Source-number-free broadband DOA estimation for acoustic arrays via covariance enhancement and subband spectrum fusion

Broadband acoustic direction-of-arrival (DOA) estimation remains challenging when the number of active sources is unknown and the available observations are affected by low signal-to-noise ratios (SNRs), limited snapshots, and reverberation. This paper proposes a source-number-free broadband DOA estimation framework based on covariance enhancement and subband spectrum fusion. The target frequency range is adaptively partitioned into multiple approximately narrowband subbands. Within each subband, a structured covariance-enhancement procedure is applied to stabilize the covariance eigenstructure under noisy and finite-sample conditions. A direction-dependent minimum-eigenvalue-difference spectrum is then constructed using the matching relationship between the minimum eigenvalues of the enhanced covariance matrix and a transformed matrix, without source-number-dependent signal- and noise-subspace partitioning. The resulting subband spectra are incoherently fused, and a median-absolute-deviation-based peak-extraction procedure jointly determines the source number and DOAs. Comparative simulations with WB-MUSIC, ERP, SRP-PHAT, and TOPS show that the proposed method provides comparable angular precision in successfully resolved trials under favorable conditions, while achieving higher localization accuracy and clearer source separation under low-SNR, limited-snapshot, and closely spaced source conditions. At an SNR of −15 dB, the proposed method achieves a localization accuracy of 86%. Ablation results indicate that trace-based loading within the covariance-enhancement procedure and cross-subband fusion provide the principal robustness improvements, whereas the Toeplitz-assisted operation has a more limited independent contribution under the investigated conditions. Experiments using open-environment and indoor circular-microphone-array recordings provide preliminary evidence of the feasibility of the proposed method under the tested broadband acoustic localization conditions.