Fast cyclostationary beamforming algorithm for localization of cyclostationary sound sources

The acoustic imaging techniques are used in areas such as diagnosing faults in rotating machinery due to the advantages of non-contact measurements. The cyclostationarity of faulty sound sources is not considered in most existing acoustic imaging techniques. As a result, cyclostationary sound sources cannot be accurately identified in their localization results. The cyclostationary conventional beamforming (CSCBF) method has been proposed to obtain the cyclic spectral correlation (CSC) at the source plane by constructing the cyclic cross-spectral matrix (CCSM) between the signals from different microphones, and its problem is the runtime limitation in the construction of the CCSM. Two fast algorithms for CSCBF imaging are proposed in this paper to speed up the construction of CCSM. The fast algorithm for CSCBF based on the Wiener prediction filter (WPF-CSCBF) is proposed to construct the CCSM by using reference microphones, and the fast algorithm for CSCBF based on snapshot matrix (SNM-CSCBF) is proposed to optimize the construction of the CCSM by the snapshot matrices of acoustic signals at different frequencies. The localization accuracy and robustness of the fast algorithms are demonstrated by comparing the runtime of the algorithms as well as the error of the sound source localization (SLE) in simulations. The speaker source localization experiments and faulty bearing localization experiments are used to examine the WPF-CSCBF and SNM-CSCBF fast algorithms enabling accurate and fast localization of cyclostationary sound sources.