A time–frequency transform that represents functions or signals using translates and dilates of a mother wavelet, providing localized analysis in both time and scale (frequency) and producing continuous or discrete coefficient sets depending on sampling.

Principle

Multiresolution and localization: a single admissible mother wavelet ψ generates an affine system ψ_{a,b}(t)=|a|^{-1/2}ψ((t-b)/a); inner products with these atoms give coefficients that capture local structure at scale a and position b; admissibility ensures invertibility (reconstruction).

Demonstration

The continuous wavelet transform of a chirp reveals time-varying instantaneous frequency by large coefficients near matching scales and positions; the discrete orthonormal wavelet transform yields sparse representations for piecewise-regular signals used in compression and denoising.

Misapplication

Choosing a non-admissible kernel and expecting perfect reconstruction, using an orthonormal wavelet's coefficients as if they were from a continuous transform without adjusting sampling, or over-interpreting coefficients without accounting for boundary effects and sampling resolution.

Consequence

Provides sparse, localized representations that excel with transient features: effective denoising, compression, edge detection, and time-varying spectral analysis; multiresolution structure facilitates hierarchical algorithms and fast transforms.

Reversal

The opposite approach emphasizes global frequency resolution (Fourier) and ignores time localization; excessive scale localization sacrifices precise frequency resolution and may miss long-range periodic structure.

Boundary

Requires admissibility (or appropriate discrete filter-bank design) and working within L2 or suitable distribution spaces; continuous vs discrete formulations differ in redundancy, invertibility, and computational implementation; choice of mother wavelet matters.

Semantic Tension

Tension between wavelet transforms and windowed Fourier methods: wavelets vary window size with frequency (good time localization at high freq, good frequency localization at low freq) whereas fixed-window STFT trades one fixed resolution for all frequencies.

Synthesis

Wavelet transforms use dilations and translations of an admissible mother wavelet to produce scale- and time-localized coefficients, enabling multiresolution analysis that captures transient and hierarchical structure for analysis, compression, and denoising.