Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

IRCNN$^{+}$: An Enhanced Iterative Residual Convolutional Neural Network for Non-stationary Signal Decomposition

Feng Zhou, Antonio Cicone, Haomin Zhou

TL;DR

This study aims to further improve IRCNN with the help of several nimble techniques from deep learning and optimization to ameliorate the method and overcome some of the limitations of this technique.

Abstract

Time-frequency analysis is an important and challenging task in many applications. Fourier and wavelet analysis are two classic methods that have achieved remarkable success in many fields. However, they also exhibit limitations when applied to nonlinear and non-stationary signals. To address this challenge, a series of nonlinear and adaptive methods, pioneered by the empirical mode decomposition method, have been proposed. The goal of these methods is to decompose a non-stationary signal into quasi-stationary components that enhance the clarity of features during time-frequency analysis. Recently, inspired by deep learning, we proposed a novel method called iterative residual convolutional neural network (IRCNN). IRCNN not only achieves more stable decomposition than existing methods but also handles batch processing of large-scale signals with low computational cost. Moreover, deep learning provides a unique perspective for non-stationary signal decomposition. In this study, we aim to further improve IRCNN with the help of several nimble techniques from deep learning and optimization to ameliorate the method and overcome some of the limitations of this technique.

IRCNN$^{+}$: An Enhanced Iterative Residual Convolutional Neural Network for Non-stationary Signal Decomposition

TL;DR

This study aims to further improve IRCNN with the help of several nimble techniques from deep learning and optimization to ameliorate the method and overcome some of the limitations of this technique.

Abstract

Time-frequency analysis is an important and challenging task in many applications. Fourier and wavelet analysis are two classic methods that have achieved remarkable success in many fields. However, they also exhibit limitations when applied to nonlinear and non-stationary signals. To address this challenge, a series of nonlinear and adaptive methods, pioneered by the empirical mode decomposition method, have been proposed. The goal of these methods is to decompose a non-stationary signal into quasi-stationary components that enhance the clarity of features during time-frequency analysis. Recently, inspired by deep learning, we proposed a novel method called iterative residual convolutional neural network (IRCNN). IRCNN not only achieves more stable decomposition than existing methods but also handles batch processing of large-scale signals with low computational cost. Moreover, deep learning provides a unique perspective for non-stationary signal decomposition. In this study, we aim to further improve IRCNN with the help of several nimble techniques from deep learning and optimization to ameliorate the method and overcome some of the limitations of this technique.
Paper Structure (11 sections, 3 equations, 6 figures, 8 tables, 2 algorithms)

This paper contains 11 sections, 3 equations, 6 figures, 8 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. IRCNN
  3. IRCNN$^+$
  4. Experiments
  5. Evaluation metrics, experimental data and setting
  6. Are TVD and multi-scale convolutional attention effective?
  7. Can IRCNN$^{+}$ outperform the state-of-the-art models?
  8. Does IRCNN$^+$ work for real-life signals?
  9. How stable is IRCNN$^+$ to the hyper-parameters?
  10. How efficient is IRCNN$^+$ model for signal decomposition?
  11. Conclusion

Figures (6)

  • Figure 1: Graphic illustration of the network structure of IRCNN.
  • Figure 2: Graphic illustration of the network structure of IRCNN$^+$.
  • Figure 3: Results of $x_1$, $x_2$ by the models trained on Dataset_1, Dataset_2.
  • Figure 4: Results of $x_1$ and $x_2$ by the top three models in terms of the overall performance in Table \ref{['tab::inputs1_2_all']}.
  • Figure 5: Performance of IRCNN$^+$ under different values of $S$ and $K$.
  • ...and 1 more figures