Narrowband multi-source direction-of-arrival estimation in the spherical harmonic domain

Area of Science:

• Acoustics and Signal Processing
• Computational Auditory Scene Analysis

Background:

• Conventional wideband multi-source (MS) direction-of-arrival (DOA) estimation relies on single-source (SS) assumptions in time-frequency (TF) bins, often using the W-disjoint orthogonality (WDO) property.
• The WDO assumption is frequently violated in scenarios with multiple simultaneously active sources, leading to failures in detecting or accurately localizing masked sources.

Purpose of the Study:

• To develop novel analytical approaches for multi-source DOA estimation that overcome the limitations of SS-based methods.
• To improve the accuracy and robustness of DOA estimation in challenging acoustic environments with masked sources.

Main Methods:

• Two analytical approaches for narrowband DOA estimation under the MS assumption in the spherical harmonic domain were proposed.
• The first method utilizes eigenvalue decomposition to transform a MS scenario into multiple SS scenarios for individual estimation.
• The second method analytically estimates two DOAs per bin, assuming the presence of two active sources.

Main Results:

• The proposed methods demonstrate significant improvements in DOA estimation accuracy, doubling the performance compared to baseline methods.
• Enhanced robustness to sensor noise was observed in the evaluated scenarios.
• Successful detection and localization of masked sources were achieved, addressing limitations of conventional SS-based techniques.

Conclusions:

• The developed analytical approaches offer a more robust and accurate solution for multi-source DOA estimation, particularly in complex acoustic scenes.
• These methods effectively handle masked sources by relaxing the strict SS assumption, advancing the field of acoustic signal processing.