A new analysis of flow noise outside the time-frequency representation using graph-based feature extraction

Among various strategies to analyze spatiotemporal aero-hydro acoustic field, complex networks are a novel technique to deal with the acoustic pressure signals (APSs) which can fully reflect the inherent structure of the time series into complex networks. In this study first, sound generated from a low Mach number and turbulent flow around a three-dimensional circular cylinder is simulated. Then, using the complex network algorithm, APSs are transformed into natural visibility graphs (NVGs), and, their hidden signatures (macroscopic and microscopic characteristics) are unveiled using the topological metrics of the NVG. Later, spatial variations of 11 network metrics associated with 76 receiver signals distributed around the cylinder are explored and a bridge is built between these variations and their physical interpretations. The results show that all network properties corresponding to receiver signals located at a constant angular position are independent of the radial direction. Moreover, irrespective of the APSs properties, the power-law degree distribution is a feature shared by all net-works, suggesting that it is a universal characteristic and highlights systems' self-organization and self-similarity characteristics in the sound field. Finally, physical interpretation of graph structures indicates that graph-based feature extraction enables identifying, distinguishing, and exploring underlying sound field dynamics in more detail.