Cylindrical cyclic acoustic imaging with a Bayesian approach for cyclostationary sources reconstruction

Abstract

Standard acoustic imaging techniques, such as beamforming or near acoustical holography, are now widely used in engineering contexts. However, large arrays of microphones are sometimes required to have a good resolution. Besides new challenges arise, particularly in the field of non stationary sources, which need to be identified and solved. Cyclostationary sound sources, a specific kind of non stationary signals, are characterized by statistical properties evolving periodically in time. In practice, the first-order statistical properties contain some periodic components while the second orders may be random with a periodic flow of energy. The present work tackles the acoustic imaging of cyclostationary sources with a scanning microphone, i.e., without any array. Cylindrical surfaces, adapted to standard rotating machines, are considered. The reconstruction difficulty of acoustic sources from discrete measurements is addressed here thanks to the cyclostationary properties. A cyclic sound field is hence extracted from the discrete measurements. Finally, a Bayesian formulation, gathering both physical and probabilistic information on this inverse problem, is used to back propagate the sound over the radiating surface.