New 2.5D method localizes moving broadband noise sources using Loève spectrum

Localizing moving sound sources with microphone arrays typically relies on modifying signals to compensate for the Doppler effect — either sample-by-sample in the time domain or with short time windows in the frequency domain. Kasess, Kreuzer, and Waubke previously developed an inverse 2.5D method for uniformly moving single-frequency sources that works directly in the spectral domain, using a modified forward model that computes motion effects at static observer positions. This eliminates the need for quasi-stationary assumptions within analysis windows. However, that approach was not directly applicable to broadband stochastic sources.

In their latest work, the authors extend the method to broadband stochastic sources by leveraging the Loève spectrum — a generalization of the cross-spectral density at static receivers. They derive the relation between the power spectral density of a moving stochastic source and the Loève spectrum in a 2.5D setting. Using simulated data with speeds up to 100 m/s, they provide a proof-of-concept based on multi-taper estimates for the Loève spectrum to localize moving broadband noise sources. The method currently requires a stationary source signal and flat spectral density around the frequency of interest; correlations between multiple sources are not yet handled. The work opens a path toward more efficient acoustic source localization without time-domain processing.