Music algorithm to localize sources with unknown directivity in acoustic imaging

The Multiple Signal Classification (MUSIC) algorithm for acoustic imaging most commonly assumes that all sources have undirectional radiation pattern. We here propose a modification of this algorithm such that the concept is applicable for arbitrary directional characteristics of the sources. This is accomplished by fitting for each frequency the real valued amplitudes of the acoustic model rather than assuming a fixed functional form. The mathematical problem can be solved analytically resulting in an eigenvalue problem of a real valued Hamiltonian matrix. The performance is illustrated in simulations using pure monopolar, dipolar and quadrupolar sources.