2D convolution model using (in)variant kernels for fast acoustic imaging

Ning Chu (1) , Nicolas Gac (1) , José Picheral (2) , Ali Mohammad-Djafari (1)

1 L2S - Laboratoire des signaux et systèmes
2 E3S - Supélec Sciences des Systèmes

Acoustic imaging is an advanced technique for acoustic source localization and power reconstruction using limited measurements at microphone sensors. The acoustic imaging methods often involve in two aspects: one is to build up a forward model of acoustic power propagation which requires tremendous matrix multiplications due to large dimension of the power propagation matrix; the other is to solve an inverse problem which is usually ill-posed and time consuming. In this paper, our main contribution is to propose to use 2D convolution model for fast acoustic imaging. We find out that power propagation ma-trix seems to be a quasi-Symmetric Toeplitz Block Toeplitz (STBT) matrix in the far-field condition, so that the (in)variant convolution kernels (sizes and values) can be well derived from this STBT matrix. For method validation, we use simulated and real data from the wind tunnel S2A (France) experiment for acoustic imaging on vehicle surface.

Domaines: Traitement du signal et de l'image [eess.SP]

Ning Chu, Nicolas Gac, José Picheral, Ali Mohammad-Djafari. 2D convolution model using (in)variant kernels for fast acoustic imaging. 5 th Berlin Beamforming Conference 2014, Feb 2014, Berlin, Germany. 15 p. ⟨hal-01083451⟩