For imaging applications in randomly inhomogeneous acoustical media, the results summarized here lead us to the following conclusions: (1) Single frequency methods (including MUSIC) are not statistically stable, and therefore cannot be used without modification in the presence of significant amounts of spatial heterogeneity in the acoustic wave speed distribution. (2) In contrast, time domain methods are statistically stable for any objective functional having the characteristic that the random Green's functions appear in Hermitian conjugate pairs of g g* (Borcea et al., 2002), because large random phases cancel precisely. This result has been shown here to be true for DOA, and is expected to be true more generally. (3) The DOA gives only cross-range information. Range information must be obtained separately.
To isolate the targets in random media, we need either multiple views (using multiple arrays) so we can triangulate, or we need to extract a direct measure of range from the data. In the SAT examples shown here, we used arrival time for the range estimation.