On broadband non-synchronous measurements beamforming: A Laplacian-enhanced low-rank tensor completion approach

Due to the physical limitations of microphone arrays, there are constraints on both the frequency range and the scale of the scanning area. While a microphone array can potentially cover a wider frequency range and provide a more comprehensive view, this requires a larger and denser array configuration, achievable through non-synchronous measurements beamforming. However, localizing multiple broadband sources remains challenging without prior information about the target signal's frequency. To address this, we propose a novel approach: a non-synchronous measurements method for broadband multiple-sound-source localization using Laplacian-enhanced low-rank tensor completion. Our method dissects the tensor data structure of broadband signals and employs an alternating direction method rooted in multiplier optimization. Simulations and experimental studies show that our Laplacian-enhanced algorithm accurately captures five distinct speech signal sources, offering a superior global view compared to other convex surrogates. Moreover, our method excels in low signal-to-noise ratio conditions, significantly reducing the side-lobe level in the sound map and outperforming previous optimization techniques. This advancement is promising for applications requiring precise localization of multiple broadband sound sources.