SIMULATED ANNEALING AS A METHOD OF DECONVOLUTION FOR ACOUSTIC TRANSIENTS MEASURED ON A VERTICAL ARRAY

The temporal signature of an acoustic pulse propagating in an ocean wa veguide is estimated from passively sensed multichannel data. A genera l formalism is provided for source signature extraction using a parall el deconvolution algorithm based on simulated annealing. The approach is tested on two sets of experimental data using a broadband signal (2 5-110 Hz) recorded on a vertical array located at distances of one and two convergence zones from the source. As a demonstration of the comp utational feasibility of this method, the algorithm is shown to be cap able of reproducing the signature using a random initial estimate of t he source time series in conjunction with a good estimate of the Green 's function. Normalized cross correlation between the known transmitte d signature and its deconvolved estimate are calculated for the two ex perimental cases and indicate that both magnitude and phase spectra ar e faithfully reconstructed. Additional constraints on the optimization process are easily incorporated, providing an extremely general proce dure for optimization quite independent of calculus-based approaches ( i.e., derivative-based methods).