Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
112 tokens/sec
GPT-4o
8 tokens/sec
Gemini 2.5 Pro Pro
47 tokens/sec
o3 Pro
5 tokens/sec
GPT-4.1 Pro
38 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

On the extraction of instantaneous frequencies from ridges in time-frequency representations of signals (1310.7276v3)

Published 27 Oct 2013 in cs.CE, math.NA, and physics.data-an

Abstract: The extraction of oscillatory components and their properties from different time-frequency representations, such as windowed Fourier transform and wavelet transform, is an important topic in signal processing. The first step in this procedure is to find an appropriate ridge curve: a sequence of amplitude peak positions (ridge points), corresponding to the component of interest. This is not a trivial issue, and the optimal method for extraction is still not settled or agreed. We discuss and develop procedures that can be used for this task and compare their performance on both simulated and real data. In particular, we propose a method which, in contrast to many other approaches, is highly adaptive so that it does not need any parameter adjustment for the signal to be analysed. Being based on dynamic path optimization and fixed point iteration, the method is very fast, and its superior accuracy is also demonstrated. In addition, we investigate the advantages and drawbacks that synchrosqueezing offers in relation to curve extraction. The codes used in this work are freely available for download.

Citations (160)

Summary

Overview of Instantaneous Frequency Extraction via Ridge Analysis

Signal processing methodologies have evolved to address the intricate task of extracting oscillatory components from signals, chiefly through time-frequency representations (TFRs) derived from techniques like the windowed Fourier transform (WFT) and wavelet transform (WT). The paper under review introduces significant advancements in the extraction of instantaneous frequencies from these TFRs by proposing a robust method for ridge curve identification. The authors, D. Iatsenko, P. V. E. McClintock, and A. Stefanovska, present an adaptive and efficient algorithm that circumvents the need for parameter adjustment, enhancing the accuracy and speed of component extraction.

Methodological Innovations

The paper acknowledges the non-trivial nature of ridge curve extraction from TFRs, where amplitude peaks denote significant oscillatory features within a signal, often obscured by noise or overlapping components. To address this, the authors propose a dynamic path optimization technique coupled with fixed point iteration, which markedly reduces computational overhead compared to traditional methods. Their approach is especially notable for its universal applicability, ensuring high precision across diverse signal types without the need for user-defined parameters.

Core Algorithms

  1. Path Optimization: The algorithm optimizes the sequence of amplitude peak positions over the entire signal duration, rather than performing local optimizations at each step. This is achieved through dynamic programming, allowing for efficient traversal through potential ridge paths, identifying the optimal sequence that best represents the signal's oscillatory components.
  2. Adaptive Methodology: The authors introduce Scheme II(α, β), an adaptive technique that utilizes normalized deviations of ridge frequencies and their derivatives. The adaptivity allows the method to dynamically adjust to various signal characteristics, making it particularly effective in reliably tracing ridge curves even amidst significant noise interference.
  3. Evaluation Against Noise: The robustness of the proposed methods is assessed against synthetic and real-world signals corrupted by colored noise. The paper details performance metrics, such as a relative error measure, to objectively quantify the accuracy of instantaneous frequency extraction.

Synchrosqueezing Considerations

The paper also addresses the implications of synchrosqueezing—a method intended to enhance TFR concentration. Despite its apparent visual advantages, synchrosqueezing does not inherently improve the resolution or extraction accuracy of signal components close in frequency or those exhibiting rapid time variability. Furthermore, the paper elucidates issues with using amplitude peaks from synchrosqueezed transforms, as they exhibit dependence on frequency discretization parameters, thereby complicating ridge extraction.

Implications and Future Directions

The research contributes practically by offering openly accessible MatLab codes that implement the discussed algorithms, providing tools for researchers to further explore signal dynamics in applications ranging from cardiovascular studies to fault diagnosis. The theoretical implications highlight the importance of adaptive, non-parametric methods in signal decomposition, suggesting avenues for exploring higher-order derivative penalizations and addressing transient component extractions.

Future advancements could focus on refining the adaptive approach to manage frequency crossings and exploring heuristics for transient signal components. Additionally, the limitations noted regarding short-lived components and component crossings could drive further research into developing termination criteria for curve extraction, thereby broadening applicability across more signal types.

In conclusion, the paper represents a significant step toward more efficient and universal signal processing techniques, pushing the boundaries of contemporary methods by focusing on adaptability and computational efficiency. As the landscape of signal processing continues to evolve, such contributions are pivotal in harnessing the capabilities of modern computational tools to manage the complexity of time-varying signals.