A comparison of tonal-broadband decomposition algorithms for propeller noise

Abstract

The usage of propellers is increasing as an alternative to traditional aircraft and has become the primary option for new urban mobility vehicles. However, noise emissions from rotating blades can be a challenge for the certification of these new vehicles. The acoustic sources from a propeller can be decomposed into deterministic and random components. Therefore, academic research has focused on developing methods to separate these components. This study compares the performance of several methods available in the literature to decompose propeller noise signals into tonal and broadband components. Initially, the methods were applied to synthetic data that represents the main characteristics of rotating blade noise. Later, the methods were tested on realistic acoustic data collected from an isolated propeller test rig. The results showed that methods based on phase-averaging recorded signals satisfactorily detected harmonic components. The method based on cross-correlation between data blocks effectively extracted the broadband spectra, but it performed poorly on harmonic detection. Additionally, the wavelet transform and Proper Orthogonal Decomposition methods were capable of reconstructing time series related to both deterministic and random parts. However, further investigation is required regarding the separation of POD modes, as unexpected trends were observed with the wavelet-based method.