Cauchy-Paul wavelet transforms revisited: A framework for intermittent non-sinusoidal oscillations
This paper revisits the continuous wavelet transform framework by establishing a rigorous physical and dimensional formulation of the Cauchy-Paul mother wavelet, tailored specifically for intermittent, non-sinusoidal electrophysiological oscillations. Departing from conventional, purely mathematical definitions, we introduce a characteristic time scale $\tau$ into the frequency-domain formulation of the mother wavelet. This parameter ensures strict dimensional consistency by maintaining dimensionless functional arguments, thereby confining the physical dimension solely to the multiplicative normalization constant under both $L^1(\mathbb{R})$ and $L^2(\mathbb{R})$ norms. A sharp dimensional and structural analysis of the resulting wavelet is conducted. We demonstrate that the spectral asymmetry inherent to the Cauchy-Paul wavelet dictates a strict mathematical hierarchy between three alternative reference frequencies: the peak ($L^\infty$), the centroid ($L^1$), and the energy-weighted ($L^2$) frequencies. Each frequency definition yields distinct quality factors ($Q$) and time-bandwidth characteristics that govern the time-frequency localization trade-off. To track multi-component EEG sleep pattern signals, a phase-based algebraic estimator is deployed alongside an advanced ridge-extraction method. The robust tracking performance and morphological adaptability of the proposed Cauchy-Paul framework are first numerically validated on synthetic transients, harmonics, and chirps, and subsequently applied to real, non-stationary EEG recordings to successfully isolate and decipher the non-sinusoidal signatures of sleep spindles.