Reconstruction of cyclostationary sound source based on a back-propagating cyclic wiener filter

The reconstruction of sound sources is recognized as a typical inverse problem consisting of back-propagating the measurements of an array onto the sources plane based on the propagation function of the sound. In this paper, the reconstruction of the cyclostationary sources in noisy environment is investigated. The cyclostationary sources assumption covers a wide range of sound sources with a cyclic transfer of energy in time, which are usually generated from the repetitive physical mechanism of a rotating machine. The rotating machines are commonly assembled with the mechanical components generating the stationary sources. The proposed back-propagating cyclic wiener filter aims at reconstructing only the cyclostationary sources from the total measurements of with stationary noises. The proposed method can be considered as an extension of the classic cyclic wiener filter, where the latter can only filter the mixed signals in the time domain whereas the former can reconstruct the spatial sources by integrating the propagation function. Thus, it extended the classic cyclic Wiener filter from the filtration of measurements to the reconstruction of sources by a cascade of the cyclic wiener filter and the Bayesian focusing. The proposed method is first validated by a laboratory experiment in a semi-anechoic chamber. Next, an industrial application involving engine bay on-site measurements is investigated to validate the proposed method: a cyclostationary source on the High Pressure Pump (HPP) is reconstructed, free of the noise interference from the engine.