisher
Elsevier B.V.
Published Version
Abstract
Extracting and isolating individual components of a multi-component signal is an important but challenging task in signal processing. Signal components can be buried in noise, time-limited without temporal overlap, and can have minimal separation in the frequency domain. This work presents two methods for decomposing multi-component signals that overcome the limitations of separating signal energy, jointly, in both time and frequency domains. These methods use established peak-tracking algorithms to extract instantaneous frequency (IF) laws from contiguous regions of local-energy concentrations in a quadratic time–frequency distribution (TFD). The novelty of this work is in the methods that convert these IF laws into signal components using either (1) time-varying filter banks or (2) sinusoidal model fitting with cross-TFD phase corrections. We validate these methods using simulated and real-world signals and compare their performance against existing techniques such as the time-varying empirical mode decomposition, variational mode decomposition, and synchrosqueezed transforms. The proposed methods achieve superior efficacy for decomposition over a range of nonstationary signals, do not require a priori knowledge, such as centre frequencies or component number, and perform well in the presence of noise.
Description
Keywords
Cross-time–frequency distributions , Quadratic time–frequency distributions , Signal decomposition , Signal modelling , Time-varying filtering
Citation
O’Toole, J. M. and Stevenson, N. J. (2025) 'Nonstationary signal decomposition using quadratic time–frequency distributions', Signal Processing, 238, 110095 (11pp). https://doi.org/10.1016/j.sigpro.2025.110095