COMPARING PARAMETER CHOICE METHODS FOR THE REGULARIZATION IN THE SONAH ALGORITHM

Jesper Skovhus Gomes (Foredragsholder)

Aktivitet: Foredrag og mundtlige bidragForedrag og præsentationer i privat eller offentlig virksomhed

Beskrivelse

Statistically Optimized Near-field Acoustical Holography (SONAH) performs a surface-to-surface projection of the sound field by developing a set of estimation coefficients. The planar version of SONAH maps plane and evanescent waves from the measurement positions to the source plane. The coefficients that perform this plane-to-plane transformation are found by solving a least squares problem, i.e. the SONAH algorithm minimizes a residual involving an infinite set of elementary waves. Since SONAH solves an inverse problem and since measurement errors are unavoidable in practice, regularization is needed. A parameter choice method based on a priori information about the signal-to-noise-ratio (SNR) in the measurement setup is often chosen. However, this parameter choice method may be undesirable since SNR is difficult to determine in practice. In this paper, data based parameter choice methods are used in order to determine a regularization parameter. Two such approaches are compared: Generalized Cross-Validation (GCV) and a trade-off curve analysis inspired by the L-curve. Results from computer simulations and from practical measurements with a two-layer microphone array are given. These results show that for the problems studied GCV seems to be the best method when choosing the regularization parameter in SONAH.
Periode1. jun. 2006
BegivenhedstitelEuronoise 2006
BegivenhedstypeKonference
PlaceringTampere, Finland