Common Feature Extraction Methods for Acoustic Radiation Noise

This document presents several widely used feature extraction methods for acoustic radiation noise. In practical engineering applications, feature extraction plays a crucial role in noise source identification and localization. Among these methods, waveform envelope features represent a common approach where analyzing the waveform envelope helps extract fundamental frequency and amplitude information of the noise signal—typically implemented using Hilbert transform or peak detection algorithms in code. Higher-order spectral features provide more precise characterization of noise frequency characteristics, utilizing bispectral or trispectral analysis techniques that are particularly valuable for noise identification and classification tasks. Fractal dimension features, based on fractal geometry theory, effectively quantify the complexity and irregularity of noise signals through algorithms like box-counting or Higuchi's method. Singular exponent features better capture the chaotic characteristics of noise signals using methods such as Lyapunov exponents or entropy measurements, which significantly contribute to noise source classification and identification. In summary, this document offers multiple feature extraction methodologies that can be selectively applied according to specific application scenarios, with each method having corresponding MATLAB or Python implementation considerations for practical deployment.