Comparison of different regularization algorithms in sound source identification: a case study

This paper examines five regularization algorithms for sound source identification, particularly in near-field acoustic holography applications that employ the equivalent source method (ESM) based on ℓ₁ norm and ℓ₂ norm. Both simulations and experimental tests were conducted to evaluate the performance of Tikhonov regularization, ℓ₁-CVX, Bregman iteration (BI), fast iterative shrinkage-thresholding algorithm for ESM (FISTESM), and iterative reweighted least squares (IRLS). These algorithms were assessed based on their accuracy in localizing both single and coherent sound sources across a range of frequencies. Findings reveal that ℓ₁-CVX and BI achieve high levels of resolution and stability, especially for coherent sources, while FISTESM proves to be highly efficient at higher frequencies. In contrast, Tikhonov regularization exhibits limitations when applied to sparse sound sources, and IRLS demonstrates particular effectiveness at lower frequencies. This comparative study provides critical guidance for selecting the most suitable algorithm according to specific frequency and source characteristics.