Discrete Wavelet Transfom for Nonstationary Signal Processing

Yansong Wang, Weiwei Wu, Qiang Zhu, Gongqi SheInTech 2011 · DOI · OpenAlex

Abstract

In engineering, digital signal processing techniques must be selected based on signal characteristics. While frequency-based techniques (FBTs) excel at stationary signal analysis, nonstationary signals—where frequency components change rapidly over time—require time-frequency techniques (TFTs). This paper explores the Discrete Wavelet Transform (DWT) and its fast version, the Mallat algorithm, as a "mathematical microscope" for extracting transient features. By applying multiresolution analysis, DWT resolves the fixed time-frequency resolution issues inherent in the Short-Time Fourier Transform (STFT) and avoids the cross-term interference seen in quadratic representations like the Wigner-Ville Distribution (WVD).

The Challenge of Nonstationary Signals

Many critical engineering signals, such as vehicle vibrations during acceleration or sudden engine faults, are nonstationary. Traditional Fourier Transforms provide a spectral overview but lose the time dimension entirely. The STFT attempts to bridge this gap by using fixed-length windows, but it faces a rigid trade-off: a narrow window provides good time resolution but poor frequency resolution, and vice versa.

Wavelet-based techniques solve this by using a mother wavelet with discrete or continuous scales. This allows for high time resolution for high-frequency components (to capture fast transients) and high frequency resolution for low-frequency components (to capture slow trends), naturally matching the physics of many mechanical systems.

Multiresolution Analysis and the Mallat Algorithm

The core of the practical DWT implementation is the Mallat algorithm, a computationally efficient filter bank approach. The signal is decomposed into "approximations" and "details" by passing it through a series of low-pass ($h(k)$) and high-pass ($g(k)$) filters followed by downsampling.

The Mallat algorithm provides an efficient framework for signal decomposition and reconstruction using filter banks.

As shown in the decomposition process, the signal $x_{j-2}(k)$ is split into a lower-resolution approximation $x_{j-1}(k)$ and a detail signal $d_{j-1}(k)$. This process can be iterated to create a multiresolution hierarchy. The reconstruction process mirrors this, using upsampling and synthesis filters to perfectly reconstruct the original signal, making it ideal for encoding, decoding, and denoising applications.

The Wavelet as a "Mathematical Microscope"

The term "mathematical microscope" refers to the wavelet's ability to "zoom in" on specific time-frequency intervals. By scaling the mother wavelet, we can analyze different frequency bands with varying levels of granularity.

Wavelets in the time and frequency domains demonstrate how scaling affects resolution.

In the figure above, we see how changing the scale of the wavelet $\psi(t)$ affects its spread in both time and frequency. A compressed wavelet (higher frequency) has a narrower time support, allowing for precise localization of transients, while a stretched wavelet (lower frequency) has a wider time support but narrower frequency bandwidth, allowing for precise identification of steady-state trends.

Practical Engineering Applications

The paper validates these wavelet-based models through several rigorous engineering tests:

  • Acoustic Denoising: Using "wavelet shrinkage" (soft thresholding), the authors effectively removed random background noise from vehicle cabin recordings. This produced cleaner spectra for Sound Quality Evaluation (SQE) without blurring important transient features.
  • Octave Band Analysis (DWT-OBA): The authors adapted DWT to replace traditional hardware filters for acoustic measurements. The DWT-OBA achieved errors within [-0.3, +0.2] dB, well within the strict ±1 dB limit required by the IEC 651 international standard.
  • Automated Fault Diagnosis (WT-NN): A hybrid model combining Wavelet Packet Analysis (WPA) and Neural Networks was used to classify engine states. By extracting a 21-point wavelet feature vector, the model achieved near-perfect classification of specific faults, such as nozzle failure or sensor breakage, outperforming models using raw data.

Conclusion

The Discrete Wavelet Transform, particularly when implemented via the Mallat algorithm, offers a superior framework for nonstationary signal processing in engineering. Its ability to provide flexible resolution and its computational efficiency make it a powerful tool for real-time monitoring systems. Future work may focus on automating the selection of the optimal "mother wavelet" to further streamline diagnostic workflows in automotive and industrial health monitoring.

Summary updated

Abstract

In engineering, digital signal processing techniques need to be carefully selected according to the characteristics of the signals of interest. The frequency-based and time-frequency techniques have been frequently mentioned in some literature (Cohen, 1995). The frequencybased techniques (FBTs) have been widely used for stationary signal analysis. For nonstationary signals, the time-frequency techniques (TFTs) in common use, such as shorttime Fourier transform (STFT), wavelet transform (WT), ambiguity function (AF) and wigner-ville distribution (WVD), etc., are usually performed for extracting transient features of the signals. These techniques use different algorithms to produce a time-frequency representation for a signal. The STFT uses a standard Fourier transform over several types of windows. Waveletbased techniques apply a mother wavelet with either discrete or continuous scales to a waveform to resolve the fixed time-frequency resolution issues inherent in STFT. In applications, the fast version of wavelet transform, that is attributed to a pair of mirror filters with variable sampling rates, is usually used for reducing the number of calculations to be done, thereby saving computer running time. AF and WVD are quadratic time-frequency representations, that use advanced techniques to combat these resolution difficulties. They have better resolution than STFT but suffer from cross-term interference and produce results with coarser granularity than wavelet techniques do. Of the wavelet-based techniques, discrete wavelet transform (DWT), especially its fast version, is usually used for encoding and decoding signals, while wavelet packet analysis (WPA) are successful in signal recognition and characteristic extraction. AF and WVD with excessive transformation durations are obviously unacceptable in the development of real-time monitoring systems. In applications, the FBTs were typically used in noise and vibration engineering (Brigham, 1988). They provide the time-averaged energy information from a signal segment in frequency domain, but remain nothing in time domain. For nonstationary signals such as vehicle noises, some implementation examples are the STFT (Hodges & Power, 1985), WVD, smoothed pseudo-WVD (Baydar & Ball, 2001) and WT (Chen, 1998). In particular, the WT as “mathematical microscope” in engineering allows the changing spectral composition of a nonstationary signal to be measured and presented in the form of a time-frequency map and thus, was suggested as an effective tool for nonstationary signal analysis.
BibTeX
@inproceedings{wang2011discrete,
  author = {Yansong Wang and Weiwei Wu and Qiang Zhu and Gongqi She},
  title = {{Discrete Wavelet Transfom for Nonstationary Signal Processing}},
  booktitle = {InTech},
  year = {2011},
  doi = {10.5772/14939},
}
Figure