What Is Wavelet Transform?

The wavelet transform is a mathematical technique that decomposes a signal into components localized in both time and frequency, using a family of scaled and translated versions of a base function called a wavelet. Unlike the Fourier transform, which represents a signal as a sum of infinite sinusoids and loses all temporal information, the wavelet transform retains information about when specific frequency components occur. This dual localization makes the method well-suited to signals that are nonstationary, contain transients, or have features that vary over time. The transform draws its theoretical foundations from harmonic analysis, functional analysis, and digital filter bank theory, and was formalized in the work of Jean Morlet and Alex Grossman in the 1980s.

A wavelet is a short, oscillating waveform with zero mean whose energy is concentrated in a finite interval. The transform correlates the signal against dilated and shifted copies of this prototype, producing coefficients that describe the signal's content at each scale and location. Large scales capture low-frequency, slowly varying features; small scales capture high-frequency transients. This multi-resolution perspective distinguishes wavelet analysis from classical spectral methods.

Continuous Wavelet Transform

The continuous wavelet transform (CWT) evaluates the correlation of a signal against a continuously parameterized family of wavelets, varying both the scale and time-shift parameters over real values. The output is a two-dimensional function of scale and translation, providing a redundant but highly interpretable time-frequency representation. The CWT is used primarily for analysis rather than compression, because its output has far more data than the original signal. Geophysicists, neuroscientists, and acoustical engineers use CWT spectrograms to identify transient events, detect periodicities of varying length, and visualize the time-varying spectral content of complex recordings. The IEEE paper "Wavelets and Signal Processing" by Rioul and Vetterli (1991) is the foundational tutorial reference on the relationship between wavelets and signal processing.

Discrete Wavelet Transform and Filter Banks

The discrete wavelet transform (DWT) evaluates the wavelet decomposition at a dyadic grid of scales and translations, making it computationally efficient and non-redundant. Stéphane Mallat's multiresolution framework, published in 1989, showed that the DWT is equivalent to passing the signal through a cascade of complementary low-pass and high-pass filters followed by downsampling. At each stage, the filter bank splits the signal into an approximation (low-frequency) component and a detail (high-frequency) component. Repeating this process on the approximation coefficients produces a hierarchical decomposition whose total number of coefficients equals the original signal length. This filter bank interpretation makes the DWT implementable with O(N) complexity, which is detailed in the IEEE Xplore book chapter on discrete wavelet transform toolbox development.

Common Wavelet Families

Specific wavelet shapes are chosen based on the properties of the signal under analysis. The Haar wavelet is the simplest, with a square-wave form suited to step-like transitions. The Daubechies family provides compact support with varying degrees of smoothness and is widely used in image processing and compression. Biorthogonal wavelets are used in JPEG 2000 because they allow symmetric filter banks without sacrificing perfect reconstruction. The Mexican hat and Morlet wavelets are preferred in geophysics and neuroscience for their similarity to natural oscillating waveforms. The IntechOpen chapter on wavelet theory and application in communication and signal processing provides a systematic comparison of these families and their deployment contexts.

Applications

Wavelet transforms have applications in a wide range of disciplines, including:

• Image compression, where the DWT underlies the JPEG 2000 standard
• Biomedical signal processing for ECG and EEG analysis and denoising
• Seismic data interpretation and earthquake detection
• Audio coding and speech enhancement systems
• Power system fault detection and power quality monitoring