# A dereverberation beamforming algorithm for noise source localization in anechoic and semi-reverberant environments

## TL;DR

This paper introduces a new beamforming algorithm that improves the accuracy of locating noise sources in both anechoic and semi-reverberant environments.

## Contribution

A novel dereverberation beamforming technique using frequency-dependent Hanning windowing of the cross-correlation matrix is proposed.

## Key findings

- The DBF algorithm localized noise sources within λ/20 uncertainty in both anechoic and semi-reverberant environments.
- Conventional beamforming showed large localization errors for complex noise sources like a mixer-grinder at high frequencies.
- Applying CLEAN-SC deconvolution further improved the resolution of DBF maps for the mixer-grinder.

## Abstract

This paper presents a dereverberation beamforming (DBF) technique based on windowing the cross-correlation matrix (CCM) to improve the localization accuracy of beamforming maps for imaging the noise sources generated by real-world applications. Following a trial-and-error procedure, an optimal frequency-dependent half-window width was determined to retain the direct field by multiplying the CCM with optimized Hanning window. A small loudspeaker and a portable mixer-grinder appliance placed in anechoic and semi-reverberant room environments were considered, and conventional beamforming (CBF) was first implemented. In the anechoic chamber, the CBF could accurately localize the loudspeaker, while for the mixer-grinder, a noticeable localization error was observed, especially at higher frequencies, which is attributed to self-scattering. In the room, the CBF delivered a small localization error for the loudspeaker, while for the mixer-grinder, it performed poorly, rendering it impossible to interpret the maps due to large side-lobes, particularly at high frequencies. By considering only the first few points at low frequencies and increasingly more points at higher frequencies in the CCM window, the DBF accurately localized the sources in both anechoic and semi-reverberant environments within λ/20 uncertainty. Additionally, the CLEAN-SC deconvolution applied to the DBF maps of the mixer-grinder significantly improved the source resolution.•DBF algorithm is presented based on CCM filtering using an optimized Hanning window to significantly improve the localization accuracy of noise sources within λ/20 uncertainty.•Trial-and-error method is used to determine the optimal Hanning window width, which depends on frequency, noise source and surroundings.•It was necessary to implement DBF even in an anechoic chamber to accurately localize noise sources generated by engineering applications.

DBF algorithm is presented based on CCM filtering using an optimized Hanning window to significantly improve the localization accuracy of noise sources within λ/20 uncertainty.

Trial-and-error method is used to determine the optimal Hanning window width, which depends on frequency, noise source and surroundings.

It was necessary to implement DBF even in an anechoic chamber to accurately localize noise sources generated by engineering applications.

Image, graphical abstract