
Wavelet transform is a signal processing method that analyzes data at multiple scales using localized wave-like functions.The secret of wavelets lies in compromise. Fourier analysis provides perfect frequency resolution but loses time information. The windowed Fourier transform fixes this but fixes the window size. Wavelets adjust the window width adaptively: for high frequencies, short windows are used; for low frequencies, long ones. This is like musical notation, where long notes represent bass sounds and short ones represent rapid trills. In practice, wavelets excel at image compression (JPEG 2000 format) and cleaning signals from noise without blurring sharp transitions—for example, in medical tomograms.
The history of wavelets began with Haar's work (1909)—the simplest step-like system. From the 1930s to the 1980s, separate ideas emerged: the Littlewood–Paley function, quadrature mirror filters by Croisier, Esteban, and Galand. A coherent theory arose in the early 1980s from the work of geophysicist Jean Morlet and mathematician Alex Grossmann, who built the continuous wavelet transform for seismic signal analysis. Then Ingrid Daubechies (1988) created smooth orthogonal wavelets with compact support, enabling a fast transform. Later, Mallat and Meyer developed multiresolution analysis, linking wavelets to signal filtering.
The scaling function φ(t) and mother wavelet ψ(t) generate a basis such that {ψ_{j,k}(t) = 2^{-j/2} ψ(2^{-j}t - k)}. Any signal can be decomposed: f(t) = Σ_k a_{j0,k} φ_{j0,k}(t) + Σ_{j≥j0} Σ_k d_{j,k} ψ_{j,k}(t). The coefficients d are obtained by the fast Mallat algorithm using cascaded filtering. At each level, the signal passes through a pair of filters—low-pass (averaging) and high-pass (detailing)—followed by decimation. This is dual to a bank of quadrature mirror filters. The orthogonality condition of the wavelet basis requires the filters to satisfy certain relations. The analysis is flexible: the wavelet shape can be tailored to the data type.
Digital fingerprints and barcodes are compressed using wavelets. Instruments that search for defects in aircraft parts listen for crack echoes with wavelets.
Human vision is somewhat similar to wavelet analysis: we can discern both the overall shape of a chair in dim light and a fine inscription on its leg—the brain processes images at multiple scales simultaneously.
| W | variable | Wavelet transform coefficient — wavelet transform coefficient |
| s | variable | Scale parameter — scale parameter |
| τ | variable | Translation parameter — translation parameter second |
| x | variable | Input signal — signal |
| t | variable | Time variable — time second |
| ψ* | variable | Complex conjugate of wavelet function — wavelet function |
| ∫ | operator | Integration over time — integral operator |
















