A comparison of background noise reduction algorithms for improving acoustic beamforming output

Abstract

This work presents a comparison of different background noise reduction algorithms available in literature at a range of signal-to-noise ratios (SNRs) for improving the conventional beamforming (CB) source maps. The algorithms include the classical background noise removal (BNR), eigenvalue identification organization and subtraction (EIOS), subspace-based background subtraction (SBS), and ensemble empirical mode decomposition (EEMD), amongst others. To assess the performance of different algorithms, three test-cases were considered: (1) localizing a loudspeaker placed in an anechoic environment in the presence of background white-noise, (2) diagnosing machine-tool noise source(s) during machining, i.e., when the workpiece and tool piece are in contact and (3) localizing airfoil trailing-edge (TE) noise sources in the presence of background tunnel noise. The analysis showed that the suitability of an algorithm depends upon the given situation – for the loudspeaker source, all algorithms deliver comparable results while for machine-tool application, the relatively simple BNR algorithm delivered the desired improvements in CB maps. For the experimental airfoil TE noise test-case, the EIOS and EEMD algorithms demonstrated substantial enhancements as compared to other methods. This investigation, therefore, suggests that for a given application, it is desirable to customize the background noise reduction algorithm to obtain optimal results.